netns: Use net_eq() to compare net-namespaces for optimization.
[safe/jmp/linux-2.6] / net / ipv4 / devinet.c
index 82def2c..2e667e2 100644 (file)
@@ -1,8 +1,6 @@
 /*
  *     NET3    IP device support routines.
  *
- *     Version: $Id: devinet.c,v 1.44 2001/10/31 21:55:54 davem Exp $
- *
  *             This program is free software; you can redistribute it and/or
  *             modify it under the terms of the GNU General Public License
  *             as published by the Free Software Foundation; either version
@@ -62,8 +60,9 @@
 #include <net/route.h>
 #include <net/ip_fib.h>
 #include <net/rtnetlink.h>
+#include <net/net_namespace.h>
 
-struct ipv4_devconf ipv4_devconf = {
+static struct ipv4_devconf ipv4_devconf = {
        .data = {
                [NET_IPV4_CONF_ACCEPT_REDIRECTS - 1] = 1,
                [NET_IPV4_CONF_SEND_REDIRECTS - 1] = 1,
@@ -82,13 +81,13 @@ static struct ipv4_devconf ipv4_devconf_dflt = {
        },
 };
 
-#define IPV4_DEVCONF_DFLT(attr) IPV4_DEVCONF(ipv4_devconf_dflt, attr)
+#define IPV4_DEVCONF_DFLT(net, attr) \
+       IPV4_DEVCONF((*net->ipv4.devconf_dflt), attr)
 
 static const struct nla_policy ifa_ipv4_policy[IFA_MAX+1] = {
        [IFA_LOCAL]             = { .type = NLA_U32 },
        [IFA_ADDRESS]           = { .type = NLA_U32 },
        [IFA_BROADCAST]         = { .type = NLA_U32 },
-       [IFA_ANYCAST]           = { .type = NLA_U32 },
        [IFA_LABEL]             = { .type = NLA_STRING, .len = IFNAMSIZ - 1 },
 };
 
@@ -163,11 +162,14 @@ static struct in_device *inetdev_init(struct net_device *dev)
        if (!in_dev)
                goto out;
        INIT_RCU_HEAD(&in_dev->rcu_head);
-       memcpy(&in_dev->cnf, &ipv4_devconf_dflt, sizeof(in_dev->cnf));
+       memcpy(&in_dev->cnf, dev_net(dev)->ipv4.devconf_dflt,
+                       sizeof(in_dev->cnf));
        in_dev->cnf.sysctl = NULL;
        in_dev->dev = dev;
        if ((in_dev->arp_parms = neigh_parms_alloc(dev, &arp_tbl)) == NULL)
                goto out_kfree;
+       if (IPV4_DEVCONF(in_dev->cnf, FORWARDING))
+               dev_disable_lro(dev);
        /* Reference in_dev->dev */
        dev_hold(dev);
        /* Account for reference dev->ip_ptr (below) */
@@ -401,17 +403,17 @@ static int inet_set_ifa(struct net_device *dev, struct in_ifaddr *ifa)
                in_dev_hold(in_dev);
                ifa->ifa_dev = in_dev;
        }
-       if (LOOPBACK(ifa->ifa_local))
+       if (ipv4_is_loopback(ifa->ifa_local))
                ifa->ifa_scope = RT_SCOPE_HOST;
        return inet_insert_ifa(ifa);
 }
 
-struct in_device *inetdev_by_index(int ifindex)
+struct in_device *inetdev_by_index(struct net *net, int ifindex)
 {
        struct net_device *dev;
        struct in_device *in_dev = NULL;
        read_lock(&dev_base_lock);
-       dev = __dev_get_by_index(&init_net, ifindex);
+       dev = __dev_get_by_index(net, ifindex);
        if (dev)
                in_dev = in_dev_get(dev);
        read_unlock(&dev_base_lock);
@@ -434,7 +436,7 @@ struct in_ifaddr *inet_ifa_byprefix(struct in_device *in_dev, __be32 prefix,
 
 static int inet_rtm_deladdr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 {
-       struct net *net = skb->sk->sk_net;
+       struct net *net = sock_net(skb->sk);
        struct nlattr *tb[IFA_MAX+1];
        struct in_device *in_dev;
        struct ifaddrmsg *ifm;
@@ -443,15 +445,12 @@ static int inet_rtm_deladdr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg
 
