Buntix edited this page Dec 28, 2014 · 2 revisions
Clone this wiki locally

How To Do Autoreplies Without The World Hating You

There is often a requirement for creating messages automatically, in response to incoming mail. On a personal level, these are often called "Out Of Office" (OoO) messages. Unix users may well refer to them as "vacation" messages, from the traditional program that implemented the functionality. Systems may automatically generate responses to incoming mail too, perhaps acknowledgements for submissions to a Helpdesk or role address.

For the purposes of this document, we refer to all automatically generically response messages as "autoreplies".

Care needs to be taken when generating autoreplies. A system which blindly autoreplies to any incoming message will likely annoy someone and cause complaints. To take a simple case, much spam forges the sender address to be that of an innocent third party. This third party may then receive autoreplies from the system for messages they didn't send in the first place. Imagine if a million spams were sent with their address forged as the sender. As well as all the delivery failure reports, they will receive many autoreply messages. This is collateral spam, or backscatter.

The following resources may be informative:

JANET CSIRT advice on collateral spam

Wikipedia page on Backscatter from email

Many spam fighters are vociferous about inappropriate autoreply mechanisms. Any autoreply mechanism must be carefully implemented to minimise the risk of a system being penalised by others for being seen as a source of backscatter.

Some autoreply systems implemented in commercial products are particularly broken, in that they will reply to messages sent to a mailing list, directing the traffic to the list posting address, so that all members of the list see the autoreply message (assuming the autoreply user's address is able to post to the list unmoderated). This is particularly abhorrent behaviour, and will often cause immediate unsubscription from the list by list owner or moderator.

Rules for Autoreply Messages

There are two basic sets of rules to follow: determining which messages to reply to, and formatting the reply message in such a way as to minimise the chances that the receiving system will itself generate an autoreply.

Selecting Which Messages to Reply To

There are a number of ways of determining whether or not a message should be responded to. This is a brief list of the very basics:

  • Only reply to email with your address in the To/Cc headers (this can help prevent you replying to mailing list stuff, as well as other things)
  • Do not reply to messages including obvious mailing list headers, or with Precedence: bulk, Precedence: list or Precedence: junk, or from addresses clearly identifiable with mailing lists
  • Do not reply to messages with null SMTP sender (mail system warning, error, and notification messages), or with from addresses such as MAILER-DAEMON, or other senders that are clearly identifiable as being connected with mailing lists or other automated systems
  • Do not reply to messages from a sender address to which you have recently sent an autoreply (ie in the last week or day - so if it goes horribly wrong you don't bury the other party under mail).

This thread from the mailing list archives details a number of checks that agents creating autoresponses should perform.

The reader should also become familiar with:

Crafting Your Reply

  • Send your replies using the null SMTP sender (ie <>) - this makes them appear like bounces, and prevents other systems trying to reply to them.
  • Add a Precedence: junk header to further reduce the chance of other systems replying to your message.
  • Add an Auto-Submitted: auto-replied header
  • Send your autoreply to the envelope sender address (as reflected in the Return-Path: header [^1] of the original message - this prevents you trying to send to a list address which may legitimately appear as the From: header. Do not try and interpret explicit Reply-To:, Resent-From: etc.

Exim Implementation

Example 1

The following is a recipe posted by Edgar Lovecraft on the mailing list. It consists of a router, which runs an exim filter file, and an autoreply transport.


  driver = redirect
  reply_transport = vacation_reply
  require_files = <location_to_user_spool_directory>/.vacation.msg
  file = <location_to_user_spool_directory>/.vacation.msg
  user = exim
  group = exim


  driver = autoreply

(Note that the autoreply transport driver automatically adds an "Auto-Submitted:" header with value "auto-replied" to the message it generates.)

Per Recipient Filter

This could be hard/symlinked if needed, or even modify the router to always use one file, but dependent on the existance of a per-user .vacation.msg file.

# Exim filter
if ($h_subject: does not contain "SPAM?" and personal) then
##### This is the only thing that a user can set when they      #####
##### decide to enable vacation messaging. The vacation.msg.txt #####
 expand file <location_to_user_spool_directory>/.vacation.msg.txt
 once <location_to_user_spool_directory>/.vacation.db
 log <location_to_user_spool_directory>/.vacation.log
 once_repeat 7d
 to $reply_address
 from $local_part\@$domain
 subject "This is an autoreply...[Re: $h_subject:]"

Note that the filter's 'personal' test has a very narrow scope; see the Exim documentation for details.

Example 2

Here is a rather more complex example, implemented slightly differently by a router and a transport, and triggered by the presence of a .vacation.msg file. It goes to great lengths to avoid generating a response to which it shouldn't reply, according to the principles discussed above. Some may consider it overkill. Likely there can be improvements to this implementation.

