gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)
authorPablo Neira <pablo@netfilter.org>
Sun, 8 May 2016 22:55:48 +0000 (00:55 +0200)
committerDavid S. Miller <davem@davemloft.net>
Tue, 10 May 2016 16:25:04 +0000 (12:25 -0400)
This is an initial implementation of a netdev driver for GTP datapath
(GTP-U) v0 and v1, according to the GSM TS 09.60 and 3GPP TS 29.060
standards. This tunneling protocol is used to prevent subscribers from
accessing mobile carrier core network infrastructure.

This implementation requires a GGSN userspace daemon that implements the
signaling protocol (GTP-C), such as OpenGGSN [1]. This userspace daemon
updates the PDP context database that represents active subscriber
sessions through a genetlink interface.

For more context on this tunneling protocol, you can check the slides
that were presented during the NetDev 1.1 [2].

Only IPv4 is supported at this time.

[1] http://git.osmocom.org/openggsn/
[2] http://www.netdevconf.org/1.1/proceedings/slides/schultz-welte-osmocom-gtp.pdf

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/Kconfig
drivers/net/Makefile
drivers/net/gtp.c [new file with mode: 0644]
include/net/gtp.h [new file with mode: 0644]
include/uapi/linux/Kbuild
include/uapi/linux/gtp.h [new file with mode: 0644]
include/uapi/linux/if_link.h
include/uapi/linux/udp.h

index befd67df08e1f26410e0817cab1506183464b522..0c5415b05ea945f39da2482ca8b34fad91ff3c31 100644 (file)
@@ -192,6 +192,23 @@ config GENEVE
          To compile this driver as a module, choose M here: the module
          will be called geneve.
 
+config GTP
+       tristate "GPRS Tunneling Protocol datapath (GTP-U)"
+       depends on INET && NET_UDP_TUNNEL
+       select NET_IP_TUNNEL
+       ---help---
+         This allows one to create gtp virtual interfaces that provide
+         the GPRS Tunneling Protocol datapath (GTP-U). This tunneling protocol
+         is used to prevent subscribers from accessing mobile carrier core
+         network infrastructure. This driver requires a userspace software that
+         implements the signaling protocol (GTP-C) to update its PDP context
+         base, such as OpenGGSN <http://git.osmocom.org/openggsn/). This
+         tunneling protocol is implemented according to the GSM TS 09.60 and
+         3GPP TS 29.060 standards.
+
+         To compile this drivers as a module, choose M here: the module
+         wil be called gtp.
+
 config MACSEC
        tristate "IEEE 802.1AE MAC-level encryption (MACsec)"
        select CRYPTO
index 1aa7cb845663152c272633667d3b1affe619f3dd..7336cbd3ef5d94d4d28a6e2913e0e8439b205bf3 100644 (file)
@@ -25,6 +25,7 @@ obj-$(CONFIG_VETH) += veth.o
 obj-$(CONFIG_VIRTIO_NET) += virtio_net.o
 obj-$(CONFIG_VXLAN) += vxlan.o
 obj-$(CONFIG_GENEVE) += geneve.o
+obj-$(CONFIG_GTP) += gtp.o
 obj-$(CONFIG_NLMON) += nlmon.o
 obj-$(CONFIG_NET_VRF) += vrf.o
 
diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c
new file mode 100644 (file)
index 0000000..8ce1104
--- /dev/null
@@ -0,0 +1,1364 @@
+/* GTP according to GSM TS 09.60 / 3GPP TS 29.060
+ *
+ * (C) 2012-2014 by sysmocom - s.f.m.c. GmbH
+ * (C) 2016 by Pablo Neira Ayuso <pablo@netfilter.org>
+ *
+ * Author: Harald Welte <hwelte@sysmocom.de>
+ *        Pablo Neira Ayuso <pablo@netfilter.org>
+ *        Andreas Schultz <aschultz@travelping.com>
+ *
+ * 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
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/module.h>
+#include <linux/version.h>
+#include <linux/skbuff.h>
+#include <linux/udp.h>
+#include <linux/rculist.h>
+#include <linux/jhash.h>
+#include <linux/if_tunnel.h>
+#include <linux/net.h>
+#include <linux/file.h>
+#include <linux/gtp.h>
+
+#include <net/net_namespace.h>
+#include <net/protocol.h>
+#include <net/ip.h>
+#include <net/udp.h>
+#include <net/udp_tunnel.h>
+#include <net/icmp.h>
+#include <net/xfrm.h>
+#include <net/genetlink.h>
+#include <net/netns/generic.h>
+#include <net/gtp.h>
+
+/* An active session for the subscriber. */
+struct pdp_ctx {
+       struct hlist_node       hlist_tid;
+       struct hlist_node       hlist_addr;
+
+       union {
+               u64             tid;
+               struct {
+                       u64     tid;
+                       u16     flow;
+               } v0;
+               struct {
+                       u32     i_tei;
+                       u32     o_tei;
+               } v1;
+       } u;
+       u8                      gtp_version;
+       u16                     af;
+
+       struct in_addr          ms_addr_ip4;
+       struct in_addr          sgsn_addr_ip4;
+
+       atomic_t                tx_seq;
+       struct rcu_head         rcu_head;
+};
+
+/* One instance of the GTP device. */
+struct gtp_dev {
+       struct list_head        list;
+
+       struct socket           *sock0;
+       struct socket           *sock1u;
+
+       struct net              *net;
+       struct net_device       *dev;
+
+       unsigned int            hash_size;
+       struct hlist_head       *tid_hash;
+       struct hlist_head       *addr_hash;
+};
+
+static int gtp_net_id __read_mostly;
+
+struct gtp_net {
+       struct list_head gtp_dev_list;
+};
+
+static u32 gtp_h_initval;
+
+static inline u32 gtp0_hashfn(u64 tid)
+{
+       u32 *tid32 = (u32 *) &tid;
+       return jhash_2words(tid32[0], tid32[1], gtp_h_initval);
+}
+
+static inline u32 gtp1u_hashfn(u32 tid)
+{
+       return jhash_1word(tid, gtp_h_initval);
+}
+
+static inline u32 ipv4_hashfn(__be32 ip)
+{
+       return jhash_1word((__force u32)ip, gtp_h_initval);
+}
+
+/* Resolve a PDP context structure based on the 64bit TID. */
+static struct pdp_ctx *gtp0_pdp_find(struct gtp_dev *gtp, u64 tid)
+{
+       struct hlist_head *head;
+       struct pdp_ctx *pdp;
+
+       head = &gtp->tid_hash[gtp0_hashfn(tid) % gtp->hash_size];
+
+       hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
+               if (pdp->gtp_version == GTP_V0 &&
+                   pdp->u.v0.tid == tid)
+                       return pdp;
+       }
+       return NULL;
+}
+
+/* Resolve a PDP context structure based on the 32bit TEI. */
+static struct pdp_ctx *gtp1_pdp_find(struct gtp_dev *gtp, u32 tid)
+{
+       struct hlist_head *head;
+       struct pdp_ctx *pdp;
+
+       head = &gtp->tid_hash[gtp1u_hashfn(tid) % gtp->hash_size];
+
+       hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
+               if (pdp->gtp_version == GTP_V1 &&
+                   pdp->u.v1.i_tei == tid)
+                       return pdp;
+       }
+       return NULL;
+}
+
+/* Resolve a PDP context based on IPv4 address of MS. */
+static struct pdp_ctx *ipv4_pdp_find(struct gtp_dev *gtp, __be32 ms_addr)
+{
+       struct hlist_head *head;
+       struct pdp_ctx *pdp;
+
+       head = &gtp->addr_hash[ipv4_hashfn(ms_addr) % gtp->hash_size];
+
+       hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
+               if (pdp->af == AF_INET &&
+                   pdp->ms_addr_ip4.s_addr == ms_addr)
+                       return pdp;
+       }
+
+       return NULL;
+}
+
+static bool gtp_check_src_ms_ipv4(struct sk_buff *skb, struct pdp_ctx *pctx,
+                                 unsigned int hdrlen)
+{
+       struct iphdr *iph;
+
+       if (!pskb_may_pull(skb, hdrlen + sizeof(struct iphdr)))
+               return false;
+
+       iph = (struct iphdr *)(skb->data + hdrlen + sizeof(struct iphdr));
+
+       return iph->saddr != pctx->ms_addr_ip4.s_addr;
+}
+
+/* Check if the inner IP source address in this packet is assigned to any
+ * existing mobile subscriber.
