icmp: move icmp_err_convert[] to .rodata
[safe/jmp/linux-2.6] / net / ipv4 / icmp.c
index 272c69e..4b4c2bc 100644 (file)
@@ -1,9 +1,7 @@
 /*
  *     NET3:   Implementation of the ICMP protocol layer.
  *
- *             Alan Cox, <alan@redhat.com>
- *
- *     Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
+ *             Alan Cox, <alan@lxorguk.ukuu.org.uk>
  *
  *     This program is free software; you can redistribute it and/or
  *     modify it under the terms of the GNU General Public License
@@ -92,6 +90,8 @@
 #include <asm/system.h>
 #include <asm/uaccess.h>
 #include <net/checksum.h>
+#include <net/xfrm.h>
+#include <net/inet_common.h>
 
 /*
  *     Build xmit assembly blocks
@@ -111,16 +111,10 @@ struct icmp_bxm {
        unsigned char  optbuf[40];
 };
 
-/*
- *     Statistics
- */
-DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
-DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
-
 /* An array of errno for error messages from dest unreach. */
 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 
-struct icmp_err icmp_err_convert[] = {
+const struct icmp_err icmp_err_convert[] = {
        {
                .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
                .fatal = 0,
@@ -187,29 +181,6 @@ struct icmp_err icmp_err_convert[] = {
        },
 };
 
-/* Control parameters for ECHO replies. */
-int sysctl_icmp_echo_ignore_all __read_mostly;
-int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
-
-/* Control parameter - ignore bogus broadcast responses? */
-int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
-
-/*
- *     Configurable global rate limit.
- *
- *     ratelimit defines tokens/packet consumed for dst->rate_token bucket
- *     ratemask defines which icmp types are ratelimited by setting
- *     it's bit position.
- *
- *     default:
- *     dest unreachable (3), source quench (4),
- *     time exceeded (11), parameter problem (12)
- */
-
-int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
-int sysctl_icmp_ratemask __read_mostly = 0x1818;
-int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
-
 /*
  *     ICMP control array. This specifies what to do with each ICMP.
  */
@@ -228,26 +199,32 @@ static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
  *
  *     On SMP we have one ICMP socket per-cpu.
  */
-static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
-#define icmp_socket    __get_cpu_var(__icmp_socket)
+static struct sock *icmp_sk(struct net *net)
+{
+       return net->ipv4.icmp_sk[smp_processor_id()];
+}
 
-static __inline__ int icmp_xmit_lock(void)
+static inline struct sock *icmp_xmit_lock(struct net *net)
 {
+       struct sock *sk;
+
        local_bh_disable();
 
-       if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
+       sk = icmp_sk(net);
+
+       if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
                /* This can happen if the output path signals a
                 * dst_link_failure() for an outgoing ICMP packet.
                 */
                local_bh_enable();
-               return 1;
+               return NULL;
        }
-       return 0;
+       return sk;
 }
 
-static void icmp_xmit_unlock(void)
+static inline void icmp_xmit_unlock(struct sock *sk)
 {
-       spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
+       spin_unlock_bh(&sk->sk_lock.slock);
 }
 
 /*
@@ -274,22 +251,24 @@ static void icmp_xmit_unlock(void)
 #define XRLIM_BURST_FACTOR 6
 int xrlim_allow(struct dst_entry *dst, int timeout)
 {
-       unsigned long now;
+       unsigned long now, token = dst->rate_tokens;
        int rc = 0;
 
        now = jiffies;
-       dst->rate_tokens += now - dst->rate_last;
+       token += now - dst->rate_last;
        dst->rate_last = now;
-       if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
-               dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
-       if (dst->rate_tokens >= timeout) {
-               dst->rate_tokens -= timeout;
+       if (token > XRLIM_BURST_FACTOR * timeout)
+               token = XRLIM_BURST_FACTOR * timeout;
+       if (token >= timeout) {
+               token -= timeout;
                rc = 1;
        }
+       dst->rate_tokens = token;
        return rc;
 }
 
