xref: /linux/drivers/net/gtp.c (revision daa2be74b1b2302004945b2a5e32424e177cc7da)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* GTP according to GSM TS 09.60 / 3GPP TS 29.060
3  *
4  * (C) 2012-2014 by sysmocom - s.f.m.c. GmbH
5  * (C) 2016 by Pablo Neira Ayuso <pablo@netfilter.org>
6  *
7  * Author: Harald Welte <hwelte@sysmocom.de>
8  *	   Pablo Neira Ayuso <pablo@netfilter.org>
9  *	   Andreas Schultz <aschultz@travelping.com>
10  */
11 
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 
14 #include <linux/module.h>
15 #include <linux/skbuff.h>
16 #include <linux/udp.h>
17 #include <linux/rculist.h>
18 #include <linux/jhash.h>
19 #include <linux/if_tunnel.h>
20 #include <linux/net.h>
21 #include <linux/file.h>
22 #include <linux/gtp.h>
23 
24 #include <net/net_namespace.h>
25 #include <net/protocol.h>
26 #include <net/ip.h>
27 #include <net/ipv6.h>
28 #include <net/udp.h>
29 #include <net/udp_tunnel.h>
30 #include <net/icmp.h>
31 #include <net/xfrm.h>
32 #include <net/genetlink.h>
33 #include <net/netns/generic.h>
34 #include <net/gtp.h>
35 
36 /* An active session for the subscriber. */
37 struct pdp_ctx {
38 	struct hlist_node	hlist_tid;
39 	struct hlist_node	hlist_addr;
40 
41 	union {
42 		struct {
43 			u64	tid;
44 			u16	flow;
45 		} v0;
46 		struct {
47 			u32	i_tei;
48 			u32	o_tei;
49 		} v1;
50 	} u;
51 	u8			gtp_version;
52 	u16			af;
53 
54 	union {
55 		struct in_addr	addr;
56 		struct in6_addr	addr6;
57 	} ms;
58 	union {
59 		struct in_addr	addr;
60 		struct in6_addr	addr6;
61 	} peer;
62 
63 	struct sock		*sk;
64 	struct net_device       *dev;
65 
66 	atomic_t		tx_seq;
67 	struct rcu_head		rcu_head;
68 };
69 
70 /* One instance of the GTP device. */
71 struct gtp_dev {
72 	struct list_head	list;
73 
74 	struct sock		*sk0;
75 	struct sock		*sk1u;
76 	u8			sk_created;
77 
78 	struct net_device	*dev;
79 	struct net		*net;
80 
81 	unsigned int		role;
82 	unsigned int		hash_size;
83 	struct hlist_head	*tid_hash;
84 	struct hlist_head	*addr_hash;
85 
86 	u8			restart_count;
87 };
88 
89 struct echo_info {
90 	u16			af;
91 	u8			gtp_version;
92 
93 	union {
94 		struct in_addr	addr;
95 	} ms;
96 	union {
97 		struct in_addr	addr;
98 	} peer;
99 };
100 
101 static unsigned int gtp_net_id __read_mostly;
102 
103 struct gtp_net {
104 	struct list_head gtp_dev_list;
105 };
106 
107 static u32 gtp_h_initval;
108 
109 static struct genl_family gtp_genl_family;
110 
111 enum gtp_multicast_groups {
112 	GTP_GENL_MCGRP,
113 };
114 
115 static const struct genl_multicast_group gtp_genl_mcgrps[] = {
116 	[GTP_GENL_MCGRP] = { .name = GTP_GENL_MCGRP_NAME },
117 };
118 
