Incident documentation/20150820-PageCategorization (Privacy)
Once the issue was confirmed and problem patch identified, the Security team immediately began the process of reverting the feature, and deleting all recent changes entries created by the patch to prevent leaking this private data further.
During the time that the patch was deployed, the API call that could have triggered this leak of private information was called 340 times (326 times on commons.wikimedia.org, 10 times on pl.wiktionary.org, and 4 times on en.wiktionary.org) by 69 different IP addresses.
- 2015-08-18 18:30 UTC: 1.26wmf19 (which contained Gerrit ) was deployed to group0 wikis (testwiki, test2wiki, mediawikiwiki, zerowiki, testwikidatawiki)
- 2015-08-19 18:06 UTC: 1.26wmf19 deployed to group1 wikis (non-Wikipedia wikis)
- 2015-08-19 21:49 UTC: User Peter Bowman reports bug T109638
- Approximately 2015-08-20 18:00 UTC: Release Engineering deploys revert (Gerrit )
- 2015-08-20 20:25 UTC: Script to delete recentchanges rows finishes
The patch that caused the issue was created, reviewed, and merged by experienced staff and volunteer developers, yet each of these developers failed to identify the potential security impact of the design. This indicates that the security ramifications of the functions involved are likely not structured or documented in a way that makes the issues clear, and this particular code pattern has not been highlighted as dangerous in documentation or training. Additionally, there are currently no automated or manual testing procedures specifically to identify privacy issues in new version of MediaWiki.
In the near term, we will add notice about User::newFromId(0) to secure code training (bug T110620), and investigate updating the function definition or documentation to make the security impact more clear to developers.
Longer term, we will investigate using static analysis tools to automatically flag this code pattern (depends on bug T110617, and look at developing security training specifically for QA testers, which would include looking for privacy issues.