[NETNS][IPV6] tcp6 - make proc per namespace
[safe/jmp/linux-2.6] / net / ipv6 / tcp_ipv6.c
index 32dc329..56d0cea 100644 (file)
@@ -59,6 +59,7 @@
 #include <net/snmp.h>
 #include <net/dsfield.h>
 #include <net/timewait_sock.h>
+#include <net/netdma.h>
 
 #include <asm/uaccess.h>
 
@@ -68,9 +69,6 @@
 #include <linux/crypto.h>
 #include <linux/scatterlist.h>
 
-/* Socket used for sending RSTs and ACKs */
-static struct socket *tcp6_socket;
-
 static void    tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
 static void    tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
 static void    tcp_v6_send_check(struct sock *sk, int len,
@@ -85,12 +83,6 @@ static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
 #endif
 
-static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
-{
-       return inet_csk_get_port(&tcp_hashinfo, sk, snum,
-                                inet6_csk_bind_conflict);
-}
-
 static void tcp_v6_hash(struct sock *sk)
 {
        if (sk->sk_state != TCP_CLOSE) {
@@ -99,7 +91,7 @@ static void tcp_v6_hash(struct sock *sk)
                        return;
                }
                local_bh_disable();
-               __inet6_hash(&tcp_hashinfo, sk);
+               __inet6_hash(sk);
                local_bh_enable();
        }
 }
@@ -264,7 +256,7 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
        if (final_p)
                ipv6_addr_copy(&fl.fl6_dst, final_p);
 
-       if ((err = __xfrm_lookup(&dst, &fl, sk, 1)) < 0) {
+       if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
                if (err == -EREMOTE)
                        err = ip6_dst_blackhole(sk, &dst, &fl);
                if (err < 0)
@@ -329,8 +321,8 @@ static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
        struct tcp_sock *tp;
        __u32 seq;
 
-       sk = inet6_lookup(&tcp_hashinfo, &hdr->daddr, th->dest, &hdr->saddr,
-                         th->source, skb->dev->ifindex);
+       sk = inet6_lookup(skb->dev->nd_net, &tcp_hashinfo, &hdr->daddr,
+                       th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 
        if (sk == NULL) {
                ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
@@ -460,8 +452,7 @@ out:
 }
 
 
-static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
-                             struct dst_entry *dst)
+static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req)
 {
        struct inet6_request_sock *treq = inet6_rsk(req);
        struct ipv6_pinfo *np = inet6_sk(sk);
@@ -469,6 +460,7 @@ static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
        struct ipv6_txoptions *opt = NULL;
        struct in6_addr * final_p = NULL, final;
        struct flowi fl;
+       struct dst_entry *dst;
        int err = -1;
 
        memset(&fl, 0, sizeof(fl));
@@ -481,24 +473,22 @@ static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
        fl.fl_ip_sport = inet_sk(sk)->sport;
        security_req_classify_flow(req, &fl);
 
-       if (dst == NULL) {
-               opt = np->opt;
-               if (opt && opt->srcrt) {
-                       struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
-                       ipv6_addr_copy(&final, &fl.fl6_dst);
-                       ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
-                       final_p = &final;
-               }
-
-               err = ip6_dst_lookup(sk, &dst, &fl);
-               if (err)
-                       goto done;
-               if (final_p)
-                       ipv6_addr_copy(&fl.fl6_dst, final_p);
-               if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
-                       goto done;
+       opt = np->opt;
+       if (opt && opt->srcrt) {
+               struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
+               ipv6_addr_copy(&final, &fl.fl6_dst);
+               ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
+               final_p = &final;
        }
 
+       err = ip6_dst_lookup(sk, &dst, &fl);
+       if (err)
+               goto done;
+       if (final_p)
+               ipv6_addr_copy(&fl.fl6_dst, final_p);
+       if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
+               goto done;
+
        skb = tcp_make_synack(sk, dst, req);
        if (skb) {
                struct tcphdr *th = tcp_hdr(skb);
@@ -519,6 +509,20 @@ done:
        return err;
 }
 