119 static void pdp_context_delete(struct pdp_ctx *pctx);
120 
121 static inline u32 gtp0_hashfn(u64 tid)
122 {
123 	u32 *tid32 = (u32 *) &tid;
124 	return jhash_2words(tid32[0], tid32[1], gtp_h_initval);
125 }
126 
127 static inline u32 gtp1u_hashfn(u32 tid)
128 {
129 	return jhash_1word(tid, gtp_h_initval);
130 }
131 
132 static inline u32 ipv4_hashfn(__be32 ip)
133 {
134 	return jhash_1word((__force u32)ip, gtp_h_initval);
135 }
136 
137 static u32 ipv6_hashfn(const struct in6_addr *ip6)
138 {
139 	return jhash_2words((__force u32)ip6->s6_addr32[0],
140 			    (__force u32)ip6->s6_addr32[1], gtp_h_initval);
141 }
142 
143 /* Resolve a PDP context structure based on the 64bit TID. */
144 static struct pdp_ctx *gtp0_pdp_find(struct gtp_dev *gtp, u64 tid, u16 family)
145 {
146 	struct hlist_head *head;
147 	struct pdp_ctx *pdp;
148 
149 	head = &gtp->tid_hash[gtp0_hashfn(tid) % gtp->hash_size];
150 
151 	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
152 		if (pdp->af == family &&
153 		    pdp->gtp_version == GTP_V0 &&
154 		    pdp->u.v0.tid == tid)
155 			return pdp;
156 	}
157 	return NULL;
158 }
159 
160 /* Resolve a PDP context structure based on the 32bit TEI. */
161 static struct pdp_ctx *gtp1_pdp_find(struct gtp_dev *gtp, u32 tid, u16 family)
162 {
163 	struct hlist_head *head;
164 	struct pdp_ctx *pdp;
165 
166 	head = &gtp->tid_hash[gtp1u_hashfn(tid) % gtp->hash_size];
167 
168 	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
169 		if (pdp->af == family &&
170 		    pdp->gtp_version == GTP_V1 &&
171 		    pdp->u.v1.i_tei == tid)
172 			return pdp;
173 	}
174 	return NULL;
175 }
176 
177 /* Resolve a PDP context based on IPv4 address of MS. */
178 static struct pdp_ctx *ipv4_pdp_find(struct gtp_dev *gtp, __be32 ms_addr)
179 {
180 	struct hlist_head *head;
181 	struct pdp_ctx *pdp;
182 
183 	head = &gtp->addr_hash[ipv4_hashfn(ms_addr) % gtp->hash_size];
184 
185 	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
186 		if (pdp->af == AF_INET &&
187 		    pdp->ms.addr.s_addr == ms_addr)
188 			return pdp;
189 	}
190 
191 	return NULL;
192 }
193 
194 /* 3GPP TS 29.060: PDN Connection: the association between a MS represented by
195  * [...] one IPv6 *prefix* and a PDN represented by an APN.
196  *
197  * Then, 3GPP TS 29.061, Section 11.2.1.3 says: The size of the prefix shall be
198  * according to the maximum prefix length for a global IPv6 address as
199  * specified in the IPv6 Addressing Architecture, see RFC 4291.
200  *
201  * Finally, RFC 4291 section 2.5.4 states: All Global Unicast addresses other
202  * than those that start with binary 000 have a 64-bit interface ID field
203  * (i.e., n + m = 64).
