OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
<?php
|
|
|
|
|
|
|
|
final class PhabricatorOAuthServerAuthController
|
2014-07-10 00:12:48 +02:00
|
|
|
extends PhabricatorAuthController {
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
|
|
|
public function shouldRequireLogin() {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
public function processRequest() {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
$request = $this->getRequest();
|
|
|
|
$viewer = $request->getUser();
|
|
|
|
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
$server = new PhabricatorOAuthServer();
|
|
|
|
$client_phid = $request->getStr('client_id');
|
2015-02-07 00:32:55 +01:00
|
|
|
$scope = $request->getStr('scope');
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
$redirect_uri = $request->getStr('redirect_uri');
|
|
|
|
$response_type = $request->getStr('response_type');
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
// state is an opaque value the client sent us for their own purposes
|
|
|
|
// we just need to send it right back to them in the response!
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
$state = $request->getStr('state');
|
|
|
|
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
if (!$client_phid) {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_request',
|
|
|
|
pht('Malformed Request'),
|
|
|
|
pht(
|
|
|
|
'Required parameter %s was not present in the request.',
|
|
|
|
phutil_tag('strong', array(), 'client_id')));
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
|
|
|
$server->setUser($viewer);
|
|
|
|
$is_authorized = false;
|
|
|
|
$authorization = null;
|
|
|
|
$uri = null;
|
|
|
|
$name = null;
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
// one giant try / catch around all the exciting database stuff so we
|
|
|
|
// can return a 'server_error' response if something goes wrong!
|
|
|
|
try {
|
2015-02-05 19:57:17 +01:00
|
|
|
try {
|
|
|
|
$client = id(new PhabricatorOAuthServerClientQuery())
|
|
|
|
->setViewer($viewer)
|
|
|
|
->withPHIDs(array($client_phid))
|
|
|
|
->executeOne();
|
|
|
|
} catch (PhabricatorPolicyException $ex) {
|
|
|
|
// We require that users must be able to see an OAuth application
|
|
|
|
// in order to authorize it. This allows an application's visibility
|
|
|
|
// policy to be used to restrict authorized users.
|
|
|
|
|
|
|
|
// None of the OAuth error responses are a perfect fit for this, but
|
|
|
|
// 'invalid_client' seems closest.
|
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_client',
|
|
|
|
pht('Not Authorized'),
|
|
|
|
pht('You are not authorized to authenticate.'));
|
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
if (!$client) {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_request',
|
|
|
|
pht('Invalid Client Application'),
|
|
|
|
pht(
|
|
|
|
'Request parameter %s does not specify a valid client application.',
|
|
|
|
phutil_tag('strong', array(), 'client_id')));
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
|
|
|
$name = $client->getName();
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
$server->setClient($client);
|
|
|
|
if ($redirect_uri) {
|
|
|
|
$client_uri = new PhutilURI($client->getRedirectURI());
|
|
|
|
$redirect_uri = new PhutilURI($redirect_uri);
|
|
|
|
if (!($server->validateSecondaryRedirectURI($redirect_uri,
|
|
|
|
$client_uri))) {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_request',
|
|
|
|
pht('Invalid Redirect URI'),
|
|
|
|
pht(
|
|
|
|
'Request parameter %s specifies an invalid redirect URI. '.
|
|
|
|
'The redirect URI must be a fully-qualified domain with no '.
|
|
|
|
'fragments, and must have the same domain and at least '.
|
|
|
|
'the same query parameters as the redirect URI the client '.
|
|
|
|
'registered.',
|
|
|
|
phutil_tag('strong', array(), 'redirect_uri')));
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
$uri = $redirect_uri;
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
} else {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
$uri = new PhutilURI($client->getRedirectURI());
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (empty($response_type)) {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_request',
|
|
|
|
pht('Invalid Response Type'),
|
|
|
|
pht(
|
|
|
|
'Required request parameter %s is missing.',
|
|
|
|
phutil_tag('strong', array(), 'response_type')));
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
if ($response_type != 'code') {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'unsupported_response_type',
|
|
|
|
pht('Unsupported Response Type'),
|
|
|
|
pht(
|
|
|
|
'Request parameter %s specifies an unsupported response type. '.
