fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / net / ipv4 / devinet.c
index ec968e4..342e641 100644 (file)
@@ -9,7 +9,7 @@
  *             2 of the License, or (at your option) any later version.
  *
  *     Derived from the IP parts of dev.c 1.0.19
- *             Authors:        Ross Biro, <bir7@leland.Stanford.Edu>
+ *             Authors:        Ross Biro
  *                             Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  *                             Mark Evans, <evansmp@uhura.aston.ac.uk>
  *
  *                                     if no match found.
  */
 
-#include <linux/config.h>
 
 #include <asm/uaccess.h>
 #include <asm/system.h>
 #include <linux/bitops.h>
+#include <linux/capability.h>
 #include <linux/module.h>
 #include <linux/types.h>
 #include <linux/kernel.h>
@@ -43,6 +43,7 @@
 #include <linux/in.h>
 #include <linux/errno.h>
 #include <linux/interrupt.h>
+#include <linux/if_addr.h>
 #include <linux/if_ether.h>
 #include <linux/inet.h>
 #include <linux/netdevice.h>
 #include <linux/sysctl.h>
 #endif
 #include <linux/kmod.h>
+#include <linux/vs_context.h>
 
+#include <net/arp.h>
 #include <net/ip.h>
 #include <net/route.h>
 #include <net/ip_fib.h>
+#include <net/netlink.h>
 
 struct ipv4_devconf ipv4_devconf = {
        .accept_redirects = 1,
@@ -77,9 +81,17 @@ static struct ipv4_devconf ipv4_devconf_dflt = {
        .accept_source_route = 1,
 };
 
-static void rtmsg_ifa(int event, struct in_ifaddr *);
+static struct nla_policy ifa_ipv4_policy[IFA_MAX+1] __read_mostly = {
+       [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 },
+};
+
+static void rtmsg_ifa(int event, struct in_ifaddr *, struct nlmsghdr *, u32);
 
-static struct notifier_block *inetaddr_chain;
+static BLOCKING_NOTIFIER_HEAD(inetaddr_chain);
 static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
                         int destroy);
 #ifdef CONFIG_SYSCTL
@@ -92,10 +104,9 @@ static void devinet_sysctl_unregister(struct ipv4_devconf *p);
 
 static struct in_ifaddr *inet_alloc_ifa(void)
 {
-       struct in_ifaddr *ifa = kmalloc(sizeof(*ifa), GFP_KERNEL);
+       struct in_ifaddr *ifa = kzalloc(sizeof(*ifa), GFP_KERNEL);
 
        if (ifa) {
-               memset(ifa, 0, sizeof(*ifa));
                INIT_RCU_HEAD(&ifa->rcu_head);
        }
 
@@ -139,10 +150,9 @@ struct in_device *inetdev_init(struct net_device *dev)
 
        ASSERT_RTNL();
 
-       in_dev = kmalloc(sizeof(*in_dev), GFP_KERNEL);
+       in_dev = kzalloc(sizeof(*in_dev), GFP_KERNEL);
        if (!in_dev)
                goto out;
-       memset(in_dev, 0, sizeof(*in_dev));
        INIT_RCU_HEAD(&in_dev->rcu_head);
        memcpy(&in_dev->cnf, &ipv4_devconf_dflt, sizeof(in_dev->cnf));
        in_dev->cnf.sysctl = NULL;
@@ -153,12 +163,11 @@ struct in_device *inetdev_init(struct net_device *dev)
        dev_hold(dev);
 #ifdef CONFIG_SYSCTL
        neigh_sysctl_register(dev, in_dev->arp_parms, NET_IPV4,
-                             NET_IPV4_NEIGH, "ipv4", NULL);
+                             NET_IPV4_NEIGH, "ipv4", NULL, NULL);
 #endif
 
-       /* Account for reference dev->ip_ptr */
+       /* Account for reference dev->ip_ptr (below) */
        in_dev_hold(in_dev);
-       rcu_assign_pointer(dev->ip_ptr, in_dev);
 
 #ifdef CONFIG_SYSCTL
        devinet_sysctl_register(in_dev, &in_dev->cnf);
@@ -166,6 +175,9 @@ struct in_device *inetdev_init(struct net_device *dev)
        ip_mc_init_dev(in_dev);
        if (dev->flags & IFF_UP)
                ip_mc_up(in_dev);
+
+       /* we can receive as soon as ip_ptr is set -- do this last */
+       rcu_assign_pointer(dev->ip_ptr, in_dev);
 out:
        return in_dev;
 out_kfree:
@@ -187,6 +199,10 @@ static void inetdev_destroy(struct in_device *in_dev)
 
        ASSERT_RTNL();
 
+       dev = in_dev->dev;
+       if (dev == &loopback_dev)
+               return;
+
        in_dev->dead = 1;
 
        ip_mc_destroy_dev(in_dev);
@@ -200,7 +216,6 @@ static void inetdev_destroy(struct in_device *in_dev)
        devinet_sysctl_unregister(&in_dev->cnf);
 #endif
 
