]> bbs.cooldavid.org Git - net-next-2.6.git/commitdiff
[NETNS][IPV6] route6 - dynamically allocate ip6_dst_ops
authorDaniel Lezcano <dlezcano@fr.ibm.com>
Tue, 4 Mar 2008 21:48:53 +0000 (13:48 -0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 4 Mar 2008 21:48:53 +0000 (13:48 -0800)
ip6_dst_ops is dynamically allocated in init and exit functions.  That
provides the ability to do multiple instanciations of this structure.

This will be needed for network namespaces, indeed dst_ops stores data
that are required to be per namespace: entries and gc_thresh.

Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/route.c

index 7ff66cebe77cff96777775a6e0bed97cd0be3f63..fa014d701c1addc87491077a9da31ea932b3cea3 100644 (file)
@@ -97,7 +97,7 @@ static struct rt6_info *rt6_get_route_info(struct net *net,
                                           struct in6_addr *gwaddr, int ifindex);
 #endif
 
-static struct dst_ops ip6_dst_ops = {
+static struct dst_ops ip6_dst_ops_template = {
        .family                 =       AF_INET6,
        .protocol               =       __constant_htons(ETH_P_IPV6),
        .gc                     =       ip6_dst_gc,
@@ -113,6 +113,8 @@ static struct dst_ops ip6_dst_ops = {
        .entries                =       ATOMIC_INIT(0),
 };
 
+static struct dst_ops *ip6_dst_ops;
+
 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
 {
 }
@@ -137,7 +139,6 @@ static struct rt6_info ip6_null_entry_template = {
                        .metrics        = { [RTAX_HOPLIMIT - 1] = 255, },
                        .input          = ip6_pkt_discard,
                        .output         = ip6_pkt_discard_out,
-                       .ops            = &ip6_dst_ops,
                }
        },
        .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
@@ -160,7 +161,6 @@ struct rt6_info ip6_prohibit_entry_template = {
                        .metrics        = { [RTAX_HOPLIMIT - 1] = 255, },
                        .input          = ip6_pkt_prohibit,
                        .output         = ip6_pkt_prohibit_out,
-                       .ops            = &ip6_dst_ops,
                }
        },
        .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
@@ -178,7 +178,6 @@ static struct rt6_info ip6_blk_hole_entry_template = {
                        .metrics        = { [RTAX_HOPLIMIT - 1] = 255, },
                        .input          = dst_discard,
                        .output         = dst_discard,
-                       .ops            = &ip6_dst_ops,
                }
        },
        .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
@@ -191,7 +190,7 @@ static struct rt6_info ip6_blk_hole_entry_template = {
 /* allocate dst with ip6_dst_ops */
 static __inline__ struct rt6_info *ip6_dst_alloc(void)
 {
-       return (struct rt6_info *)dst_alloc(&ip6_dst_ops);
+       return (struct rt6_info *)dst_alloc(ip6_dst_ops);
 }
 
 static void ip6_dst_destroy(struct dst_entry *dst)
@@ -1000,18 +999,18 @@ static int ip6_dst_gc(struct dst_ops *ops)
        unsigned long now = jiffies;
 
        if (time_after(last_gc + init_net.ipv6.sysctl.ip6_rt_gc_min_interval, now) &&
-           atomic_read(&ip6_dst_ops.entries) <= init_net.ipv6.sysctl.ip6_rt_max_size)
+           atomic_read(&ip6_dst_ops->entries) <= init_net.ipv6.sysctl.ip6_rt_max_size)
                goto out;
 
        expire++;
        fib6_run_gc(expire, &init_net);
        last_gc = now;
-       if (atomic_read(&ip6_dst_ops.entries) < ip6_dst_ops.gc_thresh)
+       if (atomic_read(&ip6_dst_ops->entries) < ip6_dst_ops->gc_thresh)
                expire = init_net.ipv6.sysctl.ip6_rt_gc_timeout>>1;
 
 out:
        expire -= expire>>init_net.ipv6.sysctl.ip6_rt_gc_elasticity;
-       return (atomic_read(&ip6_dst_ops.entries) > init_net.ipv6.sysctl.ip6_rt_max_size);
+       return (atomic_read(&ip6_dst_ops->entries) > init_net.ipv6.sysctl.ip6_rt_max_size);
 }
 
