XSS (Persistent) - Selecting role(s) for protected branches

**Summary:** When using the dropdown that selects the groups or users that are allowed to push or merge to a protected branch within a project, it is possible to trigger a XSS with a malicious user name string. **Description:** This vulnerability is similar to the recently announced CVE-2018-10379. The username input string where an attacker is able to inject a payload is in the same location, but the XSS that renders is in a different location. Since the remediation needs to be applied at the presentation layer, this is indeed a separate vulnerability and needs to be fixed separately (although Gitlab could start whitelisting characters allowed in usernames, similar to how Gitlab whitelists characters for Group or Project names). The steps to reproduce are fairly simple but there are some restrictions: * Only members of a project with Master access are able to become victims of the XSS * Only groups/members with a subscription level of Starter or higher are able to perform the XSS, since this requires the ability to restrict merge/push permissions of a branch to a specific user. This is a premium feature only allowed at Starter or higher. (https://gitlab.com/help/user/project/protected_branches#restricting-push-and-merge-access-to-certain-users-starter) ## Steps To Reproduce: 1. Set your own username as "<img src=x onerror=alert(document.domain)> foo / bar" 1. Make yourself have at least Master access to a project 1. In this project, ensure at least one branch is in the project and that branch is a "Protected Branch" 1. Under Project Settings -> Repository -> Protected Branches, select the dropdown under the "Ability to Merge" section 1. Notice that the onerror attribute from the username renders. ## Supporting Material/References: More information can be provided upon request. You'll notice the payload above is the same as the payload used in a test file (inside the ce source code repo) for the CVE I attached. I only found this vulnerability since I was testing the previous CVE on our own internal instance of Gitlab, left my username saved as that malicious string, and later found the alert() dialog popup in another area of the site even after patching to 10.7.2. ## Suggested Remediation I believe this is the offending Line: https://gitlab.com/gitlab-org/gitlab-ee/blob/master/ee/app/assets/javascripts/projects/settings/access_dropdown.js#L461 ^There could also be other XSS vulnerabilities in this JS file. Everywhere else in the app uses the underscore method `_.escape()`to escape user input, but this file doesn't. ## Impact The security impact is the same as any typical persistent xss. The hacker selected the **Cross-site Scripting (XSS) - Stored** weakness. This vulnerability type requires contextual information from the hacker. They provided the following answers: **URL** https://gitlab.com/group/project/settings/repository **Verified** Yes