Summary:
Depends on D20019. Ref T13222. Currently, TOTP uses a temporary token to make sure you've set up the app on your phone properly and that you're providing an answer to a secret which we generated (not an attacker-generated secret).
However, most factor types need some kind of sync token. SMS needs to send you a code; Duo needs to store a transaction ID. Turn this "TOTP" token into an "MFA Sync" token and lift the implementation up to the base class.
Also, slightly simplify some of the HTTP form gymnastics.
Test Plan:
- Hit the TOTP enroll screen.
- Reloaded it, got new secrets.
- Reloaded it more than 10 times, got told to stop generating new challenges.
- Answered a challenge properly, got a new TOTP factor.
- Grepped for removed class name.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D20020
Summary:
Depends on D20018. Ref T13222. When you add a new MFA configuration, you can technically (?) guess your way through it with brute force. It's not clear why this would ever really be useful (if an attacker can get here and wants to add TOTP, they can just add TOTP!) but it's probably bad, so don't let users do it.
This limit is fairly generous because I don't think this actually part of any real attack, at least today with factors we're considering.
Test Plan:
- Added TOTP, guessed wrong a ton of times, got rate limited.
- Added TOTP, guessed right, got a TOTP factor configuration added to my account.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D20019
Summary:
Depends on D20017. Ref T13222. Currently, if you:
- type some text at a TOTP gate;
- wait ~60 seconds for the challenge to expire;
- submit the form into a "Wait patiently" message; and
- mash that wait button over and over again very patiently
...you still rack up rate limiting points, because the hidden text from your original request is preserved and triggers the "is the user responding to a challenge" test. Only perform this test if we haven't already decided that we're going to make them wait.
Test Plan:
- Did the above; before patch: rate limited; after patch: not rate limited.
- Intentionally typed a bunch of bad answers which were actually evaluated: rate limited properly.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D20018
Summary:
Depends on D20016. Ref T920. This does nothing interesting on its own since the TOTP provider has no guidance/warnings, but landing it separately helps to simplify an upcoming SMS diff.
SMS will have these guidance messages:
- "Administrator: you haven't configured any mailer which can send SMS, like Twilio."
- "Administrator: SMS is weak."
- "User: you haven't configured a contact number."
Test Plan: {F6151283} {F6151284}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T920
Differential Revision: https://secure.phabricator.com/D20017
Summary:
Depends on D20011. Ref T920. This change lets a "MetaMTAMail" storage object represent various different types of messages, and makes "all" the `bin/mail` stuff "totally work" with messages of non-email types.
In practice, a lot of the related tooling needs some polish/refinement, but the basics work.
Test Plan: Used `echo beep boop | bin/mail send-test --to epriestley --type sms` to send myself SMS.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T920
Differential Revision: https://secure.phabricator.com/D20012
Summary:
Depends on D20010. Ref T920. Allow users to designate which contact number is "primary": the number we'll actually send stuff to.
Since this interacts in weird ways with "disable", just do a "when any number is touched, put all of the user's rows into the right state" sort of thing.
Test Plan:
- Added numbers, made numbers primary, disabled a primary number, un-disabled a number with no primaries. Got sensible behavior in all cases.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T920
Differential Revision: https://secure.phabricator.com/D20011
Summary: Depends on D20008. Ref T920. Continue fleshing out contact number behaviors.
Test Plan:
- Enabled and disabled a contact number.
- Saw list, detail views reflect change.
- Added number X, disabled it, added it again (allowed), enabled the disabled one ("already in use" exception).
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T920
Differential Revision: https://secure.phabricator.com/D20010
Summary:
Depends on D20006. Ref T13222. Currently, the "MFA Is Required" gate doesn't let you do anything else, but you'll need to be able to access "Contact Numbers" if an install provides SMS MFA.
Tweak this UI to give users limited access to settings, so they can set up contact numbers and change their language.
(This is a little bit fiddly, and I'm doing it early on partly so it can get more testing as these changes move forward.)
Test Plan: {F6146136}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D20008
Summary:
Ref T920. To send you SMS messages, we need to know your phone number.
This adds bare-bone basics (transactions, storage, editor, etc).
From here:
**Disabling Numbers**: I'll let you disable numbers in an upcoming diff.
**Primary Number**: I think I'm just going to let you pick a number as "primary", similar to how email works. We could imagine a world where you have one "MFA" number and one "notifications" number, but this seems unlikely-ish?
**Publishing Numbers (Profile / API)**: At some point, we could let you say that a number is public / "show on my profile" and provide API access / directory features. Not planning to touch this for now.
**Non-Phone Numbers**: Eventually this could be a list of other similar contact mechanisms (APNS/GCM devices, Whatsapp numbers, ICQ number, twitter handle so MFA can slide into your DM's?). Not planning to touch this for now, but the path should be straightforward when we get there. This is why it's called "Contact Number", not "Phone Number".