+static inline void syn_flood_warning(struct sk_buff *skb)
+{
+#ifdef CONFIG_SYN_COOKIES
+       if (sysctl_tcp_syncookies)
+               printk(KERN_INFO
+                      "TCPv6: Possible SYN flooding on port %d. "
+                      "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
+       else
+#endif
+               printk(KERN_INFO
+                      "TCPv6: Possible SYN flooding on port %d. "
+                      "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
+}
+
 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 {
        if (inet6_rsk(req)->pktopts)
@@ -560,16 +564,16 @@ static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
                             char *newkey, u8 newkeylen)
 {
        /* Add key to the list */
-       struct tcp6_md5sig_key *key;
+       struct tcp_md5sig_key *key;
        struct tcp_sock *tp = tcp_sk(sk);
        struct tcp6_md5sig_key *keys;
 
-       key = (struct tcp6_md5sig_key*) tcp_v6_md5_do_lookup(sk, peer);
+       key = tcp_v6_md5_do_lookup(sk, peer);
        if (key) {
                /* modify existing entry - just update that one */
-               kfree(key->base.key);
-               key->base.key = newkey;
-               key->base.keylen = newkeylen;
+               kfree(key->key);
+               key->key = newkey;
+               key->keylen = newkeylen;
        } else {
                /* reallocate new list if current one is full. */
                if (!tp->md5sig_info) {
@@ -580,7 +584,10 @@ static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
                        }
                        sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
                }
-               tcp_alloc_md5sig_pool();
+               if (tcp_alloc_md5sig_pool() == NULL) {
+                       kfree(newkey);
+                       return -ENOMEM;
+               }
                if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
                        keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
                                       (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
@@ -633,10 +640,6 @@ static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
                                kfree(tp->md5sig_info->keys6);
                                tp->md5sig_info->keys6 = NULL;
                                tp->md5sig_info->alloced6 = 0;
-
-                               tcp_free_md5sig_pool();
-
-                               return 0;
                        } else {
                                /* shrink the database */
                                if (tp->md5sig_info->entries6 != i)
@@ -645,6 +648,8 @@ static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
                                                (tp->md5sig_info->entries6 - i)
                                                * sizeof (tp->md5sig_info->keys6[0]));
                        }
+                       tcp_free_md5sig_pool();
+                       return 0;
                }
        }
        return -ENOENT;
@@ -731,7 +736,7 @@ static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
                                   struct in6_addr *saddr,
                                   struct in6_addr *daddr,
                                   struct tcphdr *th, int protocol,
-                                  int tcplen)
+                                  unsigned int tcplen)
 {
        struct scatterlist sg[4];
        __u16 data_len;
@@ -745,7 +750,7 @@ static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
 
        hp = tcp_get_md5sig_pool();
        if (!hp) {
-               printk(KERN_WARNING "%s(): hash pool not found...\n", __FUNCTION__);
+               printk(KERN_WARNING "%s(): hash pool not found...\n", __func__);
                goto clear_hash_noput;
        }
        bp = &hp->md5_blk.ip6;
@@ -780,22 +785,22 @@ static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
        sg_set_buf(&sg[block++], key->key, key->keylen);
        nbytes += key->keylen;
 
-       sg_mark_end(sg, block);
+       sg_mark_end(&sg[block - 1]);
 
        /* Now store the hash into the packet */
        err = crypto_hash_init(desc);
        if (err) {
-               printk(KERN_WARNING "%s(): hash_init failed\n", __FUNCTION__);
+               printk(KERN_WARNING "%s(): hash_init failed\n", __func__);
                goto clear_hash;
        }
        err = crypto_hash_update(desc, sg, nbytes);
        if (err) {
-               printk(KERN_WARNING "%s(): hash_update failed\n", __FUNCTION__);
+               printk(KERN_WARNING "%s(): hash_update failed\n", __func__);
                goto clear_hash;
        }
        err = crypto_hash_final(desc, md5_hash);
        if (err) {
-               printk(KERN_WARNING "%s(): hash_final failed\n", __FUNCTION__);
+               printk(KERN_WARNING "%s(): hash_final failed\n", __func__);
                goto clear_hash;
        }
 
@@ -816,7 +821,7 @@ static int tcp_v6_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
                                struct dst_entry *dst,
                                struct request_sock *req,
                                struct tcphdr *th, int protocol,
-                               int tcplen)
+                               unsigned int tcplen)
 {
        struct in6_addr *saddr, *daddr;
 
