* Add support for CMake
* Add support for Zephyr
* Make sure the internal key state is properly handled in case a public
key is imported into a reloaded private key.
Signed-off-by: Tobias Frauenschläger <tobias.frauenschlaeger@oth-regensburg.de>
Some fixes to better handle Dilithium keys and signatures in the TLS
handshake.
Signed-off-by: Tobias Frauenschläger <tobias.frauenschlaeger@oth-regensburg.de>
Make testing digests consistent.
Add KATs for all digests.
Check unaligned input and output works.
Perform chunking tests for all digests.
Fix Blake2b and Blake2s to checkout parameters in update and final
functions.
Fix Shake256 and Shake128 to checkout parameters in absorb and squeeze
blocks functions.
Add default digest size enums for Blake2b and Blake2s.
There are the following reasons for this
- it conflicts with the OFFSETOF macro in the OS/2 header (Open Watcom)
- it is compiler-specific and should use the C standard offsetof definition in the header file stddef.h
- it is more transparent unique name
ML-KEM/Kyber:
MakeKey call generate random once only for all data.
Allow MakeKey/Encapsulate/Decapsulate to be compiled separately.
Pull out public key decoding common to public and private key decode.
Put references to FIPS 140-3 into code. Rename variables to match FIPS
140-3.
Fix InvNTT assembly code for x64 - more reductions.
Split out ML-KEM/Kyber tests from api.c.
TLSX:
Store the object instead of the private key when WOLFSSL_MLKEM_CACHE_A
is defined or WOLFSSL_TLSX_PQC_MLKEM_STORE_OBJ. Faster decapsulation
when A is cached and object stored.
To store private key as normal define
WOLFSSL_TLSX_PQC_MLKEM_STORE_PRIV_KEY.
misc.c: when Intel x64 build, assume able to read/write unaligned
- Search certificate based on responderId
- Verify response signer is authorized for all single responses
- Align with OpenSSL behavior
- Separate wolfSSL_OCSP_basic_verify from verification done during
decoding
Matrix A is expensive to calculate.
Usage of ML-KEM/Kyber is
1. First peer generates a key and sends public to second peer.
2. Second peer encapsulates secret with public key and sends to first
peer.
3. First peer decapsulates (including encapsulating to ensure same as
seen) with key from key generation.
Caching A keeps the matrix A for encapsulation part of decapsulation.
The matrix needs to be transposed for encapsulation.
AES-GCM: don't generate M0 when using assembly unless falling back to C
and then use new assembly code.
HMAC: add option to copy hashes (--enable-hash-copy
-DWOLFSSL_HMAC_COPY_HASH) to improve performance when using the same key
for multiple operations.
XOR in random value to scalar and perform special scalar multiplication.
Multiply x3 and z3 by random value to randomize co-ordinates.
Add new APIs to support passing in an RNG.
Old APIs create a new RNG.
Only needed for the C implementations that are not small.
Modified TLS and OpenSSL compat API implementations to pass in RNG.
Fixed tests and benchmark program to pass in RNG.
Adds two features for SE050:
1. `WOLFSSL_SE050_AUTO_ERASE`. When enabled, this will automatically
erase a key from the SE050 when `wc_ecc_free()` and friends are
called.
2. `WOLFSSL_SE050_NO_RSA`. This stops RSA offloading onto the SE050,
useful for the SE050E which does not have RSA support.
AES-GCM: don't generate M0 when using assembly unless falling back to C
and then use new assembly code.
HMAC: add option to copy hashes (--enable-hash-copy
-DWOLFSSL_HMAC_COPY_HASH) to improve performance when using the same key
for multiple operations.
The SE050 port won't compile in the latest wolfSSL. This patch:
* Updates the documentation
* Fixes a missing `#ifdef` that breaks the build
* Changes the use of `mp_int` to `MATH_INT_T`
* Fixes compiler error with `ecc.c`
* Adds a tiny bit of extra debugging info
* Correct cmake script to support Open Watcom toolchain (#8167)
* Fix thread start callback prototype for Open Watcom toolchain (#8175)
* Added GitHub CI action for Windows/Linux/OS2
* Improvements for C89 compliance.
Thank you @jmalak for your contributions.
in wolfcrypt/src/coding.c, add ALIGN64 to hexDecode[], and add hexEncode[] for use by Base16_Encode();
in wolfcrypt/src/misc.c and wolfssl/wolfcrypt/misc.h:
move ctMask*() up so that min() and max() can use them, and add ctMaskWord32GTE();
add ALIGN64 to kHexChar[];
add CT implementation of CharIsWhiteSpace();
remove min_size_t() and max_size_t() recently added, but only one user (refactored).
Implemented based on the NIST Initial Public Draft "NIST SP 800-232 ipd". Testing based on KAT's available at https://github.com/ascon/ascon-c. Added configuration for testing in github action.
Fix spaces at start of copyright line in .asm files.
Changed generation of X25519 and Ed448 code to better match changes
already in C code. Fixed formatting in places.
fix internal functions wolfSSL_read_internal() and ReceiveData() to likewise accept size_t sz;
add negative sz checks where needed to other functions that call wolfSSL_read_internal() and ReceiveData();
add min_size_t() and max_size_t() to misc.c/misc.h.
- Support using MAXQ for:
- AES-ECB
- AES-CCM
- AES-CBC
- ECC Key Generation and ECDH
- in wc_ecc_import_private_key_ex():
- check to make sure devId is not invalid before calling wc_MAXQ10XX_EccSetKey().
- This is because the raspberry pi sometimes need to sign stuff.
- in aes_set_key() and ecc_set_key():
- delete a key in case it already exists; ignore error since it might not exist.
