Mail delivery failure due to GnuTLS X.509 validation

I quite commonly see undelivered mail to [email protected] on the outgoing mail queue. (DSNs and out-of-office replies with empty senders - so not critical mail.)

(The Twitter mail is recieved by en25.com/eloqua.net which is infrastructure for "Oracle Eloqua Marketing Cloud".)

This isn’t a new issue, but I thought I’d drop a note in public about it.

If I run exim4 -v -M on a delivery attempt I can see that it disconnects immediately after attempting to establish a STARTTLS session, so falls back to retrying without TLS.

  SMTP<< 220 P01SNJ018.eloqua.net Microsoft ESMTP MAIL Service, Version: 10.0.14393.4169 ready at  Tue, 10 Jan 2023 05:50:33 -0500
  SMTP>> EHLO foo.example.com
  SMTP<< 250-P01SNJ018.eloqua.net Hello [10.32.120.102]
         250-TURN
         250-SIZE 2097152
         250-ETRN
         250-PIPELINING
         250-DSN
         250-ENHANCEDSTATUSCODES
         250-8bitmime
         250-BINARYMIME
         250-CHUNKING
         250-VRFY
         250-TLS
         250-STARTTLS
         250 OK
  SMTP>> STARTTLS
  SMTP<< 220 2.0.0 SMTP server ready
  SMTP(close)>>
  TLS session: (gnutls_handshake): Key usage violation in certificate has been detected.: delivering unencrypted to H=mail.en25.com [209.167.231.14] (not in hosts_require_tls)

But, the server won't accept a bounce on an unencrypted connection.

  SMTP<< 220 P01SNJ015.eloqua.net Microsoft ESMTP MAIL Service, Version: 10.0.14393.4169 ready at  Tue, 10 Jan 2023 05:50:34 -0500
  SMTP>> EHLO foo.example.com
  SMTP<< 250-P01SNJ015.eloqua.net Hello [10.32.120.104]
         250-TURN
         250-SIZE 2097152
         250-ETRN
         250-PIPELINING
         250-DSN
         250-ENHANCEDSTATUSCODES
         250-8bitmime
         250-BINARYMIME
         250-CHUNKING
         250-VRFY
         250-TLS
         250-STARTTLS
         250 OK
  SMTP>> MAIL FROM:<> SIZE=2451
  SMTP>> RCPT TO:<[email protected]>
  SMTP>> BDAT 2092 LAST
  SMTP<< 530 5.7.0 Must issue a STARTTLS command first
  SMTP<< 530 5.7.0 Must issue a STARTTLS command first
  SMTP>> QUIT
  SMTP(close)>>

The mailserver is the Ubuntu 20.04 exim4-daemon-heavy (4.93-13ubuntu1.7) using libgnutls (3.6.13-2ubuntu1.7)

I can see using TLS debug tools that the certificate isn't signed. But certificate errors aren't uncommon for mail servers, and it doesn't usually impede delivery.

echo QUIT | gnutls-cli --starttls-proto=smtp -p 25 --no-ca-verification mail.en25.com

puts the following in stderr

|<1>| Peer's certificate does not allow digital signatures. Key usage violation detected.
*** Fatal error: Key usage violation in certificate has been detected.

If I pull out the X.509 Key Usage section

echo QUIT | gnutls-cli --starttls-proto=smtp -p 25 --no-ca-verification --insecure --save-cert=/dev/stdout mail.en25.com | certtool -i

   Key Purpose (not critical):
      TLS WWW Server.
   Key Usage (critical):
      Key encipherment.
      Data encipherment.

Checking against servers I know will accept STARTTLS, such as smtp.google.com

   Key Usage (critical):
        Digital signature.
   Key Purpose (not critical):
       TLS WWW Server.

and a server with a standard letsencrypt cert

   Key Usage (critical):
      Digital signature.
      Key encipherment.
   Key Purpose (not critical):
      TLS WWW Server.
      TLS WWW Client.

I can see that they have "Digital signature" in the X.509 key usage section.

Given other servers, eg ones using OpenSSL, will establish a connection regardless, is the hard requirement for "Digital signature" in the key usage the correct behaviour?

And, given plaintext fallback, is there any way to switch this off without recompiling?

gwire/mail_gnuttls_x509_fail.md