204  */
205 static bool ipv6_pdp_addr_equal(const struct in6_addr *a,
206 				const struct in6_addr *b)
207 {
208 	return a->s6_addr32[0] == b->s6_addr32[0] &&
209 	       a->s6_addr32[1] == b->s6_addr32[1];
210 }
211 
212 static struct pdp_ctx *ipv6_pdp_find(struct gtp_dev *gtp,
213 				     const struct in6_addr *ms_addr)
214 {
215 	struct hlist_head *head;
216 	struct pdp_ctx *pdp;
217 
218 	head = &gtp->addr_hash[ipv6_hashfn(ms_addr) % gtp->hash_size];
219 
220 	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
221 		if (pdp->af == AF_INET6 &&
222 		    ipv6_pdp_addr_equal(&pdp->ms.addr6, ms_addr))
223 			return pdp;
224 	}
225 
226 	return NULL;
227 }
228 
229 static bool gtp_check_ms_ipv4(struct sk_buff *skb, struct pdp_ctx *pctx,
230 				  unsigned int hdrlen, unsigned int role)
231 {
232 	struct iphdr *iph;
233 
234 	if (!pskb_may_pull(skb, hdrlen + sizeof(struct iphdr)))
235 		return false;
236 
237 	iph = (struct iphdr *)(skb->data + hdrlen);
238 
239 	if (role == GTP_ROLE_SGSN)
240 		return iph->daddr == pctx->ms.addr.s_addr;
241 	else
242 		return iph->saddr == pctx->ms.addr.s_addr;
243 }
244 
245 static bool gtp_check_ms_ipv6(struct sk_buff *skb, struct pdp_ctx *pctx,
246 			      unsigned int hdrlen, unsigned int role)
247 {
248 	struct ipv6hdr *ip6h;
249 	int ret;
250 
251 	if (!pskb_may_pull(skb, hdrlen + sizeof(struct ipv6hdr)))
252 		return false;
253 
254 	ip6h = (struct ipv6hdr *)(skb->data + hdrlen);
255 
256 	if ((ipv6_addr_type(&ip6h->saddr) & IPV6_ADDR_LINKLOCAL) ||
257 	    (ipv6_addr_type(&ip6h->daddr) & IPV6_ADDR_LINKLOCAL))
258 		return false;
259 
260 	if (role == GTP_ROLE_SGSN) {
261 		ret = ipv6_pdp_addr_equal(&ip6h->daddr, &pctx->ms.addr6);
262 	} else {
263 		ret = ipv6_pdp_addr_equal(&ip6h->saddr, &pctx->ms.addr6);
264 	}
265 
266 	return ret;
267 }
268 
269 /* Check if the inner IP address in this packet is assigned to any
270  * existing mobile subscriber.
271  */
272 static bool gtp_check_ms(struct sk_buff *skb, struct pdp_ctx *pctx,
273 			 unsigned int hdrlen, unsigned int role,
274 			 __u16 inner_proto)
275 {
276 	switch (inner_proto) {
277 	case ETH_P_IP:
278 		return gtp_check_ms_ipv4(skb, pctx, hdrlen, role);
279 	case ETH_P_IPV6:
280 		return gtp_check_ms_ipv6(skb, pctx, hdrlen, role);
281 	}
282 	return false;
283 }
284 
285 static int gtp_inner_proto(struct sk_buff *skb, unsigned int hdrlen,
286 			   __u16 *inner_proto)
287 {
288 	__u8 *ip_version, _ip_version;
289 
290 	ip_version = skb_header_pointer(skb, hdrlen, sizeof(*ip_version),
291 					&_ip_version);
292 	if (!ip_version)
293 		return -1;
294 
295 	switch (*ip_version & 0xf0) {
296 	case 0x40:
297 		*inner_proto = ETH_P_IP;
298 		break;
299 	case 0x60:
300 		*inner_proto = ETH_P_IPV6;
301 		break;
302 	default:
303 		return -1;
304 	}
305 
306 	return 0;
307 }
308 
309 static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb,
310 		  unsigned int hdrlen, unsigned int role, __u16 inner_proto)
311 {
312 	if (!gtp_check_ms(skb, pctx, hdrlen, role, inner_proto)) {
313 		netdev_dbg(pctx->dev, "No PDP ctx for this MS\n");
314 		return 1;
315 	}
316 
317 	/* Get rid of the GTP + UDP headers. */
318 	if (iptunnel_pull_header(skb, hdrlen, htons(inner_proto),
319 			 !net_eq(sock_net(pctx->sk), dev_net(pctx->dev)))) {
320 		pctx->dev->stats.rx_length_errors++;
321 		goto err;
322 	}
323 
324 	netdev_dbg(pctx->dev, "forwarding packet from GGSN to uplink\n");
325 
326 	/* Now that the UDP and the GTP header have been removed, set up the
327 	 * new network header. This is required by the upper layer to
328 	 * calculate the transport header.