- unlock, lock the HW mutex around ECDSA_sign() because it needs access to rng
- in wolfSSL_MAXQ10XX_CryptoDevCb:
- allow maxq1065 to call the crypto callback.
- do not set the key during signing; use pre provisioned one instead (DEVICE_KEY_PAIR_OBJ_ID)
Add assembly for Poly1305 using ARM32 NEON instruction set.
For Poly1305 ARM32 Base:
Change name from poly1305_blocks_arm32_16 to poly1305_arm32_blocks_16
poly1305.c:
ARM32 NEON - buffer up to 4 blocks
x86_64 - only calculate powers of r once after key is set.
test.c: poly1305 testing with multiple updates.
benchmark: chacha20-poly1305 now uses AAD
wolfcrypt/src/integer.c: add additional guards against OOB access from uint wraps and null derefs of mp_int.dp, and refactor mp_grow() and mp_init_size() to use XMEMSET, for the benefit of clang-tidy. in mp_grow(), fix the condition for the realloc to assure always evaluated if a->alloc == 0.
wolfcrypt/src/asn.c: fix wc_CreatePKCS8Key() so that *outSz is always assigned when LENGTH_ONLY_E is returned.
wolfcrypt/src/pkcs7.c: remove redundant inner condition in wc_PKCS7_EncodeAuthEnvelopedData(), added in previous commit and caught on review by Jacob (thanks!).
wolfcrypt/src/sp_int.c: in sp_mont_norm(), add another suppression for the same false positive in sp_mul() suppressed in previous commit.
wolfcrypt/src/srp.c: refactor SrpHashSize() to return ALGO_ID_E rather than 0 when unknown/uncompiled alg is requested.
several fixes for defects reported by cppcheck:
wolfcrypt/src/ecc.c: fix for cppcheck oppositeInnerCondition from cppcheck-2.16.0 in _ecc_make_key_ex(), and fixes for related unhandled errors discovered by manual inspection;
wolfcrypt/test/test.c: fix XREALLOC call in memcb_test() to resolve cppcheck-detected memleak.
wolfcrypt/test/test.c: in dh_ffdhe_test(), when defined(HAVE_PUBLIC_FFDHE), use wc_DhSetKey_ex() rather than wc_DhSetKey() to exercise the primality check in _DhSetKey().
AES GCM streaming - fix GHASH_ONE_BLOCK to use CPU feature information.
AES-GCM uses EOR3 (SHA-3 instruction) - split assembly code.
Kyber uses SQRDMLSH - split assembly code.
Changed define from WOLFSSL_AARCH64_NO_SQRMLSH to
WOLFSSL_AARCH64_NO_SQRDMLSH to match instruction.
Improved array data format for inline assembly code.
test.c: Dilithium private key not available in cert_test.h unless
signing is enabled.
./configure --disable-shared --enable-dilithium=make,44,65,87
./configure --disable-shared --enable-dilithium=make,sign,44,65,87
./configure --disable-shared --enable-dilithium=make,verify,44,65,87
test.c: Dilithium doesn't have decode/encode when
WOLFSSL_DILITHIUM_NO_ASN1 is defined.
./configure --disable-shared --enable-dilithium=yes
CFLAGS=-DWOLFSSL_DILITHIUM_NO_ASN1
Automated replacement with
```
git ls-files -z wolfcrypt/src/port/ IDE/ | xargs -0 pcre2grep -l '[^_]AES_BLOCK_SIZE' | xargs sed --regexp-extended --in-place 's/([^_])AES_BLOCK_SIZE/\1WC_AES_BLOCK_SIZE/g'
```
Checked for mis-transformations with
```
git ls-files -z | xargs -0 pcre2grep '[^-[()+*/[:space:]]WC_AES_BLOCK_SIZE' | less
```
Checked for residual hits with
```
git ls-files -z | xargs -0 pcre2grep '[^_]AES_BLOCK_SIZE' | less
```
Deliberately excluded:
* ChangeLog.md -- do not alter history.
* doc/ -- do not confuse documentation with newly prefixed macro, because AES_BLOCK_SIZE is available unless -DOPENSSL_COEXIST.
* tests/api.c -- the unit tests deliberately use compatibility names, and are not compatible with -DOPENSSL_COEXIST.
* wrapper/CSharp/wolfSSL_CSharp/wolfCrypt.cs -- false positive hits on C# names.
* wrapper/CSharp/wolfCrypt-Test/wolfCrypt-Test.cs -- false positive hits on C# names.
* reference in wolfssl/wolfcrypt/aes.h that defines AES_BLOCK_SIZE when -UOPENSSL_COEXIST.
* reference in wolfssl/wolfcrypt/settings.h that defines WC_AES_BLOCK_SIZE for old FIPS when -UWC_AES_BLOCK_SIZE.
When no AVX2 available, kyber_prf() is called to produce more than one
SHAKE-256 blocks worth of ouput. Otherwise only one block is needed.
Changed function to support an outlen of greater than one block.
src/x509.c: wolfssl_x509_name_entry_set() ne->object is freed if call to
wolfSSL_OBJ_nid2obj_ex() fails. Always assign directly back to
ne->object.
wolfcrypt/test/test.c: aes_ctr_test() doesn't need AES decrypt
./configure '--disable-shared' '--enable-cryptonly'
'CFLAGS=-DNO_AES_DECRYPT' '--disable-aescbc' '--disable-aesofb'
'--disable-aescfb' '--disable-aesgcm' '--disable-aesccm'
'--enable-aesctr' '--disable-aesxts' '--disable-aeseax'
tests/api.c: test_X509_STORE_InvalidCa() only defined when !NO_RSA
./configure '--disable-shared' '--enable-opensslall' '--disable-rsa'
tests/api.c: test_wolfSSL_GENERAL_NAME_print() free ridObj if not
assigned into gn.