2.6.10 dst cache overflow

2.6.10 dst cache overflow

[J. Simonetti]

Hi list,

I was directed to this list for my question. I searched many archives
but could not find any recent posts about this subject so I ask again.

I have a linux (Fedora core 3) based router. I does approximately 50Mbps
aggregated traffic. However once every week (approx) the router dies
with 'kernel: dst cache overflow' messages flooding syslog. I have
tracked it down to being an issue with the routing cache.

I was hoping someone here could give me some more information on how to
address this issue. In particular info about /proc/sys/net/route/* and
the meaning of its values (in essence the garbage collection routines).

Regards,

Jeroen Simonetti
--
Netland Internet Services
http://www.netland.nl

Never look up when dragons fly overhead.

Re: 2.6.10 dst cache overflow

[David Coulson]

J. Simonetti wrote:
> I have a linux (Fedora core 3) based router. I does approximately 50Mbps
> aggregated traffic. However once every week (approx) the router dies
> with 'kernel: dst cache overflow' messages flooding syslog. I have
> tracked it down to being an issue with the routing cache.

Funny this came up now - One of my core routers failed last night with 
the identical problem. The route cache was set to 131k routes, although 
I couldn't get a sh prompt on the box even once the switch ports were 
down... Had to sysrq-b it.

I'm running 2.6.11-rc1-bk8 right now on those boxes. 20ish day uptime 
until yesterday and handling around 12Mbit of traffic. 'route -Cn | wc 
-l' right now shows around 29k routes cached. I cranked the route cache 
up to 1m entries, but I figure that's probably excessive.


David

-- 
David J. Coulson
email: david@davidcoulson.net
web: http://www.davidcoulson.net/
phone: (216) 920-3100 / (216) 258-4942

Re: 2.6.10 dst cache overflow

[Robert Olsson]

David Coulson writes:
 > J. Simonetti wrote:
 > > I have a linux (Fedora core 3) based router. I does approximately 50Mbps
 > > aggregated traffic. However once every week (approx) the router dies
 > > with 'kernel: dst cache overflow' messages flooding syslog. I have
 > > tracked it down to being an issue with the routing cache.
 > 
 > Funny this came up now - One of my core routers failed last night with 
 > the identical problem. The route cache was set to 131k routes, although 
 > I couldn't get a sh prompt on the box even once the switch ports were 
 > down... Had to sysrq-b it.
 > 
 > I'm running 2.6.11-rc1-bk8 right now on those boxes. 20ish day uptime 
 > until yesterday and handling around 12Mbit of traffic. 'route -Cn | wc 
 > -l' right now shows around 29k routes cached. I cranked the route cache 
 > up to 1m entries, but I figure that's probably excessive.

Hello!

Very recently here on netdev "Memory leak in 2.6.11-rc1" was discussed 
this resulted in:
 
> ===== net/ipv4/ip_output.c 1.74 vs edited =====
> --- 1.74/net/ipv4/ip_output.c	2005-01-25 01:40:10 +01:00
> +++ edited/net/ipv4/ip_output.c	2005-01-30 18:54:43 +01:00
> @@ -389,6 +389,7 @@
>  	to->priority = from->priority;
>  	to->protocol = from->protocol;
>  	to->security = from->security;
> +	dst_release(to->dst);
>  	to->dst = dst_clone(from->dst);
>  	to->dev = from->dev;  

and a similar patch for ipv6. Should be a start.


						--ro

Re: 2.6.10 dst cache overflow

[David Coulson]

Robert Olsson wrote:
> Very recently here on netdev "Memory leak in 2.6.11-rc1" was discussed 
> this resulted in:

I see it's in 2.6.11-rc4 - I should probably just upgrade to that and 
see what happens :)

David

-- 
David J. Coulson
email: david@davidcoulson.net
web: http://www.davidcoulson.net/
phone: (216) 920-3100 / (216) 258-4942

Re: 2.6.10 dst cache overflow

[J. Simonetti]

On Wed, 2005-02-16 at 15:28, Robert Olsson wrote:
> Hello!
> 
> Very recently here on netdev "Memory leak in 2.6.11-rc1" was discussed 
> this resulted in:
>  
> > ===== net/ipv4/ip_output.c 1.74 vs edited =====
> > --- 1.74/net/ipv4/ip_output.c	2005-01-25 01:40:10 +01:00
> > +++ edited/net/ipv4/ip_output.c	2005-01-30 18:54:43 +01:00
> > @@ -389,6 +389,7 @@
> >  	to->priority = from->priority;
> >  	to->protocol = from->protocol;
> >  	to->security = from->security;
> > +	dst_release(to->dst);
> >  	to->dst = dst_clone(from->dst);
> >  	to->dev = from->dev;  
> 
> and a similar patch for ipv6. Should be a start.

After reading the thread and doing some research of my own it looks like
that is what I'm suffering from. I will try the patch to see if it
helps.
Thank you for the help.

Jeroen Simonetti
--
Netland Internet Services
http://www.netland.nl

Increased knowledge will help you now.  Have mate's phone bugged.