 /* Clean host part of a prefix. Not necessary in radix tree,
@@ -2408,7 +2407,7 @@ static int rt6_stats_seq_show(struct seq_file *seq, void *v)
                   net->ipv6.rt6_stats->fib_rt_alloc,
                   net->ipv6.rt6_stats->fib_rt_entries,
                   net->ipv6.rt6_stats->fib_rt_cache,
-                  atomic_read(&ip6_dst_ops.entries),
+                  atomic_read(&ip6_dst_ops->entries),
                   net->ipv6.rt6_stats->fib_discarded_routes);
 
        return 0;
@@ -2464,7 +2463,7 @@ ctl_table ipv6_route_table_template[] = {
        {
                .ctl_name       =       NET_IPV6_ROUTE_GC_THRESH,
                .procname       =       "gc_thresh",
-               .data           =       &ip6_dst_ops.gc_thresh,
+               .data           =       &ip6_dst_ops_template.gc_thresh,
                .maxlen         =       sizeof(int),
                .mode           =       0644,
                .proc_handler   =       &proc_dointvec,
@@ -2553,8 +2552,7 @@ struct ctl_table *ipv6_route_sysctl_init(struct net *net)
 
        if (table) {
                table[0].data = &net->ipv6.sysctl.flush_delay;
-               /* table[1].data will be handled when we have
-                  routes per namespace */
+               table[1].data = &ip6_dst_ops_template.gc_thresh;
                table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
                table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
                table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
@@ -2580,6 +2578,7 @@ static int ip6_route_net_init(struct net *net)
                goto out;
        net->ipv6.ip6_null_entry->u.dst.path =
                (struct dst_entry *)net->ipv6.ip6_null_entry;
+       net->ipv6.ip6_null_entry->u.dst.ops = ip6_dst_ops;
 
 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
        net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
@@ -2591,6 +2590,7 @@ static int ip6_route_net_init(struct net *net)
        }
        net->ipv6.ip6_prohibit_entry->u.dst.path =
                (struct dst_entry *)net->ipv6.ip6_prohibit_entry;
+       net->ipv6.ip6_prohibit_entry->u.dst.ops = ip6_dst_ops;
 
        net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
                                               sizeof(*net->ipv6.ip6_blk_hole_entry),
@@ -2602,6 +2602,7 @@ static int ip6_route_net_init(struct net *net)
        }
        net->ipv6.ip6_blk_hole_entry->u.dst.path =
                (struct dst_entry *)net->ipv6.ip6_blk_hole_entry;
+       net->ipv6.ip6_blk_hole_entry->u.dst.ops = ip6_dst_ops;
 #endif
 
 #ifdef CONFIG_PROC_FS
@@ -2640,13 +2641,20 @@ int __init ip6_route_init(void)
 {
        int ret;
 
-       ip6_dst_ops.kmem_cachep =
+       ip6_dst_ops = kmemdup(&ip6_dst_ops_template,
+                             sizeof(*ip6_dst_ops), GFP_KERNEL);
+       if (!ip6_dst_ops)
+               return -ENOMEM;
+
+       ret = -ENOMEM;
+       ip6_dst_ops_template.kmem_cachep =
                kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
                                  SLAB_HWCACHE_ALIGN, NULL);
-       if (!ip6_dst_ops.kmem_cachep)
-               return -ENOMEM;
+       if (!ip6_dst_ops_template.kmem_cachep)
+               goto out_ip6_dst_ops;
 
-       ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops.kmem_cachep;
+       ip6_dst_ops->kmem_cachep = ip6_dst_ops_template.kmem_cachep;
+       ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
 
        ret = register_pernet_subsys(&ip6_route_net_ops);
        if (ret)
@@ -2697,7 +2705,9 @@ out_fib6_init:
 out_register_subsys:
        unregister_pernet_subsys(&ip6_route_net_ops);
 out_kmem_cache:
-       kmem_cache_destroy(ip6_dst_ops.kmem_cachep);
+       kmem_cache_destroy(ip6_dst_ops->kmem_cachep);
+out_ip6_dst_ops:
+       kfree(ip6_dst_ops);
        goto out;
 }
 
@@ -2708,5 +2718,6 @@ void ip6_route_cleanup(void)
        xfrm6_fini();
        fib6_gc_cleanup();
        unregister_pernet_subsys(&ip6_route_net_ops);
-       kmem_cache_destroy(ip6_dst_ops.kmem_cachep);
+       kmem_cache_destroy(ip6_dst_ops->kmem_cachep);
+       kfree(ip6_dst_ops);
 }