Predictability of Android OpenSSL's pseudo random number generator

OpenSSL is the most widely used library for SSL/TLS on the Android platform. The security of OpenSSL depends greatly on the unpredictability of its Pseudo Random Number Generator (PRNG). In this paper, we reveal the vulnerability of the OpenSSL PRNG on the Android. We first analyze the architecture of the OpenSSL specific to Android, and the overall operation process of the PRNG from initialization until the session key is generated. Owing to the nature of Android, the Dalvik Virtual Machine in Zygote initializes the states of OpenSSL PRNG early upon booting, and SSL applications copy the PRNG states of Zygote when they start. Therefore, the applications that use OpenSSL generate random data from the same initial states, which is potential problem that may seriously affect the security of Android applications. Next, we investigate the possibility of recovering the initial states of the OpenSSL PRNG. To do so, we should predict the nine external entropy sources of the PRNG. However, we show that these sources can be obtained in practice if the device is fixed. For example, the complexity of the attack was O(2^{32+t}) in our smartphone, where t is the bit complexity for estimating the system boot time. In our experiments, we were able to restore the PRNG states in 74 out of 100 cases. Assuming that we knew the boot time, i.e., t=0, the average time required to restore was 35 min on a PC with four cores (eight threads). Finally, we show that it is possible to recover the PreMasterSecret of the first SSL session with O(2^{58}) computations using the restored PRNG states, if the application is implemented by utilizing org.webkit package and a key exchange scheme is RSA. It shows that the vulnerability of OpenSSL PRNG can be a real threat to the security of Android.