Philip Hazel wrote:
>
> Exim is supposed to detect the case when NFS is not there. Here is an
> extract from the code:
Roger, and I read about all the options to adjust that if needed. My NFS
mounts are indeed there, I don't think this would be the code we want.
> /* First of all, we stat the file. If this fails, we try to stat the enclosing
> directory, because a file in an unmounted NFS directory will look the same as a
Question: if you stat() the forward file and it exists, but cannot be
read for whatever reason (let's say it's root.root:0600, eg) what
happens? Does mail delivery continue, skipping the filter?
> However, this is concerned with NFS mounts, not NFS locks. In fact, I'm
> not myself clear what exactly you mean by "NFS lock".
Neither am I, I'm afraid -- this was a postulation on my part. I am
trying to debug the problem and devise possible problem areas. Here's
the layout:
An employee had their email address hijacked by a spammer. As a result,
he's now receiving a ton of incoming bounces and other junk. Now, what
I'm seeing is that randomly his user filter is being skipped -- *some*
email that's sent out to SpamAssassin and injected back in (with a
tagged subject) is not get filtered. However, if you save that
non-filtered email and run it through his filter by hand, it's a 100%
hit and would get filtered.
As mentioned, it's seemingly random in nature -- two emails could come
in one after the other, both spam. Both are piped out and injected back
in, marked as spam. The first one gets filtered, the second one skips
the filter. Odd.
So I'm a bit at a loss how to debug the random user filter skippage;
Nigel had mentioned about NFS mounts, which led me to thinking about a
locked forwardfile, or too heavy a load, or *something*. All processes
are working normally - my filter snags probably 200 spams a day, his
snags probably 500. But, a small percentage are just....missed.
Looking for any ideas or causes that could be causing this to happen.
:-/ I upgraded to 4.43, hoping it'd be a random bug that got fixed -
still the same though.
Ideas?
-te
--
Troy Engel | Systems Engineer
Fluid, Inc |
http://www.fluid.com
