Re: disregard - one more test, sorry

[Nick Drage <nickd () metastasis org uk>]

On Fri, Sep 09, 2005 at 10:02:01AM -0400, Dave Aitel wrote:

> We're testing the Bounce handler in mailman. Apparantly you can DoS[1]
> a mailman instance by signing a thousand people up to a list (which we
> did by mistake) then having those thousand people placed into the
> bounce queue. This makes mailman's bounce handler grow to  use
> approximately one gig of ram and all the CPU. To fix this, you need
> (it turns out) to set the bounce handling to "on each bounce, just
> disable/unsubscribe that person". However, while the bounce handler is
> doing it's painful dance, the administrative interface is only
> accessible by shutting down mailman (via killall -9 python).

If you set this up wouldn't it then be relatively trivial to just
unsubscribe/disable every subscriber to the list, as bounce messages are
much easier to fake than unsubscribe confirmations?

-- 
When the pin is pulled, Mr. Grenade is not our friend.