This document explains how we handle vulnerability issue triage in the x/vulndb issue tracker.
All vulnerabilities in the Go vulnerability database are currently stored as a YAML file in the data/reports or data/excluded directory.
For a detailed explanation of the report format and style guide, see doc/format.md.
Any open issue should be in one of the following states:
Maintainers of the Go vulndb move issues from one state to another. The intent behind these explicit states is to describe the (minimum) next steps required to bring the issue to resolution.
The issue has been filed by the vulndb worker or an external reporter.
The issue will have the title: x/vulndb: potential Go vuln in <module/package>: <CVE ID and or GHSA ID>
.
To transition from this state, do one of the following:
NeedsInvestigation
, and discuss the issue with the team.excluded: REASON
, and use the vulnreport create-excluded
command to create a CL.NeedsReport
, and use the vulnreport
tool to assist in creating a CL.excluded: OUT_OF_SCOPE
and close the issue.duplicate
and close the issue.Label: NeedsInvestigation
This state is used when it is not clear how to proceed. (Otherwise, an issue can move straight to one of the other states.)
Make a plan to discuss the issue with the team to determine a course of action.
Label: NeedsReport
The issue has been confirmed to be an in-scope Go vulnerability, and a report needs to be added to data/reports
.
Label: excluded: REASON
where REASON is one of the possible excluded reasons.
The issue represents a reported vulnerability, but is not in scope for the main data/reports
folder. An “excluded” report needs to be added to data/excluded
.
Label: excluded: OUT_OF_SCOPE
or duplicate
.
The issue is out of scope for both the data/reports
and data/excluded
folders. For example, it is an issue mistakenly posted to the tracker (excluded: OUT_OF_SCOPE
) or a duplicate (duplicate
) of another issue.
The issue can be closed without further action.
Clone the x/vulndb repository: git clone https://go.googlesource.com/vulndb
.
Get a GitHub access token with scope repo: public_repo
(follow instructions for “personal access token (classic)”).
Store the token in a file, e.g., ~/.github-token
, and run: export VULN_GITHUB_ACCESS_TOKEN=`cat ~/.github-token`
(you can also store this command in a ~/.bashrc
file or similar).
(To use experimental generative AI features) Get a Gemini API key.
As above, you can store the token in a file like ~/.gemini-api-key
and use the environment variable GEMINI_API_KEY
.
From the repo root, run go install ./cmd/vulnreport
to install the latest version of vulnreport tool.
NeedsReport
)Sync your git repo, re-install the vulnreport tool, and create a fresh branch.
From the repo root, run vulnreport create <GitHub issue number>
. The vulnreport
tool will create a YAML report template for the CVE or GHSA at the specified GitHub issue number.
Tips for the vulnreport create
command:
vulnreport create 99 1000-1010
would create reports for issue #99 and all issues from #1000 to #1010, skipping any that are closed, do not exist, or already have reports.)-ai
flag to automatically populate a (first-draft) AI-generated summary and description. (See Experimental Features).-symbols
flag to attempt to automatically populate vulnerable symbols. (See Experimental Features).create
command attempts to find a GHSA for the vulnerability and pull it from osv.dev. If this is not working, use the -cve
flag to use the CVE (rather than the GHSA) as the default source, or the -graphql
flag to pull GHSAs directly from Github's GraphQL API.Edit the report file template, following the guidance in doc/format.md. A few tips:
credits
field. Otherwise, delete the field.vulnerable_at
field, put the highest version just before the vuln is fixed. The pkgsite versions page can help with the list of versions. The GitHub UI also makes it easy to list tags (click “Code”, then the dropdown that shows the current branch, then “Tags”). Walk the versions backwards from the fixed one to find the highest that doesn't contain the fix. (It might not be the immediately preceding version.)symbols
list by reading the CVE, the fixing CLs, and the code at the vulnerable version you chose above.From the repo root, run vulnreport fix <GitHub issue number>
. This will lint the report, add exported symbols, and convert the YAML to OSV.
Once any errors are fixed, run vulnreport commit <GitHub issue number>
. This will create a git commit containing the new files with a standard commit message. Commits are to the local git repository. The vulnreport commit
command also accepts multiple space-separated issue numbers, and will create a separate commit for each report.
Send the commit for review and approval. See the Go contribution guide for sending a change on Gerrit.
If you make changes to the report during review, re-run vulnreport fix <GitHub issue number>
before re-mailing to update the OSV and make sure the report is still valid.
excluded: REASON
)vulnreport create-excluded
. This will batch create YAML reports for all issues with the excluded: REASON
label. If there is an error creating any given report, the skipped issue number will be printed to stdout and that issue will have to be created manually with vulnreport create <Github issue number>
. (see steps 2-4 above for more information). Additionally, create-excluded
will automatically create a single commit for all successful reports. To skip this auto-commit step, use the -dry
flag.Sometimes an issue describes a vulnerability that we already have a report for. The worker doesn't always detect this automatically, so it is a good idea to grep the /data
directory of this repo for the module path and read the report to see if the vulns are the same.
NEW: The command vulnreport duplicates
(with no args) can find likely duplicates on the issue tracker.
If the issue is indeed a duplicate:
duplicate
to the issue.data/reports
, and add the duplicate IDs to the cves
or ghsas
section, as appropriate. Running vulnreport fix
can sometimes find the IDs automatically. (If the duplicate IDs are already present, close the GH issue.)vulnreport -up commit NNN
to update generated files and create a commit. Edit the generated commit message so that it includes the words “add aliases”. You can also add “Fixes #DDDD” (the number of the duplicate issue) to the commit message, or close it manually.When adding a vulnerability report about the standard library, ensure that the references section follows this format:
references: - report: https://go.dev/issue/<#> - fix: https://go.dev/cl/<#> - web: https://groups.google.com/g/golang-announce/c/<XXX>/<YYY>
You can find these links in the golang-announce@ email for the security release fixing this vulnerability.
Report: The Github issue will be listed in the golang-announce@ email.
Fix: The PR will be a go.dev/cl/<#> link, found as a gopherbot comment on the issue for the vulnerability.
Web: The golang-announce email link.
Occasionally, we will receive new information about a Go vulnerability and want to update the existing report.
In that case, reopen the issue for the report to discuss the change, rather than create a new issue.
The command vulnreport -up commit NNN
can be used to create a more sensible commit message when committing an updated report.
The command vulnreport suggest <Github issue number>
uses Gemini to create AI-generated summaries and descriptions for a report. The -i
(interactive) flag gives the option of applying the suggestions directly to the YAML file.
The command vulnreport symbols <Github issue number>
uses the commit link(s) in the report to find a list of possibly vulnerable functions (functions that were present in the parent commit and were changed by the patch). Currently, this command cannot handle pull requests or commits with multiple parents.
This section describes frequent issues that come up when triaging vulndb reports.
When vulnreport fix
fails with an error message like
/path/to/package@v1.2.3/foo.go:1:2: could not import C (no metadata for C)
a frequent cause is the local machine missing C
library headers causing typechecking of cgo packages to fail. The easiest workaround is to use a machine with the development headers installed or to install them.
Commonly missing packages include:
When the NIST page says “AWAITING ANALYSIS”, write the report; don‘t wait for them to finish their analysis. “Awaiting analysis” just means that NVD hasn’t yet looked at the vulnerability and assigned a severity score/CWE etc. Since we don't care about those pieces of information, we can ignore that banner and just create a report if the vulnerability is in scope for our database.