**MFA-Required + SMS**: Right now, if the only MFA provider is SMS and MFA is required on the install, you can't actually get into Settings to add a contact number to configure SMS. I'll look at the best way to deal with this in an upcoming diff -- likely, giving you partial access to more of Setings before you get thorugh the MFA gate. Conceptually, it seems reasonable to let you adjust some other settings, like "Language" and "Accessibility", before you set up MFA, so if the "you need to add MFA" portal was more like a partial Settings screen, maybe that's pretty reasonable.
**Verifying Numbers**: We'll probably need to tackle this eventually, but I'm not planning to worry about it for now.
Test Plan: {F6137174}
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: avivey, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T920
Differential Revision: https://secure.phabricator.com/D19988
Summary:
Ref T13222. This updates the CLI tools and documentation for the changes in D19975.
The flags `--type` and `--all-types` retain their current meaning. In most cases, `bin/auth strip --type totp` is sufficient and you don't need to bother looking up the relevant provider PHID. The existing `bin/auth list-factors` is also unchanged.
The new `--provider` flag allows you to select configs from a particular provider in a more granular way. The new `bin/auth list-mfa-providers` provides an easy way to get PHIDs.
(In the Phacility cluster, the "Strip MFA" action just reaches into the database and deletes rows manually, so this isn't terribly important. I verified that the code should still work properly.)
Test Plan:
- Ran `bin/auth list-mfa-providers`.
- Stripped by user / type / provider.
- Grepped for `list-factors` and `auth strip`.
- Hit all (?) of the various possible error cases.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19976
Summary:
Ref T13222. Users configure "Factor Configs", which say "I have an entry on my phone for TOTP secret key XYZ".
Currently, these point at raw implementations -- always "TOTP" in practice.
To support configuring available MFA types (like "no MFA") and adding MFA types that need some options set (like "Duo", which needs API keys), bind "Factor Configs" to a "Factor Provider" instead.
In the future, several "Factors" will be available (TOTP, SMS, Duo, Postal Mail, ...). Administrators configure zero or more "MFA Providers" they want to use (e.g., "Duo" + here's my API key). Then users can add configs for these providers (e.g., "here's my Duo account").
Upshot:
- Factor: a PHP subclass, implements the technical details of a type of MFA factor (TOTP, SMS, Duo, etc).
- FactorProvider: a storage object, owned by administrators, configuration of a Factor that says "this should be available on this install", plus provides API keys, a human-readable name, etc.
- FactorConfig: a storage object, owned by a user, says "I have a factor for provider X on my phone/whatever with secret key Q / my duo account is X / my address is Y".
Couple of things not covered here:
- Statuses for providers ("Disabled", "Deprecated") don't do anything yet, but you can't edit them anyway.
- Some `bin/auth` tools need to be updated.
- When no providers are configured, the MFA panel should probably vanish.
- Documentation.
Test Plan:
- Ran migration with providers, saw configs point at the first provider.
- Ran migration without providers, saw a provider created and configs pointed at it.
- Added/removed factors and providers. Passed MFA gates. Spot-checked database for general sanity.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19975
Summary: See https://secure.phabricator.com/D18901#249481. Update the docs and a warning string to reflect the new reality that `bin/auth recover` is now able to recover any account, not just administrators.
Test Plan: Mk 1 eyeball
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D20007
Summary:
Depends on D19994. See PHI1027. If an install has customized the "Welcome Mail" message, include it in welcome mail. A special custom message from the profile screen overrides it, if provided.
(I fiddled with putting the custom message as "placeholder" text in the remarkup area as a hint, but newlines in "placeholder" text appear to have issues in Safari and Firefox. I think this is probably reasonably clear as-is.)
Make both render remarkup-into-text so things like links work properly, as it's reasonably likely that installs will want to link to things.
Test Plan:
- With custom "Welcome Mail" text, sent mail with no custom override (got custom text) and a custom override (got overridden text).
- Linked to some stuff, got sensible links in the mail (`bin/mail show-outbound`).
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19995
Summary: Depends on D19992. Ref T13222. If administrators provide a custom login message, show it on the login screen.
Test Plan:
{F6137930}
- Viewed login screen with and without a custom message.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19994
Summary:
Ref T13222. Long ago, we had a Config option (`welcome.html`) to let you dump HTML onto the login screen, but this was relatively hard to use and not good from a security perspective.
In some cases this was obsoleted by Dashboards, but there's at least some remaining set of use cases for actual login instructions on the login screen. For example, WMF has some guidance on //which// SSO mechanism to use based on what types of account you have. On `secure`, users assume they can register by clicking "Log In With GitHub" or whatever, and it might reduce frustration to tell them upfront that registration is closed.
Some other types of auth messaging could also either use customization or defaults (e.g., the invite/welcome/approve mail).
We could do this with a bunch of Config options, but I'd generally like to move to a world where there's less stuff in Config and more configuration is contextual. I think it tends to be easier to use, and we get a lot of fringe benefits (granular permissions, API, normal transaction logs, more abililty to customize workflows and provide contextual help/hints, etc). Here, for example, we can provide a remarkup preview, which would be trickier with Config.
This does not actually do anything yet.
