4.2BSD/usr/doc/sendmail/usenix.me

.nr si 3n
.he 'Mail Systems and Addressing in 4.2BSD''%'
.fo 'Version 1.3'USENIX \- Jan 83'Last Mod 3/12/83'
.if n .ls 2
.+c
.(l C
.sz 14
Mail Systems and Addressing
in 4.2BSD
.sz
.sp
Eric Allman\(dg
.sp 0.5
.i
Britton-Lee, Inc.
1919 Addison Street, Suite 105.
Berkeley, California 94704.
.sp 0.5
.r
eric@Berkeley.ARPA
ucbvax!eric
.)l
.sp
.(l F
.ce
ABSTRACT
.sp \n(psu
Routing mail through a heterogeneous internet presents many new
problems.
Among the worst of these is that of address mapping.
Historically, this has been handled on an ad hoc basis.
However,
this approach has become unmanageable as internets grow.
.sp \n(psu
Sendmail acts a unified
.q "post office"
to which all mail can be
submitted.
Address interpretation is controlled by a production
system,
which can parse both old and new format addresses.
The
new format is
.q "domain-based,"
a flexible technique that can
handle many common situations.
Sendmail is not intended to perform
user interface functions.
.sp \n(psu
Sendmail will replace delivermail in the Berkeley 4.2 distribution.
Several major hosts are now or will soon be running sendmail.
This change will affect any users that route mail through a sendmail
gateway.
The changes that will be user visible are emphasized.
.)l
.sp 2
.(f
\(dgA considerable part of this work
was done while under the employ
of the INGRES Project
at the University of California at Berkeley.
.)f
.pp
The mail system to appear in 4.2BSD
will contain a number of changes.
Most of these changes are based on the replacement of
.i delivermail
with a new module called
.i sendmail.
.i Sendmail
implements a general internetwork mail routing facility,
featuring aliasing and forwarding,
automatic routing to network gateways,
and flexible configuration.
Of key interest to the mail system user
will be the changes in the network addressing structure.
.pp
In a simple network,
each node has an address,
and resources can be identified
with a host-resource pair;
in particular,
the mail system can refer to users
using a host-username pair.
Host names and numbers have to be administered by a central authority,
but usernames can be assigned locally to each host.
.pp
In an internet,
multiple networks with different characteristics
and managements
must communicate.
In particular,
the syntax and semantics of resource identification change.
Certain special cases can be handled trivially
by
.i "ad hoc"
techniques,
such as
providing network names that appear local to hosts
on other networks,
as with the Ethernet at Xerox PARC.
However, the general case is extremely complex.
For example,
some networks require that the route the message takes
be explicitly specified by the sender,
simplifying the database update problem
since only adjacent hosts must be entered
into the system tables,
while others use logical addressing,
where the sender specifies the location of the recipient
but not how to get there.
Some networks use a left-associative syntax
and others use a right-associative syntax,
causing ambiguity in mixed addresses.
.pp
Internet standards seek to eliminate these problems.
Initially, these proposed expanding the address pairs
to address triples,
consisting of
{network, host, username}
triples.
Network numbers must be universally agreed upon,
and hosts can be assigned locally
on each network.
The user-level presentation was changed
to address domains,
comprised of a local resource identification
and a hierarchical domain specification
with a common static root.
The domain technique
separates the issue of physical versus logical addressing.
For example,
an address of the form
.q "eric@a.cc.berkeley.arpa"
describes the logical
organization of the address space
(user
.q eric
on host
.q a
in the Computer Center
at Berkeley)
but not the physical networks used
(for example, this could go over different networks
depending on whether
.q a
were on an ethernet
or a store-and-forward network).
.pp
.i Sendmail
is intended to help bridge the gap
between the totally
.i "ad hoc"
world
of networks that know nothing of each other
and the clean, tightly-coupled world
of unique network numbers.
It can accept old arbitrary address syntaxes,
resolving ambiguities using heuristics
specified by the system administrator,
as well as domain-based addressing.
It helps guide the conversion of message formats
between disparate networks.
In short,
.i sendmail
is designed to assist a graceful transition
to consistent internetwork addressing schemes.
.sp
.pp
Section 1 defines some of the terms
frequently left fuzzy
when working in mail systems.
