OBSOLETE!!! Check here instead.
The following document describes the process of working from an inbound security bug report to releasing a fix on all affected branches. You can also refer to the Security Cheat Sheets to provide a quick walk through or as a reference to check your progress.
The first step of triage is confirming that a bug does exist and it does represent a vulnerability. Most reporters provide a version in the bug report, but you’ll generally need to verify against the current stable, beta, and trunk builds. The Reporting Security Bugs page provides information on what detail is required and how to get it, so you should use that as a guide to fill in anything the report is missing. If you don’t have a particular build on hand to perform the analysis against, you can leave a comment in the bug requesting that another team member assist in verifying it. The list of currently open security (modulo Security-Severity-None) bugs can be viewed with this query.
There are a number of bugs we want to keep track of for security reasons, but shouldn't track as vulnerabilities. This includes new security features and improvements, or changes that simply have a security impact or warrant a closer inspection from a security perspective. In this case the bug should not be filed as Type-Security. Instead it should be filed as whatever type is relevant (generally Type-Bug) and also have the Feature-Security label set. This ensures the security team will still receive updates, but keeps it out of the list of open vulnerabilities.
First verify the bug, requirements to trigger, and affected platforms; if you cannot verify the bug directly (e.g. nonavailability of the affected platform) then assign or CC an owner who can. Once you’ve verified the bug, you need to determine the appropriate labels. You should follow the severity guidelines Severity Guidelines for Security Issues to determine the rating for the Security-Severity-* label. Remember to also consider any mitigating factors that might reduce the severity, such as unusual or excessive interaction, or the feasibility of reliably triggering the condition in the wild (e.g. race conditions are notoriously difficult).
Once you’ve determined the severity you can assign the milestone and priority according to the following guidelines:
Set the SecImpacts flags to identify the affected branches, so we know what needs to be merged and have historical tracking data:
The milestone targets may shift slightly to adjust for a release schedule. For example, when the current stable is at the end of its cycle you will generally bump all open bugs to target the next stable milestone.
Also, it's important to ensure that all security bugs are set to Type-Security and any security bugs with a severity other than Security-Severity-None include the Restrict-View-SecurityTeam flag.
The remaining triage follows the standard guidelines. Key points to remember are the following:
Any WebKit bugs should be upstreamed to bugs.webkit.org immediately after verification, basic analysis to isolate the issue, and reducing any test cases to a manageable size. When upstreaming WebKit security bugs, remember the following points:
Any members of the Chromium Security Team involved in bug triage will need to join the WebKit Security Group. If you're not already on WebKit Security, please ask another Chromium Security Team member to nominate you. After joining you'll need to request security bug access (this can be done by replying to your confirmation email). Once bug access is granted, you'll need to configure your account to get emailed on all inbound security bugs. You do this by adding email@example.com to the User Watching section on the Email Preferences tab on WebKit's bugzilla. (You'll probably want to create a Gmail filter on these messages as well.)
Landing WebKit Patches
To contribute a patch to WebKit you can generally just follow the instructions on webkit.org. However, one additional step must be taken to land patches for security bugs, because the WebKit Commit Bot and the WebKit Review Bot are not included in the WebKit Security Group. This means your patch will not be placed in the commit queue after receiving cq+ unless the bots are explicitly given access to the bug. To do this, add these addresses to the cc: list of the WebKit bug:
As a rule, you’ll want to ensure that the full initial triage is performed and the bug is updated within 48 hours of the initial report. During business hours the basic triage and bug update should occur within four hours at most.
Some reports are obviously not legitimate, or represent behavior that we do not intend to change. In those cases you should mark the bug as WontFix or Invalid (as appropriate) and provide a simple explanation to the reporter. For non-security bugs erroneously reported via the security template, you should remove all security labels, set the “Crash” label if appropriate, add in the appropriate area/feature labels, and set the status to “Untriaged.”
The shepherding process starts with finding an owner for a bug. If you have access to security bugs, you can find a list of all open security bugs without an owner here. When analysing a bug for triage you’ll often find that you can work out a solution yourself. If that’s the case and you have time to develop a patch, then this is usually the most expedient solution to getting a fix out.
For bugs that you can’t fix yourself, you’ll need to find an owner. If you don’t know who to assign a bug to, it’s often best to start by asking other members of the security team if they’re familiar with the code. Then they can either take the ownership of the bug, or point you to a developer who would make a good owner.
If no one on the security team has a good suggestion, you’ll generally have to use svn blame or some other means to track back one of the original developers of the code. If you don’t have a checkout handy, you can always use the annotate feature on http://src.chromium.org/.
Priorities and schedules vary between developers. So, when you assign a security bug to an owner you should make sure you convey the timeline you’re expecting for a fix. If you don’t see regular progress in the bug report, you may need to manually follow up with the owner. Oftentimes, the bug just ended up being harder to reproduce or fix than it originally appeared, so it’s a good idea to offer help whenever there are unanticipated delays.
Keep in mind that on average we get high severity bugs fixed on the release branch within 30 days of the initial report, and almost always within 60 days at the most. For critical severity bugs we have a hard deadline of 60 days from initial report to a shipping fix.
Once a fix is applied to trunk either you or the owner will change the bug status to Merge-Approved and switch the Restrict-View-SecurityTeam label to Restrict-View-SecurityNotify. At that point you’ll need to evaluate whether the bug is a good candidate for a merge.
Security fixes often need to be merged to the stable branch (this includes patches to Chrome, WebKit, or other dependencies). Generally this is done by one of the security team members. The first thing you need to do is evaluate whether a fix should be considered for merge. Generally any high or critical severity fixes should be seriously considered for merge. Medium and low-severity fixes can be considered as well if they're particularly worrisome (e.g. bad information leak), have significant non-security impact (e.g. top crasher), or are just very low risk.
One of the key considerations in determining what to merge is assessing the risk of breakage, which the bug’s owner should be able to convey. For risky fixes you’ll often want to consider bumping it to a later milestone. When you've decided to merge a fix, you should verify with a release manager that the merge window is open for any affected branches. Normally a fix will need to be merged to both the current stable and beta branches.
You should merge fixes via drover, which automates most of the process. If the drover merge does not apply cleanly you may want to reconsider whether the fix will introduce an unacceptable breakage risk. In the event that you can’t use drover to merge a fix, you still have the option of manually merging from a branch checkout. In general, most team members should have a checkout for the current stable, beta and trunk branches. So, this can generally just follow the normal patch process, but using a branch repository.
Once the fix is successfully merged you will need to update the the bug with the merge revision numbers, and change the status to FixUnreleased (or to Fixed if the stable branch is not affected) and set the merge status to Merge-Completed (if drover didn't do so). For beta merges, you should also verify that the ReleaseBlock-Stable flag is present.
After a patch has been landed (and potentially merged) you’ll need to update the security release document, which is used for tracking security bugs for release notes.
Generally, once a fix has been applied to trunk we will consider a bug for a vulnerability reward. This will generally be done by the team member shepherding the bug, or the owner if it’s a security team member. To remind yourself that you may want to consider a specific bug you can add the reward-topanel label in the tracker.
When nominating a bug for a reward, be sure to consider the overall quality, and any additional value the reporter brought to the process. For reference, the general guidelines are listed on the Vulnerability Rewards Program page.