        ASSERT_RTNL();
 
-       if (net != &init_net)
-               return -EINVAL;
-
        err = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_ipv4_policy);
        if (err < 0)
                goto errout;
 
        ifm = nlmsg_data(nlh);
-       in_dev = inetdev_by_index(ifm->ifa_index);
+       in_dev = inetdev_by_index(net, ifm->ifa_index);
        if (in_dev == NULL) {
                err = -ENODEV;
                goto errout;
@@ -482,46 +481,41 @@ errout:
        return err;
 }
 
-static struct in_ifaddr *rtm_to_ifaddr(struct nlmsghdr *nlh)
+static struct in_ifaddr *rtm_to_ifaddr(struct net *net, struct nlmsghdr *nlh)
 {
        struct nlattr *tb[IFA_MAX+1];
        struct in_ifaddr *ifa;
        struct ifaddrmsg *ifm;
        struct net_device *dev;
        struct in_device *in_dev;
-       int err = -EINVAL;
+       int err;
 
        err = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_ipv4_policy);
        if (err < 0)
                goto errout;
 
        ifm = nlmsg_data(nlh);
-       if (ifm->ifa_prefixlen > 32 || tb[IFA_LOCAL] == NULL) {
-               err = -EINVAL;
+       err = -EINVAL;
+       if (ifm->ifa_prefixlen > 32 || tb[IFA_LOCAL] == NULL)
                goto errout;
-       }
 
-       dev = __dev_get_by_index(&init_net, ifm->ifa_index);
-       if (dev == NULL) {
-               err = -ENODEV;
+       dev = __dev_get_by_index(net, ifm->ifa_index);
+       err = -ENODEV;
+       if (dev == NULL)
                goto errout;
-       }
 
        in_dev = __in_dev_get_rtnl(dev);
-       if (in_dev == NULL) {
-               err = -ENOBUFS;
+       err = -ENOBUFS;
+       if (in_dev == NULL)
                goto errout;
-       }
 
        ifa = inet_alloc_ifa();
-       if (ifa == NULL) {
+       if (ifa == NULL)
                /*
                 * A potential indev allocation can be left alive, it stays
                 * assigned to its device and is destroy with it.
                 */
-               err = -ENOBUFS;
                goto errout;
-       }
 
        ipv4_devconf_setall(in_dev);
        in_dev_hold(in_dev);
@@ -541,9 +535,6 @@ static struct in_ifaddr *rtm_to_ifaddr(struct nlmsghdr *nlh)
        if (tb[IFA_BROADCAST])
                ifa->ifa_broadcast = nla_get_be32(tb[IFA_BROADCAST]);
 
-       if (tb[IFA_ANYCAST])
-               ifa->ifa_anycast = nla_get_be32(tb[IFA_ANYCAST]);
-
        if (tb[IFA_LABEL])
                nla_strlcpy(ifa->ifa_label, tb[IFA_LABEL], IFNAMSIZ);
        else
@@ -557,15 +548,12 @@ errout:
 
 static int inet_rtm_newaddr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 {
-       struct net *net = skb->sk->sk_net;
+       struct net *net = sock_net(skb->sk);
        struct in_ifaddr *ifa;
 
        ASSERT_RTNL();
 
-       if (net != &init_net)
-               return -EINVAL;
-
-       ifa = rtm_to_ifaddr(nlh);
+       ifa = rtm_to_ifaddr(net, nlh);
        if (IS_ERR(ifa))
                return PTR_ERR(ifa);
 
@@ -580,7 +568,7 @@ static __inline__ int inet_abc_len(__be32 addr)
 {
        int rc = -1;    /* Something else, probably a multicast. */
 