Section 2 discusses the design goals for
.i sendmail .
In section 3,
the new address formats
and basic features of
.i sendmail
are described.
Section 4 discusses some of the special problems
of the UUCP network.
The differences between
.i sendmail
and
.i delivermail
are presented in section 5.
.sp
.(l F
.b DISCLAIMER:
A number of examples
in this paper
use names of actual people
and organizations.
This is not intended
to imply a commitment
or even an intellectual agreement
on the part of these people or organizations.
In particular,
Bell Telephone Laboratories (BTL),
Digital Equipment Corporation (DEC),
Lawrence Berkeley Laboratories (LBL),
Britton-Lee Incorporated (BLI),
and the University of California at Berkeley
are not committed to any of these proposals at this time.
Much of this paper
represents no more than
the personal opinions of the author.
.)l
.sh 1 "DEFINITIONS"
.pp
There are four basic concepts
that must be clearly distinguished
when dealing with mail systems:
the user (or the user's agent),
the user's identification,
the user's address,
and the route.
These are distinguished primarily by their position independence.
.sh 2 "User and Identification"
.pp
The user is the being
(a person or program)
that is creating or receiving a message.
An
.i agent
is an entity operating on behalf of the user \*-
such as a secretary who handles my mail.
or a program that automatically returns a
message such as
.q "I am at the UNICOM conference."
.pp
The identification is the tag
that goes along with the particular user.
This tag is completely independent of location.
For example,
my identification is the string
.q "Eric Allman,"
and this identification does not change
whether I am located at U.C. Berkeley,
at Britton-Lee,
or at a scientific institute in Austria.
.pp
Since the identification is frequently ambiguous
(e.g., there are two
.q "Robert Henry" s
at Berkeley)
it is common to add other disambiguating information
that is not strictly part of the identification
(e.g.,
Robert
.q "Code Generator"
Henry
versus
Robert
.q "System Administrator"
Henry).
.sh 2 "Address"
.pp
The address specifies a location.
As I move around,
my address changes.
For example,
my address might change from
.q eric@Berkeley.ARPA
to
.q eric@bli.UUCP
or
.q allman@IIASA.Austria
depending on my current affiliation.
.pp
However,
an address is independent of the location of anyone else.
That is,
my address remains the same to everyone who might be sending me mail.
For example,
a person at MIT and a person at USC
could both send to
.q eric@Berkeley.ARPA
and have it arrive to the same mailbox.
.pp
Ideally a
.q "white pages"
service would be provided to map user identifications
into addresses
(for example, see
[Solomon81]).
Currently this is handled by passing around
scraps of paper
or by calling people on the telephone
to find out their address.
.sh 2 "Route"
.pp
While an address specifies
.i where
to find a mailbox,
a route specifies
.i how
to find the mailbox.
Specifically,
it specifies a path
from sender to receiver.
As such, the route is potentially different
for every pair of people in the electronic universe.
.pp
Normally the route is hidden from the user
by the software.
However,
some networks put the burden of determining the route
onto the sender.
Although this simplifies the software,
it also greatly impairs the usability
for most users.
The UUCP network is an example of such a network.
.sh 1 "DESIGN GOALS"
.pp
Design goals for
.i sendmail \**
.(f
\**This section makes no distinction between
.i delivermail
and
.i sendmail.
.)f
include:
.np
Compatibility with the existing mail programs,
including Bell version 6 mail,
Bell version 7 mail,
Berkeley
.i Mail
[Shoens79],
BerkNet mail
[Schmidt79],
and hopefully UUCP mail
[Nowitz78].
ARPANET mail
[Crocker82]
was also required.
.np
Reliability, in the sense of guaranteeing
that every message is correctly delivered
or at least brought to the attention of a human
for correct disposal;
no message should ever be completely lost.
This goal was considered essential
because of the emphasis on mail in our environment.
It has turned out to be one of the hardest goals to satisfy,
especially in the face of the many anomalous message formats
produced by various ARPANET sites.
For example,
certain sites generate improperly formated addresses,
occasionally
causing error-message loops.
Some hosts use blanks in names,
causing problems with
mail programs that assume that an address
is one word.