-       dev = in_dev->dev;
        dev->ip_ptr = NULL;
 
 #ifdef CONFIG_SYSCTL
@@ -212,7 +227,7 @@ static void inetdev_destroy(struct in_device *in_dev)
        call_rcu(&in_dev->rcu_head, in_dev_rcu_put);
 }
 
-int inet_addr_onlink(struct in_device *in_dev, u32 a, u32 b)
+int inet_addr_onlink(struct in_device *in_dev, __be32 a, __be32 b)
 {
        rcu_read_lock();
        for_primary_ifa(in_dev) {
@@ -227,32 +242,48 @@ int inet_addr_onlink(struct in_device *in_dev, u32 a, u32 b)
        return 0;
 }
 
-static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
-                        int destroy)
+static void __inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
+                        int destroy, struct nlmsghdr *nlh, u32 pid)
 {
-       struct in_ifaddr *ifa1 = *ifap;
+       struct in_ifaddr *promote = NULL;
+       struct in_ifaddr *ifa, *ifa1 = *ifap;
+       struct in_ifaddr *last_prim = in_dev->ifa_list;
+       struct in_ifaddr *prev_prom = NULL;
+       int do_promote = IN_DEV_PROMOTE_SECONDARIES(in_dev);
 
        ASSERT_RTNL();
 
-       /* 1. Deleting primary ifaddr forces deletion all secondaries */
+       /* 1. Deleting primary ifaddr forces deletion all secondaries 
+        * unless alias promotion is set
+        **/
 
        if (!(ifa1->ifa_flags & IFA_F_SECONDARY)) {
-               struct in_ifaddr *ifa;
                struct in_ifaddr **ifap1 = &ifa1->ifa_next;
 
                while ((ifa = *ifap1) != NULL) {
+                       if (!(ifa->ifa_flags & IFA_F_SECONDARY) && 
+                           ifa1->ifa_scope <= ifa->ifa_scope)
+                               last_prim = ifa;
+
                        if (!(ifa->ifa_flags & IFA_F_SECONDARY) ||
                            ifa1->ifa_mask != ifa->ifa_mask ||
                            !inet_ifa_match(ifa1->ifa_address, ifa)) {
                                ifap1 = &ifa->ifa_next;
+                               prev_prom = ifa;
                                continue;
                        }
 
-                       *ifap1 = ifa->ifa_next;
+                       if (!do_promote) {
+                               *ifap1 = ifa->ifa_next;
 
-                       rtmsg_ifa(RTM_DELADDR, ifa);
-                       notifier_call_chain(&inetaddr_chain, NETDEV_DOWN, ifa);
-                       inet_free_ifa(ifa);
+                               rtmsg_ifa(RTM_DELADDR, ifa, nlh, pid);
+                               blocking_notifier_call_chain(&inetaddr_chain,
+                                               NETDEV_DOWN, ifa);
+                               inet_free_ifa(ifa);
+                       } else {
+                               promote = ifa;
+                               break;
+                       }
                }
        }
 
@@ -270,8 +301,29 @@ static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
           is valid, it will try to restore deleted routes... Grr.
           So that, this order is correct.
         */
-       rtmsg_ifa(RTM_DELADDR, ifa1);
-       notifier_call_chain(&inetaddr_chain, NETDEV_DOWN, ifa1);
+       rtmsg_ifa(RTM_DELADDR, ifa1, nlh, pid);
+       blocking_notifier_call_chain(&inetaddr_chain, NETDEV_DOWN, ifa1);
+
+       if (promote) {
+
+               if (prev_prom) {
+                       prev_prom->ifa_next = promote->ifa_next;
+                       promote->ifa_next = last_prim->ifa_next;
+                       last_prim->ifa_next = promote;
+               }
+
+               promote->ifa_flags &= ~IFA_F_SECONDARY;
+               rtmsg_ifa(RTM_NEWADDR, promote, nlh, pid);
+               blocking_notifier_call_chain(&inetaddr_chain,
+                               NETDEV_UP, promote);
+               for (ifa = promote->ifa_next; ifa; ifa = ifa->ifa_next) {
+                       if (ifa1->ifa_mask != ifa->ifa_mask ||
+                           !inet_ifa_match(ifa1->ifa_address, ifa))
+                                       continue;
+                       fib_add_ifaddr(ifa);
+               }
+
+       }
        if (destroy) {
                inet_free_ifa(ifa1);
 
@@ -280,7 +332,14 @@ static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
        }
 }
 