-static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
+static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
+               int type, int code)
 {
        struct dst_entry *dst = &rt->u.dst;
        int rc = 1;
@@ -306,8 +285,8 @@ static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
                goto out;
 
        /* Limit if icmp type is enabled in ratemask. */
-       if ((1 << type) & sysctl_icmp_ratemask)
-               rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
+       if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
+               rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
 out:
        return rc;
 }
@@ -315,10 +294,10 @@ out:
 /*
  *     Maintain the counters used in the SNMP statistics for outgoing ICMP
  */
-void icmp_out_count(unsigned char type)
+void icmp_out_count(struct net *net, unsigned char type)
 {
-       ICMPMSGOUT_INC_STATS(type);
-       ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
+       ICMPMSGOUT_INC_STATS(net, type);
+       ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 }
 
 /*
@@ -342,21 +321,23 @@ static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 }
 
 static void icmp_push_reply(struct icmp_bxm *icmp_param,
-                           struct ipcm_cookie *ipc, struct rtable *rt)
+                           struct ipcm_cookie *ipc, struct rtable **rt)
 {
+       struct sock *sk;
        struct sk_buff *skb;
 
-       if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
+       sk = icmp_sk(dev_net((*rt)->u.dst.dev));
+       if (ip_append_data(sk, icmp_glue_bits, icmp_param,
                           icmp_param->data_len+icmp_param->head_len,
                           icmp_param->head_len,
                           ipc, rt, MSG_DONTWAIT) < 0)
-               ip_flush_pending_frames(icmp_socket->sk);
-       else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
+               ip_flush_pending_frames(sk);
+       else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
                struct icmphdr *icmph = icmp_hdr(skb);
                __wsum csum = 0;
                struct sk_buff *skb1;
 
-               skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
+               skb_queue_walk(&sk->sk_write_queue, skb1) {
                        csum = csum_add(csum, skb1->csum);
                }
                csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
@@ -364,7 +345,7 @@ static void icmp_push_reply(struct icmp_bxm *icmp_param,
                                                 icmp_param->head_len, csum);
                icmph->checksum = csum_fold(csum);
                skb->ip_summed = CHECKSUM_NONE;
-               ip_push_pending_frames(icmp_socket->sk);
+               ip_push_pending_frames(sk);
        }
 }
 
@@ -374,23 +355,27 @@ static void icmp_push_reply(struct icmp_bxm *icmp_param,
 
 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 {
-       struct sock *sk = icmp_socket->sk;
-       struct inet_sock *inet = inet_sk(sk);
        struct ipcm_cookie ipc;
-       struct rtable *rt = (struct rtable *)skb->dst;
+       struct rtable *rt = skb_rtable(skb);
+       struct net *net = dev_net(rt->u.dst.dev);
+       struct sock *sk;
+       struct inet_sock *inet;
        __be32 daddr;
 
        if (ip_options_echo(&icmp_param->replyopts, skb))
                return;
 
-       if (icmp_xmit_lock())
+       sk = icmp_xmit_lock(net);
+       if (sk == NULL)
                return;
+       inet = inet_sk(sk);
 
        icmp_param->data.icmph.checksum = 0;
 
        inet->tos = ip_hdr(skb)->tos;
        daddr = ipc.addr = rt->rt_src;
        ipc.opt = NULL;
+       ipc.shtx.flags = 0;
        if (icmp_param->replyopts.optlen) {
                ipc.opt = &icmp_param->replyopts;
                if (ipc.opt->srr)
@@ -403,15 +388,15 @@ static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
                                                .tos = RT_TOS(ip_hdr(skb)->tos) } },
                                    .proto = IPPROTO_ICMP };
                security_skb_classify_flow(skb, &fl);
-               if (ip_route_output_key(&rt, &fl))
+               if (ip_route_output_key(net, &rt, &fl))
                        goto out_unlock;
        }
-       if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
+       if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
                               icmp_param->data.icmph.code))
-               icmp_push_reply(icmp_param, &ipc, rt);
+               icmp_push_reply(icmp_param, &ipc, &rt);
        ip_rt_put(rt);
 out_unlock:
-       icmp_xmit_unlock();
+       icmp_xmit_unlock(sk);
 }
 
