Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / drivers / net / gtp.c
blobf38e32a7ec9c979ac4524c31e09da375a6e0606c
1 /* GTP according to GSM TS 09.60 / 3GPP TS 29.060
3 * (C) 2012-2014 by sysmocom - s.f.m.c. GmbH
4 * (C) 2016 by Pablo Neira Ayuso <pablo@netfilter.org>
6 * Author: Harald Welte <hwelte@sysmocom.de>
7 * Pablo Neira Ayuso <pablo@netfilter.org>
8 * Andreas Schultz <aschultz@travelping.com>
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/module.h>
19 #include <linux/skbuff.h>
20 #include <linux/udp.h>
21 #include <linux/rculist.h>
22 #include <linux/jhash.h>
23 #include <linux/if_tunnel.h>
24 #include <linux/net.h>
25 #include <linux/file.h>
26 #include <linux/gtp.h>
28 #include <net/net_namespace.h>
29 #include <net/protocol.h>
30 #include <net/ip.h>
31 #include <net/udp.h>
32 #include <net/udp_tunnel.h>
33 #include <net/icmp.h>
34 #include <net/xfrm.h>
35 #include <net/genetlink.h>
36 #include <net/netns/generic.h>
37 #include <net/gtp.h>
39 /* An active session for the subscriber. */
40 struct pdp_ctx {
41 struct hlist_node hlist_tid;
42 struct hlist_node hlist_addr;
44 union {
45 u64 tid;
46 struct {
47 u64 tid;
48 u16 flow;
49 } v0;
50 struct {
51 u32 i_tei;
52 u32 o_tei;
53 } v1;
54 } u;
55 u8 gtp_version;
56 u16 af;
58 struct in_addr ms_addr_ip4;
59 struct in_addr peer_addr_ip4;
61 struct sock *sk;
62 struct net_device *dev;
64 atomic_t tx_seq;
65 struct rcu_head rcu_head;
68 /* One instance of the GTP device. */
69 struct gtp_dev {
70 struct list_head list;
72 struct sock *sk0;
73 struct sock *sk1u;
75 struct net_device *dev;
77 unsigned int role;
78 unsigned int hash_size;
79 struct hlist_head *tid_hash;
80 struct hlist_head *addr_hash;
83 static unsigned int gtp_net_id __read_mostly;
85 struct gtp_net {
86 struct list_head gtp_dev_list;
89 static u32 gtp_h_initval;
91 static void pdp_context_delete(struct pdp_ctx *pctx);
93 static inline u32 gtp0_hashfn(u64 tid)
95 u32 *tid32 = (u32 *) &tid;
96 return jhash_2words(tid32[0], tid32[1], gtp_h_initval);
99 static inline u32 gtp1u_hashfn(u32 tid)
101 return jhash_1word(tid, gtp_h_initval);
104 static inline u32 ipv4_hashfn(__be32 ip)
106 return jhash_1word((__force u32)ip, gtp_h_initval);
109 /* Resolve a PDP context structure based on the 64bit TID. */
110 static struct pdp_ctx *gtp0_pdp_find(struct gtp_dev *gtp, u64 tid)
112 struct hlist_head *head;
113 struct pdp_ctx *pdp;
115 head = &gtp->tid_hash[gtp0_hashfn(tid) % gtp->hash_size];
117 hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
118 if (pdp->gtp_version == GTP_V0 &&
119 pdp->u.v0.tid == tid)
120 return pdp;
122 return NULL;
125 /* Resolve a PDP context structure based on the 32bit TEI. */
126 static struct pdp_ctx *gtp1_pdp_find(struct gtp_dev *gtp, u32 tid)
128 struct hlist_head *head;
129 struct pdp_ctx *pdp;
131 head = &gtp->tid_hash[gtp1u_hashfn(tid) % gtp->hash_size];
133 hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
134 if (pdp->gtp_version == GTP_V1 &&
135 pdp->u.v1.i_tei == tid)
136 return pdp;
138 return NULL;
141 /* Resolve a PDP context based on IPv4 address of MS. */
142 static struct pdp_ctx *ipv4_pdp_find(struct gtp_dev *gtp, __be32 ms_addr)
144 struct hlist_head *head;
145 struct pdp_ctx *pdp;
147 head = &gtp->addr_hash[ipv4_hashfn(ms_addr) % gtp->hash_size];
149 hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
150 if (pdp->af == AF_INET &&
151 pdp->ms_addr_ip4.s_addr == ms_addr)
152 return pdp;
155 return NULL;
158 static bool gtp_check_ms_ipv4(struct sk_buff *skb, struct pdp_ctx *pctx,
159 unsigned int hdrlen, unsigned int role)
161 struct iphdr *iph;
163 if (!pskb_may_pull(skb, hdrlen + sizeof(struct iphdr)))
164 return false;
166 iph = (struct iphdr *)(skb->data + hdrlen);
168 if (role == GTP_ROLE_SGSN)
169 return iph->daddr == pctx->ms_addr_ip4.s_addr;
170 else
171 return iph->saddr == pctx->ms_addr_ip4.s_addr;
174 /* Check if the inner IP address in this packet is assigned to any
175 * existing mobile subscriber.