-static int inet_insert_ifa(struct in_ifaddr *ifa)
+static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
+                        int destroy)
+{
+       __inet_del_ifa(in_dev, ifap, destroy, NULL, 0);
+}
+
+static int __inet_insert_ifa(struct in_ifaddr *ifa, struct nlmsghdr *nlh,
+                            u32 pid)
 {
        struct in_device *in_dev = ifa->ifa_dev;
        struct in_ifaddr *ifa1, **ifap, **last_primary;
@@ -325,15 +384,20 @@ static int inet_insert_ifa(struct in_ifaddr *ifa)
        /* Send message first, then call notifier.
           Notifier will trigger FIB update, so that
           listeners of netlink will know about new ifaddr */
-       rtmsg_ifa(RTM_NEWADDR, ifa);
-       notifier_call_chain(&inetaddr_chain, NETDEV_UP, ifa);
+       rtmsg_ifa(RTM_NEWADDR, ifa, nlh, pid);
+       blocking_notifier_call_chain(&inetaddr_chain, NETDEV_UP, ifa);
 
        return 0;
 }
 
+static int inet_insert_ifa(struct in_ifaddr *ifa)
+{
+       return __inet_insert_ifa(ifa, NULL, 0);
+}
+
 static int inet_set_ifa(struct net_device *dev, struct in_ifaddr *ifa)
 {
-       struct in_device *in_dev = __in_dev_get(dev);
+       struct in_device *in_dev = __in_dev_get_rtnl(dev);
 
        ASSERT_RTNL();
 
@@ -368,8 +432,8 @@ struct in_device *inetdev_by_index(int ifindex)
 
 /* Called only from RTNL semaphored context. No locks. */
 
-struct in_ifaddr *inet_ifa_byprefix(struct in_device *in_dev, u32 prefix,
-                                   u32 mask)
+struct in_ifaddr *inet_ifa_byprefix(struct in_device *in_dev, __be32 prefix,
+                                   __be32 mask)
 {
        ASSERT_RTNL();
 
@@ -382,140 +446,162 @@ struct in_ifaddr *inet_ifa_byprefix(struct in_device *in_dev, u32 prefix,
 
 static int inet_rtm_deladdr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 {
-       struct rtattr **rta = arg;
+       struct nlattr *tb[IFA_MAX+1];
        struct in_device *in_dev;
-       struct ifaddrmsg *ifm = NLMSG_DATA(nlh);
+       struct ifaddrmsg *ifm;
        struct in_ifaddr *ifa, **ifap;
+       int err = -EINVAL;
 
        ASSERT_RTNL();
 
-       if ((in_dev = inetdev_by_index(ifm->ifa_index)) == NULL)
-               goto out;
+       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);
+       if (in_dev == NULL) {
+               err = -ENODEV;
+               goto errout;
+       }
+
        __in_dev_put(in_dev);
 
        for (ifap = &in_dev->ifa_list; (ifa = *ifap) != NULL;
             ifap = &ifa->ifa_next) {
-               if ((rta[IFA_LOCAL - 1] &&
-                    memcmp(RTA_DATA(rta[IFA_LOCAL - 1]),
-                           &ifa->ifa_local, 4)) ||
-                   (rta[IFA_LABEL - 1] &&
-                    rtattr_strcmp(rta[IFA_LABEL - 1], ifa->ifa_label)) ||
-                   (rta[IFA_ADDRESS - 1] &&
-                    (ifm->ifa_prefixlen != ifa->ifa_prefixlen ||
-                     !inet_ifa_match(*(u32*)RTA_DATA(rta[IFA_ADDRESS - 1]),
-                                     ifa))))
+               if (tb[IFA_LOCAL] &&
+                   ifa->ifa_local != nla_get_be32(tb[IFA_LOCAL]))
+                       continue;
+
+               if (tb[IFA_LABEL] && nla_strcmp(tb[IFA_LABEL], ifa->ifa_label))
                        continue;
-               inet_del_ifa(in_dev, ifap, 1);
+
+               if (tb[IFA_ADDRESS] &&
+                   (ifm->ifa_prefixlen != ifa->ifa_prefixlen ||
+                   !inet_ifa_match(nla_get_be32(tb[IFA_ADDRESS]), ifa)))
+                       continue;
+
+               __inet_del_ifa(in_dev, ifap, 1, nlh, NETLINK_CB(skb).pid);
                return 0;
        }
-out:
-       return -EADDRNOTAVAIL;
+
+       err = -EADDRNOTAVAIL;
+errout:
+       return err;
 }
 
-static int inet_rtm_newaddr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
+static struct in_ifaddr *rtm_to_ifaddr(struct nlmsghdr *nlh)
 {
-       struct rtattr **rta = arg;
+       struct nlattr *tb[IFA_MAX+1];
+       struct in_ifaddr *ifa;
+       struct ifaddrmsg *ifm;
        struct net_device *dev;
        struct in_device *in_dev;
-       struct ifaddrmsg *ifm = NLMSG_DATA(nlh);
-       struct in_ifaddr *ifa;
-       int rc = -EINVAL;
+       int err = -EINVAL;
 
-       ASSERT_RTNL();
+       err = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_ipv4_policy);
+       if (err < 0)
+               goto errout;
 
-       if (ifm->ifa_prefixlen > 32 || !rta[IFA_LOCAL - 1])
-               goto out;
+       ifm = nlmsg_data(nlh);
+       if (ifm->ifa_prefixlen > 32 || tb[IFA_LOCAL] == NULL) {
+               err = -EINVAL;
+               goto errout;
+       }
 
