Bloat » Cerowrt

Actually, it was this bug that was worrying me about babel
specifically, not the original forward

(the other set me off to potential trouble in the route caches)

---------- Forwarded message ----------
From: Yan, Zheng <zheng.z.yan@intel.com>
Date: Tue, Sep 6, 2011 at 12:34 AM
Subject: [PATCH] ipv6: don't use inetpeer to store metrics for routes.
To: "netdev@vger.kernel.org" <netdev@vger.kernel.org>
Cc: "davem@davemloft.net" <davem@davemloft.net>,
"eric.dumazet@gmail.com" <eric.dumazet@gmail.com>

Current IPv6 implementation uses inetpeer to store metrics for
routes. The problem of inetpeer is that it doesn't take subnet
prefix length in to consideration. If two routes have the same
address but different prefix length, they share same inetpeer.
So changing metrics of one route also affects the other. The
fix is to allocate separate metrics storage for each route.

Signed-off-by: Zheng Yan <zheng.z.yan@intel.com>
---
 net/ipv6/route.c |   33 ++++++++++++++++++++-----------
 1 files changed, 22 insertions(+), 11 deletions(-)

diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 9e69eb0..1250f90 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@ -104,6 +104,9 @ static u32 *ipv6_cow_metrics(struct dst_entry
*dst, unsigned long old)
       struct inet_peer *peer;
       u32 *p = NULL;

+       if (!(rt->dst.flags & DST_HOST))
+               return NULL;
+
       if (!rt->rt6i_peer)
               rt6_bind_peer(rt, 1);

@ -252,6 +255,9 @ static void ip6_dst_destroy(struct dst_entry *dst)
       struct inet6_dev *idev = rt->rt6i_idev;
       struct inet_peer *peer = rt->rt6i_peer;

+       if (!(rt->dst.flags & DST_HOST))
+               dst_destroy_metrics_generic(dst);
+
       if (idev != NULL) {
               rt->rt6i_idev = NULL;
               in6_dev_put(idev);
@ -723,9 +729,7 @ static struct rt6_info *rt6_alloc_cow(const struct
rt6_info *ort,
                       ipv6_addr_copy(&rt->rt6i_gateway, daddr);
               }

-               rt->rt6i_dst.plen = 128;
               rt->rt6i_flags |= RTF_CACHE;
-               rt->dst.flags |= DST_HOST;

 #ifdef CONFIG_IPV6_SUBTREES
               if (rt->rt6i_src.plen && saddr) {
@ -775,9 +779,7 @ static struct rt6_info *rt6_alloc_clone(struct
rt6_info *ort,
       struct rt6_info *rt = ip6_rt_copy(ort, daddr);

       if (rt) {
-               rt->rt6i_dst.plen = 128;
               rt->rt6i_flags |= RTF_CACHE;
-               rt->dst.flags |= DST_HOST;
               dst_set_neighbour(&rt->dst,
neigh_clone(dst_get_neighbour_raw(&ort->dst)));
       }
       return rt;
@ -1078,12 +1080,15 @ struct dst_entry *icmp6_dst_alloc(struct
net_device *dev,
                       neigh = NULL;
       }

-       rt->rt6i_idev     = idev;
+       rt->dst.flags |= DST_HOST;
+       rt->dst.output  = ip6_output;
       dst_set_neighbour(&rt->dst, neigh);
       atomic_set(&rt->dst.__refcnt, 1);
-       ipv6_addr_copy(&rt->rt6i_dst.addr, addr);
       dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 255);
-       rt->dst.output  = ip6_output;

      ipv6_addr_copy(&rt->rt6i_dst.addr, addr);
+       rt->rt6i_dst.plen = 128;
+       rt->rt6i_idev     = idev;

       spin_lock_bh(&icmp6_dst_lock);
       rt->dst.next = icmp6_dst_gc_list;
@ -1261,6 +1266,14 @ int ip6_route_add(struct fib6_config *cfg)
       if (rt->rt6i_dst.plen == 128)
              rt->dst.flags |= DST_HOST;

+       if (!(rt->dst.flags & DST_HOST) && cfg->fc_mx) {
+               u32 *metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
+               if (!metrics) {
+                       err = ENOMEM;
+                       goto out;
+               }
+               dst_init_metrics(&rt>dst, metrics, 0);
+       }
 #ifdef CONFIG_IPV6_SUBTREES
       ipv6_addr_prefix(&rt->rt6i_src.addr, &cfg->fc_src, cfg->fc_src_len);
       rt->rt6i_src.plen = cfg->fc_src_len;
@ -1607,9 +1620,6 @ void rt6_redirect(const struct in6_addr *dest,
const struct in6_addr *src,
       if (on_link)
               nrt->rt6i_flags &= ~RTF_GATEWAY;

-       nrt->rt6i_dst.plen = 128;
-       nrt->dst.flags |= DST_HOST;

       ipv6_addr_copy(&nrt>rt6i_gateway, (struct in6_addr*)neigh->primary_key);
       dst_set_neighbour(&nrt->dst, neigh_clone(neigh));

@ -1754,9 +1764,10 @ static struct rt6_info *ip6_rt_copy(const
struct rt6_info *ort,
       if (rt) {
               rt->dst.input = ort->dst.input;
               rt->dst.output = ort->dst.output;
+               rt->dst.flags |= DST_HOST;

               ipv6_addr_copy(&rt->rt6i_dst.addr, dest);
-               rt->rt6i_dst.plen = ort->rt6i_dst.plen;
+               rt->rt6i_dst.plen = 128;
               dst_copy_metrics(&rt->dst, &ort->dst);
               rt->dst.error = ort->dst.error;
               rt->rt6i_idev = ort->rt6i_idev;

well it would contain this patch if it applied against 3.0.4... looking it over now...

given the chaos in the linux kernel trees due to the linux.org server problem, I felt that fixing this bug is going to have to wait until life settles down over there some.

This patch will need to go into rc7

---------- Forwarded message ----------
From: David Miller <davem@davemloft.net>
Date: Fri, Sep 16, 2011 at 9:57 PM
Subject: Re: [PATCH] ipv6: don't use inetpeer to store metrics for routes.
To: zheng.z.yan@intel.com
Cc: netdev@vger.kernel.org, eric.dumazet@gmail.com

From: "Yan, Zheng" <zheng.z.yan@intel.com>
Date: Tue, 06 Sep 2011 15:34:30 +0800

Current IPv6 implementation uses inetpeer to store metrics for routes. The problem of inetpeer is that it doesn't take subnet prefix length in to consideration. If two routes have the same address but different prefix length, they share same inetpeer. So changing metrics of one route also affects the other. The fix is to allocate separate metrics storage for each route.

Signed-off-by: Zheng Yan <zheng.z.yan@intel.com>

Applied, thanks for fixing this bug.