Test Plan: {F6137541}
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19992
Summary:
Ref T13222. Currently, if a remote address fails a few login attempts (5) in a short period of time (15 minutes) we require a CAPTCHA for each additional attempt.
This relies on:
- Administrators configuring ReCAPTCHA, which they may just not bother with.
- Administrators being comfortable with Google running arbitrary trusted Javascript, which they may not be comfortable with.
- ReCAPTCHA actually being effective, which seems likely true for unsophisticated attackers but perhaps less true for more sophisticated attackers (see <https://github.com/ecthros/uncaptcha2>, for example).
(For unsophisticated attackers and researchers, "Rumola" has been the standard CAPTCHA bypass tool for some time. This is an extension that pays humans to solve CAPTCHAs for you. This is not practical at "brute force a strong password" scale. Google appears to have removed it from the Chrome store. The "submit the captcha back to Google's APIs" trick probably isn't practical at brute-force-scale either, but it's easier to imagine weaponizing that than weaponizing human solvers.)
Add a hard gate behind the CAPTHCA wall so that we fail into a secure state if there's no CAPTCHA or the attacker can defeat CAPTCHAs at a very low cost.
The big downside to this is that an attacker who controls your remote address (e.g., is behind the same NAT device you're behind on corpnet) can lock you out of your account. However:
- That //should// be a lot of access (although maybe this isn't that high of a barrier in many cases, since compromising a "smart fridge" or "smart water glass" or whatever might be good enough).
- You can still do "Forgot password?" and login via email link, although this may not be obvious.
Test Plan:
- Logged in normally.
- Failed many many login attempts, got hard gated.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19997
Summary:
Ref T13222. Ref T13231. See PHI912. I'm planning to turn MFA providers into concrete objects, so you can disable and configure them.
Currently, we only support TOTP, which doesn't require any configuration, but other provider types (like Duo or Yubikey OTP) do require some configuration (server URIs, API keys, etc). TOTP //could// also have some configuration, like "bits of entropy" or "allowed window size" or whatever, if we want.
Add concrete objects for this and standard transaction / policy / query support. These objects don't do anything interesting yet and don't actually interact with MFA, this is just skeleton code for now.
Test Plan:
{F6090444}
{F6090445}
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13231, T13222
Differential Revision: https://secure.phabricator.com/D19935
Summary:
Ref T12509.
- Remove the "phabricator.csrf-key" configuration option in favor of automatically generating an HMAC key.
- Upgrade two hasher callsites (one in CSRF itself, one in providing a CSRF secret for logged-out users) to SHA256.
- Extract the CSRF logic from `PhabricatorUser` to a standalone engine.
I was originally going to do this as two changes (extract logic, then upgrade hashes) but the logic had a couple of very silly pieces to it that made faithful extraction a little silly.
For example, it computed `time_block = (epoch + (offset * cycle_frequency)) / cycle_frequency` instead of `time_block = (epoch / cycle_frequency) + offset`. These are equivalent but the former was kind of silly.
It also computed `substr(hmac(substr(hmac(secret)).salt))` instead of `substr(hmac(secret.salt))`. These have the same overall effect but the former is, again, kind of silly (and a little bit materially worse, in this case).
This will cause a one-time compatibility break: pages loaded before the upgrade won't be able to submit contained forms after the upgrade, unless they're open for long enough for the Javascript to refresh the CSRF token (an hour, I think?). I'll note this in the changelog.
Test Plan:
- As a logged-in user, submitted forms normally (worked).
- As a logged-in user, submitted forms with a bad CSRF value (error, as expected).
- As a logged-out user, hit the success and error cases.
- Visually inspected tokens for correct format.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T12509
Differential Revision: https://secure.phabricator.com/D19946
Summary:
Depends on D19906. Ref T13222. This isn't going to win any design awards, but make the "wait" and "answered" elements a little more clear.
Ideally, the icon parts could be animated Google Authenticator-style timers (but I think we'd need to draw them in a `<canvas />` unless there's some clever trick that I don't know) or maybe we could just have the background be like a "water level" that empties out. Not sure I'm going to actually write the JS for either of those, but the UI at least looks a little more intentional.
Test Plan:
{F6070914}
{F6070915}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19908
Summary:
Depends on D19905. Ref T13222. In D19843, I refactored this stuff but `$jump_into_hisec` was dropped.
This is a hint to keep the upgraded session in hisec mode, which we need to do a password reset when using a recovery link. Without it, we double prompt you for MFA: first to upgrade to a full session, then to change your password.
Pass this into the engine properly to avoid the double-prompt.
Test Plan:
- Used `bin/auth recover` to get a partial session with MFA enabled and a password provider.
- Before: double MFA prompt.
- After: session stays upgraded when it becomes full, no second prompt.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19906
Summary:
Depends on D19904. Ref T13226. Ref T13222. Currently, partial sessions (where you've provided a primary auth factor like a password, but not yet provided MFA) work like normal sessions: they're good for 30 days and extend indefinitely under regular use.
