> > ip_rt_put() only frees a route if this is a) not cached and b) if
> > the reference count is exactly one. This is in the most cases correct.
> > If Philip found out all possible drops this may be a reason.
>
> Im beginning to suspect we have a kernel/interrupt context problem here and
> the reference counters arent safe
More info: I have a tcpdump during this allocation increase.
Would that be intresting?
It's about 22 seconds of recording, and 2.3Mb in size. (the first try was
68Mb... ;)
I found out yet one more thing in reproduceing this more reliably: to
make the 'somehost' a machine on the other side of the "router-firewall"
machine, more precicly the backup station (really slow thing). It might
has to do whith some packet loss, maby that the firewall gets too busy and
drops some important packets, maby in conjunction with a time window
that gets too large. Or time-window in conjunction with interrupts.
/Ryde