GitLab

* cbc.c (cbc_decrypt): For in-place operation (src == dst case),
eliminate use of src variable.
* cfb.c (cfb_decrypt): Likewise.
* gcm.c (gcm_crypt): Likewise, and replace one memxor3 by memxor.

Dated 2017-08-27.

Was broken by rename of configure variable GMP_NUMB_BITS -->
NUMB_BITS.

* aclocal.m4 (GMP_PROG_CC_FOR_BUILD): Add -g when compiling with
gcc.

* sexp-transport-format.c (base64_encode_in_place): New helper
function.
(sexp_transport_vformat): Use it.

* testsuite/base64-test.c (test_fuzz_once): Update to use char
type where appropriate.
(test_main): Use helper functions base64_encode_in_place and
base64_decode_in_place (copied to this file).

* tools/pkcs1-conv.c (base64_decode_in_place): New helper
function.
(decode_base64): Use it.

* testsuite/testutils.c (tstring_data): Use uint8_t for data
argument.
* testsuite/testutils.h (SDATA): Use US macro to cast data
argument.

* hkdf.c: Delete unneeded includes. Use Nettle licensing notice.
* hkdf.h: Include only nettle-types.h, not nettle-meta.h.

* hkdf.c (hkdf_expand): Eliminate a (signed) ssize_t variable, use
break rather than return at loop termination.

* ecc-add-eh.c (ecc_add_eh): Fix in-place operation by reordering
two multiplies. Previously, in-place operation resulted in an
invalid call to mpn_mul with overlapping operands. Reported by
Sergei Trofimovich.

* testsuite/pss-test.c: Delete magic to let valgrind to check if
pss_encode_mgf1 is side-channel silent with respect to the salt
and digest inputs. It turns out that the most significant bits of
the padded bignum, and hence its size, depends on these inputs.
Which results in a data-dependent branch in the normalization code
of at the end of gmp's mpz_import.