The semantics of some fields
are interpreted slightly differently
by different sites.
In summary,
the obscure features of the ARPANET mail protocol
really
.i are
used and
are difficult to support,
but must be supported.
.np
Existing software to do actual delivery
should be used whenever possible.
This goal derives as much from political and practical considerations
as technical.
.np
Easy expansion to
fairly complex environments,
including multiple
connections to a single network type
(such as with multiple UUCP or Ethernets).
This goal requires consideration of the contents of an address
as well as its syntax
in order to determine which gateway to use.
.np
Configuration information should not be compiled into the code.
A single compiled program should be able to run as is at any site
(barring such basic changes as the CPU type or the operating system).
We have found this seemingly unimportant goal
to be critical in real life.
Besides the simple problems that occur when any program gets recompiled
in a different environment,
many sites like to
.q fiddle
with anything that they will be recompiling anyway.
.np
.i Sendmail
must be able to let various groups maintain their own mailing lists,
and let individuals specify their own forwarding,
without modifying the system alias file.
.np
Each user should be able to specify which mailer to execute
to process mail being delivered for him.
This feature allows users who are using specialized mailers
that use a different format to build their environment
without changing the system,
and facilitates specialized functions
(such as returning an
.q "I am on vacation"
message).
.np
Network traffic should be minimized
by batching addresses to a single host where possible,
without assistance from the user.
.pp
These goals motivated the architecture illustrated in figure 1.
.(z
.hl
.ie t \
.	sp 18
.el \{\
.(c
+---------+   +---------+   +---------+
| sender1 |   | sender2 |   | sender3 |
+---------+   +---------+   +---------+
     |  	   |             |
     +----------+  +  +----------+
		|  |  |
		v  v  v
            +-------------+
            |   sendmail  |
            +-------------+
		|  |  |
     +----------+  +  +----------+
     |  	   |             |
     v             v             v
+---------+   +---------+   +---------+
| mailer1 |   | mailer2 |   | mailer3 |
+---------+   +---------+   +---------+
.)c
.\}

.ce
Figure 1 \*- Sendmail System Structure.
.hl
.)z
The user interacts with a mail generating and sending program.
When the mail is created,
the generator calls
.i sendmail ,
which routes the message to the correct mailer(s).
Since some of the senders may be network servers
and some of the mailers may be network clients,
.i sendmail
may be used as an internet mail gateway.
.sh 1 "USAGE"
.sh 2 "Address Formats"
.pp
Arguments may be flags or addresses.
Flags set various processing options.
Following flag arguments,
address arguments may be given.
Addresses follow the syntax in RFC822
[Crocker82]
for ARPANET
address formats.
In brief, the format is:
.np
Anything in parentheses is thrown away
(as a comment).
.np
Anything in angle brackets (\c
.q "<\|>" )
is preferred
over anything else.
This rule implements the ARPANET standard that addresses of the form
.(b
user name <machine-address>
.)b
will send to the electronic
.q machine-address
rather than the human
.q "user name."
.np
Double quotes
(\ "\ )
quote phrases;
backslashes quote characters.
Backslashes are more powerful
in that they will cause otherwise equivalent phrases
to compare differently \*- for example,
.i user
and
.i
"user"
.r
are equivalent,
but
.i \euser
is different from either of them.
This might be used
to avoid normal aliasing
or duplicate suppression algorithms.
.pp
Parentheses, angle brackets, and double quotes
must be properly balanced and nested.
The rewriting rules control remaining parsing\**.
.(f
\**Disclaimer: Some special processing is done
after rewriting local names; see below.
.)f
.pp
Although old style addresses are still accepted
in most cases,
the preferred address format
is based on ARPANET-style domain-based addresses
[Su82a].
These addresses are based on a hierarchical, logical decomposition
of the address space.
The addresses are hierarchical in a sense
similar to the U.S. postal addresses:
the messages may first be routed to the correct state,
with no initial consideration of the city
or other addressing details.
The addresses are logical
in that each step in the hierarchy
corresponds to a set of
.q "naming authorities"
rather than a physical network.
.pp
For example,
the address:
.(l
eric@HostA.BigSite.ARPA
.)l
would first look up the domain
BigSite
in the namespace administrated by
ARPA.
