Re: Options for temporary operational solution to DMARC problem

[Toerless Eckert <[e-mail]>]

Ted, *:

a) Fluffy's original mail also described illigitimate senders DOS
   attacks to auto-unsubscribe IETF mailing list participants.

   I only remember two answers in this long thread to that question:
   RTFML from JohnL and "have mailman do DMARC rejections itself" from mcr.

   Has this problem been observed (auto-unsubscribe because of explicit
   attacks ?) How would it change under your proposed options
   (you seem to only focus on the impact of legitimiate p=reject senders).

   To me, mcr's recommendation sounds good, and i don't believe ARC adoption
   would ever be fast/complete enough to to not do it but "just wait for ARC".

b) Given how your proposed solutions have all pro/cons, i would strongly
   suggest to also consider first options to raise awareness to the
   problem via mailman:
   
   Eg: Does IETF mailman send any mails to senders or receivers to
   inform them if/when they are DMARC victims ? And warn of potentially
   impending auto-unsubscribe to receivers ... ? Or give suggestions/hints
   to senders/receiver including eg:
     - use a DMARC-free email account for sending
     - Check email archives or add another receiver account with
       digest if you fear you're loosing emails due to DMARC issues
     - subscribe to [e-mail] if you're technically versed and want
       to contribute
   Typically all mailman error messages are totally cryptic and
   non-decipherable for non-mail-gurus.

c) Wrt to your options:

   - do DMARC check in mailman, reject p-reject failures (mcr)
     - provide good visibility on rejection
   - rewrite accepted DMARC p=reject sender address stateless, eg:
     [e-mail]
     Ideally, do this only if the sender has opted in, eg: via
     a subscription option. Of course the reply alias should only work
     for mailing list participants so it ha the same DOS protection as
     other mailman forwarders.
     Kinda a mix of your 2+4

Cheers
    Toerless

On Thu, Nov 03, 2016 at 06:21:45PM -0400, Ted Lemon wrote:
> My understanding is that there are really four possible approaches:
> 
> Bounce messages from sites that have p=REJECT; users at those sites have to use some other email address for IETF business.
> Rewrite From: header on messages from sites that have p=REJECT to point at discard address
> Rewrite all From: headers to point at discard address
> Rewrite all From: headers to reply to addresses that forward to senders for senders with p=REJECT
> 
> What it means to rewrite a From: header to point at a discard address is probably that the mailbox name will be changed to some admin address that is silently discarded, and the display name will be changed to add "via mailing-list".   So e.g., Ted Lemon <[e-mail] <mailto:[e-mail]>> would be changed to Ted Lemon via ietf <[e-mail] <mailto:[e-mail]>>.   For extra credit, [e-mail] <mailto:[e-mail]> could drop silently if the To: and Cc: headers contain multiple addresses, but generate a bounce if the _only_ address is [e-mail] <mailto:[e-mail]>.   There would probably need to be a list ID so that the bounce message could make sense???it would probably say something like "you tried to reply to sender, but it didn???t work, here???s why, here???s what to do."
> 
> Advantages of 1:
> Relatively easy
> Does not alter sender information, so no unexpected behavior in MUAs
> Supported by mailman.
> Disadvantages of 1:
> Substantial inconvenience for people who are unfortunate enough to work for orgs that use p=REJECT
> Possible snowball effect if more sites adopt p=REJECT (e.g., gmail.com <http://gmail.com/>).
> Requires checking p=REJECT when processing mail
> 
> Advantages of 2:
> Does not inconvenience senders from sites with p=REJECT
> Minimizes whose headers are rewritten
> Supported by mailman
> Disadvantages of 2:
> Lots of moving parts (e.g., have to check p=REJECT when processing mail)
> Email from senders at sites with p=REJECT can???t be sorted the same way mail from other senders can be; inconvenient for users who use rule-based sorting on IETF lists
> Harder to send off-list replies if sender is from site with p=REJECT; reply-to-sender in MUA will be handled as described above.
> 
> Advantages of 3:
> Does not inconvenience senders from sites with p=REJECT
> Does not require checking to see whether p=REJECT is in effect for a particular user
> Disadvantages of 3:
> Email from all senders can no longer be sorted as before, so everybody???s procmail filters, etc, break maximally
> Harder to send off-list replies, reply-to-sender in MUA will be handled as described above.
> 
> Advantages of 4:
> Doesn???t inconvenience any mailing list user
> Behavior is as it always was
> Disadvantages of 4:
> Requires state for each sender from a site with p=REJECT
> Possible resource exhaustion attack
> Probably requires substantial programming and testing work; risk of data loss
> Would take a long time to deploy; might not be quicker than waiting for ARC outcome
> 
> I think it is really up to the IESG to decide which of these plans to adopt.   However, if people are interested in talking about what to do, I think this is a decent enumeration of the options (please correct me if I am wrong).   My personal opinion is that option 2 is the least disruptive option, and I believe it is supported by mailman.   Option 1 is the most technically correct option that we have right now, but I think technical correctness is not the right value to optimize for _right now_.   Technical correctness can be restored once ARC completes and is deployed at the sites we are having trouble with; right now, I think the solution that removes the most pain is the best one.
> 
> I mentioned options 3 and 4 for completeness, but I think 3 is unnecessarily heavy, and option 4 is too much work.
> 

-- 
---
[e-mail]