 
@@ -431,13 +416,16 @@ void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
        struct iphdr *iph;
        int room;
        struct icmp_bxm icmp_param;
-       struct rtable *rt = (struct rtable *)skb_in->dst;
+       struct rtable *rt = skb_rtable(skb_in);
        struct ipcm_cookie ipc;
        __be32 saddr;
        u8  tos;
+       struct net *net;
+       struct sock *sk;
 
        if (!rt)
                goto out;
+       net = dev_net(rt->u.dst.dev);
 
        /*
         *      Find the original header. It is expected to be valid, of course.
@@ -501,7 +489,8 @@ void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
                }
        }
 
-       if (icmp_xmit_lock())
+       sk = icmp_xmit_lock(net);
+       if (sk == NULL)
                return;
 
        /*
@@ -512,14 +501,16 @@ void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
        if (!(rt->rt_flags & RTCF_LOCAL)) {
                struct net_device *dev = NULL;
 
-               if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
-                       dev = dev_get_by_index(&init_net, rt->fl.iif);
+               rcu_read_lock();
+               if (rt->fl.iif &&
+                       net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
+                       dev = dev_get_by_index_rcu(net, rt->fl.iif);
 
-               if (dev) {
+               if (dev)
                        saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
-                       dev_put(dev);
-               } else
+               else
                        saddr = 0;
+               rcu_read_unlock();
        }
 
        tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
@@ -540,10 +531,10 @@ void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
        icmp_param.data.icmph.checksum   = 0;
        icmp_param.skb    = skb_in;
        icmp_param.offset = skb_network_offset(skb_in);
-       icmp_out_count(icmp_param.data.icmph.type);
-       inet_sk(icmp_socket->sk)->tos = tos;
+       inet_sk(sk)->tos = tos;
        ipc.addr = iph->saddr;
        ipc.opt = &icmp_param.replyopts;
+       ipc.shtx.flags = 0;
 
        {
                struct flowi fl = {
@@ -564,12 +555,74 @@ void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
                                }
                        }
                };
+               int err;
+               struct rtable *rt2;
+
                security_skb_classify_flow(skb_in, &fl);
-               if (ip_route_output_key(&rt, &fl))
+               if (__ip_route_output_key(net, &rt, &fl))
                        goto out_unlock;
+
+               /* No need to clone since we're just using its address. */
+               rt2 = rt;
+
+               err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
+               switch (err) {
+               case 0:
+                       if (rt != rt2)
+                               goto route_done;
+                       break;
+               case -EPERM:
+                       rt = NULL;
+                       break;
+               default:
+                       goto out_unlock;
+               }
+
+               if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
+                       goto relookup_failed;
+
+               if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
+                       err = __ip_route_output_key(net, &rt2, &fl);
+               else {
+                       struct flowi fl2 = {};
+                       struct dst_entry *odst;
+
+                       fl2.fl4_dst = fl.fl4_src;
+                       if (ip_route_output_key(net, &rt2, &fl2))
+                               goto relookup_failed;
+
+                       /* Ugh! */
+                       odst = skb_dst(skb_in);
+                       err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
+                                            RT_TOS(tos), rt2->u.dst.dev);
+
+                       dst_release(&rt2->u.dst);
+                       rt2 = skb_rtable(skb_in);
+                       skb_dst_set(skb_in, odst);
+               }
+
+               if (err)
+                       goto relookup_failed;
+
+               err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
+                                 XFRM_LOOKUP_ICMP);
+               switch (err) {
+               case 0:
+                       dst_release(&rt->u.dst);
+                       rt = rt2;
+                       break;
+               case -EPERM:
+                       goto ende;
+               default:
+relookup_failed:
+                       if (!rt)
+                               goto out_unlock;
+                       break;
+               }
        }
 
