Quick Mail Transfer Protocol (QMTP)
D. J. Bernstein, djb@pobox.com
19970201
1. Introduction
The Quick Mail Transfer Protocol (QMTP) is a replacement for the
Simple Mail Transfer Protocol (SMTP). QMTP eliminates any need for
end-of-line scanning between hosts with the same end-of-line
convention. It features automatic pipelining and chunking, 8-bit
transmission, prior declaration of the message size, and efficient
batching. It is designed to be very easy to implement.
QMTP is supported by the qmail-qmtpd and maildir2qmtp programs in the
qmail package.
In this document, a string of 8-bit bytes may be written in two
different forms: as a series of hexadecimal numbers between angle
brackets, or as a sequence of ASCII characters between double quotes.
For example, <68 65 6c 6c 6f 20 77 6f 72 6c 64 21> is a string of
length 12; it is the same as the string "hello world!". Note that
these notations are part of this document, not part of the protocol.
2. Protocol
A QMTP client connects to a QMTP server, as discussed in section 7,
over a reliable stream protocol allowing transmission of 8-bit bytes.
Protocol outline: the client sends one or more packages; after each
package, the server sends back some responses.
The client begins by sending a package. A package contains a mail
message, an envelope sender address, and one or more envelope
recipient addresses. See section 4 for the format of a package.
When the server sees the end of the package, it sends back a series
of responses, one response for each envelope recipient address, in
the same order as given by the client. The server is not permitted to
change the order under any circumstances, even if two addresses are
the same. See section 5 for the format of a response.
The server is not permitted to send any portion of its responses to a
package until the client has sent the final byte of the package. The
client is permitted to close the connection before sending the final
byte of the package; in this case, the server must throw away the
package without attempting to deliver the message. However, the
server must not throw away previously accepted messages.
The client does NOT need to wait for a server response before sending
another package. The server must NOT throw away incoming data when it
sends a response. It is the client's responsibility to avoid
deadlock: if it sends a package before receiving all expected server
responses, it must continuously watch for those responses. The server
is permitted to delay its responses if further data has already shown
up from the client; while it is delaying responses, it must not pause
to wait for further data for the client.
The server is permitted to close the connection at any time, although
high-quality servers will try to avoid doing so. Any response not
received by the client indicates a temporary failure.
A QMTP session should take at most 1 hour. Both sides are expected
to close the connection after this time.
3. Messages
In this document, an ``8-bit mail message'' means a sequence of
lines. Each line is a string of zero or more 8-bit bytes.
A message is called ``safe'' if none of its bytes are <0a>.
Implementation note: Here is the intended interpretation of text
files as messages under some current operating systems. Under DOS, a
message is stored on disk as
first line, <0d 0a>, second line, <0d 0a> ... <0d 0a>, last line.
Under UNIX, a message is stored on disk as
first line, <0a>, second line, <0a> ... <0a>, last line.
Notice that both of these encodings are reversible for safe messages.
In practice, it is very common for the last line to be empty. Many
existing utilities refer to the last line as a ``partial line'' and
ignore it whether or not it is empty.
4. Packages
A package is the concatenation of three strings:
first, an encoded 8-bit mail message;
second, an encoded envelope sender address;
third, an encoded series of encoded envelope recipient addresses.
Each envelope address is a string of 8-bit bytes. The interpretation
of addresses depends on the environment in which QMTP is used and is
outside the scope of this document. Each address is encoded as a
netstring, as discussed in section 6. The series of encoded recipient
addresses is in turn encoded as a netstring.
A message is encoded as a string of 8-bit bytes in one of two ways:
Encoding #1 is <0d>, the first line, <0d 0a>, the second line,
<0d 0a>, the third line, ..., <0d 0a>, the last line.
Encoding #2 is <0a>, the first line, <0a>, the second line, <0a>,
the third line, ..., <0a>, the last line.
This string of 8-bit bytes is in turn encoded as a netstring, as
discussed in section 6.
Every server must be prepared to handle encoding #1 and encoding #2.
A server must not reject a message merely because of its encoding.