+ */
+static bool gtp_check_src_ms(struct sk_buff *skb, struct pdp_ctx *pctx,
+                            unsigned int hdrlen)
+{
+       switch (ntohs(skb->protocol)) {
+       case ETH_P_IP:
+               return gtp_check_src_ms_ipv4(skb, pctx, hdrlen);
+       }
+       return false;
+}
+
+/* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */
+static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb,
+                              bool xnet)
+{
+       unsigned int hdrlen = sizeof(struct udphdr) +
+                             sizeof(struct gtp0_header);
+       struct gtp0_header *gtp0;
+       struct pdp_ctx *pctx;
+       int ret = 0;
+
+       if (!pskb_may_pull(skb, hdrlen))
+               return -1;
+
+       gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
+
+       if ((gtp0->flags >> 5) != GTP_V0)
+               return 1;
+
+       if (gtp0->type != GTP_TPDU)
+               return 1;
+
+       rcu_read_lock();
+       pctx = gtp0_pdp_find(gtp, be64_to_cpu(gtp0->tid));
+       if (!pctx) {
+               netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
+               ret = -1;
+               goto out_rcu;
+       }
+
+       if (!gtp_check_src_ms(skb, pctx, hdrlen)) {
+               netdev_dbg(gtp->dev, "No PDP ctx for this MS\n");
+               ret = -1;
+               goto out_rcu;
+       }
+       rcu_read_unlock();
+
+       /* Get rid of the GTP + UDP headers. */
+       return iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet);
+out_rcu:
+       rcu_read_unlock();
+       return ret;
+}
+
+static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb,
+                               bool xnet)
+{
+       unsigned int hdrlen = sizeof(struct udphdr) +
+                             sizeof(struct gtp1_header);
+       struct gtp1_header *gtp1;
+       struct pdp_ctx *pctx;
+       int ret = 0;
+
+       if (!pskb_may_pull(skb, hdrlen))
+               return -1;
+
+       gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
+
+       if ((gtp1->flags >> 5) != GTP_V1)
+               return 1;
+
+       if (gtp1->type != GTP_TPDU)
+               return 1;
+
+       /* From 29.060: "This field shall be present if and only if any one or
+        * more of the S, PN and E flags are set.".
+        *
+        * If any of the bit is set, then the remaining ones also have to be
+        * set.
+        */
+       if (gtp1->flags & GTP1_F_MASK)
+               hdrlen += 4;
+
+       /* Make sure the header is larger enough, including extensions. */
+       if (!pskb_may_pull(skb, hdrlen))
+               return -1;
+
+       rcu_read_lock();
+       pctx = gtp1_pdp_find(gtp, ntohl(gtp1->tid));
+       if (!pctx) {
+               netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
+               ret = -1;
+               goto out_rcu;
+       }
+
+       if (!gtp_check_src_ms(skb, pctx, hdrlen)) {
+               netdev_dbg(gtp->dev, "No PDP ctx for this MS\n");
+               ret = -1;
+               goto out_rcu;
+       }
+       rcu_read_unlock();
+
+       /* Get rid of the GTP + UDP headers. */
+       return iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet);
+out_rcu:
+       rcu_read_unlock();
+       return ret;
+}
+
+static void gtp_encap_disable(struct gtp_dev *gtp)
+{
+       if (gtp->sock0 && gtp->sock0->sk) {
+               udp_sk(gtp->sock0->sk)->encap_type = 0;
+               rcu_assign_sk_user_data(gtp->sock0->sk, NULL);
+       }
+       if (gtp->sock1u && gtp->sock1u->sk) {
+               udp_sk(gtp->sock1u->sk)->encap_type = 0;
+               rcu_assign_sk_user_data(gtp->sock1u->sk, NULL);
+       }
+
+       gtp->sock0 = NULL;
+       gtp->sock1u = NULL;
+}
+
+static void gtp_encap_destroy(struct sock *sk)
+{
+       struct gtp_dev *gtp;
+
+       gtp = rcu_dereference_sk_user_data(sk);
+       if (gtp)
+               gtp_encap_disable(gtp);
+}
+
+/* UDP encapsulation receive handler. See net/ipv4/udp.c.
+ * Return codes: 0: success, <0: error, >0: pass up to userspace UDP socket.
+ */
+static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb)
+{
+       struct pcpu_sw_netstats *stats;
+       struct gtp_dev *gtp;
+       bool xnet;
+       int ret;
+
+       gtp = rcu_dereference_sk_user_data(sk);
+       if (!gtp)
+               return 1;
+
+       netdev_dbg(gtp->dev, "encap_recv sk=%p\n", sk);
+
+       xnet = !net_eq(gtp->net, dev_net(gtp->dev));
+
+       switch (udp_sk(sk)->encap_type) {
+       case UDP_ENCAP_GTP0:
+               netdev_dbg(gtp->dev, "received GTP0 packet\n");
+               ret = gtp0_udp_encap_recv(gtp, skb, xnet);
+               break;
+       case UDP_ENCAP_GTP1U:
+               netdev_dbg(gtp->dev, "received GTP1U packet\n");
+               ret = gtp1u_udp_encap_recv(gtp, skb, xnet);
+               break;
+       default:
+               ret = -1; /* Shouldn't happen. */
+       }
+
+       switch (ret) {
+       case 1:
+               netdev_dbg(gtp->dev, "pass up to the process\n");
+               return 1;
+       case 0:
+               netdev_dbg(gtp->dev, "forwarding packet from GGSN to uplink\n");
+               break;
+       case -1:
+               netdev_dbg(gtp->dev, "GTP packet has been dropped\n");
+               kfree_skb(skb);
+               return 0;
+       }
+
+       /* Now that the UDP and the GTP header have been removed, set up the
+        * new network header. This is required by the upper layer to
+        * calculate the transport header.
