jim@jagmac2.gsfc.nasa.gov (Jim Jagielski) (07/19/90)
Please excuse the length of this posting, but this is a response to an e-mail message that I got and could not reply to. This may also be of interest to the net:... >From claris!wet!epsilon@ames.arc.nasa.gov Thu Jul 19 08:21 EDT 1990 >Received: From [128.102.18.3] By jagmac2.gsfc.nasa.gov Using SMTP > (5.64/jj-1.6) id AA01096; Thu, 19 Jul 90 08:21:05 -0400 >From: claris!wet!epsilon@ames.arc.nasa.gov >Received: by ames.arc.nasa.gov (5.64/1.2); Thu, 19 Jul 90 05:20:56 -0700 >Received: by claris.com (4.1/SMI-3.2) > id AA10137; Thu, 19 Jul 90 05:14:44 PDT >Date: Thu, 19 Jul 90 05:14:41 PDT >Message-Id: <9007191214.AA10137@claris.com> >Apparently-To: jim@jagmac2.gsfc.nasa.gov >Status: RO > >In article <2826@dftsrv.gsfc.nasa.gov> you write: >>Recompiling the source with NAMED_BIND undefined makes everything work! >>Evidently, with this symbol defined, sendmail attempts to interface with >>the nameserver directly, using MX records and resolver routines. Obviously, >>if you don't have named running you don't need this. > >You MUST use MX records if you're Internet-connected. You don't >need named running if /etc/resolv.conf points somewhere reasonable. >In fact, I seriously doubt I'd ever run named on an A/UX machine. > >> This trick is that >>even WITH named running, you also don't need this, since the gethostent() >>routines (such as gethostbyname, etc...) will do the nameserver resolving >>for you! > >WRONG. They query A records, not MX records. The only sites >that can use this version are those running TCP/IP on a >non-Internet-connected network on which every host handles its >own mail. (e.g. this does not work if there are diskless >workstations, or firewall gateways, etc.) Most isolated IP >networks run with name service unless they're very small. > > -=EPS=- > My response is that I AM Internet connected, and compiling with NAMED_BIND undefined results in Sendmail working but compiling with NAMED_BIND defined results in Sendmail ONLY working if the host address is COMPLETELY spelled out (i.e. bar.gsfc.nasa.gov rather than just bar). Otherwise, mail is bounced with the "unknown host" error. It also is no help to comment out the RES_DEFNAMES and RES_DNSRCH option setting... although it would result in the domain being attached to all single token hosts, sendmail would STILL refuse to recognize aliases in /etc/hosts either with or without named running. As far as the comment that the only sites who could use this version are those "running TCP/IP on a non-Internet-connected network", well, all I can do is repeat that I AM Internet connected. I do, however, handle by own mail and do require a gateway to get out of gsfc.nasa.gov. named is running... It does not good to say that "You MUST use MX records" if when sendmail is compiled to do so, it refuses to do what nameserving is supposed to do and when NOT using MX records (i.e. no getmxrr()) then it DOES work. If you have any idea WHY it works out this way, let me know, but just sprouting dogma is no help. Comments are requested to help, not to inflate one's own ego on how much more you know about some- thing than someone else... have you looked at the code? Have you tried compiling and running it both ways? Do you have an CONSTRUCTIVE comments? In anycase, the focal point seems to be in the use of MX records. As far as I understand them (and I DON'T claim to be an expert) is that they are used to specify a machine that knows how to deliver mail to another machine that is not directly connected to the net... however, my experience has been that when using MX records, addresses are not resolved correctly unless totally spelled out. The version of sendmail that came with A/UX 1.1 (5.59) wasn't like that... Until then, I'll keep on distributing the port as is. People can compiled both with or without NAMED_BIND defined, but I would appreciate feedback on the difference encountered both ways. -- ======================================================================= #include <std/disclaimer.h> =:^) Jim Jagielski NASA/GSFC, Code 711.1 jim@jagmac2.gsfc.nasa.gov Greenbelt, MD 20771 "Kilimanjaro is a pretty tricky climb. Most of it's up, until you reach the very, very top, and then it tends to slope away rather sharply."
epsilon@wet.UUCP (Eric P. Scott) (07/25/90)
In article <2844@dftsrv.gsfc.nasa.gov> jim@jagmac2.gsfc.nasa.gov (Jim Jagielski) writes: >My response is that I AM Internet connected, and compiling with >NAMED_BIND undefined results in Sendmail working but compiling >with NAMED_BIND defined results in Sendmail ONLY working if the >host address is COMPLETELY spelled out (i.e. bar.gsfc.nasa.gov >rather than just bar). Otherwise, mail is bounced with the >"unknown host" error. What's in your /etc/resolv.conf file? Is there a "domain" line? Which resolver library are you using? > It also is no help to comment out the >RES_DEFNAMES and RES_DNSRCH option setting... although it >would result in the domain being attached to all single token >hosts, sendmail would STILL refuse to recognize aliases in >/etc/hosts either with or without named running. You're missing an important point: all addresses in Internet mail must go out with canonical names--no aliases! At the very least this means a DNS lookup to determine the proper name. The place for aliases isn't /etc/hosts--it's sendmail.cf or the mail user agent. >As far as the comment that the only sites who could use this version >are those "running TCP/IP on a non-Internet-connected network", well, >all I can do is repeat that I AM Internet connected. I do, however, >handle by own mail and do require a gateway to get out of gsfc.nasa.gov. >named is running... If your problem is a firewall gateway, run [one of the rewer versions of] BIND on the "red" side with the "forwarders" and "slave" options in named.boot. Then your hosts will have DNS access even without direct IP connectivity to root nameservers. >It does not good to say that "You MUST use MX records" if when >sendmail is compiled to do so, it refuses to do what nameserving >is supposed to do and when NOT using MX records (i.e. no getmxrr()) >then it DOES work. If you have any idea WHY it works out this way, >let me know, but just sprouting dogma is no help. Comments are requested >to help, not to inflate one's own ego on how much more you know about some- >thing than someone else... have you looked at the code? Have you tried >compiling and running it both ways? Do you have an CONSTRUCTIVE comments? Have you read RFC 1123? It specifies a STANDARD all Internet sites must adhere to. >Until then, I'll keep on distributing the port as is. People >can compiled both with or without NAMED_BIND defined, but I >would appreciate feedback on the difference encountered both ways. sendmail is a critical piece of software for many sites; many of the people who will run it know even less about these issues than you do. If you are willfully distributing versions that violate standards (except as "experimental software") you are probably doing the net a grave disservice, as it is those standards that keep the net functional despite its enormous size. hosts files work on isolated LANs with at most a few dozen similar machines under the same administrative control. For most of us, it is an inadequate (and often counterproductive) solution. -=EPS=-