-       if (!icmpv4_xrlim_allow(rt, type, code))
+route_done:
+       if (!icmpv4_xrlim_allow(net, rt, type, code))
                goto ende;
 
        /* RFC says return as much as we can without exceeding 576 bytes. */
@@ -585,11 +638,11 @@ void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
                icmp_param.data_len = room;
        icmp_param.head_len = sizeof(struct icmphdr);
 
-       icmp_push_reply(&icmp_param, &ipc, rt);
+       icmp_push_reply(&icmp_param, &ipc, &rt);
 ende:
        ip_rt_put(rt);
 out_unlock:
-       icmp_xmit_unlock();
+       icmp_xmit_unlock(sk);
 out:;
 }
 
@@ -603,9 +656,11 @@ static void icmp_unreach(struct sk_buff *skb)
        struct iphdr *iph;
        struct icmphdr *icmph;
        int hash, protocol;
-       struct net_protocol *ipprot;
-       struct sock *raw_sk;
+       const struct net_protocol *ipprot;
        u32 info = 0;
+       struct net *net;
+
+       net = dev_net(skb_dst(skb)->dev);
 
        /*
         *      Incomplete header ?
@@ -631,21 +686,19 @@ static void icmp_unreach(struct sk_buff *skb)
                        break;
                case ICMP_FRAG_NEEDED:
                        if (ipv4_config.no_pmtu_disc) {
-                               LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
-                                                        "fragmentation needed "
-                                                        "and DF set.\n",
-                                              NIPQUAD(iph->daddr));
+                               LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
+                                              &iph->daddr);
                        } else {
-                               info = ip_rt_frag_needed(iph,
-                                                    ntohs(icmph->un.frag.mtu));
+                               info = ip_rt_frag_needed(net, iph,
+                                                        ntohs(icmph->un.frag.mtu),
+                                                        skb->dev);
                                if (!info)
                                        goto out;
                        }
                        break;
                case ICMP_SR_FAILED:
-                       LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
-                                                "Route Failed.\n",
-                                      NIPQUAD(iph->daddr));
+                       LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
+                                      &iph->daddr);
                        break;
                default:
                        break;
@@ -673,15 +726,15 @@ static void icmp_unreach(struct sk_buff *skb)
         *      get the other vendor to fix their kit.
         */
 
-       if (!sysctl_icmp_ignore_bogus_error_responses &&
-           inet_addr_type(iph->daddr) == RTN_BROADCAST) {
+       if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
+           inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
                if (net_ratelimit())
-                       printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
+                       printk(KERN_WARNING "%pI4 sent an invalid ICMP "
                                            "type %u, code %u "
-                                           "error to a broadcast: %u.%u.%u.%u on %s\n",
-                              NIPQUAD(ip_hdr(skb)->saddr),
+                                           "error to a broadcast: %pI4 on %s\n",
+                              &ip_hdr(skb)->saddr,
                               icmph->type, icmph->code,
-                              NIPQUAD(iph->daddr),
+                              &iph->daddr,
                               skb->dev->name);
                goto out;
        }
@@ -698,21 +751,9 @@ static void icmp_unreach(struct sk_buff *skb)
        /*
         *      Deliver ICMP message to raw sockets. Pretty useless feature?
         */
+       raw_icmp_error(skb, protocol, info);
 
-       /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
        hash = protocol & (MAX_INET_PROTOS - 1);
-       read_lock(&raw_v4_lock);
-       if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
-               while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
-                                                iph->saddr,
-                                                skb->dev->ifindex)) != NULL) {
-                       raw_err(raw_sk, skb, info);
-                       raw_sk = sk_next(raw_sk);
-                       iph = (struct iphdr *)skb->data;
-               }
-       }
-       read_unlock(&raw_v4_lock);
-
        rcu_read_lock();
        ipprot = rcu_dereference(inet_protos[hash]);
        if (ipprot && ipprot->err_handler)
@@ -722,7 +763,7 @@ static void icmp_unreach(struct sk_buff *skb)
 out:
        return;
 out_err:
-       ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
+       ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
        goto out;
 }
 