+        */
+       skb_reset_network_header(skb);
+
+       skb->dev = gtp->dev;
+
+       stats = this_cpu_ptr(gtp->dev->tstats);
+       u64_stats_update_begin(&stats->syncp);
+       stats->rx_packets++;
+       stats->rx_bytes += skb->len;
+       u64_stats_update_end(&stats->syncp);
+
+       netif_rx(skb);
+
+       return 0;
+}
+
+static int gtp_dev_init(struct net_device *dev)
+{
+       struct gtp_dev *gtp = netdev_priv(dev);
+
+       gtp->dev = dev;
+
+       dev->tstats = alloc_percpu(struct pcpu_sw_netstats);
+       if (!dev->tstats)
+               return -ENOMEM;
+
+       return 0;
+}
+
+static void gtp_dev_uninit(struct net_device *dev)
+{
+       struct gtp_dev *gtp = netdev_priv(dev);
+
+       gtp_encap_disable(gtp);
+       free_percpu(dev->tstats);
+}
+
+static struct rtable *ip4_route_output_gtp(struct net *net, struct flowi4 *fl4,
+                                          const struct sock *sk, __be32 daddr)
+{
+       memset(fl4, 0, sizeof(*fl4));
+       fl4->flowi4_oif         = sk->sk_bound_dev_if;
+       fl4->daddr              = daddr;
+       fl4->saddr              = inet_sk(sk)->inet_saddr;
+       fl4->flowi4_tos         = RT_CONN_FLAGS(sk);
+       fl4->flowi4_proto       = sk->sk_protocol;
+
+       return ip_route_output_key(net, fl4);
+}
+
+static inline void gtp0_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
+{
+       int payload_len = skb->len;
+       struct gtp0_header *gtp0;
+
+       gtp0 = (struct gtp0_header *) skb_push(skb, sizeof(*gtp0));
+
+       gtp0->flags     = 0x1e; /* v0, GTP-non-prime. */
+       gtp0->type      = GTP_TPDU;
+       gtp0->length    = htons(payload_len);
+       gtp0->seq       = htons((atomic_inc_return(&pctx->tx_seq) - 1) % 0xffff);
+       gtp0->flow      = htons(pctx->u.v0.flow);
+       gtp0->number    = 0xff;
+       gtp0->spare[0]  = gtp0->spare[1] = gtp0->spare[2] = 0xff;
+       gtp0->tid       = cpu_to_be64(pctx->u.v0.tid);
+}
+
+static inline void gtp1_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
+{
+       int payload_len = skb->len;
+       struct gtp1_header *gtp1;
+
+       gtp1 = (struct gtp1_header *) skb_push(skb, sizeof(*gtp1));
+
+       /* Bits    8  7  6  5  4  3  2  1
+        *        +--+--+--+--+--+--+--+--+
+        *        |version |PT| 1| E| S|PN|
+        *        +--+--+--+--+--+--+--+--+
+        *          0  0  1  1  1  0  0  0
+        */
+       gtp1->flags     = 0x38; /* v1, GTP-non-prime. */
+       gtp1->type      = GTP_TPDU;
+       gtp1->length    = htons(payload_len);
+       gtp1->tid       = htonl(pctx->u.v1.o_tei);
+
+       /* TODO: Suppport for extension header, sequence number and N-PDU.
+        *       Update the length field if any of them is available.
+        */
+}
+
+struct gtp_pktinfo {
+       struct sock             *sk;
+       struct iphdr            *iph;
+       struct flowi4           fl4;
+       struct rtable           *rt;
+       struct pdp_ctx          *pctx;
+       struct net_device       *dev;
+       __be16                  gtph_port;
+};
+
+static void gtp_push_header(struct sk_buff *skb, struct gtp_pktinfo *pktinfo)
+{
+       switch (pktinfo->pctx->gtp_version) {
+       case GTP_V0:
+               pktinfo->gtph_port = htons(GTP0_PORT);
+               gtp0_push_header(skb, pktinfo->pctx);
+               break;
+       case GTP_V1:
+               pktinfo->gtph_port = htons(GTP1U_PORT);
+               gtp1_push_header(skb, pktinfo->pctx);
+               break;
+       }
+}
+
+static inline void gtp_set_pktinfo_ipv4(struct gtp_pktinfo *pktinfo,
+                                       struct sock *sk, struct iphdr *iph,
+                                       struct pdp_ctx *pctx, struct rtable *rt,
+                                       struct flowi4 *fl4,
+                                       struct net_device *dev)
+{
+       pktinfo->sk     = sk;
+       pktinfo->iph    = iph;
+       pktinfo->pctx   = pctx;
+       pktinfo->rt     = rt;
+       pktinfo->fl4    = *fl4;
+       pktinfo->dev    = dev;
+}
+
+static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev,
+                            struct gtp_pktinfo *pktinfo)
+{
+       struct gtp_dev *gtp = netdev_priv(dev);
+       struct pdp_ctx *pctx;
+       struct rtable *rt;
+       struct flowi4 fl4;
+       struct iphdr *iph;
+       struct sock *sk;
+       __be16 df;
+       int mtu;
+
+       /* Read the IP destination address and resolve the PDP context.
+        * Prepend PDP header with TEI/TID from PDP ctx.
