zen.org Communal Weblog

A friend of mine is set up with a satellite Internet connection to his home in a not-all-that-rural part of Ireland. He’s been hosting his domain from there, with all email traffic and such going to his local server. Until recently, it was a perfectly workable solution, even with the normal supply of spam, virus, and other junk mail arriving.

But nearly two weeks ago, his domain came under attack from a bunch of spam botnets. Uncountable messages were forged to various places, all of which set up with the Sender: header to be totally random addresses @domain.ie. Unfortunately his ISP said they would not help block the traffic. (As opposed to could not.)

The workaround we came up with pushed his traffic through a virtual-hosted system I have set up over in the US with johncompanies.com (yes, a blatant plug, but I really like their service). There were a few steps I had to take in configuring Postfix before they added the MX record for his domain to reroute everything. (This is on a system running Debian GNU/Linux version 4.0, codenamed etch, using postfix 2.3.7.)

• In main.cf, add his domain to relay_domains (which already existed for other domains I MX with).
• Since he uses a lot of different email addresses (to make it easy to catch re-use and selling of them), I didn’t set up a relay_recipient_maps hash table. That would have been even cooler with its ability to block every single address except for the few that are in fact valid. In this case, however, he had a number of variants of addresses he used so it wasn’t a practical choice.
• Add to smtpd_recipient_restrictions the line
  

  check_recipient_access hash:/etc/postfix/maps/access_recipient

  and created the file /etc/postfix/access_recipient containing

  postmaster@domain.ie  REJECT
  MAILER-DAEMON@domain.ie       REJECT

  and then ran postmap access_recipient as root. I should note I did not put a line like domain.ie OK which would have let all other mail for the domain go through—but usurped any other rules that smtpd_recipient_restrictions may try to do after my access_recipients entry.

• I created a /etc/postfix/access_sender file with the lines below. The first was used because his server will never receive mail from someone in his domain.
  

  domain.ie       REJECT
  MAILER-DAEMON@  REJECT
  MailerDaemon@   REJECT
  abuse@          REJECT
  admin@          REJECT
  Administrator@ REJECT
  autoresponder@  REJECT
  bounce@         REJECT
  info@           REJECT
  majordomo@      REJECT
  Majordomo-Owner@ REJECT
  nobody@         REJECT
  postmaster@     REJECT
  savrequest@     REJECT
  senderchallenge@ REJECT
  spam@   REJECT
  vacation@       REJECT
  

  Then I had to run postmap access_sender as root. In main.cf, for smtpd_sender_restrictions I added

  check_sender_access hash:/etc/postfix/access_sender

  as well.

• I found I wanted to add some rules that used regular expressions. After installing the postfix-pcre Debian package, I created a new file /etc/postfix/access_sender.pcre with the lines
  

  /.*bounces\@/   REJECT
  /confirm-return.*\@/    REJECT

  and in main.cf gave smtpd_sender_restrictions yet another entry of

  check_sender_access pcre:/etc/postfix/access_sender.pcre

• Following the hints from a post by Justin Mason, I created a new file /etc/postfix/header_checks and gave it the lines
  

  /^Content-Type: multipart\/report; report-type=delivery-status\;/       REJECT no third-party DSNs
  /^Content-Type: message\/delivery-status; /     REJECT no third-party DSNs

  A second file, /etc/postfix/null_sender, had

  <>      550 no third-party DSNs

  In main.cf I gave the smtpd_sender_restrictions list the new entry of

  hash:/etc/postfix/null_sender

  and also added a new line defining header_checks as

  header_checks = regexp:/etc/postfix/header_checks

  Finally I had to run postmap null_sender as root.

• In master.cf I had to adjust the smtp unix and relay unix entries to only do 2 processes, not the default of 20, since having my machine try 20 simultaneous connections to his machine wouldn’t help. Under each, respectively, I had to add
  

  -o smtp_destination_concurrency_limit=2

  and

  -o relay_destination_concurrency_limit=2

  I’m still not positive if the maximum of 2 processes would make these options necessary. I should note that this particular system I was setting up did no other mail delivery, so this change was okay. If you’re doing this on a fully production-level host, you might find a different way to throttle the delivery connections going to a specific host, instead of this change which makes all outgoing mail connections happen only two-at-a-time.