-       if (ZERONET(addr))
+       if (ipv4_is_zeronet(addr))
                rc = 0;
        else {
                __u32 haddr = ntohl(addr);
@@ -597,7 +585,7 @@ static __inline__ int inet_abc_len(__be32 addr)
 }
 
 
-int devinet_ioctl(unsigned int cmd, void __user *arg)
+int devinet_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 {
        struct ifreq ifr;
        struct sockaddr_in sin_orig;
@@ -626,7 +614,7 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
                *colon = 0;
 
 #ifdef CONFIG_KMOD
-       dev_load(&init_net, ifr.ifr_name);
+       dev_load(net, ifr.ifr_name);
 #endif
 
        switch (cmd) {
@@ -667,7 +655,7 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
        rtnl_lock();
 
        ret = -ENODEV;
-       if ((dev = __dev_get_by_name(&init_net, ifr.ifr_name)) == NULL)
+       if ((dev = __dev_get_by_name(net, ifr.ifr_name)) == NULL)
                goto done;
 
        if (colon)
@@ -753,7 +741,7 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
                                break;
                        inet_del_ifa(in_dev, ifap, 0);
                        ifa->ifa_broadcast = 0;
-                       ifa->ifa_anycast = 0;
+                       ifa->ifa_scope = 0;
                }
 
                ifa->ifa_address = ifa->ifa_local = sin->sin_addr.s_addr;
@@ -879,6 +867,7 @@ __be32 inet_select_addr(const struct net_device *dev, __be32 dst, int scope)
 {
        __be32 addr = 0;
        struct in_device *in_dev;
+       struct net *net = dev_net(dev);
 
        rcu_read_lock();
        in_dev = __in_dev_get_rcu(dev);
@@ -907,7 +896,7 @@ no_in_dev:
         */
        read_lock(&dev_base_lock);
        rcu_read_lock();
-       for_each_netdev(&init_net, dev) {
+       for_each_netdev(net, dev) {
                if ((in_dev = __in_dev_get_rcu(dev)) == NULL)
                        continue;
 
@@ -965,28 +954,25 @@ static __be32 confirm_addr_indev(struct in_device *in_dev, __be32 dst,
 
 /*
  * Confirm that local IP address exists using wildcards:
- * - dev: only on this interface, 0=any interface
+ * - in_dev: only on this interface, 0=any interface
  * - dst: only in the same subnet as dst, 0=any dst
  * - local: address, 0=autoselect the local address
  * - scope: maximum allowed scope value for the local address
  */
-__be32 inet_confirm_addr(const struct net_device *dev, __be32 dst, __be32 local, int scope)
+__be32 inet_confirm_addr(struct in_device *in_dev,
+                        __be32 dst, __be32 local, int scope)
 {
        __be32 addr = 0;
-       struct in_device *in_dev;
-
-       if (dev) {
-               rcu_read_lock();
-               if ((in_dev = __in_dev_get_rcu(dev)))
-                       addr = confirm_addr_indev(in_dev, dst, local, scope);
-               rcu_read_unlock();
+       struct net_device *dev;
+       struct net *net;
 
-               return addr;
-       }
+       if (scope != RT_SCOPE_LINK)
+               return confirm_addr_indev(in_dev, dst, local, scope);
 
+       net = dev_net(in_dev->dev);
        read_lock(&dev_base_lock);
        rcu_read_lock();
-       for_each_netdev(&init_net, dev) {
+       for_each_netdev(net, dev) {
                if ((in_dev = __in_dev_get_rcu(dev))) {
                        addr = confirm_addr_indev(in_dev, dst, local, scope);
                        if (addr)
@@ -1027,7 +1013,7 @@ static void inetdev_changename(struct net_device *dev, struct in_device *in_dev)
                memcpy(old, ifa->ifa_label, IFNAMSIZ);
                memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
                if (named++ == 0)
-                       continue;
+                       goto skip;
                dot = strchr(old, ':');
                if (dot == NULL) {
                        sprintf(old, ":%d", named);
@@ -1038,6 +1024,8 @@ static void inetdev_changename(struct net_device *dev, struct in_device *in_dev)
                } else {
                        strcpy(ifa->ifa_label + (IFNAMSIZ - strlen(dot) - 1), dot);
                }
+skip:
+               rtmsg_ifa(RTM_NEWADDR, ifa, NULL, 0);
        }
 }
 
