Support RFC 9802 LMS and XMSS in X.509 certificate and CSR generation

[Frauschi]

Overview

Builds on the existing RFC 9802 LMS/XMSS X.509 verification support by wiring HSS/LMS and XMSS/XMSS^MT keys into the certificate and PKCS#10 certificate-request generation API. These stateful hash-based schemes follow the same "sign the message directly, no pre-hash" pattern already used for Falcon / ML-DSA / SLH-DSA.

CSR support is included (per design discussion): cert signing and CSR proof-of-possession share the same MakeSignature() dispatch, and there's no reason to special-case CSRs out. The only caveat — each cert and each CSR consumes one of the key's finite one-time signatures — is inherent to the scheme and documented at the API surface.

Public API

• New CertType selectors LMS_TYPE, XMSS_TYPE, XMSSMT_TYPE accepted by wc_MakeCert_ex, wc_SignCert_ex and wc_MakeCertReq_ex.
• New cert_enums values LMS_KEY, XMSS_KEY, XMSSMT_KEY.
• New wc_LmsKey_PublicKeyToDer / wc_XmssKey_PublicKeyToDer encoders that emit the RFC 9802 SubjectPublicKeyInfo via the existing SetAsymKeyDerPublic helper and the already-registered HSS_LMSk / XMSSk / XMSSMTk key OIDs.
• Callers set cert->sigType to CTC_HSS_LMS / CTC_XMSS / CTC_XMSSMT.

LmsKey key; Cert cert; byte der[8192];
wc_LmsKey_Init(&key, heap, devId);
wc_LmsKey_SetParameters(&key, levels, height, winternitz);
wc_LmsKey_SetWriteCb/SetReadCb/SetContext(...);   /* persist state */
wc_LmsKey_MakeKey(&key, &rng);

wc_InitCert(&cert);
cert.sigType = CTC_HSS_LMS; cert.isCA = 1; cert.selfSigned = 1;
wc_MakeCert_ex(&cert, der, sizeof(der), LMS_TYPE, &key, &rng);
wc_SignCert_ex(cert.bodySz, cert.sigType, der, sizeof(der), LMS_TYPE, &key, &rng);

Implementation

• Thread LmsKey*/XmssKey* through EncodePublicKey, MakeAnyCert, MakeCertReq, MakeSignature and SignCert (ASN.1 template path), mirroring the existing PQ-key plumbing.
• SignCert sizes the signature buffer at runtime from wc_LmsKey_GetSigLen / wc_XmssKey_GetSigLen, since LMS/XMSS signatures are parameter-dependent and can exceed MAX_ENCODED_SIG_SZ. In WOLFSSL_NO_MALLOC builds (fixed-size buffer) an oversized signature fails cleanly with BUFFER_E rather than overflowing.
• SignCert validates that sigType matches the key (LMS → CTC_HSS_LMS; XMSS → CTC_XMSS/CTC_XMSSMT per is_xmssmt), and the XMSS_TYPE/XMSSMT_TYPE selector is validated against the key's tree variant — so a mismatch returns an error instead of silently emitting a malformed certificate.
• Forward-declare LmsKey/XmssKey in asn_public.h (guarded typedefs in wc_lms.h/wc_xmss.h) so the shared signatures compile regardless of feature macros.
• Enable WC_ENABLE_ASYM_KEY_EXPORT/IMPORT for non-verify-only LMS/XMSS builds so the SPKI encoder is compiled.
• LMS/XMSS cert generation is template-only (the original/non-template path has no LMS/XMSS support, consistent with verification); the non-template MakeAnyCert/MakeCertReq emit a clear diagnostic and ALGO_ID_E if handed such a key.

Tests

test_rfc9802_lms_x509_gen / test_rfc9802_xmss_x509_gen generate a self-signed root CA and a PKCS#10 CSR, then verify the cert via wolfSSL_CertManagerVerifyBuffer and the CSR proof-of-possession via wc_ParseCert. Coverage includes single-level LMS, multi-level HSS (L2), single-tree XMSS, and XMSS^MT, plus negative cases for sigType/keyType and selector mismatches.

Notes / scope

• No interop test: OpenSSL still lacks LMS/XMSS X.509 certificate signing (XMSS & LMS openssl/openssl#21360), and Bouncy Castle's encoding is not yet aligned with the final RFC 9802 — so there is no aligned third-party implementation to interoperate against.
• Verified across full, WOLFSSL_ASN_ORIGINAL (non-template), and verify-only LMS/XMSS builds; the API test suite passes with 0 failures, and testwolfcrypt passes.

https://claude.ai/code/session_01RDkoNViK5aNkUAJqnuL8Y5

---

Generated by Claude Code

[Frauschi]

can we simply remove the file and the directory instead?

[Frauschi]

Done — removed certs/xmss/include.am entirely and dropped its include certs/xmss/include.am line from certs/include.am, so the certs/xmss/ directory is gone rather than kept as an empty file.

Note this was folded into the rebase onto the updated master: upstream had moved the LMS/XMSS tests out of tests/api.c into tests/api/test_lms_xmss.c, so I re-homed all the generation test work there (api.c is now unchanged from upstream) and resolved the conflict that way. Force-pushed the rebased branch.

---

Generated by Claude Code