@@ -921,7 +926,7 @@ done_opts:
 }
 #endif
 
-static struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
+struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
        .family         =       AF_INET6,
        .obj_size       =       sizeof(struct tcp6_request_sock),
        .rtx_syn_ack    =       tcp_v6_send_synack,
@@ -983,7 +988,9 @@ static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
        struct tcphdr *th = tcp_hdr(skb), *t1;
        struct sk_buff *buff;
        struct flowi fl;
-       int tot_len = sizeof(*th);
+       struct net *net = skb->dst->dev->nd_net;
+       struct sock *ctl_sk = net->ipv6.tcp_sk;
+       unsigned int tot_len = sizeof(*th);
 #ifdef CONFIG_TCP_MD5SIG
        struct tcp_md5sig_key *key;
 #endif
@@ -1063,11 +1070,14 @@ static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
        fl.fl_ip_sport = t1->source;
        security_skb_classify_flow(skb, &fl);
 
-       /* sk = NULL, but it is safe for now. RST socket required. */
-       if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
+       /* Pass a socket to ip6_dst_lookup either it is for RST
+        * Underlying function will use this to retrieve the network
+        * namespace
+        */
+       if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
 
                if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
-                       ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
+                       ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
                        TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
                        TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
                        return;
@@ -1083,7 +1093,9 @@ static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
        struct tcphdr *th = tcp_hdr(skb), *t1;
        struct sk_buff *buff;
        struct flowi fl;
-       int tot_len = sizeof(struct tcphdr);
+       struct net *net = skb->dev->nd_net;
+       struct sock *ctl_sk = net->ipv6.tcp_sk;
+       unsigned int tot_len = sizeof(struct tcphdr);
        __be32 *topt;
 #ifdef CONFIG_TCP_MD5SIG
        struct tcp_md5sig_key *key;
@@ -1164,9 +1176,9 @@ static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
        fl.fl_ip_sport = t1->source;
        security_skb_classify_flow(skb, &fl);
 
-       if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
+       if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
                if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
-                       ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
+                       ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
                        TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
                        return;
                }
@@ -1206,9 +1218,9 @@ static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
        if (req)
                return tcp_check_req(sk, skb, req, prev);
 
-       nsk = __inet6_lookup_established(&tcp_hashinfo, &ipv6_hdr(skb)->saddr,
-                                        th->source, &ipv6_hdr(skb)->daddr,
-                                        ntohs(th->dest), inet6_iif(skb));
+       nsk = __inet6_lookup_established(sk->sk_net, &tcp_hashinfo,
+                       &ipv6_hdr(skb)->saddr, th->source,
+                       &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
 
        if (nsk) {
                if (nsk->sk_state != TCP_TIME_WAIT) {
@@ -1219,9 +1231,9 @@ static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
                return NULL;
        }
 
-#if 0 /*def CONFIG_SYN_COOKIES*/
+#ifdef CONFIG_SYN_COOKIES
        if (!th->rst && !th->syn && th->ack)
-               sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
+               sk = cookie_v6_check(sk, skb);
 #endif
        return sk;
 }
@@ -1237,6 +1249,11 @@ static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
        struct tcp_sock *tp = tcp_sk(sk);
        struct request_sock *req = NULL;
        __u32 isn = TCP_SKB_CB(skb)->when;
+#ifdef CONFIG_SYN_COOKIES
+       int want_cookie = 0;
+#else
+#define want_cookie 0
+#endif
 
        if (skb->protocol == htons(ETH_P_IP))
                return tcp_v4_conn_request(sk, skb);
@@ -1244,12 +1261,14 @@ static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
        if (!ipv6_unicast_destination(skb))
                goto drop;
 
-       /*
-        *      There are no SYN attacks on IPv6, yet...
-        */
        if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
                if (net_ratelimit())
-                       printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
+                       syn_flood_warning(skb);
+#ifdef CONFIG_SYN_COOKIES
+               if (sysctl_tcp_syncookies)
+                       want_cookie = 1;
+               else
+#endif
                goto drop;
        }
 
@@ -1270,39 +1289,51 @@ static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 
        tcp_parse_options(skb, &tmp_opt, 0);
 