A query could then be sent to
BigSite
for interpretation of
HostA.
Eventually the mail would arrive at
HostA,
which would then do final delivery
to user
.q eric.
.sh 2 "Mail to Files and Programs"
.pp
Files and programs are legitimate message recipients.
Files provide archival storage of messages,
useful for project administration and history.
Programs are useful as recipients in a variety of situations,
for example,
to maintain a public repository of systems messages
(such as the Berkeley
.i msgs
program).
.pp
Any address passing through the initial parsing algorithm
as a local address
(i.e, not appearing to be a valid address for another mailer)
is scanned for two special cases.
If prefixed by a vertical bar (\c
.q \^|\^ )
the rest of the address is processed as a shell command.
If the user name begins with a slash mark (\c
.q /\^ )
the name is used as a file name,
instead of a login name.
.sh 2 "Aliasing, Forwarding, Inclusion"
.pp
.i Sendmail
reroutes mail three ways.
Aliasing applies system wide.
Forwarding allows each user to reroute incoming mail
destined for that account.
Inclusion directs
.i sendmail
to read a file for a list of addresses,
and is normally used
in conjunction with aliasing.
.sh 3 "Aliasing"
.pp
Aliasing maps local addresses to address lists using a system-wide file.
This file is hashed to speed access.
Only addresses that parse as local
are allowed as aliases;
this guarantees a unique key
(since there are no nicknames for the local host).
.sh 3 "Forwarding"
.pp
After aliasing,
if an recipient address specifies a local user
.i sendmail
searches for a
.q .forward
file in the recipient's home directory.
If it exists,
the message is
.i not
sent to that user,
but rather to the list of addresses in that file.
Often
this list will contain only one address,
and the feature will be used for network mail forwarding.
.pp
Forwarding also permits a user to specify a private incoming mailer.
For example,
forwarding to:
.(b
"\^|\|/usr/local/newmail myname"
.)b
will use a different incoming mailer.
.sh 3 "Inclusion"
.pp
Inclusion is specified in RFC 733 [Crocker77] syntax:
.(b
:Include: pathname
.)b
An address of this form reads the file specified by
.i pathname
and sends to all users listed in that file.
.pp
The intent is
.i not
to support direct use of this feature,
but rather to use this as a subset of aliasing.
For example,
an alias of the form:
.(b
project: :include:/usr/project/userlist
.)b
is a method of letting a project maintain a mailing list
without interaction with the system administration,
even if the alias file is protected.
.pp
It is not necessary to rebuild the index on the alias database
when a :include: list is changed.
.sh 2 "Message Collection"
.pp
Once all recipient addresses are parsed and verified,
the message is collected.
The message comes in two parts:
a message header and a message body,
separated by a blank line.
The body is an uninterpreted
sequence of text lines.
.pp
The header is formated as a series of lines
of the form
.(b
	field-name: field-value
.)b
Field-value can be split across lines by starting the following
lines with a space or a tab.
Some header fields have special internal meaning,
and have appropriate special processing.
Other headers are simply passed through.
Some header fields may be added automatically,
such as time stamps.
.sh 1 "THE UUCP PROBLEM"
.pp
Of particular interest
is the UUCP network.
The explicit routing
used in the UUCP environment
causes a number of serious problems.
First,
giving out an address
is impossible
without knowing the address of your potential correspondent.
This is typically handled
by specifying the address
relative to some
.q "well-known"
host
(e.g.,
ucbvax or decvax).
Second,
it is often difficult to compute
the set of addresses
to reply to
without some knowledge
of the topology of the network.
Although it may be easy for a human being
to do this
under many circumstances,
a program does not have equally sophisticated heuristics
built in.
Third,
certain addresses will become painfully and unnecessarily long,
as when a message is routed through many hosts in the USENET.
And finally,
certain
.q "mixed domain"
addresses
are impossible to parse unambiguously \*-
e.g.,
.(l
decvax!ucbvax!lbl-h!user@LBL-CSAM
.)l
might have many possible resolutions,
depending on whether the message was first routed
to decvax
or to LBL-CSAM.
.pp
To solve this problem,
the UUCP syntax
would have to be changed to use addresses
rather than routes.