|
|
|
|
'Valid response types are: %s.',
|
|
|
|
phutil_tag('strong', array(), 'response_type'),
|
|
|
|
implode(', ', array('code'))));
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
if ($scope) {
|
|
|
|
if (!PhabricatorOAuthServerScope::validateScopesList($scope)) {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_scope',
|
|
|
|
pht('Invalid Scope'),
|
|
|
|
pht(
|
|
|
|
'Request parameter %s specifies an unsupported scope.',
|
|
|
|
phutil_tag('strong', array(), 'scope')));
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
}
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
$scope = PhabricatorOAuthServerScope::scopesListToDict($scope);
|
2015-02-07 00:32:55 +01:00
|
|
|
} else {
|
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_request',
|
|
|
|
pht('Malformed Request'),
|
|
|
|
pht(
|
|
|
|
'Required parameter %s was not present in the request.',
|
|
|
|
phutil_tag('strong', array(), 'scope')));
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
// NOTE: We're always requiring a confirmation dialog to redirect.
|
|
|
|
// Partly this is a general defense against redirect attacks, and
|
|
|
|
// partly this shakes off anchors in the URI (which are not shaken
|
|
|
|
// by 302'ing).
|
|
|
|
|
|
|
|
$auth_info = $server->userHasAuthorizedClient($scope);
|
|
|
|
list($is_authorized, $authorization) = $auth_info;
|
|
|
|
|
|
|
|
if ($request->isFormPost()) {
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
$scope = PhabricatorOAuthServerScope::getScopesFromRequest($request);
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
2012-03-05 22:27:20 +01:00
|
|
|
if ($authorization) {
|
|
|
|
$authorization->setScope($scope)->save();
|
|
|
|
} else {
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
$authorization = $server->authorizeClient($scope);
|
2012-03-05 22:27:20 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
|
|
|
|
$is_authorized = true;
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
}
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
} catch (Exception $e) {
|
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'server_error',
|
|
|
|
pht('Server Error'),
|
|
|
|
pht(
|
|
|
|
'The authorization server encountered an unexpected condition '.
|
|
|
|
'which prevented it from fulfilling the request.'));
|
|
|
|
}
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
// When we reach this part of the controller, we can be in two states:
|
|
|
|
//
|
|
|
|
// 1. The user has not authorized the application yet. We want to
|
|
|
|
// give them an "Authorize this application?" dialog.
|
|
|
|
// 2. The user has authorized the application. We want to give them
|
|
|
|
// a "Confirm Login" dialog.
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
if ($is_authorized) {
|
|
|
|
|
|
|
|
// The second case is simpler, so handle it first. The user either
|
|
|
|
// authorized the application previously, or has just authorized the
|
|
|
|
// application. Show them a confirm dialog with a normal link back to
|
|
|
|
// the application. This shakes anchors from the URI.
|
|
|
|
|
|
|
|
$unguarded = AphrontWriteGuard::beginScopedUnguardedWrites();
|
|
|
|
$auth_code = $server->generateAuthorizationCode($uri);
|
|
|
|
unset($unguarded);
|
|
|
|
|
|
|
|
$full_uri = $this->addQueryParams(
|
|
|
|
$uri,
|
|
|
|
array(
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
'code' => $auth_code->getCode(),
|
|
|
|
'scope' => $authorization->getScopeString(),
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
'state' => $state,
|
|
|
|
));
|
|
|
|
|
2015-02-09 23:23:49 +01:00
|
|
|
if ($client->getIsTrusted()) {
|
|
|
|
return id(new AphrontRedirectResponse())
|
|
|
|
->setIsExternal(true)
|
|
|
|
->setURI((string)$full_uri);
|
|
|
|
}
|
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
// TODO: It would be nice to give the user more options here, like
|
|
|
|
// reviewing permissions, canceling the authorization, or aborting
|
|
|
|
// the workflow.
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
$dialog = id(new AphrontDialogView())
|
|
|
|
->setUser($viewer)
|
|
|
|
->setTitle(pht('Authenticate: %s', $name))
|
|
|
|
->appendParagraph(
|
|
|
|
pht(
|
|
|
|
'This application ("%s") is authorized to use your Phabricator '.
|
|
|
|
'credentials. Continue to complete the authentication workflow.',
|
|
|
|
phutil_tag('strong', array(), $name)))
|
|
|
|
->addCancelButton((string)$full_uri, pht('Continue to Application'));
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return id(new AphrontDialogResponse())->setDialog($dialog);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Here, we're confirming authorization for the application.