177 static bool gtp_check_ms(struct sk_buff *skb, struct pdp_ctx *pctx,
178 unsigned int hdrlen, unsigned int role)
180 switch (ntohs(skb->protocol)) {
181 case ETH_P_IP:
182 return gtp_check_ms_ipv4(skb, pctx, hdrlen, role);
184 return false;
187 static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb,
188 unsigned int hdrlen, unsigned int role)
190 struct pcpu_sw_netstats *stats;
192 if (!gtp_check_ms(skb, pctx, hdrlen, role)) {
193 netdev_dbg(pctx->dev, "No PDP ctx for this MS\n");
194 return 1;
197 /* Get rid of the GTP + UDP headers. */
198 if (iptunnel_pull_header(skb, hdrlen, skb->protocol,
199 !net_eq(sock_net(pctx->sk), dev_net(pctx->dev))))
200 return -1;
202 netdev_dbg(pctx->dev, "forwarding packet from GGSN to uplink\n");
204 /* Now that the UDP and the GTP header have been removed, set up the
205 * new network header. This is required by the upper layer to
206 * calculate the transport header.
208 skb_reset_network_header(skb);
210 skb->dev = pctx->dev;
212 stats = this_cpu_ptr(pctx->dev->tstats);
213 u64_stats_update_begin(&stats->syncp);
214 stats->rx_packets++;
215 stats->rx_bytes += skb->len;
216 u64_stats_update_end(&stats->syncp);
218 netif_rx(skb);
219 return 0;
222 /* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */
223 static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
225 unsigned int hdrlen = sizeof(struct udphdr) +
226 sizeof(struct gtp0_header);
227 struct gtp0_header *gtp0;
228 struct pdp_ctx *pctx;
230 if (!pskb_may_pull(skb, hdrlen))
231 return -1;
233 gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
235 if ((gtp0->flags >> 5) != GTP_V0)
236 return 1;
238 if (gtp0->type != GTP_TPDU)
239 return 1;
241 pctx = gtp0_pdp_find(gtp, be64_to_cpu(gtp0->tid));
242 if (!pctx) {
243 netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
244 return 1;
247 return gtp_rx(pctx, skb, hdrlen, gtp->role);
250 static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
252 unsigned int hdrlen = sizeof(struct udphdr) +
253 sizeof(struct gtp1_header);
254 struct gtp1_header *gtp1;
255 struct pdp_ctx *pctx;
257 if (!pskb_may_pull(skb, hdrlen))
258 return -1;
260 gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
262 if ((gtp1->flags >> 5) != GTP_V1)
263 return 1;
265 if (gtp1->type != GTP_TPDU)
266 return 1;
268 /* From 29.060: "This field shall be present if and only if any one or
269 * more of the S, PN and E flags are set.".
271 * If any of the bit is set, then the remaining ones also have to be
272 * set.
274 if (gtp1->flags & GTP1_F_MASK)
275 hdrlen += 4;
277 /* Make sure the header is larger enough, including extensions. */
278 if (!pskb_may_pull(skb, hdrlen))
279 return -1;
281 gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
283 pctx = gtp1_pdp_find(gtp, ntohl(gtp1->tid));
284 if (!pctx) {
285 netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
286 return 1;
289 return gtp_rx(pctx, skb, hdrlen, gtp->role);
292 static void gtp_encap_destroy(struct sock *sk)
294 struct gtp_dev *gtp;
296 gtp = rcu_dereference_sk_user_data(sk);
297 if (gtp) {
298 udp_sk(sk)->encap_type = 0;
299 rcu_assign_sk_user_data(sk, NULL);
300 sock_put(sk);
304 static void gtp_encap_disable_sock(struct sock *sk)
306 if (!sk)
307 return;
309 gtp_encap_destroy(sk);
312 static void gtp_encap_disable(struct gtp_dev *gtp)
314 gtp_encap_disable_sock(gtp->sk0);
315 gtp_encap_disable_sock(gtp->sk1u);
318 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
319 * Return codes: 0: success, <0: error, >0: pass up to userspace UDP socket.
321 static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb)
323 struct gtp_dev *gtp;
324 int ret = 0;
326 gtp = rcu_dereference_sk_user_data(sk);
327 if (!gtp)
328 return 1;
330 netdev_dbg(gtp->dev, "encap_recv sk=%p\n", sk);
332 switch (udp_sk(sk)->encap_type) {
333 case UDP_ENCAP_GTP0:
334 netdev_dbg(gtp->dev, "received GTP0 packet\n");
335 ret = gtp0_udp_encap_recv(gtp, skb);
336 break;
337 case UDP_ENCAP_GTP1U:
338 netdev_dbg(gtp->dev, "received GTP1U packet\n");
339 ret = gtp1u_udp_encap_recv(gtp, skb);
340 break;
341 default:
342 ret = -1; /* Shouldn't happen. */
345 switch (ret) {
346 case 1:
347 netdev_dbg(gtp->dev, "pass up to the process\n");
348 break;
349 case 0:
350 break;
351 case -1:
352 netdev_dbg(gtp->dev, "GTP packet has been dropped\n");
353 kfree_skb(skb);
354 ret = 0;
355 break;
358 return ret;