@@ -762,7 +803,7 @@ static void icmp_redirect(struct sk_buff *skb)
 out:
        return;
 out_err:
-       ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
+       ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
        goto out;
 }
 
@@ -780,7 +821,10 @@ out_err:
 
 static void icmp_echo(struct sk_buff *skb)
 {
-       if (!sysctl_icmp_echo_ignore_all) {
+       struct net *net;
+
+       net = dev_net(skb_dst(skb)->dev);
+       if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
                struct icmp_bxm icmp_param;
 
                icmp_param.data.icmph      = *icmp_hdr(skb);
@@ -802,7 +846,7 @@ static void icmp_echo(struct sk_buff *skb)
  */
 static void icmp_timestamp(struct sk_buff *skb)
 {
-       struct timeval tv;
+       struct timespec tv;
        struct icmp_bxm icmp_param;
        /*
         *      Too short.
@@ -813,9 +857,9 @@ static void icmp_timestamp(struct sk_buff *skb)
        /*
         *      Fill in the current time as ms since midnight UT:
         */
-       do_gettimeofday(&tv);
-       icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
-                                        tv.tv_usec / 1000);
+       getnstimeofday(&tv);
+       icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
+                                        tv.tv_nsec / NSEC_PER_MSEC);
        icmp_param.data.times[2] = icmp_param.data.times[1];
        if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
                BUG();
@@ -830,7 +874,7 @@ static void icmp_timestamp(struct sk_buff *skb)
 out:
        return;
 out_err:
-       ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
+       ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
        goto out;
 }
 
@@ -883,7 +927,7 @@ static void icmp_address(struct sk_buff *skb)
 
 static void icmp_address_reply(struct sk_buff *skb)
 {
-       struct rtable *rt = (struct rtable *)skb->dst;
+       struct rtable *rt = skb_rtable(skb);
        struct net_device *dev = skb->dev;
        struct in_device *in_dev;
        struct in_ifaddr *ifa;
@@ -908,9 +952,8 @@ static void icmp_address_reply(struct sk_buff *skb)
                                break;
                }
                if (!ifa && net_ratelimit()) {
-                       printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
-                                        "%s/%u.%u.%u.%u\n",
-                              NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
+                       printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
+                              mp, dev->name, &rt->rt_src);
                }
        }
        rcu_read_unlock();
@@ -928,9 +971,30 @@ static void icmp_discard(struct sk_buff *skb)
 int icmp_rcv(struct sk_buff *skb)
 {
        struct icmphdr *icmph;
-       struct rtable *rt = (struct rtable *)skb->dst;
+       struct rtable *rt = skb_rtable(skb);
+       struct net *net = dev_net(rt->u.dst.dev);
 
-       ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
+       if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
+               struct sec_path *sp = skb_sec_path(skb);
+               int nh;
+
+               if (!(sp && sp->xvec[sp->len - 1]->props.flags &
+                                XFRM_STATE_ICMP))
+                       goto drop;
+
+               if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
+                       goto drop;
+
+               nh = skb_network_offset(skb);
+               skb_set_network_header(skb, sizeof(*icmph));
+
+               if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
+                       goto drop;
+
+               skb_set_network_header(skb, nh);
+       }
+
+       ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
 
        switch (skb->ip_summed) {
        case CHECKSUM_COMPLETE:
@@ -943,12 +1007,12 @@ int icmp_rcv(struct sk_buff *skb)
                        goto error;
        }
 
-       if (!pskb_pull(skb, sizeof(struct icmphdr)))
+       if (!pskb_pull(skb, sizeof(*icmph)))
                goto error;
 
        icmph = icmp_hdr(skb);
 
