We ignore do not return errors from gnutls_hash() and similar in these functions:
 - sign_outgoing_message (MD5, source4 client lib)

 - netlogon_creds_server_init ignores the errors from netlogon_creds_init_128bit() (MD5).  Consequence would be a zero session key server-side.

 - netlogon_creds_server_init ignores errors from netlogon_creds_first_step() -> netlogon_creds_step_crypt().  (AES)

 - netlogon_creds_step_crypt ignores errors from netlogon_creds_aes_encrypt() (AES)

 - SMBsesskeygen_ntv2 ignores errors from gnutls_hmac_fast
 - SMBOWFencrypt_ntv2 ignores errors from gnutls_hmac()
 - both protected because if MD5 disallowed ntv2_owf_gen() would fail first

 - encode_or_decode_arc4_passwd_buffer() returns void so would leave an unencrypted pw buffer in 4.11
 - encode_wkssvc_join_password_buffer() returns void so would leave an unencrypted pw buffer in 4.11
 - E_md5hash() returns void so would leave hash_out unchanged (used for reading legacy password history in S3 code)
 - smb_key_derivation ignores errors from gnutls_hmac_fast() in 4.11

I think we very narrowly avoid a security issue here because:
 - If FIPS mode were enabled then the server would clearly fail to operate well pretty clearly
 - none of the above allows entry to the server via zeroed out buffers.

But I would like this double-checked in 4.11 and fixes made in master.