• He’s closed port 25 on his router to try to at least stop the flood. Instead, he’s opening a random port number (like 1767) and having it listen there for new mail. I’ve made postfix deliver it by creating a /etc/postfix/transport file with the lines
  

  # 20080327 help fight the flood, tunnel the mail to its real destination, e.g., his server is 1.2.3.4
  domain.ie     :[1.2.3.4]:1767
  .domain.ie    :[1.2.3.4]:1767

  and ran postmap transport as root. Into main.cf I added

  transport_maps = hash:/etc/postfix/transport

• After all of this was done, I had to do postfix restart

The end result, with Justin’s rules in particular, has had thousands and thousands of attempts get blocked trying to get through the door. Some still trickle through, even after the amavis/clamav/spamassassin content filter has processed them.

This is the final accumulation (with a few I already had):


smtpd_sender_restrictions = check_sender_access hash:/etc/postfix/access_sender,
check_sender_access pcre:/etc/postfix/access_sender.pcre,
hash:/etc/postfix/null_sender

header_checks = regexp:/etc/postfix/header_checks

## Steps from http://www.akadia.com/services/postfix_spamassassin.html
smtpd_recipient_restrictions = permit_sasl_authenticated, permit_mynetworks,
reject_unauth_destination,
reject_unauth_pipelining,
reject_invalid_hostname,
reject_non_fqdn_hostname,
reject_non_fqdn_sender,
reject_non_fqdn_recipient,
reject_unknown_sender_domain,
reject_unknown_recipient_domain,
check_recipient_access hash:/etc/postfix/access_recipient,
check_recipient_access pcre:/etc/postfix/access_recipient.pcre,
check_policy_service inet:127.0.0.1:60000,
permit


(The check_policy_service line is for my use of postgrey, another simple step which drastically reduced the amount of spam my own server was getting.)

Please let me know if any of the instructions above prove to not work out properly for you.

P.S. A command I found handy watching the logs to see what was getting through for attempted delivery, even after everything above:

sudo tail -f /var/log/mail.log | egrep -v '((RCPT|connect(ion)?).* from |smtpd_peer_init)'

For the longest time I’ve been sticking with having to only ever visit www.ros.ie using W1ndow$ on my laptop. Being self-employed, every two months I have to give some tax to The Man.

This time, I decided to look again to see if anyone has discovered a way to do this without that other OS. Luckily, I found some notes by Andrew S. Townley explaining exactly how. He’s found the link into the ros.ie site to get at the actual KCrypto Java applet that it uses (and claims fails to start).

As described, I put it into /usr/lib/jvm/java-6-sun-1.6.0.03/jre/lib/ext and restarted Firefox. Now the login page on the site worked fine, and I could get in. Yay!

P.S. I’m doing this under Ubuntu 7.10 (Gutsy Gibbon).

A Barack Obama supporter rang my door bell a few days ago, wanted me to change my party from Green to Democratic to vote April 22. I have until March 24to change my party if I wish to vote Democratic then. I didn’t change. Over a month until Pennsylvania’s primary, with no other delegates up for grabs between now and then. Mrs. Obama is going to be twenty miles from my house today. Hillary Clinton was on the other side of Philadelphia a few days ago. They should publish better schedules, I can’t easily figure where they are going to be, though it’s easy to figure out where they have been. Vote Quimby!

We recently moved zen.org to a different server, and in the process my dump and reload of our MySQL database worked—mostly. However any posts with UTF-8 Unicode characters didn’t get displayed correctly.

After spending too much time trying to figure out how to make mysql and mysqldump help me, I realized I should look around for others who’ve had the same problem.

Voila, Jonkepon in Japan gave the fix for exactly the problem we had. The fix has to do with the collation of the entries in the database, not the actual dumping and importing of the content itself.

Since the newer WordPress already does their first step with SET TABLE, I just had to go in via phpMyAdmin. For each of post_content in wp_posts and comment_content in wp_comments, I changed the collation of each to binary (noting the type of LONGTEXT or TEXT) and saved it. Then I edited them again and set each to utf8_unicode_ci, and saved them.

Bingo! All is happy and good again. The other tables are all still latin1_swedish_ci (?!), but I’ll leave them alone until we bump into somewhere else that it’s a problem.