EAP-pwd: fix side-channel leak where 1 in 2018 handshakes fail

Previously the Hunting and Pecking algorithm of EAP-pwd aborted when
more than 10 iterations are needed. Every iteration has a 50% chance
of finding the password element. This means one in every 2048 handshakes
will fail, in which case an error frame is sent to the client. This
event leaks information that can be abused in an offline password
brute-force attack. More precisely, the adversary learns that all 10
iterations failed for the given random EAP-pwd token. Using the same
techniques as in the Dragonblood attack, this can be used to brute-force
the password.

This patch fixes the above issue by executing enough iterations such that
the password element is always found eventually.

Note that timing and cache leaks remain a risk against the current
implementation of EAP-pwd.

src/modules/rlm_eap/types/rlm_eap_pwd/eap_pwd.c

@@ -192,7 +192,7 @@ int compute_password_element (pwd_session_t *session, uint16_t grp_num,
 	}
 	ctr = 0;
 	while (1) {
-		if (ctr > 10) {
+		if (ctr > 100) {
 			DEBUG("unable to find random point on curve for group %d, something's fishy", grp_num);
 			goto fail;
 		}