Add new Firmware TPM (fwTPM)

[dgarske]

Summary

• Add firmware TPM 2.0 server (fwTPM) implementing TPM 2.0 spec v1.38 (105/113 commands, 93% coverage)
• Spec-compliant primary key derivation: deterministic RSA (iterative KDFa prime generation), ECC (KDFa scalar, Q=d*G), KEYEDHASH/SYMCIPHER keys — same seed + template always produces the same key
• ChangePPS / ChangeEPS hierarchy commands with seed regeneration and cache flush
• NV storage with TLV journal format (flash-friendly, append-only)
• Socket (mssim) and TIS/SHM transports for desktop, UART transport for embedded
• STM32 Cortex-M33 bare-metal port with TrustZone (CMSE) support
• Shared crypto refactoring: extract KDFa/KDFe, AES-CFB, HMAC/Hash to tpm2_crypto.c
• Bounds-checked TPM2_Packet_ParseU16Buf variant for safer response parsing
• Auth validation: reject oversized auth values instead of silent truncation
• SIGTERM/SIGINT signal handler for graceful NV save on server kill

fwTPM Server

Core TPM 2.0 command processing in src/fwtpm/:

• fwtpm_command.c — 105 command handlers with full auth, sessions, parameter encryption
• fwtpm_nv.c — TLV journal NV storage (file-based default, HAL-abstracted for flash)
• fwtpm_io.c — Socket transport (mssim + swtpm protocol auto-detection)
• fwtpm_tis.c / fwtpm_tis_shm.c — TIS register interface via POSIX shared memory
• fwtpm_crypto.c — Key generation, sign/verify, seed encrypt/decrypt helpers
• Clock HAL with NV-persisted clockOffset

Build: ./configure --enable-fwtpm && make

Example: wolfSSL/wolftpm-examples#1

Primary Key Derivation (TPM 2.0 Part 1 Section 26)

• RSA: iterative KDFa prime generation with labels "RSA p" / "RSA q"
• ECC: KDFa scalar derivation, public point Q = d*G
• KEYEDHASH/SYMCIPHER: KDFa with algorithm-specific labels
• hashUnique = H(sensitiveCreate.data || unique) per Section 26.1
• Primary cache retained as performance optimization, no longer required for correctness

UART Transport (--enable-swtpm=uart)

New transport option for wolfTPM client library to communicate with embedded fwTPM over serial:

• ./configure --enable-swtpm=uart — uses termios serial I/O instead of TCP sockets
• TPM2_SWTPM_HOST env var selects serial device at runtime
• Same mssim protocol as socket transport (compatible with all wolfTPM examples)

Testing

• 311 tpm2-tools compatibility tests (scripts/tpm2_tools_test.sh)
• Full wolfTPM example suite (examples/run_examples.sh)
• libFuzzer harness with corpus generator (tests/fuzz/)
• m33mu Cortex-M33 emulator CI test (scripts/fwtpm_emu_test.sh)
• ASan / UBSan clean
• 20 CI matrix configurations (pedantic gcc/clang -Werror, sanitizers, feature-disable variants)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 69 changed files in this pull request and generated 9 comments.

---

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

[wolfSSL-Fenrir-bot]

Fenrir Automated Review — PR #474

Scan targets checked: wolftpm-bugs, wolftpm-consttime, wolftpm-defaults, wolftpm-mutation, wolftpm-proptest, wolftpm-src, wolftpm-zeroize

Findings: 4
4 finding(s) posted as inline comments (see file-level comments below)

This review was generated automatically by Fenrir. Findings are non-blocking.

