The ECC key generation test was failing due not using large enough of a
buffer. Fixed to use a better size.
1. Set the shared digest/sig buffer size in
_ecc_pairwise_consistency_test() to the maximum possible based on the
math in wc_ecc_sig_sz().
configure.ac:
* fix logic in "Forcing off" test expressions, first flubbed in 19106a9510;
* fix auto-enable of compkey to exclude v5 even if v5-dev.
src/tls13.c: fix gating for HKDF _ex() variants (>=6.0, not >=5.3).
wolfcrypt/src/error.c: snip out stray spaces at start of several ECC error message strings.
wolfcrypt/test/test.c:
* in render_error_message(), use wolfSSL_ERR_reason_error_string() if available rather than wc_GetErrorString(), to render non-wolfcrypt error strings;
* in ecc_test_deterministic_k(), ecc384_test_deterministic_k(), ecc521_test_deterministic_k(), on FIPS <6.0, gate out SHA384 and SHA512 tests (FIPS v5 only supports SHA256 in wc_ecc_gen_deterministic_k());
* in cmac_test(), gate use of wc_AesCmacGenerate_ex() and wc_AesCmacVerify_ex() on >=6.0, not >=5.3.
1. Update the ECC PCT to use the key's heap to allocate any buffers for
the test. This is similar to how RSA does it.
2. Put the buffers on the stack if not using small stack option.
wolfssl/wolfcrypt/wc_port.h: add WOLFSSL_API attribute to wolfSSL_Atomic_Int_Init, wolfSSL_Atomic_Int_FetchAdd, and wolfSSL_Atomic_Int_FetchAdd, and add fallback definitions for them, allowing elimination of SINGLE_THREADED implementations of wolfSSL_Ref*(), and allowing ungated use of wolfSSL_Atomic_* calls in api.c.
wolfcrypt/src/dh.c: in wc_DhAgree_ct(), remove frivolous XMEMSET() and stray semicolon.
wolfcrypt/benchmark/benchmark.c: fix bench_rsaKeyGen() to skip tests of key sizes below RSA_MIN_SIZE, and add 4096 bit benchmark if RSA_MAX_SIZE is big enough.
tests/unit.h:
* adopt definitions of TEST_FAIL, TEST_SUCCESS, and TEST_SKIPPED from unit.c, remap TEST_SKIPPED from -7777 to 3, and add TEST_SUCCESS_NO_MSGS, TEST_SKIPPED_NO_MSGS, EXPECT_DECLS_NO_MSGS(), and EXPECT_FAILURE_CODEPOINT_ID, to support existing and future expected-particular-failure test cases without log noise.
* rename outer gate from CyaSSL_UNIT_H to TESTS_UNIT_H.
tests/api.c:
* use EXPECT_DECLS_NO_MSGS() in test_ssl_memio_setup(), test_ssl_memio_read_write(), and test_wolfSSL_client_server_nofail_memio(), and globally update affected expected error codes to correspond.
* use atomics for {client,server}SessRemCount{Malloc,free} to fix races in SessRemCtxCb() and SessRemSslSetupCb().
The function _fp_exptmod_nct() is using WOLFSSL_NO_MALLOC to guard
using stack allocation vs malloc. It's twin function _fp_exptmod_ct()
is using WOLFSSL_SMALL_STACK for this. This is causing inappropriate use
of malloc() in a small stack environment. The no-malloc case will also
be kept so static memory and no-malloc fix still works.
1. Change the guards for `#ifndef WOLFSSL_NO_MALLOC` in the function
`_fp_exptmod_nct()` to `#if defined(WOLFSSL_SMALL_STACK) &&
!defined(WOLFSSL_NO_MALLOC)`.
JacobBarthelmeh
jordan
Brett Nicholas
gojimmypi
Bill Phipps
Bill Phipps
John Safranek
Daniel Pouzzner
David Garske
Hideki Miyazaki
ZackLabPC
night1rider
msi-debian
András Fekete
Juliusz Sosinowicz
Sean Parkinson