[Raw Msg Headers][Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: POP/IMAP before SMTP



> First, I'd like to tell you guys that I have a very very basic prototype
> server working, i.e. it follows the specification.  No API so far,
> not today.

	Just a note on the protocol:
		This is an application on where we could (and likely
		SHOULD) use UDP.

		1) contruct dataset
		2) Set retry max = N
		3) Set timeout = 0.1 seconds
		4) Send the dataset to the server
		5) Wait at most 'timeout' for an ACK
		6) Got ACK ? Be happy, quit
		7) Decrease retry count
		8) Retry count <= 0 ?
			--> Be unhappy, but quit
		9) Double the timeout
		10) Go back to 4

		Timeout intervals from the first send would be:
			0.1 0.3 0.7 1.5 3.1
		I would not allow longer timeout than that..
....
> > Let me give you an example of how this can fail miserably: multi-user
> > machines (yes, they still exist, though many people seem to forget that).
> > If someone is visiting a university and uses a guest account on their large
> > UNIX system (with, say, 30,000 users) to read their mail via IMAP, does
> > that mean you want all 30,000 users to be able to bounce spam off of your
> > SMTP server?  Probably not.
> 
> That is what I thought about, and it does not bother me too much: what if
> a multiuser machine gains SMTP access to my server for a few minutes?
> It's highly unprobable that a spammer will start bombing my server right
> at that moment.

	I would not consider it to be a problem either.

> > Another real-world example.  Hotmail used to have (dunno if they still do)
> > the ability to read remote mailboxes via POP.  You probably wouldn't want
> > to open your SMTP server up to all the spam-slime on Hotmail...
> 
> This is more dangerous than the previous example.  Probably we will need
> a way to tell the smtpserver to not trust certain addresses even if there
> where POP connections from them.

	They have a few dozen in-/outbound SMTP servers, and very
	likely DIFFERENT machines for possible POP-retrieve of email.
	I would not consider Hotmail as likely example for a problem
	site.

	The smtpserver does policy database analysis by looking up
	at first specific address of the contactee, and then widening
	the scope bit by bit. If it does find specific address, it
	will never reach "wild-card" entry of  [0.0.0.0]/0  on which
	there are these various 'accept-if-..' things.

> Thanks for useful ideas.
> Eugene

/Matti Aarnio <mea@nic.funet.fi>