Pre-DATA checks on envelope about maddy HOT 7 CLOSED

Start pipeline before the body is received

+1

Requires changes to go-smtp (pre-body callback for DATA command)

That's not true, the DATA handler is called before the body has been received (directly after receiving the DATA command). and is given an io.Reader that directly reads from the underlying net.Conn.

It is not possible to report errors for individual recipients or even MAIL FROM command.

Hmm. Indeed, that's an issue.

from maddy.

That's not true, the DATA handler is called before the body has been received (directly after receiving the DATA command). and is given an io.Reader that directly reads from the underlying net.Conn.

https://github.com/emersion/go-smtp/blob/930d6dfbf447707ddcc38bbafccd8d1596cc6181/conn.go#L435

From my initial post:

Also, SMTP doesn't allow us to "abort" transaction once we sent 354 intermediate reply to the DATA command.

So at the time DATA handler gets called, we have to consume the body sent by the client, what I want to avoid:

We want to run certain checks (#41) on the envelope before the message body is sent to us to not waste traffic

from maddy.

Oh, I see. You're right.

from maddy.

Still no ideas other/better than n. 1. Though interfaces may be split into three (EHLO/MAIL FROM/RCPT TO checks). Probably we should just go with it, wdyt?

from maddy.

Another (relatively) small change to envelope checks is to hide them behind regular filter directive.
If you, say, write filter check_source_hostname, it will actually add a check to the EHLO command.
But there are several problems with this "attempt to hide complexity": Envelope checks are executed in a different order. Say, with filter check_source_mx and then filter check_source_mx user could expect that the checks will run in that order, but in fact, they will be executed in reverse because first "filter" checks RCPT and second checks EHLO. So I proposed to add explicit groups of checks. The same approach used by postifx and it is actually recommended by RFC 5321 (the allowed way is to accept all messages and then send bounce messages if something is wrong, but it creates more problems on its own).

from maddy.

Alright, a decision is needed on this one.

We are going with a modified version of the first solution.

type EhloFilter interface {
  Ehlo(ehlo string) error
}
type FromFilter interface {
  From(to string) (string, error)
}
type RcptFilter interface {
  Rcpt(to string) (string, error)
}

smtp ... {
  ehlo {
    check_source_hostname
    check_source_rdns
  }
  rcpt {
    check_source_mx
  }
}

from maddy.

Closing in favor of the design change proposed in #101.

from maddy.