361 static int gtp_dev_init(struct net_device *dev)
363 struct gtp_dev *gtp = netdev_priv(dev);
365 gtp->dev = dev;
367 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
368 if (!dev->tstats)
369 return -ENOMEM;
371 return 0;
374 static void gtp_dev_uninit(struct net_device *dev)
376 struct gtp_dev *gtp = netdev_priv(dev);
378 gtp_encap_disable(gtp);
379 free_percpu(dev->tstats);
382 static struct rtable *ip4_route_output_gtp(struct flowi4 *fl4,
383 const struct sock *sk,
384 __be32 daddr)
386 memset(fl4, 0, sizeof(*fl4));
387 fl4->flowi4_oif = sk->sk_bound_dev_if;
388 fl4->daddr = daddr;
389 fl4->saddr = inet_sk(sk)->inet_saddr;
390 fl4->flowi4_tos = RT_CONN_FLAGS(sk);
391 fl4->flowi4_proto = sk->sk_protocol;
393 return ip_route_output_key(sock_net(sk), fl4);
396 static inline void gtp0_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
398 int payload_len = skb->len;
399 struct gtp0_header *gtp0;
401 gtp0 = skb_push(skb, sizeof(*gtp0));
403 gtp0->flags = 0x1e; /* v0, GTP-non-prime. */
404 gtp0->type = GTP_TPDU;
405 gtp0->length = htons(payload_len);
406 gtp0->seq = htons((atomic_inc_return(&pctx->tx_seq) - 1) % 0xffff);
407 gtp0->flow = htons(pctx->u.v0.flow);
408 gtp0->number = 0xff;
409 gtp0->spare[0] = gtp0->spare[1] = gtp0->spare[2] = 0xff;
410 gtp0->tid = cpu_to_be64(pctx->u.v0.tid);
413 static inline void gtp1_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
415 int payload_len = skb->len;
416 struct gtp1_header *gtp1;
418 gtp1 = skb_push(skb, sizeof(*gtp1));
420 /* Bits 8 7 6 5 4 3 2 1
421 * +--+--+--+--+--+--+--+--+
422 * |version |PT| 0| E| S|PN|
423 * +--+--+--+--+--+--+--+--+
424 * 0 0 1 1 1 0 0 0
426 gtp1->flags = 0x30; /* v1, GTP-non-prime. */
427 gtp1->type = GTP_TPDU;
428 gtp1->length = htons(payload_len);
429 gtp1->tid = htonl(pctx->u.v1.o_tei);
431 /* TODO: Suppport for extension header, sequence number and N-PDU.
432 * Update the length field if any of them is available.
436 struct gtp_pktinfo {
437 struct sock *sk;
438 struct iphdr *iph;
439 struct flowi4 fl4;
440 struct rtable *rt;
441 struct pdp_ctx *pctx;
442 struct net_device *dev;
443 __be16 gtph_port;
446 static void gtp_push_header(struct sk_buff *skb, struct gtp_pktinfo *pktinfo)
448 switch (pktinfo->pctx->gtp_version) {
449 case GTP_V0:
450 pktinfo->gtph_port = htons(GTP0_PORT);
451 gtp0_push_header(skb, pktinfo->pctx);
452 break;
453 case GTP_V1:
454 pktinfo->gtph_port = htons(GTP1U_PORT);
455 gtp1_push_header(skb, pktinfo->pctx);
456 break;
460 static inline void gtp_set_pktinfo_ipv4(struct gtp_pktinfo *pktinfo,
461 struct sock *sk, struct iphdr *iph,
462 struct pdp_ctx *pctx, struct rtable *rt,
463 struct flowi4 *fl4,
464 struct net_device *dev)
466 pktinfo->sk = sk;
467 pktinfo->iph = iph;
468 pktinfo->pctx = pctx;
469 pktinfo->rt = rt;
470 pktinfo->fl4 = *fl4;
471 pktinfo->dev = dev;
474 static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev,
475 struct gtp_pktinfo *pktinfo)
477 struct gtp_dev *gtp = netdev_priv(dev);
478 struct pdp_ctx *pctx;
479 struct rtable *rt;
480 struct flowi4 fl4;
481 struct iphdr *iph;
482 __be16 df;
483 int mtu;
485 /* Read the IP destination address and resolve the PDP context.
486 * Prepend PDP header with TEI/TID from PDP ctx.
488 iph = ip_hdr(skb);
489 if (gtp->role == GTP_ROLE_SGSN)
490 pctx = ipv4_pdp_find(gtp, iph->saddr);
491 else
492 pctx = ipv4_pdp_find(gtp, iph->daddr);
494 if (!pctx) {
495 netdev_dbg(dev, "no PDP ctx found for %pI4, skip\n",
496 &iph->daddr);
497 return -ENOENT;
499 netdev_dbg(dev, "found PDP context %p\n", pctx);
501 rt = ip4_route_output_gtp(&fl4, pctx->sk, pctx->peer_addr_ip4.s_addr);
502 if (IS_ERR(rt)) {
503 netdev_dbg(dev, "no route to SSGN %pI4\n",
504 &pctx->peer_addr_ip4.s_addr);
505 dev->stats.tx_carrier_errors++;
506 goto err;
509 if (rt->dst.dev == dev) {
510 netdev_dbg(dev, "circular route to SSGN %pI4\n",
511 &pctx->peer_addr_ip4.s_addr);
512 dev->stats.collisions++;
513 goto err_rt;
516 skb_dst_drop(skb);
518 /* This is similar to tnl_update_pmtu(). */
519 df = iph->frag_off;
520 if (df) {
521 mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
522 sizeof(struct iphdr) - sizeof(struct udphdr);
523 switch (pctx->gtp_version) {
524 case GTP_V0:
525 mtu -= sizeof(struct gtp0_header);
526 break;
527 case GTP_V1:
528 mtu -= sizeof(struct gtp1_header);
529 break;
531 } else {
532 mtu = dst_mtu(&rt->dst);
535 rt->dst.ops->update_pmtu(&rt->dst, NULL, skb, mtu);