-       rc = -ENODEV;
-       if ((dev = __dev_get_by_index(ifm->ifa_index)) == NULL)
-               goto out;
+       dev = __dev_get_by_index(ifm->ifa_index);
+       if (dev == NULL) {
+               err = -ENODEV;
+               goto errout;
+       }
 
-       rc = -ENOBUFS;
-       if ((in_dev = __in_dev_get(dev)) == NULL) {
+       in_dev = __in_dev_get_rtnl(dev);
+       if (in_dev == NULL) {
                in_dev = inetdev_init(dev);
-               if (!in_dev)
-                       goto out;
+               if (in_dev == NULL) {
+                       err = -ENOBUFS;
+                       goto errout;
+               }
        }
 
-       if ((ifa = inet_alloc_ifa()) == NULL)
-               goto out;
+       ifa = inet_alloc_ifa();
+       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;
+       }
+
+       in_dev_hold(in_dev);
+
+       if (tb[IFA_ADDRESS] == NULL)
+               tb[IFA_ADDRESS] = tb[IFA_LOCAL];
 
-       if (!rta[IFA_ADDRESS - 1])
-               rta[IFA_ADDRESS - 1] = rta[IFA_LOCAL - 1];
-       memcpy(&ifa->ifa_local, RTA_DATA(rta[IFA_LOCAL - 1]), 4);
-       memcpy(&ifa->ifa_address, RTA_DATA(rta[IFA_ADDRESS - 1]), 4);
        ifa->ifa_prefixlen = ifm->ifa_prefixlen;
        ifa->ifa_mask = inet_make_mask(ifm->ifa_prefixlen);
-       if (rta[IFA_BROADCAST - 1])
-               memcpy(&ifa->ifa_broadcast,
-                      RTA_DATA(rta[IFA_BROADCAST - 1]), 4);
-       if (rta[IFA_ANYCAST - 1])
-               memcpy(&ifa->ifa_anycast, RTA_DATA(rta[IFA_ANYCAST - 1]), 4);
        ifa->ifa_flags = ifm->ifa_flags;
        ifa->ifa_scope = ifm->ifa_scope;
-       in_dev_hold(in_dev);
-       ifa->ifa_dev   = in_dev;
-       if (rta[IFA_LABEL - 1])
-               rtattr_strlcpy(ifa->ifa_label, rta[IFA_LABEL - 1], IFNAMSIZ);
+       ifa->ifa_dev = in_dev;
+
+       ifa->ifa_local = nla_get_be32(tb[IFA_LOCAL]);
+       ifa->ifa_address = nla_get_be32(tb[IFA_ADDRESS]);
+
+       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
                memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
 
-       rc = inet_insert_ifa(ifa);
-out:
-       return rc;
+       return ifa;
+
+errout:
+       return ERR_PTR(err);
+}
+
+static int inet_rtm_newaddr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
+{
+       struct in_ifaddr *ifa;
+
+       ASSERT_RTNL();
+
+       ifa = rtm_to_ifaddr(nlh);
+       if (IS_ERR(ifa))
+               return PTR_ERR(ifa);
+
+       return __inet_insert_ifa(ifa, nlh, NETLINK_CB(skb).pid);
 }
 
 /*
  *     Determine a default network mask, based on the IP address.
  */
 
-static __inline__ int inet_abc_len(u32 addr)
+static __inline__ int inet_abc_len(__be32 addr)
 {
        int rc = -1;    /* Something else, probably a multicast. */
 
        if (ZERONET(addr))
                rc = 0;
        else {
-               addr = ntohl(addr);
+               __u32 haddr = ntohl(addr);
 
-               if (IN_CLASSA(addr))
+               if (IN_CLASSA(haddr))
                        rc = 8;
-               else if (IN_CLASSB(addr))
+               else if (IN_CLASSB(haddr))
                        rc = 16;
-               else if (IN_CLASSC(addr))
+               else if (IN_CLASSC(haddr))
                        rc = 24;
        }
 
        return rc;
 }
 
-/*
-       Check that a device is not member of the ipv4root assigned to the process
-       Return true if this is the case
-
-       If the process is not bound to specific IP, then it returns 0 (all
-       interface are fine).
-*/
-static inline int devinet_notiproot (struct in_ifaddr *ifa)
-{
-       int ret = 0;
-       struct nx_info *nxi;
-
-       if ((nxi = current->nx_info)) {
-               int i;
-               int nbip = nxi->nbipv4;
-               __u32 addr = ifa->ifa_local;
-               ret = 1;
-               for (i=0; i<nbip; i++) {
-                       if(nxi->ipv4[i] == addr) {
-                               ret = 0;
-                               break;
-                       }
-               }
-       }
-       return ret;
-}
-
 
 int devinet_ioctl(unsigned int cmd, void __user *arg)
 {
@@ -593,7 +679,10 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
        if (colon)
                *colon = ':';
 