This behavior is convenient for full sessions, but normal users don't ever spend 30 minutes answering MFA, so there's no real reason to do it for partial sessions. If we add login alerts in the future, limiting partial sessions to a short lifetime will make them more useful, since an attacker can't get one partial session and keep extending it forever while waiting for an opportunity to get past your MFA.
Test Plan:
- Did a partial login (to the MFA prompt), checked database, saw a ~29 minute partial session.
- Did a full login, saw session extend to ~30 days.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13226, T13222
Differential Revision: https://secure.phabricator.com/D19905
Summary:
Depends on D19903. Ref T13222. This was a Facebook-specific thing from D6202 that I believe no other install ever used, and I'm generally trying to move away from the old "event" system (the more modern modular/engine patterns generally replace it).
Just drop support for this. Since the constant is being removed, anything that's actually using it should break in an obvious way, and I'll note this in the changelog.
There's no explicit replacement but I don't think this hook is useful for anything except "being Facebook in 2013".
Test Plan:
- Grepped for `TYPE_AUTH_WILLLOGIN`.
- Logged in.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19904
Summary:
Depends on D19902. Ref T13222. This is mostly a "while I'm in here..." change since MFA is getting touched so much anyway.
Doing cluster support, I sometimes need to log into user accounts on instances that have MFA. I currently accomplish this by doing `bin/auth recover`, getting a parital session, and then forcing it into a full session in the database. This is inconvenient and somewhat dangerous.
Instead, allow `bin/auth recover` to generate a link that skips the "partial session" stage: adding required MFA, providing MFA, and signing legalpad documents.
Anyone who can run `bin/auth recover` can do this anyway, this just reduces the chance I accidentally bypass MFA on the wrong session when doing support stuff.
Test Plan:
- Logged in with `bin/auth recover`, was prompted for MFA.
- Logged in with `bin/auth recover --force-full-session`, was not prompted for MFA.
- Did a password reset, followed reset link, was prompted for MFA.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19903
Summary:
Depends on D19899. Ref T13222. When we prompt you for one-shot MFA, we currently give you a lot of misleading text about your session staying in "high security mode".
Differentiate between one-shot and session upgrade MFA, and give the user appropriate cues and explanatory text.
Test Plan:
- Hit one-shot MFA on an "mfa" task in Maniphest.
- Hit session upgrade MFA in Settings > Multi-Factor.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19900
Summary:
Depends on D19897. Ref T13222. See some discussion in D19890.
- Only rate limit users if they're actually answering a challenge, not if they're just clicking "Wait Patiently".
- Reduce the number of allowed attempts per hour from 100 back to 10.
- Reduce the TOTP window from +/- 2 timesteps (allowing ~60 seconds of skew) to +/- 1 timestep (allowing ~30 seconds of skew).
- Change the window where a TOTP response remains valid to a flat 60 seconds instead of a calculation based on windows and timesteps.
Test Plan:
- Hit an MFA prompt.
- Without typing in any codes, mashed "submit" as much as I wanted (>>10 times / hour).
- Answered prompt correctly.
- Mashed "Wait Patiently" as much as I wanted (>>10 times / hour).
- Guessed random numbers, was rate limited after 10 attempts.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19898
Summary:
Depends on D19896. Ref T13222. See PHI873. Add a core "Sign With MFA" transaction type which prompts you for MFA and marks your transactions as MFA'd.
This is a one-shot gate and does not keep you in MFA.
Test Plan:
- Used "Sign with MFA", got prompted for MFA, answered MFA, saw transactions apply with MFA metadata and markers.
- Tried to sign alone, got appropriate errors.
- Tried to sign no-op changes, got appropriate errors.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19897
Summary:
Depends on D19919. Ref T11351. This method appeared in D8802 (note that "get...Object" was renamed to "get...Transaction" there, so this method was actually "new" even though a method of the same name had existed before).
The goal at the time was to let Harbormaster post build results to Diffs and have them end up on Revisions, but this eventually got a better implementation (see below) where the Harbormaster-specific code can just specify a "publishable object" where build results should go.
The new `get...Object` semantics ultimately broke some stuff, and the actual implementation in Differential was removed in D10911, so this method hasn't really served a purpose since December 2014. I think that broke the Harbormaster thing by accident and we just lived with it for a bit, then Harbormaster got some more work and D17139 introduced "publishable" objects which was a better approach. This was later refined by D19281.
So: the original problem (sending build results to the right place) has a good solution now, this method hasn't done anything for 4 years, and it was probably a bad idea in the first place since it's pretty weird/surprising/fragile.
Note that `Comment` objects still have an unrelated method with the same name. In that case, the method ties the `Comment` storage object to the related `Transaction` storage object.
Test Plan: Grepped for `getApplicationTransactionObject`, verified that all remaining callsites are related to `Comment` objects.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T11351
Differential Revision: https://secure.phabricator.com/D19920
Summary:
Depends on D19918. Ref T11351. In D19918, I removed all calls to this method. Now, remove all implementations.