@@ -1049,9 +1037,6 @@ static int inetdev_event(struct notifier_block *this, unsigned long event,
        struct net_device *dev = ptr;
        struct in_device *in_dev = __in_dev_get_rtnl(dev);
 
-       if (dev->nd_net != &init_net)
-               return NOTIFY_DONE;
-
        ASSERT_RTNL();
 
        if (!in_dev) {
@@ -1125,7 +1110,6 @@ static inline size_t inet_nlmsg_size(void)
               + nla_total_size(4) /* IFA_ADDRESS */
               + nla_total_size(4) /* IFA_LOCAL */
               + nla_total_size(4) /* IFA_BROADCAST */
-              + nla_total_size(4) /* IFA_ANYCAST */
               + nla_total_size(IFNAMSIZ); /* IFA_LABEL */
 }
 
@@ -1155,9 +1139,6 @@ static int inet_fill_ifaddr(struct sk_buff *skb, struct in_ifaddr *ifa,
        if (ifa->ifa_broadcast)
                NLA_PUT_BE32(skb, IFA_BROADCAST, ifa->ifa_broadcast);
 
-       if (ifa->ifa_anycast)
-               NLA_PUT_BE32(skb, IFA_ANYCAST, ifa->ifa_anycast);
-
        if (ifa->ifa_label[0])
                NLA_PUT_STRING(skb, IFA_LABEL, ifa->ifa_label);
 
@@ -1170,19 +1151,16 @@ nla_put_failure:
 
 static int inet_dump_ifaddr(struct sk_buff *skb, struct netlink_callback *cb)
 {
-       struct net *net = skb->sk->sk_net;
+       struct net *net = sock_net(skb->sk);
        int idx, ip_idx;
        struct net_device *dev;
        struct in_device *in_dev;
        struct in_ifaddr *ifa;
        int s_ip_idx, s_idx = cb->args[0];
 
-       if (net != &init_net)
-               return 0;
-
        s_ip_idx = ip_idx = cb->args[1];
        idx = 0;
-       for_each_netdev(&init_net, dev) {
+       for_each_netdev(net, dev) {
                if (idx < s_idx)
                        goto cont;
                if (idx > s_idx)
@@ -1216,7 +1194,9 @@ static void rtmsg_ifa(int event, struct in_ifaddr* ifa, struct nlmsghdr *nlh,
        struct sk_buff *skb;
        u32 seq = nlh ? nlh->nlmsg_seq : 0;
        int err = -ENOBUFS;
+       struct net *net;
 
+       net = dev_net(ifa->ifa_dev->dev);
        skb = nlmsg_new(inet_nlmsg_size(), GFP_KERNEL);
        if (skb == NULL)
                goto errout;
@@ -1228,41 +1208,43 @@ static void rtmsg_ifa(int event, struct in_ifaddr* ifa, struct nlmsghdr *nlh,
                kfree_skb(skb);
                goto errout;
        }
-       err = rtnl_notify(skb, &init_net, pid, RTNLGRP_IPV4_IFADDR, nlh, GFP_KERNEL);
+       err = rtnl_notify(skb, net, pid, RTNLGRP_IPV4_IFADDR, nlh, GFP_KERNEL);
 errout:
        if (err < 0)
-               rtnl_set_sk_err(&init_net, RTNLGRP_IPV4_IFADDR, err);
+               rtnl_set_sk_err(net, RTNLGRP_IPV4_IFADDR, err);
 }
 
