Invalid signatures produced using gnupg-pkcs11-scd

[Moritz Bechler]

After upgrading from wily to xenial (and at the same time having to move from gpg to gpg2) I can no longer produce signatures using gnupg-pkcs11-scd.

Debugging this I found that the algorithm prefix is now included twice in the signed data, making the signature self-test fail.

Here we have the data to sign, including the algorithm prefix (3031300D0609608648016503004020):

2016-05-03 16:33:56 gpg-agent[18007] DBG: chan_6 -> SETDATA 3031300D0609608648016503040201050004207B1F9A47922DEDFA9E7A430B4191A1ED2474BE21A
48B8BCA9FE278DD586882C2
2016-05-03 16:33:56 gpg-agent[18007] DBG: chan_6 <- OK

Calling PKSIGN with the hash argument will cause gnupg-pkcs11-scd to add another copy of the algorithm prefix:
2016-05-03 16:33:56 gpg-agent[18007] DBG: chan_6 -> PKSIGN --hash=sha256 SafeNet\x20Inc\x2E/eToken/0020f8ec/mb/01

The signed data, showing the duplicated algorithm prefix under rsa_verify cmp is attached.

Not sure how a backward-compatible fix would look like (probably would have to check whether this prefix is already present), but forcing inject = INJECT_NONE in cmd_pksign seems to fix the issue for me.

Moritz