+        */
+       iph = ip_hdr(skb);
+       pctx = ipv4_pdp_find(gtp, iph->daddr);
+       if (!pctx) {
+               netdev_dbg(dev, "no PDP ctx found for %pI4, skip\n",
+                          &iph->daddr);
+               return -ENOENT;
+       }
+       netdev_dbg(dev, "found PDP context %p\n", pctx);
+
+       switch (pctx->gtp_version) {
+       case GTP_V0:
+               if (gtp->sock0)
+                       sk = gtp->sock0->sk;
+               else
+                       sk = NULL;
+               break;
+       case GTP_V1:
+               if (gtp->sock1u)
+                       sk = gtp->sock1u->sk;
+               else
+                       sk = NULL;
+               break;
+       default:
+               return -ENOENT;
+       }
+
+       if (!sk) {
+               netdev_dbg(dev, "no userspace socket is available, skip\n");
+               return -ENOENT;
+       }
+
+       rt = ip4_route_output_gtp(sock_net(sk), &fl4, gtp->sock0->sk,
+                                 pctx->sgsn_addr_ip4.s_addr);
+       if (IS_ERR(rt)) {
+               netdev_dbg(dev, "no route to SSGN %pI4\n",
+                          &pctx->sgsn_addr_ip4.s_addr);
+               dev->stats.tx_carrier_errors++;
+               goto err;
+       }
+
+       if (rt->dst.dev == dev) {
+               netdev_dbg(dev, "circular route to SSGN %pI4\n",
+                          &pctx->sgsn_addr_ip4.s_addr);
+               dev->stats.collisions++;
+               goto err_rt;
+       }
+
+       skb_dst_drop(skb);
+
+       /* This is similar to tnl_update_pmtu(). */
+       df = iph->frag_off;
+       if (df) {
+               mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
+                       sizeof(struct iphdr) - sizeof(struct udphdr);
+               switch (pctx->gtp_version) {
+               case GTP_V0:
+                       mtu -= sizeof(struct gtp0_header);
+                       break;
+               case GTP_V1:
+                       mtu -= sizeof(struct gtp1_header);
+                       break;
+               }
+       } else {
+               mtu = dst_mtu(&rt->dst);
+       }
+
+       rt->dst.ops->update_pmtu(&rt->dst, NULL, skb, mtu);
+
+       if (!skb_is_gso(skb) && (iph->frag_off & htons(IP_DF)) &&
+           mtu < ntohs(iph->tot_len)) {
+               netdev_dbg(dev, "packet too big, fragmentation needed\n");
+               memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
+               icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
+                         htonl(mtu));
+               goto err_rt;
+       }
+
+       gtp_set_pktinfo_ipv4(pktinfo, sk, iph, pctx, rt, &fl4, dev);
+       gtp_push_header(skb, pktinfo);
+
+       return 0;
+err_rt:
+       ip_rt_put(rt);
+err:
+       return -EBADMSG;
+}
+
+static netdev_tx_t gtp_dev_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       unsigned int proto = ntohs(skb->protocol);
+       struct gtp_pktinfo pktinfo;
+       int err;
+
+       /* Ensure there is sufficient headroom. */
+       if (skb_cow_head(skb, dev->needed_headroom))
+               goto tx_err;
+
+       skb_reset_inner_headers(skb);
+
+       /* PDP context lookups in gtp_build_skb_*() need rcu read-side lock. */
+       rcu_read_lock();
+       switch (proto) {
+       case ETH_P_IP:
+               err = gtp_build_skb_ip4(skb, dev, &pktinfo);
+               break;
+       default:
+               err = -EOPNOTSUPP;
+               break;
+       }
+       rcu_read_unlock();
+
+       if (err < 0)
+               goto tx_err;
+
+       switch (proto) {
+       case ETH_P_IP:
+               netdev_dbg(pktinfo.dev, "gtp -> IP src: %pI4 dst: %pI4\n",
+                          &pktinfo.iph->saddr, &pktinfo.iph->daddr);
+               udp_tunnel_xmit_skb(pktinfo.rt, pktinfo.sk, skb,
+                                   pktinfo.fl4.saddr, pktinfo.fl4.daddr,
+                                   pktinfo.iph->tos,
+                                   ip4_dst_hoplimit(&pktinfo.rt->dst),
+                                   htons(IP_DF),
+                                   pktinfo.gtph_port, pktinfo.gtph_port,
+                                   true, false);
+               break;
+       }
+
+       return NETDEV_TX_OK;
+tx_err:
+       dev->stats.tx_errors++;
+       dev_kfree_skb(skb);
+       return NETDEV_TX_OK;
+}
+
+static const struct net_device_ops gtp_netdev_ops = {
+       .ndo_init               = gtp_dev_init,
+       .ndo_uninit             = gtp_dev_uninit,
+       .ndo_start_xmit         = gtp_dev_xmit,
+       .ndo_get_stats64        = ip_tunnel_get_stats64,
+};
+
+static void gtp_link_setup(struct net_device *dev)
+{
+       dev->netdev_ops         = &gtp_netdev_ops;
+       dev->destructor         = free_netdev;
+
+       dev->hard_header_len = 0;
+       dev->addr_len = 0;
+
+       /* Zero header length. */
+       dev->type = ARPHRD_NONE;
+       dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
+
+       dev->priv_flags |= IFF_NO_QUEUE;
+       dev->features   |= NETIF_F_LLTX;
+       netif_keep_dst(dev);
+
+       /* Assume largest header, ie. GTPv0. */
+       dev->needed_headroom    = LL_MAX_HEADER +
+                                 sizeof(struct iphdr) +
+                                 sizeof(struct udphdr) +
+                                 sizeof(struct gtp0_header);
+}
+
+static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize);
+static void gtp_hashtable_free(struct gtp_dev *gtp);
+static int gtp_encap_enable(struct net_device *dev, struct gtp_dev *gtp,
+                           int fd_gtp0, int fd_gtp1, struct net *src_net);
+
+static int gtp_newlink(struct net *src_net, struct net_device *dev,
+                       struct nlattr *tb[], struct nlattr *data[])
+{
+       int hashsize, err, fd0, fd1;
+       struct gtp_dev *gtp;
+       struct gtp_net *gn;
+
+       if (!data[IFLA_GTP_FD0] || !data[IFLA_GTP_FD1])
+               return -EINVAL;
+
+       gtp = netdev_priv(dev);
+
+       fd0 = nla_get_u32(data[IFLA_GTP_FD0]);
+       fd1 = nla_get_u32(data[IFLA_GTP_FD1]);
+
+       err = gtp_encap_enable(dev, gtp, fd0, fd1, src_net);
+       if (err < 0)
+               goto out_err;
+
+       if (!data[IFLA_GTP_PDP_HASHSIZE])
+               hashsize = 1024;
+       else
+               hashsize = nla_get_u32(data[IFLA_GTP_PDP_HASHSIZE]);
+
+       err = gtp_hashtable_new(gtp, hashsize);
+       if (err < 0)
+               goto out_encap;
+
+       err = register_netdevice(dev);
+       if (err < 0) {
+               netdev_dbg(dev, "failed to register new netdev %d\n", err);
+               goto out_hashtable;
+       }
+
+       gn = net_generic(dev_net(dev), gtp_net_id);
+       list_add_rcu(&gtp->list, &gn->gtp_dev_list);
+
+       netdev_dbg(dev, "registered new GTP interface\n");
+
+       return 0;
+
+out_hashtable:
+       gtp_hashtable_free(gtp);
+out_encap:
+       gtp_encap_disable(gtp);
+out_err:
+       return err;
+}
+
+static void gtp_dellink(struct net_device *dev, struct list_head *head)
+{
+       struct gtp_dev *gtp = netdev_priv(dev);
+
+       gtp_encap_disable(gtp);
+       gtp_hashtable_free(gtp);
+       list_del_rcu(&gtp->list);
+       unregister_netdevice_queue(dev, head);
+}
+
+static const struct nla_policy gtp_policy[IFLA_GTP_MAX + 1] = {
+       [IFLA_GTP_FD0]                  = { .type = NLA_U32 },
+       [IFLA_GTP_FD1]                  = { .type = NLA_U32 },
+       [IFLA_GTP_PDP_HASHSIZE]         = { .type = NLA_U32 },
+};
+
+static int gtp_validate(struct nlattr *tb[], struct nlattr *data[])
+{
+       if (!data)
+               return -EINVAL;
+
+       return 0;
+}
+
+static size_t gtp_get_size(const struct net_device *dev)
+{
+       return nla_total_size(sizeof(__u32));   /* IFLA_GTP_PDP_HASHSIZE */
+}
+
+static int gtp_fill_info(struct sk_buff *skb, const struct net_device *dev)
+{
+       struct gtp_dev *gtp = netdev_priv(dev);
+
+       if (nla_put_u32(skb, IFLA_GTP_PDP_HASHSIZE, gtp->hash_size))
+               goto nla_put_failure;
+
+       return 0;
+
+nla_put_failure:
+       return -EMSGSIZE;
+}
+
+static struct rtnl_link_ops gtp_link_ops __read_mostly = {
+       .kind           = "gtp",
+       .maxtype        = IFLA_GTP_MAX,
+       .policy         = gtp_policy,
+       .priv_size      = sizeof(struct gtp_dev),
+       .setup          = gtp_link_setup,
+       .validate       = gtp_validate,
+       .newlink        = gtp_newlink,
+       .dellink        = gtp_dellink,
+       .get_size       = gtp_get_size,
+       .fill_info      = gtp_fill_info,
+};
+
+static struct net *gtp_genl_get_net(struct net *src_net, struct nlattr *tb[])
+{
+       struct net *net;
+
+       /* Examine the link attributes and figure out which network namespace
+        * we are talking about.
