In particular, if interruptible ECDSA is supported but not the deterministic
variant, detect this in psa_sign_hash_start(), whereas before start() would
succeed and psa_sign_hash_complete() would fail. This avoids an
inconsistency between psa_sign_hash() and psa_sign_hash_start() that would
be annoying to handle in test_suite_psa_crypto_op_fail.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
ECDSA has two variants: deterministic (PSA_ALG_DETERMINISTIC_ECDSA) and
randomized (PSA_ALG_ECDSA). The two variants are different for signature but
identical for verification. Mbed TLS accepts either variant as the algorithm
parameter for verification even when only the other variant is supported,
so we need to handle this as a special case when generating not-supported
test cases.
In this commit:
* Add manually written not-supported test cases for the signature
operation when exactly one variant is supported.
* Add manually written positive test cases for the verification
operation when exactly one variant is supported.
* Register that !ECDSA but DETERMINISTIC_ECDSA is not tested yet
(https://github.com/Mbed-TLS/mbedtls/issues/9592).
A commit in the framework will take care of automatically generated test cases.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Following "PSA sign/verify: more uniform error on an unsupported hash", some
error cases are detected earlier, so there is some sloppiness in test case
dependencies that is not longer acceptable.
* In test_suite_psa_crypto, one test case for a hash+sign algorithm now
returns NOT_SUPPORTED rather than INVALID_ARGUMENT when the hash is not
supported and the key is invalid.
* In test_suite_psa_crypto_se_driver_hal_mocks, some test cases now error
out before reaching the mocks rather than after when they attempt to
use an unsupported hash.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Allow imports of an ECC public key on an unsupported curve to return
INVALID_ARGUMENT rather than NOT_SUPPORTED. This can happen in our library
code in edge cases when only certain curve families are supported, and it's
acceptable.
The new code does not trigger yet, but it will be useful for a future commit
"Do run not-supported test cases on not-implemented mechanisms"
(forward port of 995d7d4c15).
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
This fixes accesses to uninitialized memory in test code if
`psa_crypto_init()` fails.
A lot of those were pointed out by Coverity. I quickly reviewed all calls to
`MD_PSA_INIT()` manually, rather than follow any particular list.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
The existing check only took care of CONFIG_H. This was both not enough
and too much:
- not enough because config.py can also modify CRYPTO_CONFIG_H and we
want to know about it just as much as CONFIG_H;
- too much because CONFIG_H does not exist in tf-psa-crypto.
Check a list of files instead of a single one, and adjust that list.
Also update an outdated comment about Makefiles
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
In preparation for adding tf-psa-crypto/test/scripts/all.sh which will
run from tf-psa-crypto.
Use paths relative to the currently sourced file when including common
files (ie, those that will soon be moved to the framework). Otherwise,
use paths relative to the current directory, aka project's root.
Document that test/script/all.sh must be invoked from the project's root
(that was already the case, but implicit so far).
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Skip 6144-bit and 8192-bit FFDH tests in MSan or Valgrind test
configurations. These test cases are slow and sometimes trigger GnuTLS's
built-in timeout (which cannot be configured until more recent versions).
Resolves#9742.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
generate_test_code.py doesn't support UTF-8 in .function files (but does
in .data files) when run in a non-UTF-8 locale with Python <=3.6.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Run the function on a few different curves with shorter, just-right and
larger buffer sizes. Perform some basic sanity checks on the output (which
is random).
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
The Thumb-1 tests are by far the slowest, so split the -O0 and -Os tests
into their own components.
Signed-off-by: Bence Szépkúti <bence.szepkuti@arm.com>
Cleanly reject bits == 0 when calling mbedtls_ecdsa_raw_to_der() and
mbedtls_ecdsa_der_to_raw(). This can plausibly happen when bits is
user-provided data that the calling application doesn't check.
Before this patch, there was typically-benign undefined behavior, such as
adding 0 to a null pointer or calling memcpy on a null pointer with a size
of 0.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Use TEST_CALLOC instead of a fixed-size buffer, so that Asan/Valgrind builds
will detect a buffer overflow.
Honor output_size regardless of the value of the number.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Since PSA is always on in 4.x, *_PSA_INIT() and *_PSA_DONE() can be
simply aliased to PSA_INIT() and PSA_DONE() until such a time as we
remove them.
Simplify the login of these PSA_INIT/DONE variants by aliasing them in
4.x and keeping the more complex 3.6 logic entirely separate.
Signed-off-by: David Horstmann <david.horstmann@arm.com>
Variables that are in any way destructed on exit should be initialised
prior to any tests that might jump to exit, to save potential
uninitialised memory accesses.
Signed-off-by: Paul Elliott <paul.elliott@arm.com>