All of these implementations just `return $timeline`, only the three sites in D19918 did anything interesting.
Test Plan: Used `grep willRenderTimeline` to find callsites, found none.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T11351
Differential Revision: https://secure.phabricator.com/D19919
Summary:
Depends on D19894. Ref T13222. See PHI873. When you provide a correct response to an MFA challenge, we mark it as "answered".
Currently, we never let you reuse an "answered" token. That's usually fine, but if you have 2+ factors on your account and get one or more (but fewer than all of them) right when you submit the form, you need to answer them all again, possibly after waiting for a lockout period. This is needless.
When you answer a challenge correctly, add a hidden input with a code proving you got it right so you don't need to provide another answer for a little while.
Why not just put your response in a form input, e.g. `<input type="hidden" name="totp-response" value="123456" />`?
- We may allow the "answered" response to be valid for a different amount of time than the actual answer. For TOTP, we currently allow a response to remain valid for 60 seconds, but the actual code you entered might expire sooner.
- In some cases, there's no response we can provide (with push + approve MFA, you don't enter a code, you just tap "yes, allow this" on your phone). Conceivably, we may not be able to re-verify a push+approve code if the remote implements one-shot answers.
- The "responseToken" stuff may end up embedded in normal forms in some cases in the future, and this approach just generally reduces the amount of plaintext MFA we have floating around.
Test Plan:
- Added 2 MFA tokens to my account.
- Hit the MFA prompt.
- Provided one good response and one bad response.
- Submitted the form.
- Old behavior: good response gets locked out for ~120 seconds.
- New behavior: good response is marked "answered", fixing the other response lets me submit the form.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19895
Summary:
Depends on D19893. Ref T13222. See PHI873. A challenge is "answered" if you provide a valid response. A challenge is "completed" if we let you through the MFA check and do whatever actual action the check is protecting.
If you only have one MFA factor, challenges will be "completed" immediately after they are "answered". However, if you have two or more factors, it's possible to "answer" one or more prompts, but fewer than all of the prompts, and end up with "answered" challenges that are not "completed".
In the future, it may also be possible to answer all the challenges but then have an error occur before they are marked "completed" (for example, a unique key collision in the transaction code). For now, nothing interesting happens between "answered" and "completed". This would take the form of the caller explicitly providing flags like "wait to mark the challenges as completed until I do something" and "okay, mark the challenges as completed now".
This change prevents all token reuse, even on the same workflow. Future changes will let the answered challenges "stick" to the client form so you don't have to re-answer challenges for a short period of time if you hit a unique key collision.
Test Plan:
- Used a token to get through an MFA gate.
- Tried to go through another gate, was told to wait for a long time for the next challenge window.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19894
Summary:
Depends on D19890. Ref T13222. See PHI873. Currently, we only validate TOTP responses against the current (realtime) timestep. Instead, also validate them against a specific challenge.
This mostly just moves us toward more specifically preventing responses from being reused, and supporting flows which must look more like this (SMS/push).
One rough edge here is that during the T+3 and T+4 windows (you request a prompt, then wait 60-120 seconds to respond) only past responses actually work (the current code on your device won't). For example:
- At T+0, you request MFA. We issue a T+0 challenge that accepts codes T-2, T-1, T+0, T+1, and T+2. The challenge locks out T+3 and T+4 to prevent the window from overlapping with the next challenge we may issue (see D19890).
- If you wait 60 seconds until T+3 to actually submit a code, the realtime valid responses are T+1, T+2, T+3, T+4, T+5. The challenge valid responses are T-2, T-1, T+0, T+1, and T+2. Only T+1 and T+2 are in the intersection. Your device is showing T+3 if the clock is right, so if you type in what's shown on your device it won't be accepted.
- This //may// get refined in future changes, but, in the worst case, it's probably fine if it doesn't. Beyond 120s you'll get a new challenge and a full [-2, ..., +2] window to respond, so this lockout is temporary even if you manage to hit it.
- If this //doesn't// get refined, I'll change the UI to say "This factor recently issued a challenge which has expired, wait N seconds." to smooth this over a bit.
Test Plan:
- Went through MFA.
- Added a new TOTP factor.
- Hit some error cases on purpose.
- Tried to use an old code a moment after it expired, got rejected.
- Waited 60+ seconds, tried to use the current displayed factor, got rejected (this isn't great, but currently expected).
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19893
Summary:
Depends on D19889. Ref T13222. Some of this logic is either not-quite-right or a little more complicated than it needs to be.
Currently, we TTL TOTP challenges after three timesteps -- once the current code could no longer be used. But we actually have to TTL it after five timesteps -- once the most-future acceptable code could no longer be used. Otherwise, you can enter the most-future code now (perhaps the attacker compromises NTP and skews the server clock back by 75 seconds) and then an attacker can re-use it in three timesteps.
Generally, simplify things a bit and trust TTLs more. This also makes the "wait" dialog friendlier since we can give users an exact number of seconds.
