(In reply to Miro Hrončok from comment #0)

> 
> The build stared failing after this:
> 
> https://src.fedoraproject.org/rpms/openssl/c/
> 1aaf4073e34a620385e878140b048aadd09c6a85?branch=master
> 
> Not sure if that is the direct cause.

That does not seem to me to be like so. From the Koschei it looks like it started after the update to openssl-1.1.1b-6.fc31, the next build. That one brought various changes from the upstream 1.1.1 branch, unfortunately it will be quite hard to isolate which particular change was exactly the cause. If there was some isolated testcase or what exact error is left on the error stack it would definitely help with debugging the issue.

My fault (was looking at old version instead of new).

So that is:

https://src.fedoraproject.org/rpms/openssl/c/5c7382cd79a3ba0832c4c77f875c0feb4ea1b13d?branch=master
https://src.fedoraproject.org/rpms/openssl/c/f5bba4a2a3ffffae130376212ff8ccf48ee13974?branch=master

It would be super useful to see what the actual openssl error is.

I think that code is actually trying to report that back, but all it gets is an empty list from _consume_errors(lib).

Working on upstream reproducer over a rawhide docker image, so we can get their help.

https://github.com/pyca/cryptography/issues/4884

Upstream cryptography guesses crypto/evp/evp_enc.c changes in OpenSSL might be good place to look into.

The Fedora commit says "apply new bugfixes from upstream 1.1.1 branch" - Tomáš, do you please at least have a list of bugs and their fixes? The patch is gigantic.

Note: This is one of the 2 major blockers for the Python 3.8 rebuild in a rawhide side tag that is planned after the 3.8.0 beta1 release (scheduled upstream for 2019-05-31).

This gigantic patch is basically all the relevant fixes from the current 1.1.1 upstream branch. I.E. not including things fixing Windows or other irrelevant target fixes.

I am just now building a new build that should hopefully fix the AES-CCM related regression.

The build is openssl-1.1.1b-9.fc31

Thanks.

https://koji.fedoraproject.org/koji/buildinfo?buildID=1266116

openssl-libs-1.1.1b-9.fc31 does not fix the issue.

Interesting comment: https://github.com/pyca/cryptography/issues/4884#issuecomment-491279382

The openssl-1.1.1b-10.fc31 fixes the C test sample issue for me. Not sure if it fully fixes the python-cryptography build though

Most of them. One remains:

=================================== FAILURES ===================================
_________________ test_buffer_protocol_alternate_modes[mode5] __________________

mode = <cryptography.hazmat.primitives.ciphers.modes.XTS object at 0x7f6d75211690>
backend = <cryptography.hazmat.backends.openssl.backend.Backend object at 0x7f6d7efeb650>

    @pytest.mark.parametrize(
        "mode",
        [
            modes.CBC(bytearray(b"\x00" * 16)),
            modes.CTR(bytearray(b"\x00" * 16)),
            modes.OFB(bytearray(b"\x00" * 16)),
            modes.CFB(bytearray(b"\x00" * 16)),
            modes.CFB8(bytearray(b"\x00" * 16)),
            modes.XTS(bytearray(b"\x00" * 16)),
        ]
    )
    @pytest.mark.requires_backend_interface(interface=CipherBackend)
    def test_buffer_protocol_alternate_modes(mode, backend):
        data = bytearray(b"sixteen_byte_msg")
        cipher = base.Cipher(
            algorithms.AES(bytearray(b"\x00" * 32)), mode, backend
        )
>       enc = cipher.encryptor()

tests/hazmat/primitives/test_aes.py:495: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
src/cryptography/hazmat/primitives/ciphers/base.py:121: in encryptor
    self.algorithm, self.mode
src/cryptography/hazmat/backends/openssl/backend.py:295: in create_symmetric_encryption_ctx
    return _CipherContext(self, cipher, mode, _CipherContext._ENCRYPT)
src/cryptography/hazmat/backends/openssl/ciphers.py:116: in __init__
    self._backend.openssl_assert(res != 0)
src/cryptography/hazmat/backends/openssl/backend.py:125: in openssl_assert
    return binding._openssl_assert(self._lib, ok)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

lib = <module 'lib' (built-in)>, ok = False

    def _openssl_assert(lib, ok):
        if not ok:
            errors = _consume_errors(lib)
            errors_with_text = []
            for err in errors:
                buf = ffi.new("char[]", 256)
                lib.ERR_error_string_n(err.code, buf, len(buf))
                err_text_reason = ffi.string(buf)
    
                errors_with_text.append(
                    _OpenSSLErrorWithText(
                        err.code, err.lib, err.func, err.reason, err_text_reason
                    )
                )
    
            raise InternalError(
                "Unknown OpenSSL error. This error is commonly encountered when "
                "another library is not cleaning up the OpenSSL error stack. If "
                "you are using cryptography with another library that uses "
                "OpenSSL try disabling it before reporting a bug. Otherwise "
                "please file an issue at https://github.com/pyca/cryptography/"
                "issues with information on how to reproduce "
                "this. ({0!r})".format(errors_with_text),
>               errors_with_text
            )
E           InternalError: Unknown OpenSSL error. This error is commonly encountered when another library is not cleaning up the OpenSSL error stack. If you are using cryptography with another library that uses OpenSSL try disabling it before reporting a bug. Otherwise please file an issue at https://github.com/pyca/cryptography/issues with information on how to reproduce this. ([_OpenSSLErrorWithText(code=101617856L, lib=6, func=233, reason=192, reason_text='error:060E90C0:digital envelope routines:aesni_xts_init_key:xts duplicated keys')])

src/cryptography/hazmat/bindings/openssl/binding.py:78: InternalError
=========== 1 failed, 100943 passed, 6032 skipped in 323.44 seconds ============


But this one actually has the error.

Yes, this one is actually real error in the py-cryptography test. It should not try AES-XTS with both halves of the key being the same.

https://github.com/pyca/cryptography/issues/4885

https://src.fedoraproject.org/rpms/python-cryptography/pull-request/7