Bugzilla Bug 88971

package update due to GLSA 200504-10 to version 1.5 introduces a new feature
"MXGREY" which is very useful, but if activated via "MXGREY=0" in gld.conf, it
breaks gld totally, postfix isn

package update due to GLSA 200504-10 to version 1.5 introduces a new feature
"MXGREY" which is very useful, but if activated via "MXGREY=0" in gld.conf, it
breaks gld totally, postfix isn´t able to communicate to the policy daemon
process anymore:
[postfix/smtpd] warning: premature end-of-input on 127.0.0.1:2525 while reading
input attribute name
[postfix/smtpd] warning: problem talking to server 127.0.0.1:2525: Success

Then it rejects mail with server configuration problem.


Reproducible: Always
Steps to Reproduce:
1. update to gld 1.5
2. activate feature
3. watch postfix log

Actual Results:  
Postfix rejects email

Expected Results:  
communicate with postfix

The feature "MXGREY" is activated via "MXGREY=1" of course... sorry for hasty
cut & paste...
Further digging in logfiles shows segfaults by gld, it seems to be the cause
for lost input postfix is complaining...

As I just tested, gld-1.5 ebuild only installs /etc/gld.conf.sample, with
MXGREY set to 0.

That

That´s correct, sample conf sets "MXGREY=0", but I think a advertized and very
useful feature should not cause segfaults of the daemon process and postfix
rejects email because of that.
This version is heavily broken if this could happen...
If time permits, I will do a debug build and a backtrace when this error is not
        comprehensible by everyone.

Confirmed I can replicate this bug. Will provide more info as I get it.

Further debugging shows that it is breaking right when gld tries to connect to
the database. I think this is an upstream bug.

Attaching a patch which when applied stops the behaviour. I am not 100% sure
that the application is functioning normally, but my tests showed it was sane.
YMMV

Created an attachment (id=56769) [edit]
patch to fix mxgrey=1 crash

The author has confirmed my patch is ok, its been incorporated and a new
version has been released. This should be closed and this version masked imo.

The new version is in CVS.

Ben: I'm not going to mask the old versions because if people notice a problem they will probably go with the upgrade (or be following this bug already ;) ).

Thanks everyone, resolving.

--Tom

This obviously fixed all issues and runs stable for several weeks on different
systems. If no one complains, this really should be marked stable and maybe the
versions with known issues should be dropped...