Postfix 2.5 with MailScanner? Bug found, fix attached...

[Glenn Steen]

On 23/03/2008, Glenn Steen <glenn.steen at gmail.com> wrote:
> On 23/03/2008, Gerard <gerard at seibercom.net> wrote:
>
> > On Sun, 23 Mar 2008 17:28:44 +0100
>  >
>  > "Glenn Steen" <glenn.steen at gmail.com> wrote:
>  >
>  >
>  > > So you're skipping 2.5 altogether? I *might* be able tosqueeze this
>  >  > in, but ... far from sure:-(. Would be great if someone already had
>  >  > tried it all:-).
>  >
>  >
>  > Actually, I am setting up a small system to test various software.
>  >  Since Postfix-2.5.x is all ready released as a stable product, testing
>  >  it in various configurations and against different software would seem
>  >  like a waste of time.
>  :-)
>
>  >  I do know that Postfix made several changes to it's 'milter' support in
>  >  the 2.5.x' release. Not sure about the other stuff though. It is a well
>
> But AFAICT nothing that would break the milter handling in
>  MailScanner... We should "survive" those changes. The queue files I've
>  seen from Alex looks ... sane... from that perspective.
>
>
>  >  known fact that Wietse Venema wants his program to be accessed via
>  >  standard SMTP and other documented protocols. Maybe he is atempting to
>  >  enforce that behavior. That would only be a guess. More than likely,
>
> I don't think so, no.
>  Might be some "off by one" obscurity on my part, but I doubt that too.
>
>
>  >  the changes he made in the final 2.5.x release simply caused
>  >  unintentional problems with MailScanner.
>
> Actually we're not exactly sure here:-). I need some logs from Alex
>  before I can tell for sure what's up... If anything.
>
>
>  Cheers


Well, I'm now sure of what happens,and I have a pretty certain view of
why it happens now and not before... So this is good news:-). Sad
thing is it happened to be a genuine bug by yours truly.

With 2.5/2.5.1 the (false) assumption that I could happily ignore w
and p records while positioning to read the actual body into the
message object isn't true anymore. So one by one the MailScanner
children would enter a forever loop in the Start function of the Body
class in PFDiskStore (one child/new message with a milter action...
With milters that don't change many messages, this could take a while
to be known... And a restart of MS would "clear  things up" for a
while. Sigh. I feel a ripe fool...). The fix is pretty
straightforward: Just make sure we do the same in the Start functions
while loop as we do for the Next action when actually reading the
message body.
Earlier the logic for inserting the pointer records were slightly
different (in PF). Which would explain why it worked then (more or
less by accident... By design, but a flawed design:-).

Anyway... I'll attach a patch for PFDiskStore.pm that should be clean
against the latest stable.

Alex, please apply and report back any findings...
Actually.... This goes for anyone using Postfix, especially if you are
using a milter... We need handle this bug ASAP.... For diverse
reasons.

Jules, could we have a beta with this in?

-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PFDiskStore.pm.milterfix.patch.gz
Type: application/x-gzip
Size: 472 bytes
Desc: not available
Url : http://lists.mailscanner.info/pipermail/mailscanner/attachments/20080324/b7ecc7f0/PFDiskStore.pm.milterfix.patch.gz