-       if ((in_dev = __in_dev_get(dev)) != NULL) {
+       if ((in_dev = __in_dev_get_rtnl(dev)) != NULL) {
+               struct nx_info *nxi = current->nx_info;
+               int hide_netif = vx_flags(VXF_HIDE_NETIF, 0);
+
                if (tryaddrmatch) {
                        /* Matthias Andree */
                        /* compare label and address (4.4BSD style) */
@@ -602,6 +691,8 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
                           This is checked above. */
                        for (ifap = &in_dev->ifa_list; (ifa = *ifap) != NULL;
                             ifap = &ifa->ifa_next) {
+                               if (hide_netif && !ifa_in_nx_info(ifa, nxi))
+                                       continue;
                                if (!strcmp(ifr.ifr_name, ifa->ifa_label) &&
                                    sin_orig.sin_addr.s_addr ==
                                                        ifa->ifa_address) {
@@ -614,18 +705,18 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
                   comparing just the label */
                if (!ifa) {
                        for (ifap = &in_dev->ifa_list; (ifa = *ifap) != NULL;
-                            ifap = &ifa->ifa_next)
+                            ifap = &ifa->ifa_next) {
+                               if (hide_netif && !ifa_in_nx_info(ifa, nxi))
+                                       continue;
                                if (!strcmp(ifr.ifr_name, ifa->ifa_label))
                                        break;
+                       }
                }
        }
 
        ret = -EADDRNOTAVAIL;
        if (!ifa && cmd != SIOCSIFADDR && cmd != SIOCSIFFLAGS)
                goto done;
-       if (vx_flags(VXF_HIDE_NETIF, 0) &&
-               !ifa_in_nx_info(ifa, current->nx_info))
-               goto done;
 
        switch(cmd) {
        case SIOCGIFADDR:       /* Get interface address */
@@ -727,6 +818,7 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
                        break;
                ret = 0;
                if (ifa->ifa_mask != sin->sin_addr.s_addr) {
+                       __be32 old_mask = ifa->ifa_mask;
                        inet_del_ifa(in_dev, ifap, 0);
                        ifa->ifa_mask = sin->sin_addr.s_addr;
                        ifa->ifa_prefixlen = inet_mask_len(ifa->ifa_mask);
@@ -740,7 +832,7 @@ int devinet_ioctl(unsigned int cmd, void __user *arg)
                        if ((dev->flags & IFF_BROADCAST) &&
                            (ifa->ifa_prefixlen < 31) &&
                            (ifa->ifa_broadcast ==
-                            (ifa->ifa_local|~ifa->ifa_mask))) {
+                            (ifa->ifa_local|~old_mask))) {
                                ifa->ifa_broadcast = (ifa->ifa_local |
                                                      ~sin->sin_addr.s_addr);
                        }
@@ -760,7 +852,7 @@ rarok:
 
 static int inet_gifconf(struct net_device *dev, char __user *buf, int len)
 {
-       struct in_device *in_dev = __in_dev_get(dev);
+       struct in_device *in_dev = __in_dev_get_rtnl(dev);
        struct in_ifaddr *ifa;
        struct ifreq ifr;
        int done = 0;
@@ -800,13 +892,13 @@ out:
        return done;
 }
 
-u32 inet_select_addr(const struct net_device *dev, u32 dst, int scope)
+__be32 inet_select_addr(const struct net_device *dev, __be32 dst, int scope)
 {
-       u32 addr = 0;
+       __be32 addr = 0;
        struct in_device *in_dev;
 
        rcu_read_lock();
-       in_dev = __in_dev_get(dev);
+       in_dev = __in_dev_get_rcu(dev);
        if (!in_dev)
                goto no_in_dev;
 
@@ -833,7 +925,7 @@ no_in_dev:
        read_lock(&dev_base_lock);
        rcu_read_lock();
        for (dev = dev_base; dev; dev = dev->next) {
-               if ((in_dev = __in_dev_get(dev)) == NULL)
+               if ((in_dev = __in_dev_get_rcu(dev)) == NULL)
                        continue;
 
                for_primary_ifa(in_dev) {
@@ -851,11 +943,11 @@ out:
        return addr;
 }
 
-static u32 confirm_addr_indev(struct in_device *in_dev, u32 dst,
-                             u32 local, int scope)
+static __be32 confirm_addr_indev(struct in_device *in_dev, __be32 dst,
+                             __be32 local, int scope)
 {
        int same = 0;
-       u32 addr = 0;
+       __be32 addr = 0;
 
        for_ifa(in_dev) {
                if (!addr &&
@@ -895,14 +987,14 @@ static u32 confirm_addr_indev(struct in_device *in_dev, u32 dst,
  * - local: address, 0=autoselect the local address
  * - scope: maximum allowed scope value for the local address
  */
-u32 inet_confirm_addr(const struct net_device *dev, u32 dst, u32 local, int scope)
+__be32 inet_confirm_addr(const struct net_device *dev, __be32 dst, __be32 local, int scope)
 {
-       u32 addr = 0;
+       __be32 addr = 0;
        struct in_device *in_dev;
 
        if (dev) {
                rcu_read_lock();
-               if ((in_dev = __in_dev_get(dev)))
+               if ((in_dev = __in_dev_get_rcu(dev)))
                        addr = confirm_addr_indev(in_dev, dst, local, scope);
                rcu_read_unlock();
 