Implementation note: The intent of encoding #1 and encoding #2 is to
allow very straightforward handling of text files under DOS and UNIX
respectively. The programmer can print <0d> or <0a> and then simply
copy the file.
5. Responses
Each response is a nonempty string of 8-bit bytes, encoded as a
netstring. The first byte of the string is one of the following:
"K" The message has been accepted for delivery to this envelope
recipient. This is morally equivalent to the 250 response to
DATA in SMTP; it is subject to the reliability requirements
of RFC 1123, section 5.3.3.
"Z" Temporary failure. The client should try again later.
"D" Permanent failure.
The remaining bytes are a description of what happened. It is
expected that the description, when interpreted as UTF-2 characters,
(1) will be human-readable, (2) will not repeat the envelope
recipient address, and (3) will not include formatting characters
other than <20>. However, these expectations are not requirements,
and the client should be ready for arbitrary bytes from the server.
Descriptions beginning with <20> are reserved for future extensions.
In descriptions not beginning with <20>, the character "#" must not
appear except in HCMSSC codes.
A server must NOT accept a safe message unless it can store the
message without corruption. More precisely: if the encoded message
sent by the client matches the encoding of some safe message M, then
acceptance means that the server is accepting responsibility to
deliver M to the envelope recipient. (There is at most one
possibility for M, since encodings are reversible on safe messages.)
Deletion of nulls is NOT permissible; a server that deletes nulls
must reject any message containing nulls. Folding of long lines and
high-bit stripping are also NOT permissible.
Servers are permitted to change unsafe messages.
6. Netstrings
Any string of 8-bit bytes may be encoded as [len]":"[string]",".
Here [string] is the string and [len] is a nonempty sequence of ASCII
digits giving the length of [string] in decimal. The ASCII digits are
<30> for 0, <31> for 1, and so on up through <39> for 9. Extra zeros
at the front of [len] are prohibited: [len] begins with <30> exactly
when [string] is empty.
For example, the string "hello world!" is encoded as <31 32 3a 68
65 6c 6c 6f 20 77 6f 72 6c 64 21 2c>, i.e., "12:hello world!,". The
empty string is encoded as "0:,".
[len]":"[string]"," is called a netstring. [string] is called the
interpretation of the netstring.
7. Encapsulation
QMTP may be used on top of TCP. A QMTP-over-TCP server listens for
TCP connections on port 209.
8. Examples
A client opens a connection and sends the concatenation of the
following strings:
"246:" <0a>
"Received: (qmail-queue invoked by uid 0);"
" 29 Jul 1996 09:36:40 -0000" <0a>
"Date: 29 Jul 1996 11:35:35 -0000" <0a>
"Message-ID: <19960729113535.375.qmail@heaven.af.mil>" <0a>
"From: God@heaven.af.mil" <0a>
"To: djb@silverton.berkeley.edu (D. J. Bernstein)" <0a>
<0a>
"This is a test." <0a> ","
"24:" "God-DSN-37@heaven.af.mil" ","
"30:" "26:djb@silverton.berkeley.edu," ","
"356:" <0d>
"From: MAILER-DAEMON@heaven.af.mil" <0d 0a>
"To:" <0d 0a>
" Hate." <22> "The Quoting" <22>
"@SILVERTON.berkeley.edu," <0d 0a>
" " <22> "\\Backslashes!" <22>
"@silverton.BERKELEY.edu" <0d 0a>
<0d 0a>
"The recipient addresses here could"
" have been encoded in SMTP as" <0d 0a>
"" <0d 0a>
" RCPT TO:<Hate.The\ Quoting@silverton.berkeley.EDU>" <0d 0a>
" RCPT TO:<\\Backslashes!@silverton.berkeley.edu>" <0d 0a>
<0d 0a>
"This ends with a partial last line, right here" ","
"0:" ","
"83:" "39:Hate.The Quoting@silverton.berkeley.edu,"
"36:\Backslashes!@silverton.berkeley.EDU," ","
The server sends the following response, indicating acceptance:
"21:Kok 838640135 qp 1390,"
"21:Kok 838640135 qp 1391,"
"21:Kok 838640135 qp 1391,"
The client closes the connection.