Impact

The code that performs decryption and padding check in RSA PKCS#1 v1.5 decryption is data dependant. In particular, code in current (as of 0.8.0-alpha38) master https://github.com/tlsfuzzer/tlslite-ng/blob/0812ed60860fa61a6573b2c0e18771414958f46d/tlslite/utils/rsakey.py#L407-L441 and code in 0.7.5 branch https://github.com/tlsfuzzer/tlslite-ng/blob/acdde3161124d6ae37c506b3476aea9996d12e97/tlslite/utils/rsakey.py#L394-L425 has multiple ways in which it leaks information (for one, it aborts as soon as the plaintext doesn't start with 0x00, 0x02) about the decrypted ciphertext (both the bit length of the decrypted message as well as where the first unexpected byte lays).

All TLS servers that enable RSA key exchange as well as applications that use the RSA decryption API directly are vulnerable.

All previous versions of tlslite-ng are vulnerable.

Patches

The patches to fix it are proposed in https://github.com/tlsfuzzer/tlslite-ng/pull/438 https://github.com/tlsfuzzer/tlslite-ng/pull/439

Note: the patches depend on Python processing the individual bytes in side-channel free manner, this is known to not be the case: https://securitypitfalls.wordpress.com/2018/08/03/constant-time-compare-in-python/ As such, users that require side-channel resistance are recommended to use different TLS implementations, as stated in the security policy of tlslite-ng.

Workarounds

There is no way to workaround this issue.

References

https://securitypitfalls.wordpress.com/2018/08/03/constant-time-compare-in-python/

For more information

If you have any questions or comments about this advisory please open an issue in tlslite-ng.