|
2015-02-07 00:32:55 +01:00
|
|
|
if ($authorization) {
|
|
|
|
$desired_scopes = array_merge($scope, $authorization->getScope());
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
} else {
|
2015-02-07 00:32:55 +01:00
|
|
|
$desired_scopes = $scope;
|
|
|
|
}
|
|
|
|
if (!PhabricatorOAuthServerScope::validateScopesDict($desired_scopes)) {
|
|
|
|
return $this->buildErrorResponse(
|
|
|
|
'invalid_scope',
|
|
|
|
pht('Invalid Scope'),
|
|
|
|
pht('The requested scope is invalid, unknown, or malformed.'));
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
}
|
|
|
|
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
$form = id(new AphrontFormView())
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
->addHiddenInput('client_id', $client_phid)
|
|
|
|
->addHiddenInput('redirect_uri', $redirect_uri)
|
|
|
|
->addHiddenInput('response_type', $response_type)
|
|
|
|
->addHiddenInput('state', $state)
|
|
|
|
->addHiddenInput('scope', $request->getStr('scope'))
|
|
|
|
->setUser($viewer)
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
->appendChild(
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
PhabricatorOAuthServerScope::getCheckboxControl($desired_scopes));
|
|
|
|
|
|
|
|
$cancel_msg = pht('The user declined to authorize this application.');
|
|
|
|
$cancel_uri = $this->addQueryParams(
|
|
|
|
$uri,
|
|
|
|
array(
|
|
|
|
'error' => 'access_denied',
|
|
|
|
'error_description' => $cancel_msg,
|
|
|
|
));
|
|
|
|
|
|
|
|
$dialog = id(new AphrontDialogView())
|
|
|
|
->setUser($viewer)
|
|
|
|
->setTitle(pht('Authorize "%s"?', $name))
|
|
|
|
->setSubmitURI($request->getRequestURI()->getPath())
|
|
|
|
->setWidth(AphrontDialogView::WIDTH_FORM)
|
|
|
|
->appendParagraph(
|
|
|
|
pht(
|
|
|
|
'Do you want to authorize the external application "%s" to '.
|
|
|
|
'access your Phabricator account data?',
|
|
|
|
phutil_tag('strong', array(), $name)))
|
|
|
|
->appendChild($form->buildLayoutView())
|
|
|
|
->addSubmitButton(pht('Authorize Access'))
|
|
|
|
->addCancelButton((string)$cancel_uri, pht('Do Not Authorize'));
|
|
|
|
|
|
|
|
return id(new AphrontDialogResponse())->setDialog($dialog);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
private function buildErrorResponse($code, $title, $message) {
|
|
|
|
$viewer = $this->getRequest()->getUser();
|
|
|
|
|
|
|
|
$dialog = id(new AphrontDialogView())
|
|
|
|
->setUser($viewer)
|
|
|
|
->setTitle(pht('OAuth: %s', $title))
|
|
|
|
->appendParagraph($message)
|
|
|
|
->appendParagraph(
|
|
|
|
pht('OAuth Error Code: %s', phutil_tag('tt', array(), $code)))
|
|
|
|
->addCancelButton('/', pht('Alas!'));
|
|
|
|
|
|
|
|
return id(new AphrontDialogResponse())->setDialog($dialog);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
private function addQueryParams(PhutilURI $uri, array $params) {
|
|
|
|
$full_uri = clone $uri;
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
foreach ($params as $key => $value) {
|
|
|
|
if (strlen($value)) {
|
|
|
|
$full_uri->setQueryParam($key, $value);
|
|
|
|
}
|
|
|
|
}
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
|
Fix an anchor redirect issue with OAuth server, plus modernize the application a bit
Summary:
Ref T4593. Via HackerOne. An attacker can use the anchor reattachment, combined with the Facebook token workflow, combined with redirection on OAuth errors to capture access tokens. The attack works roughly like this:
- Create an OAuth application on Phabricator.
- Set the domain to `evil.com`.
- Grab the OAuth URI for it (something like `https://phabricator.com/oauthserver/auth/?redirect_uri=http://evil.com&...`).
- Add an invalid `scope` parameter (`scope=xyz`).
- Use //that// URI to build a Facebook OAuth URI (something like `https://facebook.com/oauth/?redirect_uri=http://phabricator.com/...&response_type=token`).
- After the user authorizes the application on Facebook (or instantly if they've already authorized it), they're redirected to the OAuth server, which processes the request. Since this is the 'token' workflow, it has auth information in the URL anchor/fragment.
- The OAuth server notices the `scope` error and 302's to the attacker's domain, preserving the anchor in most browsers through anchor reattachment.
- The attacker reads the anchor in JS and can do client workflow stuff.
To fix this, I've made several general changes/modernizations:
- Add a new application and make it beta. This is mostly cleanup, but also turns the server off for typical installs (it's not generally useful quite yet).
- Add a "Console" page to make it easier to navigate.
- Modernize some of the UI, since I was touching most of it anyways.