537 if (!skb_is_gso(skb) && (iph->frag_off & htons(IP_DF)) &&
538 mtu < ntohs(iph->tot_len)) {
539 netdev_dbg(dev, "packet too big, fragmentation needed\n");
540 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
541 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
542 htonl(mtu));
543 goto err_rt;
546 gtp_set_pktinfo_ipv4(pktinfo, pctx->sk, iph, pctx, rt, &fl4, dev);
547 gtp_push_header(skb, pktinfo);
549 return 0;
550 err_rt:
551 ip_rt_put(rt);
552 err:
553 return -EBADMSG;
556 static netdev_tx_t gtp_dev_xmit(struct sk_buff *skb, struct net_device *dev)
558 unsigned int proto = ntohs(skb->protocol);
559 struct gtp_pktinfo pktinfo;
560 int err;
562 /* Ensure there is sufficient headroom. */
563 if (skb_cow_head(skb, dev->needed_headroom))
564 goto tx_err;
566 skb_reset_inner_headers(skb);
568 /* PDP context lookups in gtp_build_skb_*() need rcu read-side lock. */
569 rcu_read_lock();
570 switch (proto) {
571 case ETH_P_IP:
572 err = gtp_build_skb_ip4(skb, dev, &pktinfo);
573 break;
574 default:
575 err = -EOPNOTSUPP;
576 break;
578 rcu_read_unlock();
580 if (err < 0)
581 goto tx_err;
583 switch (proto) {
584 case ETH_P_IP:
585 netdev_dbg(pktinfo.dev, "gtp -> IP src: %pI4 dst: %pI4\n",
586 &pktinfo.iph->saddr, &pktinfo.iph->daddr);
587 udp_tunnel_xmit_skb(pktinfo.rt, pktinfo.sk, skb,
588 pktinfo.fl4.saddr, pktinfo.fl4.daddr,
589 pktinfo.iph->tos,
590 ip4_dst_hoplimit(&pktinfo.rt->dst),
592 pktinfo.gtph_port, pktinfo.gtph_port,
593 true, false);
594 break;
597 return NETDEV_TX_OK;
598 tx_err:
599 dev->stats.tx_errors++;
600 dev_kfree_skb(skb);
601 return NETDEV_TX_OK;
604 static const struct net_device_ops gtp_netdev_ops = {
605 .ndo_init = gtp_dev_init,
606 .ndo_uninit = gtp_dev_uninit,
607 .ndo_start_xmit = gtp_dev_xmit,
608 .ndo_get_stats64 = ip_tunnel_get_stats64,
611 static void gtp_link_setup(struct net_device *dev)
613 dev->netdev_ops = &gtp_netdev_ops;
614 dev->needs_free_netdev = true;
616 dev->hard_header_len = 0;
617 dev->addr_len = 0;
619 /* Zero header length. */
620 dev->type = ARPHRD_NONE;
621 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
623 dev->priv_flags |= IFF_NO_QUEUE;
624 dev->features |= NETIF_F_LLTX;
625 netif_keep_dst(dev);
627 /* Assume largest header, ie. GTPv0. */
628 dev->needed_headroom = LL_MAX_HEADER +
629 sizeof(struct iphdr) +
630 sizeof(struct udphdr) +
631 sizeof(struct gtp0_header);
634 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize);
635 static void gtp_hashtable_free(struct gtp_dev *gtp);
636 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]);
638 static int gtp_newlink(struct net *src_net, struct net_device *dev,
639 struct nlattr *tb[], struct nlattr *data[],
640 struct netlink_ext_ack *extack)
642 struct gtp_dev *gtp;
643 struct gtp_net *gn;
644 int hashsize, err;
646 if (!data[IFLA_GTP_FD0] && !data[IFLA_GTP_FD1])
647 return -EINVAL;
649 gtp = netdev_priv(dev);
651 err = gtp_encap_enable(gtp, data);
652 if (err < 0)
653 return err;
655 if (!data[IFLA_GTP_PDP_HASHSIZE])
656 hashsize = 1024;
657 else
658 hashsize = nla_get_u32(data[IFLA_GTP_PDP_HASHSIZE]);
660 err = gtp_hashtable_new(gtp, hashsize);
661 if (err < 0)
662 goto out_encap;
664 err = register_netdevice(dev);
665 if (err < 0) {
666 netdev_dbg(dev, "failed to register new netdev %d\n", err);
667 goto out_hashtable;
670 gn = net_generic(dev_net(dev), gtp_net_id);
671 list_add_rcu(&gtp->list, &gn->gtp_dev_list);
673 netdev_dbg(dev, "registered new GTP interface\n");
675 return 0;
677 out_hashtable:
678 gtp_hashtable_free(gtp);
679 out_encap:
680 gtp_encap_disable(gtp);
681 return err;
684 static void gtp_dellink(struct net_device *dev, struct list_head *head)
686 struct gtp_dev *gtp = netdev_priv(dev);
688 gtp_encap_disable(gtp);
689 gtp_hashtable_free(gtp);
690 list_del_rcu(&gtp->list);
691 unregister_netdevice_queue(dev, head);
694 static const struct nla_policy gtp_policy[IFLA_GTP_MAX + 1] = {
695 [IFLA_GTP_FD0] = { .type = NLA_U32 },
696 [IFLA_GTP_FD1] = { .type = NLA_U32 },
697 [IFLA_GTP_PDP_HASHSIZE] = { .type = NLA_U32 },
698 [IFLA_GTP_ROLE] = { .type = NLA_U32 },
701 static int gtp_validate(struct nlattr *tb[], struct nlattr *data[],
702 struct netlink_ext_ack *extack)
704 if (!data)
705 return -EINVAL;
707 return 0;
710 static size_t gtp_get_size(const struct net_device *dev)
712 return nla_total_size(sizeof(__u32)); /* IFLA_GTP_PDP_HASHSIZE */