+        */
+       if (tb[GTPA_NET_NS_FD])
+               net = get_net_ns_by_fd(nla_get_u32(tb[GTPA_NET_NS_FD]));
+       else
+               net = get_net(src_net);
+
+       return net;
+}
+
+static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize)
+{
+       int i;
+
+       gtp->addr_hash = kmalloc(sizeof(struct hlist_head) * hsize, GFP_KERNEL);
+       if (gtp->addr_hash == NULL)
+               return -ENOMEM;
+
+       gtp->tid_hash = kmalloc(sizeof(struct hlist_head) * hsize, GFP_KERNEL);
+       if (gtp->tid_hash == NULL)
+               goto err1;
+
+       gtp->hash_size = hsize;
+
+       for (i = 0; i < hsize; i++) {
+               INIT_HLIST_HEAD(&gtp->addr_hash[i]);
+               INIT_HLIST_HEAD(&gtp->tid_hash[i]);
+       }
+       return 0;
+err1:
+       kfree(gtp->addr_hash);
+       return -ENOMEM;
+}
+
+static void gtp_hashtable_free(struct gtp_dev *gtp)
+{
+       struct pdp_ctx *pctx;
+       int i;
+
+       for (i = 0; i < gtp->hash_size; i++) {
+               hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i], hlist_tid) {
+                       hlist_del_rcu(&pctx->hlist_tid);
+                       hlist_del_rcu(&pctx->hlist_addr);
+                       kfree_rcu(pctx, rcu_head);
+               }
+       }
+       synchronize_rcu();
+       kfree(gtp->addr_hash);
+       kfree(gtp->tid_hash);
+}
+
+static int gtp_encap_enable(struct net_device *dev, struct gtp_dev *gtp,
+                           int fd_gtp0, int fd_gtp1, struct net *src_net)
+{
+       struct udp_tunnel_sock_cfg tuncfg = {NULL};
+       struct socket *sock0, *sock1u;
+       int err;
+
+       netdev_dbg(dev, "enable gtp on %d, %d\n", fd_gtp0, fd_gtp1);
+
+       sock0 = sockfd_lookup(fd_gtp0, &err);
+       if (sock0 == NULL) {
+               netdev_dbg(dev, "socket fd=%d not found (gtp0)\n", fd_gtp0);
+               return -ENOENT;
+       }
+
+       if (sock0->sk->sk_protocol != IPPROTO_UDP) {
+               netdev_dbg(dev, "socket fd=%d not UDP\n", fd_gtp0);
+               err = -EINVAL;
+               goto err1;
+       }
+
+       sock1u = sockfd_lookup(fd_gtp1, &err);
+       if (sock1u == NULL) {
+               netdev_dbg(dev, "socket fd=%d not found (gtp1u)\n", fd_gtp1);
+               err = -ENOENT;
+               goto err1;
+       }
+
+       if (sock1u->sk->sk_protocol != IPPROTO_UDP) {
+               netdev_dbg(dev, "socket fd=%d not UDP\n", fd_gtp1);
+               err = -EINVAL;
+               goto err2;
+       }
+
+       netdev_dbg(dev, "enable gtp on %p, %p\n", sock0, sock1u);
+
+       gtp->sock0 = sock0;
+       gtp->sock1u = sock1u;
+       gtp->net = src_net;
+
+       tuncfg.sk_user_data = gtp;
+       tuncfg.encap_rcv = gtp_encap_recv;
+       tuncfg.encap_destroy = gtp_encap_destroy;
+
+       tuncfg.encap_type = UDP_ENCAP_GTP0;
+       setup_udp_tunnel_sock(sock_net(gtp->sock0->sk), gtp->sock0, &tuncfg);
+
+       tuncfg.encap_type = UDP_ENCAP_GTP1U;
+       setup_udp_tunnel_sock(sock_net(gtp->sock1u->sk), gtp->sock1u, &tuncfg);
+
+       err = 0;
+err2:
+       sockfd_put(sock1u);
+err1:
+       sockfd_put(sock0);
+       return err;
+}
+
+static struct net_device *gtp_find_dev(struct net *net, int ifindex)
+{
+       struct gtp_net *gn = net_generic(net, gtp_net_id);
+       struct gtp_dev *gtp;
+
+       list_for_each_entry_rcu(gtp, &gn->gtp_dev_list, list) {
+               if (ifindex == gtp->dev->ifindex)
+                       return gtp->dev;
+       }
+       return NULL;
+}
+
+static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
+{
+       pctx->gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]);
+       pctx->af = AF_INET;
+       pctx->sgsn_addr_ip4.s_addr =
+               nla_get_be32(info->attrs[GTPA_SGSN_ADDRESS]);
+       pctx->ms_addr_ip4.s_addr =
+               nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
+
+       switch (pctx->gtp_version) {
+       case GTP_V0:
+               /* According to TS 09.60, sections 7.5.1 and 7.5.2, the flow
+                * label needs to be the same for uplink and downlink packets,
+                * so let's annotate this.
