http://bugzilla.spamassassin.org/show_bug.cgi?id=4546





------- Additional Comments From [EMAIL PROTECTED]  2005-08-22 02:58 -------
Duncan, Ok, I think I understand. With DBM or SDBM users each have their own
database file, so running sa-learn can give them no access to another user's
database. With SQL, there is only one shared database, so access via sa-learn
implies access to all users' data. With DBM and SDBM you can get security by
setting things up so users can run sa-learn and either not allow spamc -L or set
up --auth-ident properly. With SQL you can't get security if they can run
sa-learn. But you still would have to either not allow spamc -L or else set up
--auth-ident properly.

In any case, I agree that doc can be dealt with later and the voting for the
patch can proceed without it being perfect. I'll vote on this as is after I get
a chance to build and run with the patch.

Loren, --allow-tell is not an option that means
"--allow-malicious-data-corruption". It means allow spamc to tell spamd the
spam/ham status of the message being sent for the purpose of bayes learning. It
just so happens that there is a side effect that allows any user to affect the
Bayes database of any other user. The name reflects what it is for, not the side
effect. In any case, I think that getting this fix in is what is important. If
we figure out a way to get real user authentication for version 3.2, this option
could change anyway.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

Reply via email to