715 static int gtp_fill_info(struct sk_buff *skb, const struct net_device *dev)
717 struct gtp_dev *gtp = netdev_priv(dev);
719 if (nla_put_u32(skb, IFLA_GTP_PDP_HASHSIZE, gtp->hash_size))
720 goto nla_put_failure;
722 return 0;
724 nla_put_failure:
725 return -EMSGSIZE;
728 static struct rtnl_link_ops gtp_link_ops __read_mostly = {
729 .kind = "gtp",
730 .maxtype = IFLA_GTP_MAX,
731 .policy = gtp_policy,
732 .priv_size = sizeof(struct gtp_dev),
733 .setup = gtp_link_setup,
734 .validate = gtp_validate,
735 .newlink = gtp_newlink,
736 .dellink = gtp_dellink,
737 .get_size = gtp_get_size,
738 .fill_info = gtp_fill_info,
741 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize)
743 int i;
745 gtp->addr_hash = kmalloc(sizeof(struct hlist_head) * hsize, GFP_KERNEL);
746 if (gtp->addr_hash == NULL)
747 return -ENOMEM;
749 gtp->tid_hash = kmalloc(sizeof(struct hlist_head) * hsize, GFP_KERNEL);
750 if (gtp->tid_hash == NULL)
751 goto err1;
753 gtp->hash_size = hsize;
755 for (i = 0; i < hsize; i++) {
756 INIT_HLIST_HEAD(&gtp->addr_hash[i]);
757 INIT_HLIST_HEAD(&gtp->tid_hash[i]);
759 return 0;
760 err1:
761 kfree(gtp->addr_hash);
762 return -ENOMEM;
765 static void gtp_hashtable_free(struct gtp_dev *gtp)
767 struct pdp_ctx *pctx;
768 int i;
770 for (i = 0; i < gtp->hash_size; i++)
771 hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i], hlist_tid)
772 pdp_context_delete(pctx);
774 synchronize_rcu();
775 kfree(gtp->addr_hash);
776 kfree(gtp->tid_hash);
779 static struct sock *gtp_encap_enable_socket(int fd, int type,
780 struct gtp_dev *gtp)
782 struct udp_tunnel_sock_cfg tuncfg = {NULL};
783 struct socket *sock;
784 struct sock *sk;
785 int err;
787 pr_debug("enable gtp on %d, %d\n", fd, type);
789 sock = sockfd_lookup(fd, &err);
790 if (!sock) {
791 pr_debug("gtp socket fd=%d not found\n", fd);
792 return NULL;
795 if (sock->sk->sk_protocol != IPPROTO_UDP) {
796 pr_debug("socket fd=%d not UDP\n", fd);
797 sk = ERR_PTR(-EINVAL);
798 goto out_sock;
801 if (rcu_dereference_sk_user_data(sock->sk)) {
802 sk = ERR_PTR(-EBUSY);
803 goto out_sock;
806 sk = sock->sk;
807 sock_hold(sk);
809 tuncfg.sk_user_data = gtp;
810 tuncfg.encap_type = type;
811 tuncfg.encap_rcv = gtp_encap_recv;
812 tuncfg.encap_destroy = gtp_encap_destroy;
814 setup_udp_tunnel_sock(sock_net(sock->sk), sock, &tuncfg);
816 out_sock:
817 sockfd_put(sock);
818 return sk;
821 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[])
823 struct sock *sk1u = NULL;
824 struct sock *sk0 = NULL;
825 unsigned int role = GTP_ROLE_GGSN;
827 if (data[IFLA_GTP_FD0]) {
828 u32 fd0 = nla_get_u32(data[IFLA_GTP_FD0]);
830 sk0 = gtp_encap_enable_socket(fd0, UDP_ENCAP_GTP0, gtp);
831 if (IS_ERR(sk0))
832 return PTR_ERR(sk0);
835 if (data[IFLA_GTP_FD1]) {
836 u32 fd1 = nla_get_u32(data[IFLA_GTP_FD1]);
838 sk1u = gtp_encap_enable_socket(fd1, UDP_ENCAP_GTP1U, gtp);
839 if (IS_ERR(sk1u)) {
840 if (sk0)
841 gtp_encap_disable_sock(sk0);
842 return PTR_ERR(sk1u);
846 if (data[IFLA_GTP_ROLE]) {
847 role = nla_get_u32(data[IFLA_GTP_ROLE]);
848 if (role > GTP_ROLE_SGSN)
849 return -EINVAL;
852 gtp->sk0 = sk0;
853 gtp->sk1u = sk1u;
854 gtp->role = role;
856 return 0;
859 static struct gtp_dev *gtp_find_dev(struct net *src_net, struct nlattr *nla[])
861 struct gtp_dev *gtp = NULL;
862 struct net_device *dev;
863 struct net *net;
865 /* Examine the link attributes and figure out which network namespace
866 * we are talking about.
868 if (nla[GTPA_NET_NS_FD])
869 net = get_net_ns_by_fd(nla_get_u32(nla[GTPA_NET_NS_FD]));
870 else
871 net = get_net(src_net);
873 if (IS_ERR(net))
874 return NULL;
876 /* Check if there's an existing gtpX device to configure */
877 dev = dev_get_by_index_rcu(net, nla_get_u32(nla[GTPA_LINK]));
878 if (dev && dev->netdev_ops == &gtp_netdev_ops)
879 gtp = netdev_priv(dev);
881 put_net(net);
882 return gtp;
885 static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
887 pctx->gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]);
888 pctx->af = AF_INET;
889 pctx->peer_addr_ip4.s_addr =
890 nla_get_be32(info->attrs[GTPA_PEER_ADDRESS]);
891 pctx->ms_addr_ip4.s_addr =
892 nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
894 switch (pctx->gtp_version) {
895 case GTP_V0:
896 /* According to TS 09.60, sections 7.5.1 and 7.5.2, the flow
897 * label needs to be the same for uplink and downlink packets,
898 * so let's annotate this.