+                */
+               pctx->u.v0.tid = nla_get_u64(info->attrs[GTPA_TID]);
+               pctx->u.v0.flow = nla_get_u16(info->attrs[GTPA_FLOW]);
+               break;
+       case GTP_V1:
+               pctx->u.v1.i_tei = nla_get_u32(info->attrs[GTPA_I_TEI]);
+               pctx->u.v1.o_tei = nla_get_u32(info->attrs[GTPA_O_TEI]);
+               break;
+       default:
+               break;
+       }
+}
+
+static int ipv4_pdp_add(struct net_device *dev, struct genl_info *info)
+{
+       struct gtp_dev *gtp = netdev_priv(dev);
+       u32 hash_ms, hash_tid = 0;
+       struct pdp_ctx *pctx;
+       bool found = false;
+       __be32 ms_addr;
+
+       ms_addr = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
+       hash_ms = ipv4_hashfn(ms_addr) % gtp->hash_size;
+
+       hlist_for_each_entry_rcu(pctx, &gtp->addr_hash[hash_ms], hlist_addr) {
+               if (pctx->ms_addr_ip4.s_addr == ms_addr) {
+                       found = true;
+                       break;
+               }
+       }
+
+       if (found) {
+               if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
+                       return -EEXIST;
+               if (info->nlhdr->nlmsg_flags & NLM_F_REPLACE)
+                       return -EOPNOTSUPP;
+
+               ipv4_pdp_fill(pctx, info);
+
+               if (pctx->gtp_version == GTP_V0)
+                       netdev_dbg(dev, "GTPv0-U: update tunnel id = %llx (pdp %p)\n",
+                                  pctx->u.v0.tid, pctx);
+               else if (pctx->gtp_version == GTP_V1)
+                       netdev_dbg(dev, "GTPv1-U: update tunnel id = %x/%x (pdp %p)\n",
+                                  pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
+
+               return 0;
+
+       }
+
+       pctx = kmalloc(sizeof(struct pdp_ctx), GFP_KERNEL);
+       if (pctx == NULL)
+               return -ENOMEM;
+
+       ipv4_pdp_fill(pctx, info);
+       atomic_set(&pctx->tx_seq, 0);
+
+       switch (pctx->gtp_version) {
+       case GTP_V0:
+               /* TS 09.60: "The flow label identifies unambiguously a GTP
+                * flow.". We use the tid for this instead, I cannot find a
+                * situation in which this doesn't unambiguosly identify the
+                * PDP context.
+                */
+               hash_tid = gtp0_hashfn(pctx->u.v0.tid) % gtp->hash_size;
+               break;
+       case GTP_V1:
+               hash_tid = gtp1u_hashfn(pctx->u.v1.i_tei) % gtp->hash_size;
+               break;
+       }
+
+       hlist_add_head_rcu(&pctx->hlist_addr, &gtp->addr_hash[hash_ms]);
+       hlist_add_head_rcu(&pctx->hlist_tid, &gtp->tid_hash[hash_tid]);
+
+       switch (pctx->gtp_version) {
+       case GTP_V0:
+               netdev_dbg(dev, "GTPv0-U: new PDP ctx id=%llx ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
+                          pctx->u.v0.tid, &pctx->sgsn_addr_ip4,
+                          &pctx->ms_addr_ip4, pctx);
+               break;
+       case GTP_V1:
+               netdev_dbg(dev, "GTPv1-U: new PDP ctx id=%x/%x ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
+                          pctx->u.v1.i_tei, pctx->u.v1.o_tei,
+                          &pctx->sgsn_addr_ip4, &pctx->ms_addr_ip4, pctx);
+               break;
+       }
+
+       return 0;
+}
+
+static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
+{
+       struct net_device *dev;
+       struct net *net;
+
+       if (!info->attrs[GTPA_VERSION] ||
+           !info->attrs[GTPA_LINK] ||
+           !info->attrs[GTPA_SGSN_ADDRESS] ||
+           !info->attrs[GTPA_MS_ADDRESS])
+               return -EINVAL;
+
+       switch (nla_get_u32(info->attrs[GTPA_VERSION])) {
+       case GTP_V0:
+               if (!info->attrs[GTPA_TID] ||
+                   !info->attrs[GTPA_FLOW])
+                       return -EINVAL;
+               break;
+       case GTP_V1:
+               if (!info->attrs[GTPA_I_TEI] ||
+                   !info->attrs[GTPA_O_TEI])
+                       return -EINVAL;
+               break;
+
+       default:
+               return -EINVAL;
+       }
+
+       net = gtp_genl_get_net(sock_net(skb->sk), info->attrs);
+       if (IS_ERR(net))
+               return PTR_ERR(net);
+
+       /* Check if there's an existing gtpX device to configure */
+       dev = gtp_find_dev(net, nla_get_u32(info->attrs[GTPA_LINK]));
+       if (dev == NULL)
+               return -ENODEV;
+
+       return ipv4_pdp_add(dev, info);
+}
+
+static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
+{
+       struct net_device *dev;
+       struct pdp_ctx *pctx;
+       struct gtp_dev *gtp;
+       struct net *net;
+
+       if (!info->attrs[GTPA_VERSION] ||
+           !info->attrs[GTPA_LINK])
+               return -EINVAL;
+
+       net = gtp_genl_get_net(sock_net(skb->sk), info->attrs);
+       if (IS_ERR(net))
+               return PTR_ERR(net);
+
+       /* Check if there's an existing gtpX device to configure */
+       dev = gtp_find_dev(net, nla_get_u32(info->attrs[GTPA_LINK]));
+       if (dev == NULL)
+               return -ENODEV;
+
+       gtp = netdev_priv(dev);
+
+       switch (nla_get_u32(info->attrs[GTPA_VERSION])) {
+       case GTP_V0:
+               if (!info->attrs[GTPA_TID])
+                       return -EINVAL;
+               pctx = gtp0_pdp_find(gtp, nla_get_u64(info->attrs[GTPA_TID]));
+               break;
+       case GTP_V1:
+               if (!info->attrs[GTPA_I_TEI])
+                       return -EINVAL;
+               pctx = gtp1_pdp_find(gtp, nla_get_u64(info->attrs[GTPA_I_TEI]));
+               break;
+
+       default:
+               return -EINVAL;
+       }
+
+       if (pctx == NULL)
+               return -ENOENT;
+
+       if (pctx->gtp_version == GTP_V0)
+               netdev_dbg(dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n",
+                          pctx->u.v0.tid, pctx);
+       else if (pctx->gtp_version == GTP_V1)