 #ifdef CONFIG_SYSCTL
 
-static void devinet_copy_dflt_conf(int i)
+static void devinet_copy_dflt_conf(struct net *net, int i)
 {
        struct net_device *dev;
 
        read_lock(&dev_base_lock);
-       for_each_netdev(&init_net, dev) {
+       for_each_netdev(net, dev) {
                struct in_device *in_dev;
                rcu_read_lock();
                in_dev = __in_dev_get_rcu(dev);
                if (in_dev && !test_bit(i, in_dev->cnf.state))
-                       in_dev->cnf.data[i] = ipv4_devconf_dflt.data[i];
+                       in_dev->cnf.data[i] = net->ipv4.devconf_dflt->data[i];
                rcu_read_unlock();
        }
        read_unlock(&dev_base_lock);
 }
 
-static void inet_forward_change(void)
+static void inet_forward_change(struct net *net)
 {
        struct net_device *dev;
-       int on = IPV4_DEVCONF_ALL(FORWARDING);
+       int on = IPV4_DEVCONF_ALL(net, FORWARDING);
 
-       IPV4_DEVCONF_ALL(ACCEPT_REDIRECTS) = !on;
-       IPV4_DEVCONF_DFLT(FORWARDING) = on;
+       IPV4_DEVCONF_ALL(net, ACCEPT_REDIRECTS) = !on;
+       IPV4_DEVCONF_DFLT(net, FORWARDING) = on;
 
        read_lock(&dev_base_lock);
-       for_each_netdev(&init_net, dev) {
+       for_each_netdev(net, dev) {
                struct in_device *in_dev;
+               if (on)
+                       dev_disable_lro(dev);
                rcu_read_lock();
                in_dev = __in_dev_get_rcu(dev);
                if (in_dev)
@@ -1270,8 +1252,6 @@ static void inet_forward_change(void)
                rcu_read_unlock();
        }
        read_unlock(&dev_base_lock);
-
-       rt_cache_flush(0);
 }
 
 static int devinet_conf_proc(ctl_table *ctl, int write,
@@ -1282,12 +1262,13 @@ static int devinet_conf_proc(ctl_table *ctl, int write,
 
        if (write) {
                struct ipv4_devconf *cnf = ctl->extra1;
+               struct net *net = ctl->extra2;
                int i = (int *)ctl->data - cnf->data;
 
                set_bit(i, cnf->state);
 
-               if (cnf == &ipv4_devconf_dflt)
-                       devinet_copy_dflt_conf(i);
+               if (cnf == net->ipv4.devconf_dflt)
+                       devinet_copy_dflt_conf(net, i);
        }
 
        return ret;
@@ -1298,6 +1279,7 @@ static int devinet_conf_sysctl(ctl_table *table, int __user *name, int nlen,
                               void __user *newval, size_t newlen)
 {
        struct ipv4_devconf *cnf;
+       struct net *net;
        int *valp = table->data;
        int new;
        int i;
@@ -1333,12 +1315,13 @@ static int devinet_conf_sysctl(ctl_table *table, int __user *name, int nlen,
        *valp = new;
 
        cnf = table->extra1;
+       net = table->extra2;
        i = (int *)table->data - cnf->data;
 
        set_bit(i, cnf->state);
 
-       if (cnf == &ipv4_devconf_dflt)
-               devinet_copy_dflt_conf(i);
+       if (cnf == net->ipv4.devconf_dflt)
+               devinet_copy_dflt_conf(net, i);
 
        return 1;
 }
@@ -1352,10 +1335,21 @@ static int devinet_sysctl_forward(ctl_table *ctl, int write,
        int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 
        if (write && *valp != val) {
-               if (valp == &IPV4_DEVCONF_ALL(FORWARDING))
-                       inet_forward_change();
-               else if (valp != &IPV4_DEVCONF_DFLT(FORWARDING))
-                       rt_cache_flush(0);
+               struct net *net = ctl->extra2;
+
+               if (valp != &IPV4_DEVCONF_DFLT(net, FORWARDING)) {
+                       rtnl_lock();
+                       if (valp == &IPV4_DEVCONF_ALL(net, FORWARDING)) {
+                               inet_forward_change(net);
+                       } else if (*valp) {
+                               struct ipv4_devconf *cnf = ctl->extra1;
+                               struct in_device *idev =
+                                       container_of(cnf, struct in_device, cnf);
+                               dev_disable_lro(idev->dev);
+                       }
+                       rtnl_unlock();
+                       rt_cache_flush(net, 0);
+               }
        }
 