The overall behavior here is still a little odd because we don't actually require you to respond to the challenge you were issued (right now, we check that the response is valid whenever you submit it, not that it's a valid response to the challenge we issued), but that will change in a future diff. This is just moving us generally in the right direction, and doesn't yet lock everything down properly.
Test Plan:
- Added a little snippet to the control caption to list all the valid codes to make this easier:
```
$key = new PhutilOpaqueEnvelope($config->getFactorSecret());
$valid = array();
foreach ($this->getAllowedTimesteps() as $step) {
$valid[] = self::getTOTPCode($key, $step);
}
$control->setCaption(
pht(
'Valid Codes: '.implode(', ', $valid)));
```
- Used the most-future code to sign `L3`.
- Verified that `L4` did not unlock until the code for `L3` left the activation window.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19890
Summary:
Depends on D19888. Ref T13222. When we issue an MFA challenge, prevent the user from responding to it in the context of a different workflow: if you ask for MFA to do something minor (award a token) you can't use the same challenge to do something more serious (launch nukes).
This defuses highly-hypothetical attacks where the attacker:
- already controls the user's session (since the challenge is already bound to the session); and
- can observe MFA codes.
One version of this attack is the "spill coffee on the victim when the code is shown on their phone, then grab their phone" attack. This whole vector really strains the bounds of plausibility, but it's easy to lock challenges to a workflow and it's possible that there's some more clever version of the "spill coffee" attack available to more sophisticated social engineers or with future MFA factors which we don't yet support.
The "spill coffee" attack, in detail, is:
- Go over to the victim's desk.
- Ask them to do something safe and nonsuspicious that requires MFA (sign `L123 Best Friendship Agreement`).
- When they unlock their phone, spill coffee all over them.
- Urge them to go to the bathroom to clean up immediately, leaving their phone and computer in your custody.
- Type the MFA code shown on the phone into a dangerous MFA prompt (sign `L345 Eternal Declaration of War`).
- When they return, they may not suspect anything (it would be normal for the MFA token to have expired), or you can spill more coffee on their computer now to destroy it, and blame it on the earlier spill.
Test Plan:
- Triggered signatures for two different documents.
- Got prompted in one, got a "wait" in the other.
- Backed out of the good prompt, returned, still prompted.
- Answered the good prompt.
- Waited for the bad prompt to expire.
- Went through the bad prompt again, got an actual prompt this time.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19889
Summary:
Depends on D19886. Ref T13222. Clean up MFA challenges after they expire.
(There's maybe some argument to keeping these around for a little while for debugging/forensics, but I suspect it would never actually be valuable and figure we can cross that bridge if we come to it.)
Test Plan:
- Ran `bin/garbage collect --collector ...` and saw old MFA challenges collected.
- Triggered a new challenge, GC'd again, saw it survive GC while still active.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19888
Summary:
Ref T13222. See PHI873. Ref T9770.
Currently, we support only TOTP MFA. For some MFA (SMS and "push-to-app"-style MFA) we may need to keep track of MFA details (e.g., the code we SMS'd you). There isn't much support for that yet.
We also currently allow free reuse of TOTP responses across sessions and workflows. This hypothetically enables some "spyglass" attacks where you look at someone's phone and type the code in before they do. T9770 discusses this in more detail, but is focused on an attack window starting when the user submits the form. I claim the attack window opens when the TOTP code is shown on their phone, and the window between the code being shown and being submitted is //much// more interesting than the window after it is submitted.
To address both of these cases, start tracking MFA "Challenges". These are basically a record that we asked you to give us MFA credentials.
For TOTP, the challenge binds a particular timestep to a given session, so an attacker can't look at your phone and type the code into their browser before (or after) you do -- they have a different session. For now, this means that codes are reusable in the same session, but that will be refined in the future.
For SMS / push, the "Challenge" would store the code we sent you so we could validate it.
This is mostly a step on the way toward one-shot MFA, ad-hoc MFA in comment action stacks, and figuring out what's going on with Duo.
Test Plan:
- Passed MFA normally.
- Passed MFA normally, simultaneously, as two different users.
- With two different sessions for the same user:
- Opened MFA in A, opened MFA in B. B got a "wait".
- Submitted MFA in A.
- Clicked "Wait" a bunch in B.
- Submitted MFA in B when prompted.
- Passed MFA normally, then passed MFA normally again with the same code in the same session. (This change does not prevent code reuse.)
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13222, T9770
Differential Revision: https://secure.phabricator.com/D19886
Summary:
Ref T13222. See PHI873. Currently, MFA implementations return this weird sort of ad-hoc dictionary from validation, which is later used to render form/control stuff.
I want to make this more formal to handle token reuse / session binding cases, and let MFA factors share more code around challenges. Formalize this into a proper object instead of an ad-hoc bundle of properties.
Test Plan:
- Answered a TOTP MFA prompt wrong (nothing, bad value).
- Answered a TOTP MFA prompt properly.
- Added new TOTP MFA, survived enrollment.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19885
Summary:
Ref T13222. Ref T12509. When you add a new MFA TOTP authenticator, we generate a temporary token to make sure you're actually adding the key we generated and not picking your own key.