900 pctx->u.v0.tid = nla_get_u64(info->attrs[GTPA_TID]);
901 pctx->u.v0.flow = nla_get_u16(info->attrs[GTPA_FLOW]);
902 break;
903 case GTP_V1:
904 pctx->u.v1.i_tei = nla_get_u32(info->attrs[GTPA_I_TEI]);
905 pctx->u.v1.o_tei = nla_get_u32(info->attrs[GTPA_O_TEI]);
906 break;
907 default:
908 break;
912 static int ipv4_pdp_add(struct gtp_dev *gtp, struct sock *sk,
913 struct genl_info *info)
915 struct net_device *dev = gtp->dev;
916 u32 hash_ms, hash_tid = 0;
917 struct pdp_ctx *pctx;
918 bool found = false;
919 __be32 ms_addr;
921 ms_addr = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
922 hash_ms = ipv4_hashfn(ms_addr) % gtp->hash_size;
924 hlist_for_each_entry_rcu(pctx, &gtp->addr_hash[hash_ms], hlist_addr) {
925 if (pctx->ms_addr_ip4.s_addr == ms_addr) {
926 found = true;
927 break;
931 if (found) {
932 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
933 return -EEXIST;
934 if (info->nlhdr->nlmsg_flags & NLM_F_REPLACE)
935 return -EOPNOTSUPP;
937 ipv4_pdp_fill(pctx, info);
939 if (pctx->gtp_version == GTP_V0)
940 netdev_dbg(dev, "GTPv0-U: update tunnel id = %llx (pdp %p)\n",
941 pctx->u.v0.tid, pctx);
942 else if (pctx->gtp_version == GTP_V1)
943 netdev_dbg(dev, "GTPv1-U: update tunnel id = %x/%x (pdp %p)\n",
944 pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
946 return 0;
950 pctx = kmalloc(sizeof(struct pdp_ctx), GFP_KERNEL);
951 if (pctx == NULL)
952 return -ENOMEM;
954 sock_hold(sk);
955 pctx->sk = sk;
956 pctx->dev = gtp->dev;
957 ipv4_pdp_fill(pctx, info);
958 atomic_set(&pctx->tx_seq, 0);
960 switch (pctx->gtp_version) {
961 case GTP_V0:
962 /* TS 09.60: "The flow label identifies unambiguously a GTP
963 * flow.". We use the tid for this instead, I cannot find a
964 * situation in which this doesn't unambiguosly identify the
965 * PDP context.
967 hash_tid = gtp0_hashfn(pctx->u.v0.tid) % gtp->hash_size;
968 break;
969 case GTP_V1:
970 hash_tid = gtp1u_hashfn(pctx->u.v1.i_tei) % gtp->hash_size;
971 break;
974 hlist_add_head_rcu(&pctx->hlist_addr, &gtp->addr_hash[hash_ms]);
975 hlist_add_head_rcu(&pctx->hlist_tid, &gtp->tid_hash[hash_tid]);
977 switch (pctx->gtp_version) {
978 case GTP_V0:
979 netdev_dbg(dev, "GTPv0-U: new PDP ctx id=%llx ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
980 pctx->u.v0.tid, &pctx->peer_addr_ip4,
981 &pctx->ms_addr_ip4, pctx);
982 break;
983 case GTP_V1:
984 netdev_dbg(dev, "GTPv1-U: new PDP ctx id=%x/%x ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
985 pctx->u.v1.i_tei, pctx->u.v1.o_tei,
986 &pctx->peer_addr_ip4, &pctx->ms_addr_ip4, pctx);
987 break;
990 return 0;
993 static void pdp_context_free(struct rcu_head *head)
995 struct pdp_ctx *pctx = container_of(head, struct pdp_ctx, rcu_head);
997 sock_put(pctx->sk);
998 kfree(pctx);
1001 static void pdp_context_delete(struct pdp_ctx *pctx)
1003 hlist_del_rcu(&pctx->hlist_tid);
1004 hlist_del_rcu(&pctx->hlist_addr);
1005 call_rcu(&pctx->rcu_head, pdp_context_free);
1008 static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
1010 unsigned int version;
1011 struct gtp_dev *gtp;
1012 struct sock *sk;
1013 int err;
1015 if (!info->attrs[GTPA_VERSION] ||
1016 !info->attrs[GTPA_LINK] ||
1017 !info->attrs[GTPA_PEER_ADDRESS] ||
1018 !info->attrs[GTPA_MS_ADDRESS])
1019 return -EINVAL;
1021 version = nla_get_u32(info->attrs[GTPA_VERSION]);
1023 switch (version) {
1024 case GTP_V0:
1025 if (!info->attrs[GTPA_TID] ||
1026 !info->attrs[GTPA_FLOW])
1027 return -EINVAL;
1028 break;
1029 case GTP_V1:
1030 if (!info->attrs[GTPA_I_TEI] ||
1031 !info->attrs[GTPA_O_TEI])
1032 return -EINVAL;
1033 break;
1035 default:
1036 return -EINVAL;
1039 rcu_read_lock();
1041 gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
1042 if (!gtp) {
1043 err = -ENODEV;
1044 goto out_unlock;
1047 if (version == GTP_V0)
1048 sk = gtp->sk0;
1049 else if (version == GTP_V1)
1050 sk = gtp->sk1u;
1051 else
1052 sk = NULL;
1054 if (!sk) {
1055 err = -ENODEV;
