soter: make sure we have the RSA private key before trying to decrypt…

[ignatk]

… data

Some OpenSSL versions do not handle gracefully, when you try to do RSA
decryption with a public key (no private part in the RSA structure) and just
crash the whole process. So we add an explicit check we have the private part
before attempting the decryption.

[ignatk]

Fixes #332

The crash.c program does not crash now, however, fails with 11 (which is just THEMIS_FAIL). So the proper SOTER_INVALID_PARAMETER error is not propagated properly.

[vixentael]

@secumod thank you for this fix!

Could you please update code because tests are crashing at this point.

src/soter/openssl/soter_asym_cipher.c:249:2: error: implicit declaration of function 'RSA_get0_key' [-Werror=implicit-function-declaration]
  RSA_get0_key(rsa, NULL, NULL, &d);
  ^
cc1: all warnings being treated as errors
Makefile:294: recipe for target 'build/obj/soter/openssl/soter_asym_cipher.o' failed```

[ignatk]

Yes, I saw that - older OpenSSL libs (< 1.1.0) do not have this accessor function.

[ilammy]

I've tested this change with OpenSSL 1.0.2, OpenSSL 1.1.0, and LibreSSL 2.8.2. The code from linked issue no longer crashes for me as well. Themis behavior is now identical across all supported crypto backends (BoringSSL already returns THEMIS_FAIL in this case).

LGTM 👍

I guess this resolves the issue. Thank you, @secumod!

[ignatk]

Thanks for testing. I guess we should decide separately, if we should fix the return codes. Any opinions @vixentael ?

[vixentael]

LGTM

Thank you @secumod and @ilammy for such quick testing and fixing

[vixentael]

This fix will be part of 0.11.0 which will be released eventually (I think it makes sense to fix all issues found by @ilammy before release).

[vixentael]

I guess we should decide separately, if we should fix the return codes.

@secumod agree, I think matching Soter and Themis error codes requires more discussion and is definitely a different issue. Shall we open a new one?