That is, if we just put inputs in the form like `key=123, response=456`, users could pick their own keys by changing the value of `key` and then generating the correct `response`. That's probably fine, but maybe attackers could somehow force users to pick known keys in combination with other unknown vulnerabilities that might exist in the future. Instead, we generate a random key and keep track of it to make sure nothing funny is afoot.
As an additional barrier, we do the standard "store the digest, not the real key" sort of thing so you can't force a known value even if you can read the database (although this is mostly pointless since you can just read TOTP secrets directly if you can read the database). But it's pretty standard and doesn't hurt anything.
Update this from SHA1 to SHA256. This will break any TOTP factors which someone was in the middle of adding during a Phabricator upgrade, but that seems reasonable. They'll get a sensible failure mode.
Test Plan: Added a new TOTP factor.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222, T12509
Differential Revision: https://secure.phabricator.com/D19884
Summary:
Ref T13222. Ref T13225. We store a digest of the session key in the session table (not the session key itself) so that users with access to this table can't easily steal sessions by just setting their cookies to values from the table.
Users with access to the database can //probably// do plenty of other bad stuff (e.g., T13134 mentions digesting Conduit tokens) but there's very little cost to storing digests instead of live tokens.
We currently digest session keys with HMAC-SHA1. This is fine, but HMAC-SHA256 is better. Upgrade:
- Always write new digests.
- We still match sessions with either digest.
- When we read a session with an old digest, upgrade it to a new digest.
In a few months we can throw away the old code. When we do, installs that skip upgrades for a long time may suffer a one-time logout, but I'll note this in the changelog.
We could avoid this by storing `hmac256(hmac1(key))` instead and re-hashing in a migration, but I think the cost of a one-time logout for some tiny subset of users is very low, and worth keeping things simpler in the long run.
Test Plan:
- Hit a page with an old session, got a session upgrade.
- Reviewed sessions in Settings.
- Reviewed user logs.
- Logged out.
- Logged in.
- Terminated other sessions individually.
- Terminated all other sessions.
- Spot checked session table for general sanity.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13225, T13222
Differential Revision: https://secure.phabricator.com/D19883
Summary:
Ref T13222. See PHI873. I'm preparing to introduce a new MFA "Challenge" table which stores state about challenges we've issued (to bind challenges to sessions and prevent most challenge reuse).
This table will reference sessions (since each challenge will be bound to a particular session) but sessions currently don't have PHIDs. Give them PHIDs and slightly modernize some related code.
Test Plan:
- Ran migrations.
- Verified table got PHIDs.
- Used `var_dump()` to dump an organic user session.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19881
Summary:
Ref T13222. See PHI873. Currently, when applications prompt users to enter MFA, their session upgrades as a side effect.
In some cases (like managing your email addresses) it makes sense to upgrade your session for a little while since it's common to make multiple edits in sequence (add a new address, make it primary, remove an old address). We generally want MFA to stay out of the way and not feel annoying.
In other cases, we don't expect multiple high-security actions in a row. Notably, PHI873 looks at more "one-shot" use cases where a prompt is answering a specific workflow. We already have at least one of these in the upstream: answering an MFA prompt when signing a Legalpad document.
Introduce a "token" workflow (in contrast to the existing "session") workflow that just does a one-shot prompt without upgrading your session statefully. Then, make Legalpad use this new workflow.
Note that this workflow has a significant problem: if the form submission is invalid for some other reason, we re-prompt you on resubmit. In Legalpad, this workflow looks like:
- Forget to check the "I agree" checkbox.
- Submit the form.
- Get prompted for MFA.
- Answer MFA prompt.
- Get dumped back to the form with an error.
- When you fix the error and submit again, you have to do another MFA check.
This isn't a fatal flaw in Legalpad, but would become a problem with wider adoption. I'll work on fixing this (so the MFA token sticks to the form) in the next set of changes.
Roughly, this is headed toward "MFA sticks to the form/workflow" instead of "MFA sticks to the user/session".
Test Plan:
- Signed a legalpad document with MFA enabled.
- Was prompted for MFA.
- Session no longer upgraded (no purple "session in high security" badge).
- Submitted form with error, answered MFA, fixed error, submitted form again.
- Bad behavior: got re-prompted for MFA. In the future, MFA should stick to the form.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13222
Differential Revision: https://secure.phabricator.com/D19843
Summary:
Ref T6960. Ref T13217. Ref T13216. Depends on D19811. Use the recently-introduced "%P" conversion ("Password/Secret") to load sessions in SessionEngine.
This secret isn't critical to protect (it's the //hash// of the actual secret and not useful to attackers on its own) but it shows up on every page in DarkConsole and is an obvious case where `%P` is a more appropriate conversion.
Test Plan:
Note "*********" in the middle of the output here, instead of a session key hash:
{F6012805}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13217, T13216, T6960
Differential Revision: https://secure.phabricator.com/D19812
Summary: Depends on D19789. Ref T13217. Continue updating things to use the new %Q-flavored conversions instead of smushing a bunch of strings together.
