Fixed failing cert verification with nonblocking-OCSP and low MFL

[MatanRad]

Description

Fixes a bug that happens when using nonblocking OCSP and having a low MFL configured.
This is caused by the Certificate handshake message being fragmented and the OCSP_WANT_READ error being ignored, causing the library to skip over the current pendingMsg.

Testing

Ran locally with a nonblocking OCSP setup and a configured MFL of 1024

[wolfSSL-Bot]

Can one of the admins verify this patch?

[Copilot]

Pull request overview

Fixes a TLS 1.3 handshake bug where, under nonblocking OCSP with a small Max Fragment Length (MFL), a fragmented Certificate handshake message can complete assembly but then return OCSP_WANT_READ; the current fragment handling would ignore that and move past the assembled pendingMsg.

Changes:

• Treat OCSP_WANT_READ like WC_PENDING_E when processing a fully reassembled fragmented handshake message, rewinding state to retry the fragment.
• Expand the preprocessor guard around the retry/rewind logic to include nonblocking OCSP builds.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The rollback for OCSP_WANT_READ is currently compiled only when WOLFSSL_NONBLOCK_OCSP or WOLFSSL_ASYNC_CRYPT is defined. OCSP_WANT_READ can also be returned from CRL I/O paths (HAVE_CRL_IO) during cert verification, and this fragment-buffer code would still free pendingMsg and advance *inOutIdx in that case, reintroducing the same “skip pendingMsg” behavior. Consider keying this block off WOLFSSL_ASYNC_IO (or additionally HAVE_CRL_IO) to ensure any nonblocking I/O path returning OCSP_WANT_READ preserves pendingMsg for retry.

Suggested change

	#if defined(WOLFSSL_ASYNC_CRYPT) || defined(WOLFSSL_NONBLOCK_OCSP)
	if (ret == WC_NO_ERR_TRACE(WC_PENDING_E) || ret == WC_NO_ERR_TRACE(OCSP_WANT_READ)) {
	#if defined(WOLFSSL_ASYNC_IO) || defined(HAVE_CRL_IO) || \
	defined(WOLFSSL_ASYNC_CRYPT) || defined(WOLFSSL_NONBLOCK_OCSP)
	if (ret == WC_NO_ERR_TRACE(WC_PENDING_E) ||
	ret == WC_NO_ERR_TRACE(OCSP_WANT_READ)) {

[Copilot]

The combined condition ret == ...WC_PENDING_E || ret == ...OCSP_WANT_READ is written on a single long line, whereas nearby code in this file typically wraps multi-term conditions across lines for readability/consistency. Please reformat this if to match the surrounding style (e.g., split across lines like other WC_PENDING_E/OCSP_WANT_READ checks).

Suggested change

	if (ret == WC_NO_ERR_TRACE(WC_PENDING_E) || ret == WC_NO_ERR_TRACE(OCSP_WANT_READ)) {
	if (ret == WC_NO_ERR_TRACE(WC_PENDING_E) ||
	ret == WC_NO_ERR_TRACE(OCSP_WANT_READ)) {

[Copilot]

This change adds special handling for OCSP_WANT_READ when completing a fragmented TLS 1.3 handshake message. There doesn’t appear to be an automated regression test exercising TLS 1.3 fragmentation together with a nonblocking OCSP callback that returns WANT_READ (e.g., similar to the existing TLS1.3 fragment test in tests/test-tls13.conf, but with WOLFSSL_NONBLOCK_OCSP/OCSP callback configured). Adding such a test would help prevent regressions in this retry/rewind logic.