Score:0

Spamassassin unintentionally honoring user_prefs

br flag

Postfix is configured to use Spamassassin as a milter, and spamd runs as user debian_spamd (homedir: /var/lib/spamassassin) with no per-user configuration. All mail is forwarded to a remote server. Every time spamd encounters a message for an existing username (e.g. [email protected]), it does this:

Jan 30 07:00:22 eden spamd[385091]: spamd: creating default_prefs: /home/existinguser/.spamassassin/user_prefs
Jan 30 07:00:22 eden spamd[385091]: spamd: failed to create readable default_prefs: /home/existinguser/.spamassassin/user_prefs

Stranger even, when a message arrives without a corresponding local user (e.g. for [email protected]), spamd tries to access the existing user's home dir to read/write the bayes files:

Jan 30 07:00:59 eden spamd[385091]: spamd: handle_user (userdir) unable to find user: 'nonexistinguser'
Jan 30 07:00:59 eden spamd[385091]: spamd: processing message <220130080050QP.30506@msbatch05> for nonexistinguser:111
Jan 30 07:01:00 eden spamd[385091]: plugin: eval failed: bayes: (in learn) locker: safe_lock: cannot create tmp lockfile /home/existinguser/.spamassassin/bayes.lock.localhost.385091 for /home/existinguser/.spamassassin/bayes.lock: Permission denied

I realized that I had to add --nouser-config to the Spamassassin options in /etc/default/spamassassin, although allow_user_rules defaults to 0/off (as per the documentation and this thread).

So the problem is solved, but I still wonder why

  • I had to add an option to achieve the default behavior
  • spamd tried to use an existing user's .spamassassin subdirectory for non-existing users

I am grateful for any hints.

Thanks, Jan

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.