For example,
the address
.q decvax!ucbvax!eric
might be expressed as
.q eric@ucbvax.UUCP
(with the hop through decvax implied).
This address would itself be a domain-based address;
for example,
an address might be of the form:
.(l
mark@d.cbosg.btl.UUCP
.)l
Hosts outside of Bell Telephone Laboratories
would then only need to know
how to get to a designated BTL relay,
and the BTL topology
would only be maintained inside Bell.
.pp
There are three major problems
associated with turning UUCP addresses
into something reasonable:
defining the namespace,
creating and propagating the necessary software,
and building and maintaining the database.
.sh 2 "Defining the Namespace"
.pp
Putting all UUCP hosts into a flat namespace
(e.g.,
.q \&...@host.UUCP )
is not practical for a number of reasons.
First,
with over 1600 sites already,
and (with the increasing availability of inexpensive microcomputers
and autodialers)
several thousand more coming within a few years,
the database update problem
is simply intractable
if the namespace is flat.
Second,
there are almost certainly name conflicts today.
Third,
as the number of sites grow
the names become ever less mnemonic.
.pp
It seems inevitable
that there be some sort of naming authority
for the set of top level names
in the UUCP domain,
as unpleasant a possibility
as that may seem.
It will simply not be possible
to have one host resolving all names.
It may however be possible
to handle this
in a fashion similar to that of assigning names of newsgroups
in USENET.
However,
it will be essential to encourage everyone
to become subdomains of an existing domain
whenever possible \*-
even though this will certainly bruise some egos.
For example,
if a new host named
.q blid
were to be added to the UUCP network,
it would probably actually be addressed as
.q d.bli.UUCP
(i.e.,
as host
.q d
in the pseudo-domain
.q bli
rather than as host
.q blid
in the UUCP domain).
.sh 2 "Creating and Propagating the Software"
.pp
The software required to implement a consistent namespace
is relatively trivial.
Two modules are needed,
one to handle incoming mail
and one to handle outgoing mail.
.pp
The incoming module
must be prepared to handle either old or new style addresses.
New-style addresses
can be passed through unchanged.
Old style addresses
must be turned into new style addresses
where possible.
.pp
The outgoing module
is slightly trickier.
It must do a database lookup on the recipient addresses
(passed on the command line)
to determine what hosts to send the message to.
If those hosts do not accept new-style addresses,
it must transform all addresses in the header of the message
into old style using the database lookup.
.pp
Both of these modules
are straightforward
except for the issue of modifying the header.
It seems prudent to choose one format
for the message headers.
For a number of reasons,
Berkeley has elected to use the ARPANET protocols
for message formats.
However,
this protocol is somewhat difficult to parse.
.pp
Propagation is somewhat more difficult.
There are a large number of hosts
connected to UUCP
that will want to run completely standard systems
(for very good reasons).
The strategy is not to convert the entire network \*-
only enough of it it alleviate the problem.
.sh 2 "Building and Maintaining the Database"
.pp
This is by far the most difficult problem.
A prototype for this database
already exists,
but it is maintained by hand
and does not pretend to be complete.
.pp
This problem will be reduced considerably
if people choose to group their hosts
into subdomains.
This would require a global update
only when a new top level domain
joined the network.
A message to a host in a subdomain
could simply be routed to a known domain gateway
for further processing.
For example,
the address
.q eric@a.bli.UUCP
might be routed to the
.q bli
gateway
for redistribution;
new hosts could be added
within BLI
without notifying the rest of the world.
Of course,
other hosts
.i could
be notified as an efficiency measure.
.pp
There may be more than one domain gateway.
A domain such as BTL,
for instance,
might have a dozen gateways to the outside world;
a non-BTL site
could choose the closest gateway.
The only restriction
would be that all gateways
maintain a consistent view of the domain
they represent.
.sh 2 "Logical Structure"
.pp
Logically,
domains are organized into a tree.
There need not be a host actually associated
with each level in the tree \*-
for example,
there will be no host associated with the name
.q UUCP.
Similarly,
an organization might group names together for administrative reasons;
for example,
the name
.(l
CAD.research.BigCorp.UUCP
.)l
might not actually have a host representing
.q research.