Note that at present, it doesn't perform a check for the presence of a user's address or any of their aliases in the To: or Cc: headers. The interested reader may like to make this addition.


  driver = accept
  domains = +local_domains
  condition = ${if or { \
    { match {$h_precedence:} {(?i)junk|bulk|list} } \
    { eq {$sender_address} {} } \
    { def:header_X-Cron-Env: } \
    { def:header_Auto-Submitted: } \
    { def:header_List-Id: } \
    { def:header_List-Help: } \
    { def:header_List-Unsubscribe:} \
    { def:header_List-Subscribe: } \
    { def:header_List-Owner: } \
    { def:header_List-Post: } \
    { def:header_List-Archive: } \
    { def:header_Autorespond: } \
    { def:header_X-Autoresponse: } \
    { def:header_X-Autoreply-From: } \
    { def:header_X-eBay-MailTracker: } \
    { def:header_X-MaxCode-Template: } \
    { match {$h_X-Auto-Response-Suppress: } {OOF} } \
    { match {$h_X-OS:} {HP Onboard Administrator} } \
    { match {$h_X-MimeOLE:} {\N^Produced By phpBB2$\N} } \
    { match {$h_Subject:} {\N^Yahoo! Auto Response$\N} } \
    { match {$h_Subject:} {\N^ezmlm warning$\N} } \
    { match {$h_X-FC-MachineGenerated:} {true} } \
    { match {$message_body} {\N^Your \"cron\" job on\N} } \
    { match {$h_Subject:} {\N^Out of Office\N} } \
    { match {$h_Subject:} {\N^Auto-Reply:\N} } \
    { match {$h_Subject:} {\N^Autoresponse:\N} } \
    { match {$h_Subject:} {\N(Auto Reply)$\N} } \
    { match {$h_Subject:} {\N(Out of Office)$\N} } \
    { match {$h_Subject:} {\Nis out of the office.$\N} } \
    { match {$h_From:} {\N(via the vacation program)\N } } \
    } \
    {no} {yes} \
  require_files = <location_of_user_home_directory>/.vacation.msg
  user = ${lc:$local_part}
  senders = !+noautoreply_senders
  transport = vacation_transport

Depending on your environment, you may receive the non-English equivalent of a typical OoO's Subject: "XYZ is out of the office" -- for example, I have seen the German equivalent "XYZ ist ausser haus". Matching these possibilities could be more fun, especially considering that they may be marked up with different charsets, etc. Unless you receive a lot of non-English mail, checking for these as well probably really is overkill: the chances are some of the other more generic rules will get them anyway).

You may also like to add a check for any mail that was detected and marked up as spam by local checking facilities, for example:

{ match {$h_X-Spam-Flag:} {\N^yes\N} } \

  driver = autoreply
  log = <location_of_user_home_directory>/.vacation.log
  once = <location_of_user_home_directory>/.vacation.once
  once_repeat = 7d
  # Errors-To: is deprecated
  # There are arguments over whether this should send to the SMTP sender, or
  # to a From:, Reply-To: or Resent-From: header
  to = "${if def:h_Errors-To: {$h_Errors-To:} {$sender_address}}"
  file =  <location_of_user_home_directory>/.vacation.msg
  subject = ${if def:h_subject: \
                {Auto: Re: ${rfc2047:${quote:${escape:${length_60:$h_subject:}} }} }\
                {Auto: I am away from my mail} \
  user = ${lc:$local_part}

You will also need the following list establishing:

addresslist  noautoreply_senders = <location_of_exim_data_files>/autorep.noanswer

The file autorep.noanswer contains patterns of sender addresses, mail from which should not be responded to. Here is a start, ordered roughly from very generically useful, to likely to be quite site-specific; some are commented out, you may like to think a bit harder about using them:


Some people argue that autoreply messages should be sent to email that comes from role accounts, like info@..., webmaster@... and so on. Your own circumstances will dictate whether that seems reasonable, modify the above list to suit. You will wish to consider the risks of sending an autoreply message inappropriately, verses any inconvenience caused by a genuine message sender (as opposed to an automated system) not receiving an autoreply.

What Will Happen If You Ignore These Recommendations

If you send an autoreply to one of the exim mailing lists you will be unsubscribed. If you have caught the list admins in a good mood then this is all that happens.... otherwise....


[^1]: Some people disagree with this - feel free to add your reasoning to this page. One thread starts [http://lists.exim.org/lurker/message/20040123.060645.9aa2db18.en.html](http://lists.exim.org/lurker/message/20040123.060645.9aa2db18.en.html); there are others