        return ret;
@@ -1368,9 +1362,10 @@ int ipv4_doint_and_flush(ctl_table *ctl, int write,
        int *valp = ctl->data;
        int val = *valp;
        int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
+       struct net *net = ctl->extra2;
 
        if (write && *valp != val)
-               rt_cache_flush(0);
+               rt_cache_flush(net, 0);
 
        return ret;
 }
@@ -1381,9 +1376,10 @@ int ipv4_doint_and_flush_strategy(ctl_table *table, int __user *name, int nlen,
 {
        int ret = devinet_conf_sysctl(table, name, nlen, oldval, oldlenp,
                                      newval, newlen);
+       struct net *net = table->extra2;
 
        if (ret == 1)
-               rt_cache_flush(0);
+               rt_cache_flush(net, 0);
 
        return ret;
 }
@@ -1477,6 +1473,7 @@ static int __devinet_sysctl_register(struct net *net, char *dev_name,
        for (i = 0; i < ARRAY_SIZE(t->devinet_vars) - 1; i++) {
                t->devinet_vars[i].data += (char *)p - (char *)&ipv4_devconf;
                t->devinet_vars[i].extra1 = p;
+               t->devinet_vars[i].extra2 = net;
        }
 
        /*
@@ -1491,7 +1488,7 @@ static int __devinet_sysctl_register(struct net *net, char *dev_name,
        devinet_ctl_path[DEVINET_CTL_PATH_DEV].procname = t->dev_name;
        devinet_ctl_path[DEVINET_CTL_PATH_DEV].ctl_name = ctl_name;
 
-       t->sysctl_header = register_sysctl_paths(devinet_ctl_path,
+       t->sysctl_header = register_net_sysctl_table(net, devinet_ctl_path,
                        t->devinet_vars);
        if (!t->sysctl_header)
                goto free_procname;
@@ -1524,8 +1521,8 @@ static void devinet_sysctl_register(struct in_device *idev)
 {
        neigh_sysctl_register(idev->dev, idev->arp_parms, NET_IPV4,
                        NET_IPV4_NEIGH, "ipv4", NULL, NULL);
-       __devinet_sysctl_register(idev->dev->name, idev->dev->ifindex,
-                       &idev->cnf);
+       __devinet_sysctl_register(dev_net(idev->dev), idev->dev->name,
+                       idev->dev->ifindex, &idev->cnf);
 }
 
 static void devinet_sysctl_unregister(struct in_device *idev)
@@ -1533,7 +1530,6 @@ static void devinet_sysctl_unregister(struct in_device *idev)
        __devinet_sysctl_unregister(&idev->cnf);
        neigh_sysctl_unregister(idev->arp_parms);
 }
-#endif
 
 static struct ctl_table ctl_forward_entry[] = {
        {
@@ -1546,31 +1542,122 @@ static struct ctl_table ctl_forward_entry[] = {
                .proc_handler   = devinet_sysctl_forward,
                .strategy       = devinet_conf_sysctl,
                .extra1         = &ipv4_devconf,
+               .extra2         = &init_net,
        },
        { },
 };
 