1056 goto out_unlock;
1059 err = ipv4_pdp_add(gtp, sk, info);
1061 out_unlock:
1062 rcu_read_unlock();
1063 return err;
1066 static struct pdp_ctx *gtp_find_pdp_by_link(struct net *net,
1067 struct nlattr *nla[])
1069 struct gtp_dev *gtp;
1071 gtp = gtp_find_dev(net, nla);
1072 if (!gtp)
1073 return ERR_PTR(-ENODEV);
1075 if (nla[GTPA_MS_ADDRESS]) {
1076 __be32 ip = nla_get_be32(nla[GTPA_MS_ADDRESS]);
1078 return ipv4_pdp_find(gtp, ip);
1079 } else if (nla[GTPA_VERSION]) {
1080 u32 gtp_version = nla_get_u32(nla[GTPA_VERSION]);
1082 if (gtp_version == GTP_V0 && nla[GTPA_TID])
1083 return gtp0_pdp_find(gtp, nla_get_u64(nla[GTPA_TID]));
1084 else if (gtp_version == GTP_V1 && nla[GTPA_I_TEI])
1085 return gtp1_pdp_find(gtp, nla_get_u32(nla[GTPA_I_TEI]));
1088 return ERR_PTR(-EINVAL);
1091 static struct pdp_ctx *gtp_find_pdp(struct net *net, struct nlattr *nla[])
1093 struct pdp_ctx *pctx;
1095 if (nla[GTPA_LINK])
1096 pctx = gtp_find_pdp_by_link(net, nla);
1097 else
1098 pctx = ERR_PTR(-EINVAL);
1100 if (!pctx)
1101 pctx = ERR_PTR(-ENOENT);
1103 return pctx;
1106 static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
1108 struct pdp_ctx *pctx;
1109 int err = 0;
1111 if (!info->attrs[GTPA_VERSION])
1112 return -EINVAL;
1114 rcu_read_lock();
1116 pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
1117 if (IS_ERR(pctx)) {
1118 err = PTR_ERR(pctx);
1119 goto out_unlock;
1122 if (pctx->gtp_version == GTP_V0)
1123 netdev_dbg(pctx->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n",
1124 pctx->u.v0.tid, pctx);
1125 else if (pctx->gtp_version == GTP_V1)
1126 netdev_dbg(pctx->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n",
1127 pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
1129 pdp_context_delete(pctx);
1131 out_unlock:
1132 rcu_read_unlock();
1133 return err;
1136 static struct genl_family gtp_genl_family;
1138 static int gtp_genl_fill_info(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
1139 u32 type, struct pdp_ctx *pctx)
1141 void *genlh;
1143 genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, 0,
1144 type);
1145 if (genlh == NULL)
1146 goto nlmsg_failure;
1148 if (nla_put_u32(skb, GTPA_VERSION, pctx->gtp_version) ||
1149 nla_put_be32(skb, GTPA_PEER_ADDRESS, pctx->peer_addr_ip4.s_addr) ||
1150 nla_put_be32(skb, GTPA_MS_ADDRESS, pctx->ms_addr_ip4.s_addr))
1151 goto nla_put_failure;
1153 switch (pctx->gtp_version) {
1154 case GTP_V0:
1155 if (nla_put_u64_64bit(skb, GTPA_TID, pctx->u.v0.tid, GTPA_PAD) ||
1156 nla_put_u16(skb, GTPA_FLOW, pctx->u.v0.flow))
1157 goto nla_put_failure;
1158 break;
1159 case GTP_V1:
1160 if (nla_put_u32(skb, GTPA_I_TEI, pctx->u.v1.i_tei) ||
1161 nla_put_u32(skb, GTPA_O_TEI, pctx->u.v1.o_tei))
1162 goto nla_put_failure;
1163 break;
1165 genlmsg_end(skb, genlh);
1166 return 0;
1168 nlmsg_failure:
1169 nla_put_failure:
1170 genlmsg_cancel(skb, genlh);
1171 return -EMSGSIZE;
1174 static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info)
1176 struct pdp_ctx *pctx = NULL;
1177 struct sk_buff *skb2;
1178 int err;
1180 if (!info->attrs[GTPA_VERSION])
1181 return -EINVAL;
1183 rcu_read_lock();
1185 pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
1186 if (IS_ERR(pctx)) {
1187 err = PTR_ERR(pctx);
1188 goto err_unlock;
1191 skb2 = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
1192 if (skb2 == NULL) {
1193 err = -ENOMEM;
1194 goto err_unlock;
1197 err = gtp_genl_fill_info(skb2, NETLINK_CB(skb).portid,
1198 info->snd_seq, info->nlhdr->nlmsg_type, pctx);
1199 if (err < 0)
1200 goto err_unlock_free;
1202 rcu_read_unlock();
1203 return genlmsg_unicast(genl_info_net(info), skb2, info->snd_portid);
1205 err_unlock_free:
1206 kfree_skb(skb2);
1207 err_unlock:
1208 rcu_read_unlock();
1209 return err;
1212 static int gtp_genl_dump_pdp(struct sk_buff *skb,
1213 struct netlink_callback *cb)
1215 struct gtp_dev *last_gtp = (struct gtp_dev *)cb->args[2], *gtp;
1216 struct net *net = sock_net(skb->sk);