+               netdev_dbg(dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n",
+                          pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
+
+       hlist_del_rcu(&pctx->hlist_tid);
+       hlist_del_rcu(&pctx->hlist_addr);
+       kfree_rcu(pctx, rcu_head);
+
+       return 0;
+}
+
+static struct genl_family gtp_genl_family = {
+       .id             = GENL_ID_GENERATE,
+       .name           = "gtp",
+       .version        = 0,
+       .hdrsize        = 0,
+       .maxattr        = GTPA_MAX,
+       .netnsok        = true,
+};
+
+static int gtp_genl_fill_info(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
+                             u32 type, struct pdp_ctx *pctx)
+{
+       void *genlh;
+
+       genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, 0,
+                           type);
+       if (genlh == NULL)
+               goto nlmsg_failure;
+
+       if (nla_put_u32(skb, GTPA_VERSION, pctx->gtp_version) ||
+           nla_put_be32(skb, GTPA_SGSN_ADDRESS, pctx->sgsn_addr_ip4.s_addr) ||
+           nla_put_be32(skb, GTPA_MS_ADDRESS, pctx->ms_addr_ip4.s_addr))
+               goto nla_put_failure;
+
+       switch (pctx->gtp_version) {
+       case GTP_V0:
+               if (nla_put_u64_64bit(skb, GTPA_TID, pctx->u.v0.tid, GTPA_PAD) ||
+                   nla_put_u16(skb, GTPA_FLOW, pctx->u.v0.flow))
+                       goto nla_put_failure;
+               break;
+       case GTP_V1:
+               if (nla_put_u32(skb, GTPA_I_TEI, pctx->u.v1.i_tei) ||
+                   nla_put_u32(skb, GTPA_O_TEI, pctx->u.v1.o_tei))
+                       goto nla_put_failure;
+               break;
+       }
+       genlmsg_end(skb, genlh);
+       return 0;
+
+nlmsg_failure:
+nla_put_failure:
+       genlmsg_cancel(skb, genlh);
+       return -EMSGSIZE;
+}
+
+static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info)
+{
+       struct pdp_ctx *pctx = NULL;
+       struct net_device *dev;
+       struct sk_buff *skb2;
+       struct gtp_dev *gtp;
+       u32 gtp_version;
+       struct net *net;
+       int err;
+
+       if (!info->attrs[GTPA_VERSION] ||
+           !info->attrs[GTPA_LINK])
+               return -EINVAL;
+
+       gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]);
+       switch (gtp_version) {
+       case GTP_V0:
+       case GTP_V1:
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       net = gtp_genl_get_net(sock_net(skb->sk), info->attrs);
+       if (IS_ERR(net))
+               return PTR_ERR(net);
+
+       /* Check if there's an existing gtpX device to configure */
+       dev = gtp_find_dev(net, nla_get_u32(info->attrs[GTPA_LINK]));
+       if (dev == NULL)
+               return -ENODEV;
+
+       gtp = netdev_priv(dev);
+
+       rcu_read_lock();
+       if (gtp_version == GTP_V0 &&
+           info->attrs[GTPA_TID]) {
+               u64 tid = nla_get_u64(info->attrs[GTPA_TID]);
+
+               pctx = gtp0_pdp_find(gtp, tid);
+       } else if (gtp_version == GTP_V1 &&
+                info->attrs[GTPA_I_TEI]) {
+               u32 tid = nla_get_u32(info->attrs[GTPA_I_TEI]);
+
+               pctx = gtp1_pdp_find(gtp, tid);
+       } else if (info->attrs[GTPA_MS_ADDRESS]) {
+               __be32 ip = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
+
+               pctx = ipv4_pdp_find(gtp, ip);
+       }
+
+       if (pctx == NULL) {
+               err = -ENOENT;
+               goto err_unlock;
+       }
+
+       skb2 = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
+       if (skb2 == NULL) {
+               err = -ENOMEM;
+               goto err_unlock;
+       }
+
+       err = gtp_genl_fill_info(skb2, NETLINK_CB(skb).portid,
+                                info->snd_seq, info->nlhdr->nlmsg_type, pctx);
+       if (err < 0)
+               goto err_unlock_free;
+
+       rcu_read_unlock();
+       return genlmsg_unicast(genl_info_net(info), skb2, info->snd_portid);
+
+err_unlock_free:
+       kfree_skb(skb2);
+err_unlock:
+       rcu_read_unlock();
+       return err;
+}
+
+static int gtp_genl_dump_pdp(struct sk_buff *skb,
+                               struct netlink_callback *cb)
+{
+       struct gtp_dev *last_gtp = (struct gtp_dev *)cb->args[2], *gtp;
+       struct net *net = sock_net(skb->sk);
+       struct gtp_net *gn = net_generic(net, gtp_net_id);
+       unsigned long tid = cb->args[1];
+       int i, k = cb->args[0], ret;
+       struct pdp_ctx *pctx;
+
+       if (cb->args[4])
+               return 0;
+
+       list_for_each_entry_rcu(gtp, &gn->gtp_dev_list, list) {
+               if (last_gtp && last_gtp != gtp)
+                       continue;
+               else
+                       last_gtp = NULL;
+
+               for (i = k; i < gtp->hash_size; i++) {
+                       hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i], hlist_tid) {
+                               if (tid && tid != pctx->u.tid)
+                                       continue;
+                               else
+                                       tid = 0;
+
+                               ret = gtp_genl_fill_info(skb,
+                                                        NETLINK_CB(cb->skb).portid,
+                                                        cb->nlh->nlmsg_seq,
+                                                        cb->nlh->nlmsg_type, pctx);
+                               if (ret < 0) {
+                                       cb->args[0] = i;
+                                       cb->args[1] = pctx->u.tid;
+                                       cb->args[2] = (unsigned long)gtp;
+                                       goto out;
+                               }
+                       }
+               }
+       }
+       cb->args[4] = 1;
+out:
+       return skb->len;
+}
+
+static struct nla_policy gtp_genl_policy[GTPA_MAX + 1] = {
+       [GTPA_LINK]             = { .type = NLA_U32, },
+       [GTPA_VERSION]          = { .type = NLA_U32, },