[Copilot]

Pull request overview

Copilot reviewed 53 out of 71 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (1)

src/tpm2_swtpm.c:199

• For UART mode, read() can legally return 0 due to VMIN=0/VTIME>0 timeout (not EOF). Treating wrc==0 as EOF will break long-running fwTPM operations over UART (e.g., RSA keygen) and prevents the cumulative timeout logic below from ever triggering. Fix by handling wrc==0 differently under WOLFTPM_SWTPM_UART (e.g., continue polling until the cumulative timeout elapses), while keeping the EOF behavior for sockets.

    while (bytes_remaining > 0) {
        wrc = read(ctx->tcpCtx.fd, ptr, bytes_remaining);
        if (wrc < 0) {
            /* Retry on EINTR; treat EAGAIN/EWOULDBLOCK as transient too. */
            if (errno == EINTR
                #ifdef EAGAIN
                    || errno == EAGAIN
                #endif
                #if defined(EWOULDBLOCK) && (!defined(EAGAIN) || EWOULDBLOCK != EAGAIN)
                    || errno == EWOULDBLOCK
                #endif
            ) {
                continue;
            }
            #ifdef DEBUG_WOLFTPM
            printf("Failed to read from TPM socket %d, got errno %d"
                   " = %s\n", ctx->tcpCtx.fd, errno, strerror(errno));
            #endif
            rc = TPM_RC_FAILURE;
            break;
        }
        if (wrc == 0) {
            #ifdef DEBUG_WOLFTPM
            printf("Failed to read from TPM socket: EOF\n");
            #endif
            rc = TPM_RC_FAILURE;
            break;
        }

---

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

[Copilot]

Pull request overview

Copilot reviewed 53 out of 71 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

src/tpm2_swtpm.c:199

• In UART mode, termios is configured with VMIN=0 and VTIME>0, where read() returning 0 indicates a timeout (no bytes yet), not EOF. Treating wrc==0 as EOF will cause spurious failures for long-running TPM operations. For WOLFTPM_SWTPM_UART, handle wrc==0 as a transient timeout and continue waiting until the cumulative timeout is reached (the CLOCK_MONOTONIC block you added is a good place to enforce the overall limit). Keep the EOF behavior for socket mode.

        if (wrc == 0) {
            #ifdef DEBUG_WOLFTPM
            printf("Failed to read from TPM socket: EOF\n");
            #endif
            rc = TPM_RC_FAILURE;
            break;
        }

---

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

[Copilot]

Truncating handle->name.size here can leave the session in a state where session->name no longer matches the real object name. That can break authorization/HMAC computations in later command flows and can also hide an upstream size/encoding error. Prefer keeping the previous behavior: validate handle->name.size against sizeof(session->name.name) and return BUFFER_E rather than truncating.

[Copilot]

This read path only zeroes len bytes (clamped to 4), but the client-side HAL copies the originally requested size bytes from reg_data and does not know about the server-side clamp. That can leak stale shared-memory contents beyond the first 4 bytes. To prevent data disclosure, zero the entire regs->reg_data buffer (or at least regs->reg_len bytes as requested) before writing the scalar register bytes.

[Copilot]

This script runs with set -e, so if the grep pattern matches nothing (e.g., on early boot failure or different output), grep exits 1 and the whole script aborts before the intended result checks and log tail. Consider appending || true (or temporarily disabling set -e around this grep) so the script can still report failures and show diagnostics consistently.

Suggested change

	grep -E "^===|^fwTPM|^ |^Running|^ Startup|^ SelfTest|^ GetRandom|^ GetCapability|^ Random|^All self|^SELF-TEST" "$LOG"
	grep -E "^===|^fwTPM|^ |^Running|^ Startup|^ SelfTest|^ GetRandom|^ GetCapability|^ Random|^All self|^SELF-TEST" "$LOG" || true

[Copilot]

fcntl(fd, F_SETFD, FD_CLOEXEC) overwrites existing FD flags instead of preserving them. Use F_GETFD and OR-in FD_CLOEXEC (as done elsewhere in the PR) so you don't accidentally clear other descriptor flags.

Suggested change

	(void)fcntl(fd, F_SETFD, FD_CLOEXEC);
	{
	int fdFlags = fcntl(fd, F_GETFD);
	if (fdFlags >= 0) {
	(void)fcntl(fd, F_SETFD, fdFlags | FD_CLOEXEC);
	}
	}