Test Plan: Browsed around, far fewer errors. These changes are largely mechanical in nature.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13217
Differential Revision: https://secure.phabricator.com/D19790
Summary: Depends on D19785. Ref T13217. This converts many of the most common clause construction pathways to the new %Q / %LQ / %LO / %LA / %LJ semantics.
Test Plan: Browsed around a bunch, saw fewer warnings and no obvious behavioral errors. The transformations here are generally mechanical (although I did them by hand).
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: hach-que
Maniphest Tasks: T13217
Differential Revision: https://secure.phabricator.com/D19789
Summary:
See <https://hackerone.com/reports/435648>. We currently use 80-bit TOTP keys. The RFC suggests 128 as a minimum and recommends 160.
The math suggests that doing the hashing for an 80-bit key is hard (slightly beyond the reach of a highly motivated state actor, today) but there's no reason not to use 160 bits instead to put this completely out of reach.
See some additional discussion on the HackerOne report about enormous key sizes, number of required observations, etc.
Test Plan: Added a new 160-bit TOTP factor to Google Authenticator without issue.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19792
Summary:
Ref T13216. We occasionally receive HackerOne reports concerned that you can select your username as a password. I suspect very few users actually do this and that this is mostly a compliance/checklist sort of issue, not a real security issue.
On this install, we have about 41,000 user accounts. Of these, 100 have their username as a password (account or VCS). A substantial subset of these are either explicitly intentional ("demo", "bugmenot") or obvious test accounts ("test" in name, or name is a nonsensical string of gibberish, or looks like "tryphab" or similar) or just a bunch of numbers (?), or clearly a "researcher" doing this on purpose (e.g., name includes "pentest" or "xss" or similar).
So I'm not sure real users are actually very inclined to do this, and we can't really ever stop them from picking awful passwords anyway. But we //can// stop researchers from reporting that this is an issue.
Don't allow users to select passwords which contain words in a blocklist: their username, real name, email addresses, or the install's domain name. These words also aren't allowed to contain the password (that is, neither your password nor your username may be a substring of the other one). We also do a little normalization to try to split apart email addresses, domains, and real names, so I can't have "evan1234" as my password.
Test Plan:
- Added unit tests and made them pass.
- Tried to set my password to a bunch of variations of my username / email / domain name / real name / etc, got rejected.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13216
Differential Revision: https://secure.phabricator.com/D19776
Summary:
Ref T13202. See <https://discourse.phabricator-community.org/t/2fa-input-box-isnt-hinted-as-a-password-so-browsers-suggest-auto-fills/1959>.
If browsers are autofilling this, I think browser behavior here is bad, but behavior is probably better on the balance if we hint this as `autocomplete="off"` and this is a minor concesssion.
Test Plan:
- I couldn't immediately get any browser to try to autofill this field (perhaps I've disabled autofill, or just not enabled it aggressively?), but this change didn't break anything.
- After the change, answered a TOTP prompt normally.
- After the change, inspected page content and saw `autocomplete="off"` on the `<input />` node.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13202
Differential Revision: https://secure.phabricator.com/D19722
Summary:
Ref T13197. See PHI873. Record when a user has MFA'd and add a little icon to the transaction, similar to the exiting "Silent" icon.
For now, this just makes this stuff more auditable. Future changes may add ways to require MFA for certain specific transactions, outside of the ones that already always require MFA (like revealing credentials).
Test Plan: {F5877960}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13197
Differential Revision: https://secure.phabricator.com/D19665
Summary:
Depends on D19585. Ref T13164. This is a precursor for D19586, which causes Editors to start doing more explicit CAN_EDIT checks.
Passwords have an Editor, but don't actually define a CAN_EDIT capability. Define one (you can edit a password if you can edit the object the password is associated with).
(Today, this object is always a User -- this table just unified VCS passwords and Account passwords so they can be handled more consistently.)
Test Plan:
- With D19586, ran unit tests and got a pass.
- Edited my own password.
- Tried to edit another user's password and wasn't permitted to.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13164
Differential Revision: https://secure.phabricator.com/D19592
Summary:
Fixes T12397. Ref T13164. See PHI801.
Several installs have hit various use cases where the path on disk where Phabricator lives changes at runtime. Currently, `bin/ssh-auth` caches a flat file which includes the path to `bin/ssh-exec`, so this may fall out of date if `phabricator/` moves.
These use cases have varying strengths of legitimacy, but "we're migrating to a new set of hosts and the pool is half old machines and half new machines" seems reasonably compelling and not a problem entirely of one's own making.
Test Plan:
- Compared output on `master` to output after change, found them byte-for-byte identical.
- Moved `phabricator/` to `phabricator2/`, ran `bin/ssh-auth`, got updated output.
- Added a new SSH key, saw it appear in the output.
- Grepped for `AUTHFILE_CACHEKEY` (no hits).
- Dropped the cache, verified that the file regenerates cleanly.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13164, T12397
Differential Revision: https://secure.phabricator.com/D19568