-       ICMPMSGIN_INC_STATS_BH(icmph->type);
+       ICMPMSGIN_INC_STATS_BH(net, icmph->type);
        /*
         *      18 is the highest 'known' ICMP type. Anything else is a mystery
         *
@@ -972,7 +1036,7 @@ int icmp_rcv(struct sk_buff *skb)
                 */
                if ((icmph->type == ICMP_ECHO ||
                     icmph->type == ICMP_TIMESTAMP) &&
-                   sysctl_icmp_echo_ignore_broadcasts) {
+                   net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
                        goto error;
                }
                if (icmph->type != ICMP_ECHO &&
@@ -989,7 +1053,7 @@ drop:
        kfree_skb(skb);
        return 0;
 error:
-       ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
+       ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
        goto drop;
 }
 
@@ -1067,42 +1131,90 @@ static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
        },
 };
 
-void __init icmp_init(struct net_proto_family *ops)
+static void __net_exit icmp_sk_exit(struct net *net)
 {
-       struct inet_sock *inet;
        int i;
 
-       for_each_possible_cpu(i) {
-               int err;
+       for_each_possible_cpu(i)
+               inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
+       kfree(net->ipv4.icmp_sk);
+       net->ipv4.icmp_sk = NULL;
+}
+
+static int __net_init icmp_sk_init(struct net *net)
+{
+       int i, err;
+
+       net->ipv4.icmp_sk =
+               kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
+       if (net->ipv4.icmp_sk == NULL)
+               return -ENOMEM;
 
-               err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
-                                      &per_cpu(__icmp_socket, i));
+       for_each_possible_cpu(i) {
+               struct sock *sk;
 
+               err = inet_ctl_sock_create(&sk, PF_INET,
+                                          SOCK_RAW, IPPROTO_ICMP, net);
                if (err < 0)
-                       panic("Failed to create the ICMP control socket.\n");
+                       goto fail;
 
-               per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
+               net->ipv4.icmp_sk[i] = sk;
 
                /* Enough space for 2 64K ICMP packets, including
                 * sk_buff struct overhead.
                 */
-               per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
+               sk->sk_sndbuf =
                        (2 * ((64 * 1024) + sizeof(struct sk_buff)));
 
-               inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
-               inet->uc_ttl = -1;
-               inet->pmtudisc = IP_PMTUDISC_DONT;
-
-               /* Unhash it so that IP input processing does not even
-                * see it, we do not wish this socket to see incoming
-                * packets.
+               /*
+                * Speedup sock_wfree()
                 */
-               per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
+               sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
+               inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
        }
+
+       /* Control parameters for ECHO replies. */
+       net->ipv4.sysctl_icmp_echo_ignore_all = 0;
+       net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
+
+       /* Control parameter - ignore bogus broadcast responses? */
+       net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
+
+       /*
+        *      Configurable global rate limit.
+        *
+        *      ratelimit defines tokens/packet consumed for dst->rate_token
+        *      bucket ratemask defines which icmp types are ratelimited by
+        *      setting it's bit position.
+        *
+        *      default:
+        *      dest unreachable (3), source quench (4),
+        *      time exceeded (11), parameter problem (12)
+        */
+
+       net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
+       net->ipv4.sysctl_icmp_ratemask = 0x1818;
+       net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
+
+       return 0;
+
+fail:
+       for_each_possible_cpu(i)
+               inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
+       kfree(net->ipv4.icmp_sk);
+       return err;
+}
+
+static struct pernet_operations __net_initdata icmp_sk_ops = {
+       .init = icmp_sk_init,
+       .exit = icmp_sk_exit,
+};
+
+int __init icmp_init(void)
+{
+       return register_pernet_subsys(&icmp_sk_ops);
 }
 
 EXPORT_SYMBOL(icmp_err_convert);
 EXPORT_SYMBOL(icmp_send);
-EXPORT_SYMBOL(icmp_statistics);
-EXPORT_SYMBOL(icmpmsg_statistics);
 EXPORT_SYMBOL(xrlim_allow);