Although all my basic instincts are in agreement with Keith's position -- the only way you will ever persuade message senders to comply more fully with the standards is to reject their non-compliant messages -- my recent experience has turned me 180 degrees and convinced me that this position is hopelessly idealistic.
As some of you know, I have only recently returned to work at a company (Mimecast) whose primary business includes the maintenance and operation of a large-scale MTA. The situation has changed a lot in the last ten (not to say twenty) years, and in many ways not for the better. Today's commercial reality is a race to the bottom that more or less guarantees that MTAs will do precisely what Keith says they should not do, which means, to my mind, that it behooves us to document our collective wisdom on the least harmful and most consistent way to do it.
A single example should make this clear: Nearly every business now receives its mail via some third party email security mechanism -- perhaps an appliance, perhaps locally installed code, and increasingly a cloud-based service. Competition among these third parties is fierce, and inevitably standards compliance is far less of a driver of business decisions than customer satisfaction. Imagine, then, that company X, which enforces the standards very strictly, as Keith advocates, manages to "steal" a customer from company Y, which does not. All of a sudden -- and I assure you this really happens -- a host of slightly-misformatted but customer-desired messages which were delivered by company Y will start getting blocked by company X. The customer, inevitably, will scream bloody murder. They won't care that company X is doing a better job of enforcing the standards, they will only care that company X is -- wrongly, in their eyes -- blocking desirable streams of messages that the previous vendor delivered with no visible problem. The generalization is clear: the financial and business incentive is to deliver every message that you can possibly figure out how to deliver, so that your service doesn't appear "inferior" to customers who don't give a rat's rump about the details of standards compliance.
I'm not sure why this came as such a surprise to me, as it is actually just another instance of Postel's Law. Being liberal in what we accept means, in this case, accepting and delivering any message where we believe, with a high level of confidence, that the sender's intentions are clear despite its standards violations.
I can tell you that nothing the IETF says is likely to have any effect on whether or not Mimecast tries to "fix" and deliver such messages -- we do and we will, as do nearly all of our competitors, because although we care what the IETF says, we have no choice but to care what our paying customers say even more. However, if the IETF offered guidance on the least harmful way to do this, the odds are good that we would follow it. And I think it would be better if vendors who felt the need to "fix" messages would at least be mutually consistent in how they fix them.
One (probably controversial) idea that might reduce our collective angst would be to specify that, when an MTA fixes such a message, it also generates an explanatory/warning message back to the sender or the sender's postmaster (perhaps limited to 1 message per day per malformation type). That way we might not be exacerbating existing problems quite as much as Keith and other (myself included) fear. And the senders of misformatted messages might eventually fix their code, if only to shut up the annoying warning messages. -- Nathaniel
On Apr 15, 2011, at 1:50 PM, Keith Moore wrote:
I'm strongly opposed to MTAs "fixing" malformed messages (other than submission servers fixing a small number of known problems caused by broken mail clients).
If an MTA does anything at all when it thinks that a message is malformed, it should be to bounce it _exactly as it received it originally_.
MTAs trying to fix malformed messages, at best, mask problems further upstream that should be fixed. At worst, they exacerbate existing problems and make such problems harder to diagnose.
On Apr 14, 2011, at 3:07 PM, Murray S. Kucherawy wrote:
<ATT00001..txt> This is some work starting up in the APPS area. Please comment on the apps-discuss list if you’re interested in participating. Hi,
Having read the Malformed Message BCP draft I am interested in getting some discussion going on its content and to find the best way forward.
For those who missed it, the draft is at:
I have a few comments on it.
One thing that concerns me is the proposal that processing should stop when certain malformations are identified.
For example it is proposed that should a badly wrapped header field be found (quite how we define this is left open, I presume a line that does not start with a valid header field name followed by a colon) then the processing agent should treat this as the end of the header. My feeling is that this opens up a greater potential hole than the one closed and that can be exploited.
An example of the type of issue this could is cause is that should such a malformation occur before the MIME header fields in the header then this would cause the rest of the header and the message body to be treated as plain text. This could cause content analysis system to fail as they would not interpret the MIME content in the way that was presumably intended.
Given that these recommendations are unlikely to be followed by all clients and servers, I feel that this suggestion will allow content through an agent without suitable processing. My preference on this particular malformation would be to continue processing the header fields, this is based on the assumption that what follows the malformed header field is more likely to be further header fields and not body content. What we do with the malformed header field I am less certain about. We could just ignore it or we could treat it as part of the previous header field - both will be right as often as they wrong. I would welcome some additional thoughts on this.
I have similar feelings about some of the other suggestions including the lack of a MIME-Version header. We cannot ignore intended meaning just because a non-compliant application made a small error in the header fields. Everyone will be best served if we subject such messages to more analysis, not less.
On the whole I think a set of guidelines in this area is good but it will be hard to reach consensus without agreement on some basic underlying principles. I would suggest that one such principle is to try to do what the intended recipient would most likely prefer, which is generally to fix and deliver non-spam messages.
I would also propose some additions to the draft. At Mimecast we see a lot of messages generated by all sorts of systems and amongst these we see a lot of different kinds of message malformations.
I'll suggest more as I think of them but for starters here are a few:
1. Excessively long lines in both headers and body. I commonly see lines that are several hundred Kbs in length. These are often valid messages in the sense that the content is desired by the receiver and in all respects other than line length are well formed. Obviously a limit has to be enforced and I would like to find a consensus on what sort of limit is reasonable. Initially I felt 8K was a good limit - it is after all 8 times the limit in RFC 5321. But it appears that this is too small a limit in real situations. When the limit is exceeded, what is the best option – a rejection or forced line wrap. I am open to both.
2. Invalid characters in headers. I often see headers with un-encoded 8bit characters. These are often displayed correctly to the recipient where the client happens upon the correct character set by virtue of chance.
3. 8bit characters in MIME sections with a content-transfer-encoding of 7bit.
If you have read this far then I think you will agree with me that Murray has made a good start on a much needed set of guidelines. Now let's see if we can support him to expand on the work he has done and reach a consensus on the best approaches.