.pp
However,
it may frequently be convenient to have a host
or hosts
that
.q represent
a domain.
For example,
if a single host exists that
represents
Berkeley,
then mail from outside Berkeley
can forward mail to that host
for further resolution
without knowing Berkeley's
(rather volatile)
topology.
This is not unlike the operation
of the telephone network.
.pp
This may also be useful
inside certain large domains.
For example,
at Berkeley it may be presumed
that most hosts know about other hosts
inside the Berkeley domain.
But if they process an address
that is unknown,
they can pass it
.q upstairs
for further examination.
Thus as new hosts are added
only one host
(the domain master)
.i must
be updated immediately;
other hosts can be updated as convenient.
.pp
Ideally this name resolution process
would be performed by a name server
(e.g., [Su82b])
to avoid unnecessary copying
of the message.
However,
in a batch network
such as UUCP
this could result in unnecessary delays.
.sh 1 "COMPARISON WITH DELIVERMAIL"
.pp
.i Sendmail
is an outgrowth of
.i delivermail .
The primary differences are:
.np
Configuration information is not compiled in.
This change simplifies many of the problems
of moving to other machines.
It also allows easy debugging of new mailers.
.np
Address parsing is more flexible.
For example,
.i delivermail
only supported one gateway to any network,
whereas
.i sendmail
can be sensitive to host names
and reroute to different gateways.
.np
Forwarding and
:include:
features eliminate the requirement that the system alias file
be writable by any user
(or that an update program be written,
or that the system administration make all changes).
.np
.i Sendmail
supports message batching across networks
when a message is being sent to multiple recipients.
.np
A mail queue is provided in
.i sendmail.
Mail that cannot be delivered immediately
but can potentially be delivered later
is stored in this queue for a later retry.
The queue also provides a buffer against system crashes;
after the message has been collected
it may be reliably redelivered
even if the system crashes during the initial delivery.
.np
.i Sendmail
uses the networking support provided by 4.2BSD
to provide a direct interface networks such as the ARPANET
and/or Ethernet
using SMTP (the Simple Mail Transfer Protocol)
over a TCP/IP connection.
.+c
.ce
REFERENCES
.nr ii 1.5i
.ip [Crocker77]
Crocker, D. H.,
Vittal, J. J.,
Pogran, K. T.,
and
Henderson, D. A. Jr.,
.ul
Standard for the Format of ARPA Network Text Messages.
RFC 733,
NIC 41952.
In [Feinler78].
November 1977.
.ip [Crocker82]
Crocker, D. H.,
.ul
Standard for the Format of Arpa Internet Text Messages.
RFC 822.
Network Information Center,
SRI International,
Menlo Park, California.
August 1982.
.ip [Feinler78]
Feinler, E.,
and
Postel, J.
(eds.),
.ul
ARPANET Protocol Handbook.
NIC 7104,
Network Information Center,
SRI International,
Menlo Park, California.
1978.
.ip [Nowitz78]
Nowitz, D. A.,
and
Lesk, M. E.,
.ul
A Dial-Up Network of UNIX Systems.
Bell Laboratories.
In
UNIX Programmer's Manual, Seventh Edition,
Volume 2.
August, 1978.
.ip [Schmidt79]
Schmidt, E.,
.ul
An Introduction to the Berkeley Network.
University of California, Berkeley California.
1979.
.ip [Shoens79]
Shoens, K.,
.ul
Mail Reference Manual.
University of California, Berkeley.
In UNIX Programmer's Manual,
Seventh Edition,
Volume 2C.
December 1979.
.ip [Solomon81]
Solomon, M.,
Landweber, L.,
and
Neuhengen, D.,
.ul
The Design of the CSNET Name Server.
CS-DN-2.
University of Wisconsin,
Madison.
October 1981.
.ip [Su82a]
Su, Zaw-Sing,
and
Postel, Jon,
.ul
The Domain Naming Convention for Internet User Applications.
RFC819.
Network Information Center,
SRI International,
Menlo Park, California.
August 1982.
.ip [Su82b]
Su, Zaw-Sing,
.ul
A Distributed System for Internet Name Service.
RFC830.
Network Information Center,
SRI International,
Menlo Park, California.
October 1982.