net: fix route cache rebuilds

We added an automatic route cache rebuilding in commit 1080d709fb9d8cd43
but had to correct few bugs. One of the assumption of original patch,
was that entries where kept sorted in a given way.

This assumption is known to be wrong (commit 1ddbcb005c395518 gave an
explanation of this and corrected a leak) and expensive to respect.

Paweł Staszewski reported to me one of his machine got its routing cache
disabled after few messages like :

[ 2677.850065] Route hash chain too long!
[ 2677.850080] Adjust your secret_interval!
[82839.662993] Route hash chain too long!
[82839.662996] Adjust your secret_interval!
[155843.731650] Route hash chain too long!
[155843.731664] Adjust your secret_interval!
[155843.811881] Route hash chain too long!
[155843.811891] Adjust your secret_interval!
[155843.858209] vlan0811: 5 rebuilds is over limit, route caching
disabled
[155843.858212] Route hash chain too long!
[155843.858213] Adjust your secret_interval!

This is because rt_intern_hash() might be fooled when computing a chain
length, because multiple entries with same keys can differ because of
TOS (or mark/oif) bits.

In the rare case the fast algorithm see a too long chain, and before
taking expensive path, we call a helper function in order to not count
duplicates of same routes, that only differ with tos/mark/oif bits. This
helper works with data already in cpu cache and is not be very
expensive, despite its O(N^2) implementation.

Paweł Staszewski sucessfully tested this patch on his loaded router.

Reported-and-tested-by: Paweł Staszewski <pstaszewski@itcare.pl>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
 net/ipv4/route.c |   50 ++++++++++++++++++++++++++++++++++-----------
 1 file changed, 38 insertions(+), 12 deletions(-)





--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Mon, Mar 08, 2010 at 02:20:00PM +0100, Eric Dumazet wrote:
> We added an automatic route cache rebuilding in commit 1080d709fb9d8cd43
> but had to correct few bugs. One of the assumption of original patch,
> was that entries where kept sorted in a given way.
> 
> This assumption is known to be wrong (commit 1ddbcb005c395518 gave an
> explanation of this and corrected a leak) and expensive to respect.
> 
> Paweł Staszewski reported to me one of his machine got its routing cache
> disabled after few messages like :
> 
> [ 2677.850065] Route hash chain too long!
> [ 2677.850080] Adjust your secret_interval!
> [82839.662993] Route hash chain too long!
> [82839.662996] Adjust your secret_interval!
> [155843.731650] Route hash chain too long!
> [155843.731664] Adjust your secret_interval!
> [155843.811881] Route hash chain too long!
> [155843.811891] Adjust your secret_interval!
> [155843.858209] vlan0811: 5 rebuilds is over limit, route caching
> disabled
> [155843.858212] Route hash chain too long!
> [155843.858213] Adjust your secret_interval!
> 
> This is because rt_intern_hash() might be fooled when computing a chain
> length, because multiple entries with same keys can differ because of
> TOS (or mark/oif) bits.
> 
> In the rare case the fast algorithm see a too long chain, and before
> taking expensive path, we call a helper function in order to not count
> duplicates of same routes, that only differ with tos/mark/oif bits. This
> helper works with data already in cpu cache and is not be very
> expensive, despite its O(N^2) implementation.
> 
> Paweł Staszewski sucessfully tested this patch on his loaded router.
> 
> Reported-and-tested-by: Paweł Staszewski <pstaszewski@itcare.pl>
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> ---
>  net/ipv4/route.c |   50 ++++++++++++++++++++++++++++++++++-----------
>  1 file changed, 38 insertions(+), 12 deletions(-)
> 


This looks pretty reasonable to me, Thanks Eric!
Acked-by: Neil Horman <nhorman@tuxdriver.com>

> diff --git a/net/ipv4/route.c b/net/ipv4/route.c
> index b2ba558..d9b4024 100644
> --- a/net/ipv4/route.c
> +++ b/net/ipv4/route.c
> @@ -146,7 +146,6 @@ static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
>  static void		 ipv4_link_failure(struct sk_buff *skb);
>  static void		 ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
>  static int rt_garbage_collect(struct dst_ops *ops);
> -static void rt_emergency_hash_rebuild(struct net *net);
>  
>  
>  static struct dst_ops ipv4_dst_ops = {
> @@ -780,11 +779,30 @@ static void rt_do_flush(int process_context)
>  #define FRACT_BITS 3
>  #define ONE (1UL << FRACT_BITS)
>  
> +/*
> + * Given a hash chain and an item in this hash chain,
> + * find if a previous entry has the same hash_inputs
> + * (but differs on tos, mark or oif)
> + * Returns 0 if an alias is found.
> + * Returns ONE if rth has no alias before itself.
> + */
> +static int has_noalias(const struct rtable *head, const struct rtable *rth)
> +{
> +	const struct rtable *aux = head;
> +
> +	while (aux != rth) {
> +		if (compare_hash_inputs(&aux->fl, &rth->fl))
> +			return 0;
> +		aux = aux->u.dst.rt_next;
> +	}
> +	return ONE;
> +}
> +
>  static void rt_check_expire(void)
>  {
>  	static unsigned int rover;
>  	unsigned int i = rover, goal;
> -	struct rtable *rth, *aux, **rthp;
> +	struct rtable *rth, **rthp;
>  	unsigned long samples = 0;
>  	unsigned long sum = 0, sum2 = 0;
>  	unsigned long delta;
> @@ -835,15 +853,7 @@ nofree:
>  					 * attributes don't unfairly skew
>  					 * the length computation
>  					 */
> -					for (aux = rt_hash_table[i].chain;;) {
> -						if (aux == rth) {
> -							length += ONE;
> -							break;
> -						}
> -						if (compare_hash_inputs(&aux->fl, &rth->fl))
> -							break;
> -						aux = aux->u.dst.rt_next;
> -					}
> +					length += has_noalias(rt_hash_table[i].chain, rth);
>  					continue;
>  				}
>  			} else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
> @@ -1073,6 +1083,21 @@ work_done:
>  out:	return 0;
>  }
>  
> +/*
> + * Returns number of entries in a hash chain that have different hash_inputs
> + */
> +static int slow_chain_length(const struct rtable *head)
> +{
> +	int length = 0;
> +	const struct rtable *rth = head;
> +
> +	while (rth) {
> +		length += has_noalias(head, rth);
> +		rth = rth->u.dst.rt_next;
> +	}
> +	return length >> FRACT_BITS;
> +}
> +
>  static int rt_intern_hash(unsigned hash, struct rtable *rt,
>  			  struct rtable **rp, struct sk_buff *skb)
>  {
> @@ -1185,7 +1210,8 @@ restart:
>  			rt_free(cand);
>  		}
>  	} else {
> -		if (chain_length > rt_chain_length_max) {
> +		if (chain_length > rt_chain_length_max &&
> +		    slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
>  			struct net *net = dev_net(rt->u.dst.dev);
>  			int num = ++net->ipv4.current_rt_cache_rebuild_count;
>  			if (!rt_caching(dev_net(rt->u.dst.dev))) {
> 
> 
> 
> 
> 
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

From: Neil Horman <nhorman@tuxdriver.com>
Date: Mon, 8 Mar 2010 13:06:03 -0500

> On Mon, Mar 08, 2010 at 02:20:00PM +0100, Eric Dumazet wrote:
 ...
>> Reported-and-tested-by: Paweł Staszewski <pstaszewski@itcare.pl>
>> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
 ...
> This looks pretty reasonable to me, Thanks Eric!
> Acked-by: Neil Horman <nhorman@tuxdriver.com>

Applied, thanks guys.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html