@@ -912,7 +1004,7 @@ u32 inet_confirm_addr(const struct net_device *dev, u32 dst, u32 local, int scop
        read_lock(&dev_base_lock);
        rcu_read_lock();
        for (dev = dev_base; dev; dev = dev->next) {
-               if ((in_dev = __in_dev_get(dev))) {
+               if ((in_dev = __in_dev_get_rcu(dev))) {
                        addr = confirm_addr_indev(in_dev, dst, local, scope);
                        if (addr)
                                break;
@@ -930,12 +1022,12 @@ u32 inet_confirm_addr(const struct net_device *dev, u32 dst, u32 local, int scop
 
 int register_inetaddr_notifier(struct notifier_block *nb)
 {
-       return notifier_chain_register(&inetaddr_chain, nb);
+       return blocking_notifier_chain_register(&inetaddr_chain, nb);
 }
 
 int unregister_inetaddr_notifier(struct notifier_block *nb)
 {
-       return notifier_chain_unregister(&inetaddr_chain, nb);
+       return blocking_notifier_chain_unregister(&inetaddr_chain, nb);
 }
 
 /* Rename ifa_labels for a device name change. Make some effort to preserve existing
@@ -972,12 +1064,20 @@ static int inetdev_event(struct notifier_block *this, unsigned long event,
                         void *ptr)
 {
        struct net_device *dev = ptr;
-       struct in_device *in_dev = __in_dev_get(dev);
+       struct in_device *in_dev = __in_dev_get_rtnl(dev);
 
        ASSERT_RTNL();
 
-       if (!in_dev)
+       if (!in_dev) {
+               if (event == NETDEV_REGISTER && dev == &loopback_dev) {
+                       in_dev = inetdev_init(dev);
+                       if (!in_dev)
+                               panic("devinet: Failed to create loopback\n");
+                       in_dev->cnf.no_xfrm = 1;
+                       in_dev->cnf.no_policy = 1;
+               }
                goto out;
+       }
 
        switch (event) {
        case NETDEV_REGISTER:
@@ -1000,8 +1100,6 @@ static int inetdev_event(struct notifier_block *this, unsigned long event,
                                memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
                                inet_insert_ifa(ifa);
                        }
-                       in_dev->cnf.no_xfrm = 1;
-                       in_dev->cnf.no_policy = 1;
                }
                ip_mc_up(in_dev);
                break;
@@ -1025,7 +1123,7 @@ static int inetdev_event(struct notifier_block *this, unsigned long event,
                devinet_sysctl_unregister(&in_dev->cnf);
                neigh_sysctl_unregister(in_dev->arp_parms);
                neigh_sysctl_register(dev, in_dev->arp_parms, NET_IPV4,
-                                     NET_IPV4_NEIGH, "ipv4", NULL);
+                                     NET_IPV4_NEIGH, "ipv4", NULL, NULL);
                devinet_sysctl_register(in_dev, &in_dev->cnf);
 #endif
                break;
@@ -1038,38 +1136,52 @@ static struct notifier_block ip_netdev_notifier = {
        .notifier_call =inetdev_event,
 };
 
+static inline size_t inet_nlmsg_size(void)
+{
+       return NLMSG_ALIGN(sizeof(struct ifaddrmsg))
+              + 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 */
+}
+
 static int inet_fill_ifaddr(struct sk_buff *skb, struct in_ifaddr *ifa,
-                           u32 pid, u32 seq, int event)
+                           u32 pid, u32 seq, int event, unsigned int flags)
 {
        struct ifaddrmsg *ifm;
        struct nlmsghdr  *nlh;
-       unsigned char    *b = skb->tail;
 