+       if (want_cookie) {
+               tcp_clear_options(&tmp_opt);
+               tmp_opt.saw_tstamp = 0;
+       }
+
        tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
        tcp_openreq_init(req, &tmp_opt, skb);
 
        treq = inet6_rsk(req);
        ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
        ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
-       TCP_ECN_create_request(req, tcp_hdr(skb));
        treq->pktopts = NULL;
-       if (ipv6_opt_accepted(sk, skb) ||
-           np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
-           np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
-               atomic_inc(&skb->users);
-               treq->pktopts = skb;
-       }
-       treq->iif = sk->sk_bound_dev_if;
+       if (!want_cookie)
+               TCP_ECN_create_request(req, tcp_hdr(skb));
+
+       if (want_cookie) {
+               isn = cookie_v6_init_sequence(sk, skb, &req->mss);
+       } else if (!isn) {
+               if (ipv6_opt_accepted(sk, skb) ||
+                   np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
+                   np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
+                       atomic_inc(&skb->users);
+                       treq->pktopts = skb;
+               }
+               treq->iif = sk->sk_bound_dev_if;
 
-       /* So that link locals have meaning */
-       if (!sk->sk_bound_dev_if &&
-           ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
-               treq->iif = inet6_iif(skb);
+               /* So that link locals have meaning */
+               if (!sk->sk_bound_dev_if &&
+                   ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
+                       treq->iif = inet6_iif(skb);
 
-       if (isn == 0)
                isn = tcp_v6_init_sequence(skb);
+       }
 
        tcp_rsk(req)->snt_isn = isn;
 
        security_inet_conn_request(sk, skb, req);
 
-       if (tcp_v6_send_synack(sk, req, NULL))
+       if (tcp_v6_send_synack(sk, req))
                goto drop;
 
-       inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
-       return 0;
+       if (!want_cookie) {
+               inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
+               return 0;
+       }
 
 drop:
        if (req)
@@ -1502,8 +1533,8 @@ static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
        }
 #endif
 
-       __inet6_hash(&tcp_hashinfo, newsk);
-       inet_inherit_port(&tcp_hashinfo, sk, newsk);
+       __inet6_hash(newsk);
+       inet_inherit_port(sk, newsk);
 
        return newsk;
 
@@ -1708,9 +1739,10 @@ static int tcp_v6_rcv(struct sk_buff *skb)
        TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
        TCP_SKB_CB(skb)->sacked = 0;
 
-       sk = __inet6_lookup(&tcp_hashinfo, &ipv6_hdr(skb)->saddr, th->source,
-                           &ipv6_hdr(skb)->daddr, ntohs(th->dest),
-                           inet6_iif(skb));
+       sk = __inet6_lookup(skb->dev->nd_net, &tcp_hashinfo,
+                       &ipv6_hdr(skb)->saddr, th->source,
+                       &ipv6_hdr(skb)->daddr, ntohs(th->dest),
+                       inet6_iif(skb));
 
        if (!sk)
                goto no_tcp_socket;
@@ -1732,6 +1764,8 @@ process:
        if (!sock_owned_by_user(sk)) {
 #ifdef CONFIG_NET_DMA
                struct tcp_sock *tp = tcp_sk(sk);
+               if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
+                       tp->ucopy.dma_chan = get_softnet_dma();
                if (tp->ucopy.dma_chan)
                        ret = tcp_v6_do_rcv(sk, skb);
                else
@@ -1788,7 +1822,7 @@ do_time_wait:
        {
                struct sock *sk2;
 
-               sk2 = inet6_lookup_listener(&tcp_hashinfo,
+               sk2 = inet6_lookup_listener(skb->dev->nd_net, &tcp_hashinfo,
                                            &ipv6_hdr(skb)->daddr,
                                            ntohs(th->dest), inet6_iif(skb));
                if (sk2 != NULL) {
@@ -1828,6 +1862,7 @@ static struct inet_connection_sock_af_ops ipv6_specific = {
        .getsockopt        = ipv6_getsockopt,
        .addr2sockaddr     = inet6_csk_addr2sockaddr,
        .sockaddr_len      = sizeof(struct sockaddr_in6),
+       .bind_conflict     = inet6_csk_bind_conflict,
 #ifdef CONFIG_COMPAT
        .compat_setsockopt = compat_ipv6_setsockopt,
        .compat_getsockopt = compat_ipv6_getsockopt,
@@ -1859,6 +1894,7 @@ static struct inet_connection_sock_af_ops ipv6_mapped = {
        .getsockopt        = ipv6_getsockopt,
        .addr2sockaddr     = inet6_csk_addr2sockaddr,
        .sockaddr_len      = sizeof(struct sockaddr_in6),
+       .bind_conflict     = inet6_csk_bind_conflict,
 #ifdef CONFIG_COMPAT
        .compat_setsockopt = compat_ipv6_setsockopt,
        .compat_getsockopt = compat_ipv6_getsockopt,
@@ -2093,17 +2129,19 @@ static struct tcp_seq_afinfo tcp6_seq_afinfo = {
        .seq_fops       = &tcp6_seq_fops,
 };
 
