Re: Queued Mail or Unreturnable Mail?
2008-05-02 15:49:39
On Apr 30, 2008, at 7:02 PM, Frank Ellermann wrote:
2821 and 2821bis have it clear that the mere presence of a listening
process does not say anything about the intent to receive mail for
the corresponding domains. The MX can say something else, without
MX a listening process at A or AAAA can reject clients or individual
mails as it sees fit.
The problematic situation is the absence of MX and listening
process, senders are forced to wait 100 hours, somebody might find
the reboot button of the affected host. Another problematic
situation is mail claiming to be "from" this domain without MX,
receivers cannot wait until somebody finds the reboot button for a
quick plausibility test, they are forced to reject such suspicious
mails temporarily.
Since DARPA's development of a "network of networks" became known as
the Internet, and intercommunication processes known as Internet
protocols (IP), a means to relate hostnames and services with IP
addresses was needed. Jon Postel publish a concept of Sockets
(combing network address and port) in use with assignments for
standard programs, laying a foundation on efforts made by Steve Carr
from Utah, Jeff Rulifson from SRI, and Steve Crocker of UCLA, who
introduced documents describing host networking almost four decades ago.
The next decade introduced RFC761 TCP, RFC768 UDP, and RFC790 Assigned
numbers. Within a few years, RFC1034 DNS supplanted host files. With
DNS, in addition to port assignments, service support could be
declared using specialized DNS resource records. RFC974 recommended
WKS (well known service) resource records to confirm support of SMTP.
RFC1123 renounced WKS, as few domains published these records. The
resource record concept did not mesh with host files, or NetBIOS
adaptation of WINS's offering referrals from specific service names.
To help transitioning from WINS to DNS, SRV records were created.
Even with SRV records, Window's adoption of DNS remained limited.
Within a Window's enterprise environment, DNS might be converted into
RPC. Avoiding proprietary development for this environment, protocols
often employ only the limited list of resource records having data
structures defined for Windows OS APIs. While API provisions now
accommodate undefined resources records, a lack of OS data structure
definitions remains an impediment.
Resolving services and hostname addresses, generic resource records
holding signatures and policies continues to impact SMTP. Although
SMTP adopted MX resource records within DNS, publishing MX records was
unfortunately never mandated. While ensuring the viability of local
hostname resolution remains important, undesired traffic associated
with spoofed email addresses represents an ever increasing problem.
However, there does not appear to be an easy method to differentiate
name services when resolving an address. Facing massive amounts of
undesired traffic, SMTP reception typically employes extensive
examination of SMTP client IP addresses and return-path email-addresses.
Therefore, it should be reasonable to publish an RFC that warns of
possible future changes where message acceptance may require return-
path email-address domains to publish MX records. Mandating MX
records be published by domains supporting SMTP better protects
domains not supporting SMTP, and better ensures integration of IPv6
addresses. The MX mandate also replaces unofficial recommendations
that domains not supporting SMTP publish bogus MX records targeting
roots. This mandate would help establish conventions for receiving
SMTP servers to predicate additional transactions upon first finding
MX records for associated email-address domains. The receiving SMTP
server is also best equipped at making necessary exceptions where
needed.
In essence, SMTP discovery resource record mandate essentially impacts
message reception. Justification for this mandate could be to support
the obligation accepted by receiving SMTP servers of ensuring delivery
or offering notification upon a delivery failure.
-Doug
|
|