-       nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*ifm));
-       if (pid) nlh->nlmsg_flags |= NLM_F_MULTI;
-       ifm = NLMSG_DATA(nlh);
+       nlh = nlmsg_put(skb, pid, seq, event, sizeof(*ifm), flags);
+       if (nlh == NULL)
+               return -ENOBUFS;
+
+       ifm = nlmsg_data(nlh);
        ifm->ifa_family = AF_INET;
        ifm->ifa_prefixlen = ifa->ifa_prefixlen;
        ifm->ifa_flags = ifa->ifa_flags|IFA_F_PERMANENT;
        ifm->ifa_scope = ifa->ifa_scope;
        ifm->ifa_index = ifa->ifa_dev->dev->ifindex;
+
        if (ifa->ifa_address)
-               RTA_PUT(skb, IFA_ADDRESS, 4, &ifa->ifa_address);
+               NLA_PUT_BE32(skb, IFA_ADDRESS, ifa->ifa_address);
+
        if (ifa->ifa_local)
-               RTA_PUT(skb, IFA_LOCAL, 4, &ifa->ifa_local);
+               NLA_PUT_BE32(skb, IFA_LOCAL, ifa->ifa_local);
+
        if (ifa->ifa_broadcast)
-               RTA_PUT(skb, IFA_BROADCAST, 4, &ifa->ifa_broadcast);
+               NLA_PUT_BE32(skb, IFA_BROADCAST, ifa->ifa_broadcast);
+
        if (ifa->ifa_anycast)
-               RTA_PUT(skb, IFA_ANYCAST, 4, &ifa->ifa_anycast);
+               NLA_PUT_BE32(skb, IFA_ANYCAST, ifa->ifa_anycast);
+
        if (ifa->ifa_label[0])
-               RTA_PUT(skb, IFA_LABEL, IFNAMSIZ, &ifa->ifa_label);
-       nlh->nlmsg_len = skb->tail - b;
-       return skb->len;
+               NLA_PUT_STRING(skb, IFA_LABEL, ifa->ifa_label);
+
+       return nlmsg_end(skb, nlh);
 
-nlmsg_failure:
-rtattr_failure:
-       skb_trim(skb, b - skb->data);
-       return -1;
+nla_put_failure:
+       return nlmsg_cancel(skb, nlh);
 }
 
 static int inet_dump_ifaddr(struct sk_buff *skb, struct netlink_callback *cb)
@@ -1089,7 +1201,7 @@ static int inet_dump_ifaddr(struct sk_buff *skb, struct netlink_callback *cb)
                if (idx > s_idx)
                        s_ip_idx = 0;
                rcu_read_lock();
-               if ((in_dev = __in_dev_get(dev)) == NULL) {
+               if ((in_dev = __in_dev_get_rcu(dev)) == NULL) {
                        rcu_read_unlock();
                        continue;
                }
@@ -1103,7 +1215,7 @@ static int inet_dump_ifaddr(struct sk_buff *skb, struct netlink_callback *cb)
                                continue;
                        if (inet_fill_ifaddr(skb, ifa, NETLINK_CB(cb->skb).pid,
                                             cb->nlh->nlmsg_seq,
-                                            RTM_NEWADDR) <= 0) {
+                                            RTM_NEWADDR, NLM_F_MULTI) <= 0) {
                                rcu_read_unlock();
                                goto done;
                        }
@@ -1119,33 +1231,37 @@ done:
        return skb->len;
 }
 
-static void rtmsg_ifa(int event, struct in_ifaddr* ifa)
+static void rtmsg_ifa(int event, struct in_ifaddr* ifa, struct nlmsghdr *nlh,
+                     u32 pid)
 {
-       int size = NLMSG_SPACE(sizeof(struct ifaddrmsg) + 128);
-       struct sk_buff *skb = alloc_skb(size, GFP_KERNEL);
-
-       if (!skb)
-               netlink_set_err(rtnl, 0, RTMGRP_IPV4_IFADDR, ENOBUFS);
-       else if (inet_fill_ifaddr(skb, ifa, 0, 0, event) < 0) {
-               kfree_skb(skb);
-               netlink_set_err(rtnl, 0, RTMGRP_IPV4_IFADDR, EINVAL);
-       } else {
-               NETLINK_CB(skb).dst_groups = RTMGRP_IPV4_IFADDR;
-               netlink_broadcast(rtnl, skb, 0, RTMGRP_IPV4_IFADDR, GFP_KERNEL);
-       }
+       struct sk_buff *skb;
+       u32 seq = nlh ? nlh->nlmsg_seq : 0;
+       int err = -ENOBUFS;
+
+       skb = nlmsg_new(inet_nlmsg_size(), GFP_KERNEL);
+       if (skb == NULL)
+               goto errout;
+
+       err = inet_fill_ifaddr(skb, ifa, pid, seq, event, 0);
+       /* failure implies BUG in inet_nlmsg_size() */
+       BUG_ON(err < 0);
+
+       err = rtnl_notify(skb, pid, RTNLGRP_IPV4_IFADDR, nlh, GFP_KERNEL);
+errout:
+       if (err < 0)
+               rtnl_set_sk_err(RTNLGRP_IPV4_IFADDR, err);
 }
 
