wrong RD/RA flags cause exim4 to fail if no MX Record for a domain is set

[Dr. Volker Jaenisch]

Hi Maradns-list!

We use maradns on large exim mail-server-clusters and we like it.

The stabe version on Debian 1.2.12.04 as well as the
sourcecode for 1.3.11 shows the same behavior:
MaraDNS does not return the RD and sets does not set RA Flag in case
of a RECORD NOT FOUND case. (See attached maradns and bind output of
such a case, below)

In general the reflection of the RD/RA flags have been fixed with the
last patches in the stable and developement branch.
For this seldom case the existing bugfix should be extended by the
following patch (against 1.3.11).

--- MaraDNS.c~  2008-03-24 16:52:28.000000000 +0100
+++ MaraDNS.c   2008-08-01 15:49:05.000000000 +0200
@@ -1892,8 +1892,8 @@
     header.qr = 1;
     header.opcode = 0;
     header.tc = 0; /* To do: truncation handling */
-    header.rd = 0; /* RDBUG udpnotfound */
-    header.ra = 0;
+    header.rd = rd_val; /* RDBUG udpnotfound */
+    header.ra = rd_val;
     header.z = 0;
     /* Code that verifies that this host does not exist in
        any class.  If so, then we give them a rcode of NXDOMAIN_RCODE.

This patch cures the exim problem. But I am pretty sure that it is not
complete. Also there may be a conflict with
the RDBUG udpnotfound bugfix.

Background:
If a domain has no MX entry exim decides to use the IP of the A entry.
But exim have to be resonable sure that no other DNS error has ocurred.
Since maradns does not return the RD and RA flags properly exim supposes
an dns malfunction:

I did send a recursive MX question and get a nonrecursive answer > the
DNS may be broken

and tries again and again to get a recursive answer. The bitter
consequence is that emails to domains without MX records
will completely fail on exim systems.

It is no RFC violation not to set an MX record for a maildomain.
Let's take a look at what RFC 2821 says:
- Once the domain is identified SMTP must perform a DNS lookup and first
attempt to locate an MX record.
- If no MX records are found but an A record is found, it is treated as
if it was associated with an implicit MX record with a preference of 0
pointing to that host - also known as the "implicit MX" rule.

With maradns the following happens:

skylla:~/maradns-1.3.11# dig @10.1.0.8 -tmx +recurse mail.tropos.de
; <<>> DiG 9.3.4-P1.1 <<>> @10.1.0.8 -tmx +recurse mail.tropos.de
; (1 server found)
;; global options:  printcmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 24702
;; flags: qr; QUERY: 1, ANSWER: 0, AUTHORITY: 1, ADDITIONAL: 0

;; QUESTION SECTION:
;mail.tropos.de.                        IN      MX

;; AUTHORITY SECTION:
mail.tropos.de.         85665   IN      SOA     sana.tropos.de.
jagemann.tropos.de. 200807312 86400 3600 3600000 86400

Compared to a bind DNS-server:

skylla:~/maradns-1.3.11# dig @193.239.30.4 -tmx +recurse mail.tropos.de

; <<>> DiG 9.3.4-P1.1 <<>> @193.239.30.4 -tmx +recurse mail.tropos.de
; (1 server found)
;; global options:  printcmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 61493
;; flags: qr rd ra; QUERY: 1, ANSWER: 0, AUTHORITY: 1, ADDITIONAL: 0

;; QUESTION SECTION:
;mail.tropos.de.                        IN      MX

;; AUTHORITY SECTION:
tropos.de.              10800   IN      SOA     sana.tropos.de.
jagemann.tropos.de. 200807312 86400 3600 3600000 86400

Best regards

Volker

-- 
====================================================
   inqbus it-consulting      +49 ( 341 )  5643800
   Dr.  Volker Jaenisch      http://www.inqbus.de
   Herloßsohnstr.    12      0 4 1 5 5    Leipzig
   N  O  T -  F Ä L L E      +49 ( 170 )  3113748
====================================================