The Root That Brought Its Own Generator — Java Bug Hunt
Modelled on "Curveball" (CVE-2020-0601, patched by Microsoft in January 2020 after a report from the NSA).
- Language: Java
- Layer: Backend
- Difficulty: Hard
- Concepts: Security, Validation
- Modelled on: Windows CryptoAPI · CVE-2020-0601
- Visible tests: the genuine root is trusted; a root's key with a substituted generator is not trusted
- Reward: 50 XP for a complete fix
Incident
- Severity: SEV-1 (priority: Critical (signature spoofing))
- Ticket: BUG-CURVEBALL
- Reported by: security
- Time to fix: 35 minutes, the target for a hard incident
Failing checks on the shipped build
- a root's key with a substituted generator is not trusted — same key, attacker-chosen generator: expected condition to hold
How to work this incident
- Read the report: write down the expected behaviour and the actual one in a sentence each.
- Reproduce it: run the visible tests and find the failing check that matches the report.
- Read the logs and the failing checks for the first value that is wrong — the earliest wrong value is nearer the cause than the loudest error.
- Follow the request through the handler, the validation and the service it calls; find the first step whose output disagrees with the report.
- Fix the cause rather than the symptom, run the visible tests again, and submit against the hidden ones — then write the root cause down.
Briefing
Modelled on "Curveball" (CVE-2020-0601, patched by Microsoft in January 2020 after a report from the NSA). Windows' CryptoAPI accepted elliptic-curve certificates that spell out their curve parameters explicitly. When checking whether a certificate was a trusted root, it matched on the public key but did not check that the curve parameters — in particular the generator point — were the standard ones. An attacker could keep a trusted root's public key, pick a different generator for which they knew the matching private key, and have their forged "root" trusted for TLS and code signing.
ChainValidator.java decides whether a certificate is one of the trusted roots. Curves can be named (resolved through Curves.named) or given explicitly.
Fix isTrustedRoot so a match requires the same public key on the same curve.
Bug report
BUG-CURVEBALL · Priority: Critical (signature spoofing) · Reported by: security
ChainValidator.isTrustedRoot(cert, store) is true only when some root in store.roots has:
- the same public key point (qx, qy), AND
- identical domain parameters: p, a, b, gx, gy and n all equal
A key's parameters are key.explicit when present, otherwise Curves.named(key.curveName). A curve that resolves to nothing (unknown name, no name and no parameters) never matches. Explicitly spelled-out parameters equal to a named curve's values count as that curve.
Observed: a certificate carrying a trusted root's public key with its own explicit generator is accepted as that root.
Logs
[chain] leaf "update.example" -> root "Trusted Root CA" (explicit params, gx=555 gy=777)
[chain] root matched by public key -> TRUSTEDThe code as shipped
src/crypto/ChainValidator.java (editable)
class ChainValidator {
// A certificate is anchored when it is one of the trusted roots.
static boolean isTrustedRoot(Certificate cert, RootStore store) {
for (Certificate root : store.roots) {
if (root.key.qx == cert.key.qx && root.key.qy == cert.key.qy) {
return true;
}
}
return false;
}
}Read-only context: src/crypto/Certificate.java, src/crypto/EcParams.java.
Open the hunt to start the incident clock, edit the files, run the visible tests and submit against the hidden ones, then write the root cause down. More Java bug hunts.