1217 struct gtp_net *gn = net_generic(net, gtp_net_id);
1218 unsigned long tid = cb->args[1];
1219 int i, k = cb->args[0], ret;
1220 struct pdp_ctx *pctx;
1222 if (cb->args[4])
1223 return 0;
1225 list_for_each_entry_rcu(gtp, &gn->gtp_dev_list, list) {
1226 if (last_gtp && last_gtp != gtp)
1227 continue;
1228 else
1229 last_gtp = NULL;
1231 for (i = k; i < gtp->hash_size; i++) {
1232 hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i], hlist_tid) {
1233 if (tid && tid != pctx->u.tid)
1234 continue;
1235 else
1236 tid = 0;
1238 ret = gtp_genl_fill_info(skb,
1239 NETLINK_CB(cb->skb).portid,
1240 cb->nlh->nlmsg_seq,
1241 cb->nlh->nlmsg_type, pctx);
1242 if (ret < 0) {
1243 cb->args[0] = i;
1244 cb->args[1] = pctx->u.tid;
1245 cb->args[2] = (unsigned long)gtp;
1246 goto out;
1251 cb->args[4] = 1;
1252 out:
1253 return skb->len;
1256 static struct nla_policy gtp_genl_policy[GTPA_MAX + 1] = {
1257 [GTPA_LINK] = { .type = NLA_U32, },
1258 [GTPA_VERSION] = { .type = NLA_U32, },
1259 [GTPA_TID] = { .type = NLA_U64, },
1260 [GTPA_PEER_ADDRESS] = { .type = NLA_U32, },
1261 [GTPA_MS_ADDRESS] = { .type = NLA_U32, },
1262 [GTPA_FLOW] = { .type = NLA_U16, },
1263 [GTPA_NET_NS_FD] = { .type = NLA_U32, },
1264 [GTPA_I_TEI] = { .type = NLA_U32, },
1265 [GTPA_O_TEI] = { .type = NLA_U32, },
1268 static const struct genl_ops gtp_genl_ops[] = {
1270 .cmd = GTP_CMD_NEWPDP,
1271 .doit = gtp_genl_new_pdp,
1272 .policy = gtp_genl_policy,
1273 .flags = GENL_ADMIN_PERM,
1276 .cmd = GTP_CMD_DELPDP,
1277 .doit = gtp_genl_del_pdp,
1278 .policy = gtp_genl_policy,
1279 .flags = GENL_ADMIN_PERM,
1282 .cmd = GTP_CMD_GETPDP,
1283 .doit = gtp_genl_get_pdp,
1284 .dumpit = gtp_genl_dump_pdp,
1285 .policy = gtp_genl_policy,
1286 .flags = GENL_ADMIN_PERM,
1290 static struct genl_family gtp_genl_family __ro_after_init = {
1291 .name = "gtp",
1292 .version = 0,
1293 .hdrsize = 0,
1294 .maxattr = GTPA_MAX,
1295 .netnsok = true,
1296 .module = THIS_MODULE,
1297 .ops = gtp_genl_ops,
1298 .n_ops = ARRAY_SIZE(gtp_genl_ops),
1301 static int __net_init gtp_net_init(struct net *net)
1303 struct gtp_net *gn = net_generic(net, gtp_net_id);
1305 INIT_LIST_HEAD(&gn->gtp_dev_list);
1306 return 0;
1309 static void __net_exit gtp_net_exit(struct net *net)
1311 struct gtp_net *gn = net_generic(net, gtp_net_id);
1312 struct gtp_dev *gtp;
1313 LIST_HEAD(list);
1315 rtnl_lock();
1316 list_for_each_entry(gtp, &gn->gtp_dev_list, list)
1317 gtp_dellink(gtp->dev, &list);
1319 unregister_netdevice_many(&list);
1320 rtnl_unlock();
1323 static struct pernet_operations gtp_net_ops = {
1324 .init = gtp_net_init,
1325 .exit = gtp_net_exit,
1326 .id = &gtp_net_id,
1327 .size = sizeof(struct gtp_net),
1330 static int __init gtp_init(void)
1332 int err;
1334 get_random_bytes(&gtp_h_initval, sizeof(gtp_h_initval));
1336 err = rtnl_link_register(&gtp_link_ops);
1337 if (err < 0)
1338 goto error_out;
1340 err = genl_register_family(&gtp_genl_family);
1341 if (err < 0)
1342 goto unreg_rtnl_link;
1344 err = register_pernet_subsys(&gtp_net_ops);
1345 if (err < 0)
1346 goto unreg_genl_family;
1348 pr_info("GTP module loaded (pdp ctx size %zd bytes)\n",
1349 sizeof(struct pdp_ctx));
1350 return 0;
1352 unreg_genl_family:
1353 genl_unregister_family(&gtp_genl_family);
1354 unreg_rtnl_link:
1355 rtnl_link_unregister(&gtp_link_ops);
1356 error_out:
1357 pr_err("error loading GTP module loaded\n");
1358 return err;
1360 late_initcall(gtp_init);
1362 static void __exit gtp_fini(void)
1364 unregister_pernet_subsys(&gtp_net_ops);
1365 genl_unregister_family(&gtp_genl_family);
1366 rtnl_link_unregister(&gtp_link_ops);
1368 pr_info("GTP module unloaded\n");
1370 module_exit(gtp_fini);
1372 MODULE_LICENSE("GPL");
1373 MODULE_AUTHOR("Harald Welte <hwelte@sysmocom.de>");
1374 MODULE_DESCRIPTION("Interface driver for GTP encapsulated traffic");
1375 MODULE_ALIAS_RTNL_LINK("gtp");
1376 MODULE_ALIAS_GENL_FAMILY("gtp");