[dgarske]

Three RSA 4096 issues in fwTPM were fixed (all undetected at RSA 2048):

• Buffer Under-allocation & Name Collisions: TPM2_Packet_AppendBytes() skips oversized fields instead of truncating. At RSA 4096, an RSA public area requires 604 bytes, but FWTPM_MAX_PUB_BUF was capped at 512. The omitted field was the modulus, causing every object to hash to the same Name. Similarly, FWTPM_NV_PUBAREA_EST was capped at 600. Both buffers now dynamically size their RSA arms using MAX_RSA_KEY_BITS, leaving RSA 2048 output byte-identical.
• Unchecked Overflow & Uninitialized Memory: FwComputePublicName(), FwSpdmTpmKeyName(), and Name calculations in FwCmd_Create() / FwCmd_Import() used packet.pos without checking packet.overflow (matching wolfTPM2_ComputeName()). Two of those packets were also left uninitialized.
• Benchmark Mislabeling: examples/bench/bench.c sent a fixed 256-byte unpadded RSA message and labeled all results as 2048-bit, even though unpadded RSA requires a payload width matching the modulus.

Verification: Built and tested at RSA 2048, RSA 4096 (--enable-infineon=slb9672), and --disable-rsa (--enable-pqc was skipped; requires ML-DSA wolfSSL). An additional RSA 4096 failure in tests/unit_tests.c (FWTPM_MAX_PRIVKEY_DER) is resolved by #628.

[aidangarske]

Skoll Code Review

Scan type: review

Overall recommendation: COMMENT
Findings: 2 total — 2 posted, 0 skipped
2 finding(s) posted as inline comments (see file-level comments below)

Posted findings

• [Medium] RSA-disabled builds still pay the RSA-4096 buffer cost — wolftpm/fwtpm/fwtpm.h:302-303; wolftpm/fwtpm/fwtpm_nv.h:60-66
• [Medium] Regression test does not exercise the changed failure paths — tests/fwtpm_unit_tests.c:18176-18219

Review generated by Skoll