Then I've made specific security-focused changes:
- In the web-based OAuth workflow, send back a human-readable page when errors occur. I //think// this is universally correct. Previously, humans would get a blob of JSON if they entered an invalid URI, etc. This type of response is correct for the companion endpoint ("ServerTokenController") since it's called by programs, but I believe not correct for this endpoint ("AuthController") since it's used by humans. Most of this is general cleanup (give humans human-readable errors instead of JSON blobs).
- Never 302 off this endpoint automatically. Previously, a small set of errors (notably, bad `scope`) would cause a 302 with 'error'. This exposes us to anchor reattachment, and isn't generally helpful to anyone, since the requesting application did something wrong and even if it's prepared to handle the error, it can't really do anything better than we can.
- The only time we'll 'error' back now from this workflow is if a user explicitly cancels the workflow. This isn't a 302, but a normal link (the cancel button), so the anchor is lost.
- Even if the application is already approved, don't blindly 302. Instead, show the user a confirmation dialog with a 'continue' link. This is perhaps slightly less user-friendly than the straight redirect, but I think it's pretty reasonable in general, and it gives us a lot of protection against these classes of attack. This redirect is then through a link, not a 302, so the anchor is again detached.
-
Test Plan: I attempted to hit everything I touched. See screenshots.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8517
2014-03-13 20:59:10 +01:00
|
|
|
return $full_uri;
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
}
|
OAuthServer polish and random sauce
Summary:
This diff makes the OAuthServer more compliant with the spec by
- making it return well-formatted error codes with error types from the spec.
- making it respect the "state" variable, which is a transparent variable the
client passes and the server passes back
- making it be super, duper compliant with respect to redirect uris
-- if specified in authorization step, check if its valid relative to the client
registered URI and if so save it
-- if specified in authorization step, check if its been specified in the access
step and error if it doesn't match or doesn't exist
-- note we don't make any use of it in the access step which seems strange but
hey, that's what the spec says!
This diff makes the OAuthServer suck less by
- making the "cancel" button do something in the user authorization flow
- making the client list view and client edit view be a bit more usable around
client secrets
- fixing a few bugs I managed to introduce along the way
Test Plan:
- create a test phabricator client, updated my conf, and then linked and
unlinked phabricator to itself
- wrote some tests for PhabricatorOAuthServer -- they pass!
-- these validate the various validate URI checks
- tried a few important authorization calls
--
http://phabricator.dev/oauthserver/auth/?client_id=X&state=test&redirect_uri=http://www.evil.com
--- verified error'd from mismatching redirect uri's
--- verified state parameter in response
--- verified did not redirect to client redirect uri
-- http://phabricator.dev/oauthserver/auth/?client_id=X w/ existing
authorization
--- got redirected to proper client url with error that response_type not
specified
-- http://phabricator.dev/oauthserver/auth/?client_id=X&response_type=code w/
existing authorization
--- got redirected to proper client url with pertinent code!
- tried a few important access calls
-- verified appropriate errors if missing any required parameters
-- verified good access code with appropriate other variables resulted in an
access token
- verified that if redirect_uri set correctly in authorization required for
access and errors if differs at all / only succeeds if exactly the same
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley, ajtrichards
Maniphest Tasks: T889, T906, T897
Differential Revision: https://secure.phabricator.com/D1727
2012-03-01 23:46:18 +01:00
|
|
|
|
OAuth - Phabricator OAuth server and Phabricator client for new Phabricator OAuth Server
Summary:
adds a Phabricator OAuth server, which has three big commands:
- auth - allows $user to authorize a given client or application. if $user has already authorized, it hands an authoization code back to $redirect_uri
- token - given a valid authorization code, this command returns an authorization token
- whoami - Conduit.whoami, all nice and purdy relative to the oauth server.
Also has a "test" handler, which I used to create some test data. T850 will
delete this as it adds the ability to create this data in the Phabricator
product.
This diff also adds the corresponding client in Phabricator for the Phabricator
OAuth Server. (Note that clients are known as "providers" in the Phabricator
codebase but client makes more sense relative to the server nomenclature)
Also, related to make this work well
- clean up the diagnostics page by variabilizing the provider-specific
information and extending the provider classes as appropriate.
- augment Conduit.whoami for more full-featured OAuth support, at least where
the Phabricator client is concerned
What's missing here... See T844, T848, T849, T850, and T852.
Test Plan:
- created a dummy client via the test handler. setup development.conf to have
have proper variables for this dummy client. went through authorization and
de-authorization flows
- viewed the diagnostics page for all known oauth providers and saw
provider-specific debugging information
Reviewers: epriestley
CC: aran, epriestley
Maniphest Tasks: T44, T797
Differential Revision: https://secure.phabricator.com/D1595
2012-02-04 01:21:40 +01:00
|
|
|
}
|