-static __initdata struct ctl_path net_ipv4_path[] = {
+static __net_initdata struct ctl_path net_ipv4_path[] = {
        { .procname = "net", .ctl_name = CTL_NET, },
        { .procname = "ipv4", .ctl_name = NET_IPV4, },
        { },
 };
+#endif
+
+static __net_init int devinet_init_net(struct net *net)
+{
+       int err;
+       struct ipv4_devconf *all, *dflt;
+#ifdef CONFIG_SYSCTL
+       struct ctl_table *tbl = ctl_forward_entry;
+       struct ctl_table_header *forw_hdr;
+#endif
+
+       err = -ENOMEM;
+       all = &ipv4_devconf;
+       dflt = &ipv4_devconf_dflt;
+
+       if (net != &init_net) {
+               all = kmemdup(all, sizeof(ipv4_devconf), GFP_KERNEL);
+               if (all == NULL)
+                       goto err_alloc_all;
+
+               dflt = kmemdup(dflt, sizeof(ipv4_devconf_dflt), GFP_KERNEL);
+               if (dflt == NULL)
+                       goto err_alloc_dflt;
+
+#ifdef CONFIG_SYSCTL
+               tbl = kmemdup(tbl, sizeof(ctl_forward_entry), GFP_KERNEL);
+               if (tbl == NULL)
+                       goto err_alloc_ctl;
+
+               tbl[0].data = &all->data[NET_IPV4_CONF_FORWARDING - 1];
+               tbl[0].extra1 = all;
+               tbl[0].extra2 = net;
+#endif
+       }
+
+#ifdef CONFIG_SYSCTL
+       err = __devinet_sysctl_register(net, "all",
+                       NET_PROTO_CONF_ALL, all);
+       if (err < 0)
+               goto err_reg_all;
+
+       err = __devinet_sysctl_register(net, "default",
+                       NET_PROTO_CONF_DEFAULT, dflt);
+       if (err < 0)
+               goto err_reg_dflt;
+
+       err = -ENOMEM;
+       forw_hdr = register_net_sysctl_table(net, net_ipv4_path, tbl);
+       if (forw_hdr == NULL)
+               goto err_reg_ctl;
+       net->ipv4.forw_hdr = forw_hdr;
+#endif
+
+       net->ipv4.devconf_all = all;
+       net->ipv4.devconf_dflt = dflt;
+       return 0;
+
+#ifdef CONFIG_SYSCTL
+err_reg_ctl:
+       __devinet_sysctl_unregister(dflt);
+err_reg_dflt:
+       __devinet_sysctl_unregister(all);
+err_reg_all:
+       if (tbl != ctl_forward_entry)
+               kfree(tbl);
+err_alloc_ctl:
+#endif
+       if (dflt != &ipv4_devconf_dflt)
+               kfree(dflt);
+err_alloc_dflt:
+       if (all != &ipv4_devconf)
+               kfree(all);
+err_alloc_all:
+       return err;
+}
+
+static __net_exit void devinet_exit_net(struct net *net)
+{
+#ifdef CONFIG_SYSCTL
+       struct ctl_table *tbl;
+
+       tbl = net->ipv4.forw_hdr->ctl_table_arg;
+       unregister_net_sysctl_table(net->ipv4.forw_hdr);
+       __devinet_sysctl_unregister(net->ipv4.devconf_dflt);
+       __devinet_sysctl_unregister(net->ipv4.devconf_all);
+       kfree(tbl);
+#endif
+       kfree(net->ipv4.devconf_dflt);
+       kfree(net->ipv4.devconf_all);
+}
+
+static __net_initdata struct pernet_operations devinet_ops = {
+       .init = devinet_init_net,
+       .exit = devinet_exit_net,
+};
 
 void __init devinet_init(void)
 {
+       register_pernet_subsys(&devinet_ops);
+
        register_gifconf(PF_INET, inet_gifconf);
        register_netdevice_notifier(&ip_netdev_notifier);
 
        rtnl_register(PF_INET, RTM_NEWADDR, inet_rtm_newaddr, NULL);
        rtnl_register(PF_INET, RTM_DELADDR, inet_rtm_deladdr, NULL);
        rtnl_register(PF_INET, RTM_GETADDR, NULL, inet_dump_ifaddr);
-#ifdef CONFIG_SYSCTL
-       __devinet_sysctl_register("all", NET_PROTO_CONF_ALL,
-                       &ipv4_devconf);
-       __devinet_sysctl_register("default", NET_PROTO_CONF_DEFAULT,
-                       &ipv4_devconf_dflt);
-       register_sysctl_paths(net_ipv4_path, ctl_forward_entry);
-#endif
 }
 
 EXPORT_SYMBOL(in_dev_finish_destroy);