In April 2026 we reported an unauthenticated SQL injection in Metacat 2.x to a customer through the Synack platform. The flaw sits in a legacy registration endpoint, needs no credentials to reach, and gives an attacker full access in the context of the database account Metacat itself uses. We found it together with our colleagues at LVL 0x00, disclosed it privately to the Metacat maintainers, and it is now public as CVE-2026-48114 with a CVSS 3.1 score of 9.8.
Metacat is an open-source data repository platform from the National Center for Ecological Analysis and Synthesis (NCEAS) at UC Santa Barbara, used to store and share research data and its metadata. The vulnerable code was removed in Metacat 3.0.0, so current releases are not affected. This post covers what was wrong, how to tell whether you are exposed, and what the episode says about software that has been retired but not switched off.
- Identifier
- CVE-2026-48114 (GHSA-wrc6-rc34-hrcg)
- Product
- Metacat, a Java servlet application from NCEAS
- Affected
- 2.0.0 through 2.19.1
- Fixed in
- 3.0.0 and later
- Entry point
/harvesterRegistration- Severity
- 9.8 Critical,
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H - Weaknesses
- CWE-89 (SQL Injection), CWE-287 (Improper Authentication)
If you run Metacat 2.x
Start here, then read on for the detail.
Upgrade. Metacat 3.0.0 removed the vulnerable code, and the advisory recommends moving to the current 3.x release (3.4.0 or later at the time it was published). The maintainers have not backported a fix to 2.x, which is past its supported lifetime.
If you cannot upgrade today, the advisory describes a workaround: in Tomcat, remove or restrict the servlet and servlet-mapping entries in web.xml for HarvesterRegistration, HarvesterRegistrationLogin, HarvesterServlet and DataProvider, then restart the container. This can reduce functionality for clients that depend on the legacy harvester or the OAI-PMH handler. Blocking those paths at your reverse proxy is a reasonable stopgap while you make the change.
Check whether it was used. If the endpoint was ever reachable from an untrusted network, look for requests to /harvesterRegistration and /harvesterRegistrationLogin from addresses you do not recognize. Standard access logs record the request line but not form bodies, so you will usually see that the endpoint was hit rather than what was sent. Also look for unexpected or malformed rows in the HARVEST_SITE_SCHEDULE table. If anything looks wrong, treat the database as compromised: rotate the Metacat database credentials and any secrets stored alongside the data.
What was wrong
Three weaknesses combined. Each is common on its own; together they gave an anonymous caller full access in the context of Metacat's database account.
1. SQL built by string concatenation
The registration servlet's dbInsert() method builds an INSERT into HARVEST_SITE_SCHEDULE by joining strings. Values go through a helper named quoteString(), which sounds like a defense but only wraps its input in single quotes. It never escapes them. The pattern looks like this:
// Illustrative pattern only, not copied from the Metacat source
String quoteString(String value) {
return "'" + value + "'"; // adds quotes, never escapes them
}
String sql = "INSERT INTO harvest_site_schedule (...) VALUES ("
+ quoteString(unit) + ", "
+ quoteString(contactEmail) + ", "
+ quoteString(documentListURL) + ", ...)";
statement.executeUpdate(sql);
A value that contains a single quote ends the string literal early, and everything after it is parsed as SQL. Three request parameters reached this code: unit, contactEmail and documentListURL.
2. Stacked queries turn one bad INSERT into database access
PostgreSQL accepts several statements, separated by semicolons, in a single executeUpdate() call. An attacker is therefore not limited to the INSERT the developer intended. Whatever the Metacat database account is permitted to do, the attacker can do: read data, write data, and run functions. In a Metacat database that includes stored credentials, session material and the repository's document corpus.
3. A session is not a login
The registration servlet's only gate was a check that a session exists. It never verified that the session belonged to an authenticated user. Metacat's main servlet creates a session for any caller, including an unauthenticated one, so the check was satisfied by anyone who asked for a cookie. That is CWE-287, and it is why the CVSS vector reads PR:N: no privileges required.
How the pieces fit together
A second, delayed injection point
The documentListURL value is not only stored. A scheduled harvester later fetches that URL and parses what comes back. A <scope> element from that attacker-controlled document flowed, unsanitized, into a query in HarvestDocument.metacatHighestRevision(). A fix that cleaned up only the registration form would have left this second path open. The reliable fix is to parameterize every query, whatever the source of the value: a form field, a database row or a document fetched from the network.
About the proof of concept
The proof of concept we shared with the maintainers used a time-delay function: a request that makes the database sleep for several seconds is enough to show that injected statements execute. We are deliberately not publishing a ready-to-send request. The fix has been available since 3.0.0, but not every deployment has migrated, and the advisory gives defenders what they need to act.
The fix, and why there was no advisory
The vulnerable code lived in Metacat's legacy harvester client. Metacat 3.0.0 removed that package (commit 820d595), which fixed the flaw as a side effect of removing old code. It was not handled as a security fix, so no advisory was issued, and 2.x was never patched. Anyone still running 2.x had no signal that their version carried a critical, unauthenticated vulnerability.
This is not a criticism of the Metacat team. Retiring legacy code is the right instinct, and maintaining old branches is expensive for any open-source project. What costs almost nothing is publishing an advisory once you learn that the code you removed was vulnerable, so the people still running it can find out.
To their credit, the maintainers took the report seriously. They confirmed the issue within a week, set up GitHub security advisories for the project, and reached out to groups they knew were still on 2.x before the details became public.
Disclosure timeline
- We submit the finding through the Synack platform to a customer program, together with LVL 0x00.
- Synack approves the finding and the reward is paid.
- We report privately to the Metacat maintainers by email and propose 30 days before public disclosure.
- The maintainers confirm the issue.
- We reply and offer to review draft advisories.
- We verify the remediation on the affected customer deployment.
- GitHub, acting as CNA, reserves CVE-2026-48114.
- The maintainers publish GitHub advisory GHSA-wrc6-rc34-hrcg.
- The CVE record and NVD entry are published. CISA's assessment at that point listed exploitation as "none", automatable as "yes" and technical impact as "total".
The customer's deployment had been remediated, and the fix verified, before the identifier was assigned.
What to take from this
- End of support is not end of life in production. Retired versions keep running. Know which ones you operate, and plan their replacement before the vendor stops publishing fixes.
- A fix by removal is still a fix for a vulnerability. If you maintain software and learn that code you deleted was exploitable, publish an advisory. If you run software, watch release notes for removed components as well as patched ones.
- A session is not authentication. Check that the caller has an authenticated identity, not that a cookie exists.
- Parameterize everything. That includes values you stored earlier and data you fetched from elsewhere. Quoting helpers that do not escape are worse than nothing, because they look like protection.
- Remove entry points you no longer use. The vulnerable servlets stayed mapped and reachable on 2.x installs long after 3.x replaced that API.
Credits
The GitHub advisory credits wwallace (Fjord.AI) and lvl0x00 as reporters. Thank you to the Metacat team, credited on the advisory as taojing2002 (remediation developer), artntek (remediation verifier) and mbjones (coordinator), and to Synack and the customer's security team for their quick triage and remediation. We also reported a separate path traversal issue in Metacat, tracked as CVE-2026-47754, which we cover in a companion write-up.
References
- GitHub advisory GHSA-wrc6-rc34-hrcg
- CVE record for CVE-2026-48114
- NVD entry for CVE-2026-48114
- Fix commit 820d595
Find it before an adversary does
We find new vulnerabilities in the software you depend on, prove them with working proofs of concept, and coordinate disclosure with the vendor.
Start an engagement