+       [GTPA_TID]              = { .type = NLA_U64, },
+       [GTPA_SGSN_ADDRESS]     = { .type = NLA_U32, },
+       [GTPA_MS_ADDRESS]       = { .type = NLA_U32, },
+       [GTPA_FLOW]             = { .type = NLA_U16, },
+       [GTPA_NET_NS_FD]        = { .type = NLA_U32, },
+       [GTPA_I_TEI]            = { .type = NLA_U32, },
+       [GTPA_O_TEI]            = { .type = NLA_U32, },
+};
+
+static const struct genl_ops gtp_genl_ops[] = {
+       {
+               .cmd = GTP_CMD_NEWPDP,
+               .doit = gtp_genl_new_pdp,
+               .policy = gtp_genl_policy,
+               .flags = GENL_ADMIN_PERM,
+       },
+       {
+               .cmd = GTP_CMD_DELPDP,
+               .doit = gtp_genl_del_pdp,
+               .policy = gtp_genl_policy,
+               .flags = GENL_ADMIN_PERM,
+       },
+       {
+               .cmd = GTP_CMD_GETPDP,
+               .doit = gtp_genl_get_pdp,
+               .dumpit = gtp_genl_dump_pdp,
+               .policy = gtp_genl_policy,
+               .flags = GENL_ADMIN_PERM,
+       },
+};
+
+static int __net_init gtp_net_init(struct net *net)
+{
+       struct gtp_net *gn = net_generic(net, gtp_net_id);
+
+       INIT_LIST_HEAD(&gn->gtp_dev_list);
+       return 0;
+}
+
+static void __net_exit gtp_net_exit(struct net *net)
+{
+       struct gtp_net *gn = net_generic(net, gtp_net_id);
+       struct gtp_dev *gtp;
+       LIST_HEAD(list);
+
+       rtnl_lock();
+       list_for_each_entry(gtp, &gn->gtp_dev_list, list)
+               gtp_dellink(gtp->dev, &list);
+
+       unregister_netdevice_many(&list);
+       rtnl_unlock();
+}
+
+static struct pernet_operations gtp_net_ops = {
+       .init   = gtp_net_init,
+       .exit   = gtp_net_exit,
+       .id     = &gtp_net_id,
+       .size   = sizeof(struct gtp_net),
+};
+
+static int __init gtp_init(void)
+{
+       int err;
+
+       get_random_bytes(&gtp_h_initval, sizeof(gtp_h_initval));
+
+       err = rtnl_link_register(&gtp_link_ops);
+       if (err < 0)
+               goto error_out;
+
+       err = genl_register_family_with_ops(&gtp_genl_family, gtp_genl_ops);
+       if (err < 0)
+               goto unreg_rtnl_link;
+
+       err = register_pernet_subsys(&gtp_net_ops);
+       if (err < 0)
+               goto unreg_genl_family;
+
+       pr_info("GTP module loaded (pdp ctx size %Zd bytes)\n",
+               sizeof(struct pdp_ctx));
+       return 0;
+
+unreg_genl_family:
+       genl_unregister_family(&gtp_genl_family);
+unreg_rtnl_link:
+       rtnl_link_unregister(&gtp_link_ops);
+error_out:
+       pr_err("error loading GTP module loaded\n");
+       return err;
+}
+late_initcall(gtp_init);
+
+static void __exit gtp_fini(void)
+{
+       unregister_pernet_subsys(&gtp_net_ops);
+       genl_unregister_family(&gtp_genl_family);
+       rtnl_link_unregister(&gtp_link_ops);
+
+       pr_info("GTP module unloaded\n");
+}
+module_exit(gtp_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Harald Welte <hwelte@sysmocom.de>");
+MODULE_DESCRIPTION("Interface driver for GTP encapsulated traffic");
+MODULE_ALIAS_RTNL_LINK("gtp");
diff --git a/include/net/gtp.h b/include/net/gtp.h
new file mode 100644 (file)
index 0000000..894a37b
--- /dev/null
@@ -0,0 +1,34 @@
+#ifndef _GTP_H_
+#define _GTP_H
+
+/* General GTP protocol related definitions. */
+
+#define GTP0_PORT      3386
+#define GTP1U_PORT     2152
+
+#define GTP_TPDU       255
+
+struct gtp0_header {   /* According to GSM TS 09.60. */
+       __u8    flags;
+       __u8    type;
+       __be16  length;
+       __be16  seq;
+       __be16  flow;
+       __u8    number;
+       __u8    spare[3];
+       __be64  tid;
+} __attribute__ ((packed));
+
+struct gtp1_header {   /* According to 3GPP TS 29.060. */
+       __u8    flags;
+       __u8    type;
+       __be16  length;
+       __be32  tid;
+} __attribute__ ((packed));
+
+#define GTP1_F_NPDU    0x01
+#define GTP1_F_SEQ     0x02
+#define GTP1_F_EXTHDR  0x04
+#define GTP1_F_MASK    0x07
+
+#endif
index 813ffb2e22c9a5a9d9049aadb2aa467d734b734c..8bdae34d1f9add25d968182c6540c431d994c7ce 100644 (file)
@@ -141,6 +141,7 @@ header-y += gfs2_ondisk.h
 header-y += gigaset_dev.h
 header-y += gpio.h
 header-y += gsmmux.h
+header-y += gtp.h
 header-y += hdlcdrv.h
 header-y += hdlc.h
 header-y += hdreg.h
diff --git a/include/uapi/linux/gtp.h b/include/uapi/linux/gtp.h
new file mode 100644 (file)
index 0000000..ca1054d
--- /dev/null
@@ -0,0 +1,33 @@
+#ifndef _UAPI_LINUX_GTP_H_
+#define _UAPI_LINUX_GTP_H__
+
+enum gtp_genl_cmds {
+       GTP_CMD_NEWPDP,
+       GTP_CMD_DELPDP,
+       GTP_CMD_GETPDP,
+
+       GTP_CMD_MAX,
+};
+
+enum gtp_version {
+       GTP_V0 = 0,
+       GTP_V1,
+};
+
+enum gtp_attrs {
+       GTPA_UNSPEC = 0,
+       GTPA_LINK,
+       GTPA_VERSION,
+       GTPA_TID,       /* for GTPv0 only */
+       GTPA_SGSN_ADDRESS,
+       GTPA_MS_ADDRESS,
+       GTPA_FLOW,
+       GTPA_NET_NS_FD,
+       GTPA_I_TEI,     /* for GTPv1 only */
+       GTPA_O_TEI,     /* for GTPv1 only */
+       GTPA_PAD,
+       __GTPA_MAX,
+};
+#define GTPA_MAX (__GTPA_MAX + 1)
+
+#endif /* _UAPI_LINUX_GTP_H_ */
index d2d7fd4ba5f565cdd0cd7183805e2263ac76e631..bb36bd5675a7ceb641a597aa4d82d428cdaaf529 100644 (file)
@@ -529,6 +529,16 @@ enum {
 };
 #define IFLA_PPP_MAX (__IFLA_PPP_MAX - 1)
 
+/* GTP section */
+enum {
+       IFLA_GTP_UNSPEC,
+       IFLA_GTP_FD0,
+       IFLA_GTP_FD1,
+       IFLA_GTP_PDP_HASHSIZE,
+       __IFLA_GTP_MAX,
+};
+#define IFLA_GTP_MAX (__IFLA_GTP_MAX - 1)
+
 /* Bonding section */
 
 enum {
index 16574ea18f0cf62d743287779ffaed6205d120d0..2c8180f9156faf8c6429bb738e22101ceec11660 100644 (file)
@@ -36,6 +36,7 @@ struct udphdr {
 #define UDP_ENCAP_ESPINUDP_NON_IKE     1 /* draft-ietf-ipsec-nat-t-ike-00/01 */
 #define UDP_ENCAP_ESPINUDP     2 /* draft-ietf-ipsec-udp-encaps-06 */
 #define UDP_ENCAP_L2TPINUDP    3 /* rfc2661 */
-
+#define UDP_ENCAP_GTP0         4 /* GSM TS 09.60 */
+#define UDP_ENCAP_GTP1U                5 /* 3GPP TS 29.060 */
 
 #endif /* _UAPI_LINUX_UDP_H */