Postfix snapshot 20180826 introduces server support for RFC 3030
CHUNKING (the BDAT command) without BINARYMIME, in both smtpd(8)
and postscreen(8).

Impact on existing configurations:
----------------------------------

- There are no changes for smtpd_mumble_restrictions, smtpd_proxy_filter,
  smtpd_milters, or for postscreen settings, except for the additional
  option to suppress the SMTP server's CHUNKING service announcement,
  for example, with:

  /etc/postfix/main.cf:
      smtpd_discard_ehlo_keywords = chunking

- There are no changes in the Postfix queue file content, no changes
  for down-stream SMTP servers or after-queue content filters, and
  no changes in the envelope or message content that Milters will
  receive.

Example SMTP session
--------------------

The main differences are that the Postfix SMTP server announces
"CHUNKING" support in the EHLO response, and that instead of sending
one DATA request, the remote SMTP client may send one or more BDAT
requests. In the example below, "S:" indicates server responses,
and "C:" indicates client requests.

    S: 220 server.example.com
    C: EHLO client.example.com
    S: 250-server.example.com
    S: 250-PIPELINING
    S: 250-...more announcements...
    S: 250 CHUNKING
    C: MAIL FROM:<sen...@example.com>
    S: 250 2.1.0 Ok
    C: RCPT TO:<recipi...@example.com>
    S: 250 2.1.5 Ok
    C: BDAT 10000
    C: ..followed by 10000 bytes...
    S: 250 2.0.0 Ok: 10000 bytes
    C: BDAT 123
    C: ..followed by 123 bytes...
    S: 250 2.0.0 Ok: 123 bytes
    C: BDAT 0 LAST
    S: 250 2.0.0 Ok: 10123 bytes queued as 41yYhh41qmznjbD
    C: QUIT
    S: 221 2.0.0 Bye

Internally in Postfix, there is no difference between mail that was
received with BDAT or with DATA. Postfix smtpd_mumble_restrictions,
policy delegation queries, smtpd_proxy_filter and Milters all behave
as if Postfix received (MAIL + RCPT + DATA + end-of-data). However,
Postfix will log BDAT-related failures as "xxx after BDAT" to avoid
complicating troubleshooting (xxx = 'lost connection' or 'timeout'),
and will log a warning when a client sends a malformed BDAT command.

Benefits of CHUNKING (BDAT) support without BINARYMIME:
-------------------------------------------------------

Support for CHUNKING (BDAT) was added to improve interoperability
with some clients, a benefit that would reportedly exist even without
Postfix support for BINARYMIME.

Postfix does not support BINARYMIME at this time because:

- BINARYMIME support would require moderately invasive changes to
  support email content that is not line-oriented. With BINARYMIME,
  the Content-Length: header specifies the length of arbitrary
  content that has no line boundaries. Without BINARYMIME, binary
  content is base64-encoded, and formatted as lines of text.

- There is no conversion of BINARYMIME to a line-oriented 8BITMIME
  form that is compatible with legacy systems including UNIX mbox.
  The available options are to convert binary content into one of
  the 7bit forms (base64 or quoted-printable), or to return email
  as undeliverable. Any conversion would break digital signatures,
  so it would have to happen before signing.

Downsides of CHUNKING (BDAT) support:
-------------------------------------

The RFC 3030 authors did not specify any limitations on how clients
may pipeline commands (i.e. send commands without waiting for a
server response). If a server announces PIPELINING support, like
Postfix does, then a remote SMTP client can pipeline all commands
following EHLO, for example, MAIL/RCPT/BDAT/BDAT/MAIL/RCPT/BDAT,
without ever having to wait for a server response. This means that
with BDAT, the Postfix SMTP server cannot distinguish between a
well-behaved client and a spambot, based on their command pipelining
behavior. If you require "reject_unauth_pipelining" to block spambots,
turn off the CHUNKING support announcement as described above.

        Wietse

Reply via email to