-int __init tcp6_proc_init(void)
+int tcp6_proc_init(struct net *net)
 {
-       return tcp_proc_register(&tcp6_seq_afinfo);
+       return tcp_proc_register(net, &tcp6_seq_afinfo);
 }
 
-void tcp6_proc_exit(void)
+void tcp6_proc_exit(struct net *net)
 {
-       tcp_proc_unregister(&tcp6_seq_afinfo);
+       tcp_proc_unregister(net, &tcp6_seq_afinfo);
 }
 #endif
 
+DEFINE_PROTO_INUSE(tcpv6)
+
 struct proto tcpv6_prot = {
        .name                   = "TCPv6",
        .owner                  = THIS_MODULE,
@@ -2120,8 +2158,8 @@ struct proto tcpv6_prot = {
        .recvmsg                = tcp_recvmsg,
        .backlog_rcv            = tcp_v6_do_rcv,
        .hash                   = tcp_v6_hash,
-       .unhash                 = tcp_unhash,
-       .get_port               = tcp_v6_get_port,
+       .unhash                 = inet_unhash,
+       .get_port               = inet_csk_get_port,
        .enter_memory_pressure  = tcp_enter_memory_pressure,
        .sockets_allocated      = &tcp_sockets_allocated,
        .memory_allocated       = &tcp_memory_allocated,
@@ -2134,10 +2172,12 @@ struct proto tcpv6_prot = {
        .obj_size               = sizeof(struct tcp6_sock),
        .twsk_prot              = &tcp6_timewait_sock_ops,
        .rsk_prot               = &tcp6_request_sock_ops,
+       .hashinfo               = &tcp_hashinfo,
 #ifdef CONFIG_COMPAT
        .compat_setsockopt      = compat_tcp_setsockopt,
        .compat_getsockopt      = compat_tcp_getsockopt,
 #endif
+       REF_PROTO_INUSE(tcpv6)
 };
 
 static struct inet6_protocol tcpv6_protocol = {
@@ -2159,14 +2199,60 @@ static struct inet_protosw tcpv6_protosw = {
                                INET_PROTOSW_ICSK,
 };
 
-void __init tcpv6_init(void)
+static int tcpv6_net_init(struct net *net)
+{
+       int err;
+       struct socket *sock;
+       struct sock *sk;
+
+       err = inet_csk_ctl_sock_create(&sock, PF_INET6, SOCK_RAW, IPPROTO_TCP);
+       if (err)
+               return err;
+
+       net->ipv6.tcp_sk = sk = sock->sk;
+       sk_change_net(sk, net);
+       return err;
+}
+
+static void tcpv6_net_exit(struct net *net)
+{
+       sk_release_kernel(net->ipv6.tcp_sk);
+}
+
+static struct pernet_operations tcpv6_net_ops = {
+       .init = tcpv6_net_init,
+       .exit = tcpv6_net_exit,
+};
+
+int __init tcpv6_init(void)
 {
+       int ret;
+
+       ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
+       if (ret)
+               goto out;
+
        /* register inet6 protocol */
-       if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
-               printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
-       inet6_register_protosw(&tcpv6_protosw);
+       ret = inet6_register_protosw(&tcpv6_protosw);
+       if (ret)
+               goto out_tcpv6_protocol;
+
+       ret = register_pernet_subsys(&tcpv6_net_ops);
+       if (ret)
+               goto out_tcpv6_protosw;
+out:
+       return ret;
 
-       if (inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6, SOCK_RAW,
-                                    IPPROTO_TCP) < 0)
-               panic("Failed to create the TCPv6 control socket.\n");
+out_tcpv6_protocol:
+       inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
+out_tcpv6_protosw:
+       inet6_unregister_protosw(&tcpv6_protosw);
+       goto out;
+}
+
+void tcpv6_exit(void)
+{
+       unregister_pernet_subsys(&tcpv6_net_ops);
+       inet6_unregister_protosw(&tcpv6_protosw);
+       inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
 }