-static struct rtnetlink_link inet_rtnetlink_table[RTM_MAX - RTM_BASE + 1] = {
-        [4] = { .doit   = inet_rtm_newaddr,  },
-        [5] = { .doit   = inet_rtm_deladdr,  },
-        [6] = { .dumpit = inet_dump_ifaddr,  },
-        [8] = { .doit   = inet_rtm_newroute, },
-        [9] = { .doit   = inet_rtm_delroute, },
-       [10] = { .doit   = inet_rtm_getroute, .dumpit = inet_dump_fib, },
+static struct rtnetlink_link inet_rtnetlink_table[RTM_NR_MSGTYPES] = {
+       [RTM_NEWADDR  - RTM_BASE] = { .doit     = inet_rtm_newaddr,     },
+       [RTM_DELADDR  - RTM_BASE] = { .doit     = inet_rtm_deladdr,     },
+       [RTM_GETADDR  - RTM_BASE] = { .dumpit   = inet_dump_ifaddr,     },
+       [RTM_NEWROUTE - RTM_BASE] = { .doit     = inet_rtm_newroute,    },
+       [RTM_DELROUTE - RTM_BASE] = { .doit     = inet_rtm_delroute,    },
+       [RTM_GETROUTE - RTM_BASE] = { .doit     = inet_rtm_getroute,
+                                     .dumpit   = inet_dump_fib,        },
 #ifdef CONFIG_IP_MULTIPLE_TABLES
-       [16] = { .doit   = inet_rtm_newrule, },
-       [17] = { .doit   = inet_rtm_delrule, },
-       [18] = { .dumpit = inet_dump_rules,  },
+       [RTM_GETRULE  - RTM_BASE] = { .dumpit   = fib4_rules_dump,      },
 #endif
 };
 
@@ -1163,7 +1279,7 @@ void inet_forward_change(void)
        for (dev = dev_base; dev; dev = dev->next) {
                struct in_device *in_dev;
                rcu_read_lock();
-               in_dev = __in_dev_get(dev);
+               in_dev = __in_dev_get_rcu(dev);
                if (in_dev)
                        in_dev->cnf.forwarding = on;
                rcu_read_unlock();
@@ -1207,8 +1323,7 @@ int ipv4_doint_and_flush(ctl_table *ctl, int write,
 
 int ipv4_doint_and_flush_strategy(ctl_table *table, int __user *name, int nlen,
                                  void __user *oldval, size_t __user *oldlenp,
-                                 void __user *newval, size_t newlen, 
-                                 void **context)
+                                 void __user *newval, size_t newlen)
 {
        int *valp = table->data;
        int new;
@@ -1249,7 +1364,7 @@ int ipv4_doint_and_flush_strategy(ctl_table *table, int __user *name, int nlen,
 
 static struct devinet_sysctl_table {
        struct ctl_table_header *sysctl_header;
-       ctl_table               devinet_vars[20];
+       ctl_table               devinet_vars[__NET_IPV4_CONF_MAX];
        ctl_table               devinet_dev[2];
        ctl_table               devinet_conf_dir[2];
        ctl_table               devinet_proto_dir[2];
@@ -1384,6 +1499,14 @@ static struct devinet_sysctl_table {
                        .mode           = 0644,
                        .proc_handler   = &proc_dointvec,
                },
+               {
+                       .ctl_name       = NET_IPV4_CONF_ARP_ACCEPT,
+                       .procname       = "arp_accept",
+                       .data           = &ipv4_devconf.arp_accept,
+                       .maxlen         = sizeof(int),
+                       .mode           = 0644,
+                       .proc_handler   = &proc_dointvec,
+               },
                {
                        .ctl_name       = NET_IPV4_CONF_NOXFRM,
                        .procname       = "disable_xfrm",
@@ -1411,6 +1534,15 @@ static struct devinet_sysctl_table {
                        .proc_handler   = &ipv4_doint_and_flush,
                        .strategy       = &ipv4_doint_and_flush_strategy,
                },
+               {
+                       .ctl_name       = NET_IPV4_CONF_PROMOTE_SECONDARIES,
+                       .procname       = "promote_secondaries",
+                       .data           = &ipv4_devconf.promote_secondaries,
+                       .maxlen         = sizeof(int),
+                       .mode           = 0644,
+                       .proc_handler   = &ipv4_doint_and_flush,
+                       .strategy       = &ipv4_doint_and_flush_strategy,
+               },
        },
        .devinet_dev = {
                {
@@ -1451,12 +1583,12 @@ static void devinet_sysctl_register(struct in_device *in_dev,
 {
        int i;
        struct net_device *dev = in_dev ? in_dev->dev : NULL;
-       struct devinet_sysctl_table *t = kmalloc(sizeof(*t), GFP_KERNEL);
+       struct devinet_sysctl_table *t = kmemdup(&devinet_sysctl, sizeof(*t),
+                                                GFP_KERNEL);
        char *dev_name = NULL;
 
        if (!t)
                return;
-       memcpy(t, &devinet_sysctl, sizeof(*t));
        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].de = NULL;
@@ -1475,7 +1607,7 @@ static void devinet_sysctl_register(struct in_device *in_dev,
         * by sysctl and we wouldn't want anyone to change it under our feet
         * (see SIOCSIFNAME).
         */     
-       dev_name = net_sysctl_strdup(dev_name);
+       dev_name = kstrdup(dev_name, GFP_KERNEL);
        if (!dev_name)
            goto free;
 
@@ -1528,7 +1660,6 @@ void __init devinet_init(void)
 #endif
 }
 
-EXPORT_SYMBOL(devinet_ioctl);
 EXPORT_SYMBOL(in_dev_finish_destroy);
 EXPORT_SYMBOL(inet_select_addr);
 EXPORT_SYMBOL(inetdev_by_index);