Re: OOM when adding ipv6 route: How to make available more per-cpu memory?

[Eric Dumazet <eric.dumazet@gmail.com>]

Le lundi 08 novembre 2010 à 13:27 -0800, Ben Greear a écrit :
> On 11/08/2010 03:02 AM, Eric Dumazet wrote:
> > Le samedi 06 novembre 2010 à 10:08 -0700, Ben Greear a écrit :
> >
> >> At least I don't see any percpu dumps in dmesg.  I vaguely remember
> >> someone posting some ipv6 address scalability patches some time back.
> >> I think they had to hack on /proc fs as well.  I'll see if I can
> >> dig those up.
> >>
> >>> Make sure udev / hotplug is not the problem, if you create your devices
> >>> very fast.
> >>
> >> We can create the macvlans w/out problem, though I'm sure that could
> >> be sped up.  The problem is when we try to add IPv6 addresses to
> >> them.
> >
> > I see. Did you check /proc/sys/net/ipv6/ tunables ?
> >
> > For example, I bet you need to make route/max_size a bigger value than
> > default (4096)
> 
> I'm having a hard time figuring out how this actually causes something
> to fail.
> 
> It seems that having max_size too small would cause the garbage collector
> logic to happen more often, but I don't see any place it actually
> removes or limits entries, and I can't find any other references to
> ip6_rt_max_size.
> 
> I must just be missing something..so if you have any ideas where to
> look, I'd love to hear it!
> 

ipv6_add_addr() calls addrconf_dst_alloc
addrconf_dst_alloc() calls dst_alloc
dst_alloc() calls ip6_dst_gc
ip6_dst_gc() calls fib6_run_gc
fib6_run_gc() calls icmp6_dst_gc
fib6_run_gc() calls fib6_clean_all

a bit later, we return to ipv6_add_addr() (line 644) with an error

You want to add a warning in net/ipv6/route.c, line 1949

Thanks