329 	 */
330 	skb_reset_network_header(skb);
331 	skb_reset_mac_header(skb);
332 
333 	skb->dev = pctx->dev;
334 
335 	dev_sw_netstats_rx_add(pctx->dev, skb->len);
336 
337 	__netif_rx(skb);
338 	return 0;
339 
340 err:
341 	pctx->dev->stats.rx_dropped++;
342 	return -1;
343 }
344 
345 static struct rtable *ip4_route_output_gtp(struct flowi4 *fl4,
346 					   const struct sock *sk,
347 					   __be32 daddr, __be32 saddr)
348 {
349 	memset(fl4, 0, sizeof(*fl4));
350 	fl4->flowi4_oif		= sk->sk_bound_dev_if;
351 	fl4->daddr		= daddr;
352 	fl4->saddr		= saddr;
353 	fl4->flowi4_tos		= ip_sock_rt_tos(sk);
354 	fl4->flowi4_scope	= ip_sock_rt_scope(sk);
355 	fl4->flowi4_proto	= sk->sk_protocol;
356 
357 	return ip_route_output_key(sock_net(sk), fl4);
358 }
359 
360 static struct rt6_info *ip6_route_output_gtp(struct net *net,
361 					     struct flowi6 *fl6,
362 					     const struct sock *sk,
363 					     const struct in6_addr *daddr,
364 					     struct in6_addr *saddr)
365 {
366 	struct dst_entry *dst;
367 
368 	memset(fl6, 0, sizeof(*fl6));
369 	fl6->flowi6_oif		= sk->sk_bound_dev_if;
370 	fl6->daddr		= *daddr;
371 	fl6->saddr		= *saddr;
372 	fl6->flowi6_proto	= sk->sk_protocol;
373 
374 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sk, fl6, NULL);
375 	if (IS_ERR(dst))
376 		return ERR_PTR(-ENETUNREACH);
377 
378 	return (struct rt6_info *)dst;
379 }
380 
381 /* GSM TS 09.60. 7.3
382  * In all Path Management messages:
383  * - TID: is not used and shall be set to 0.
384  * - Flow Label is not used and shall be set to 0
385  * In signalling messages:
386  * - number: this field is not yet used in signalling messages.
387  *   It shall be set to 255 by the sender and shall be ignored
388  *   by the receiver
389  * Returns true if the echo req was correct, false otherwise.
390  */
391 static bool gtp0_validate_echo_hdr(struct gtp0_header *gtp0)
392 {
393 	return !(gtp0->tid || (gtp0->flags ^ 0x1e) ||
394 		gtp0->number != 0xff || gtp0->flow);
395 }
396 
397 /* msg_type has to be GTP_ECHO_REQ or GTP_ECHO_RSP */
398 static void gtp0_build_echo_msg(struct gtp0_header *hdr, __u8 msg_type)
399 {
400 	int len_pkt, len_hdr;
401 
402 	hdr->flags = 0x1e; /* v0, GTP-non-prime. */
403 	hdr->type = msg_type;
404 	/* GSM TS 09.60. 7.3 In all Path Management Flow Label and TID
405 	 * are not used and shall be set to 0.
406 	 */
407 	hdr->flow = 0;
408 	hdr->tid = 0;
409 	hdr->number = 0xff;
410 	hdr->spare[0] = 0xff;
411 	hdr->spare[1] = 0xff;
412 	hdr->spare[2] = 0xff;
413 
414 	len_pkt = sizeof(struct gtp0_packet);
415 	len_hdr = sizeof(struct gtp0_header);
416 
417 	if (msg_type == GTP_ECHO_RSP)
418 		hdr->length = htons(len_pkt - len_hdr);
419 	else
420 		hdr->length = 0;
421 }
422 
423 static int gtp0_send_echo_resp_ip(struct gtp_dev *gtp, struct sk_buff *skb)
424 {
425 	struct iphdr *iph = ip_hdr(skb);
426 	struct flowi4 fl4;
427 	struct rtable *rt;
428 
429 	/* find route to the sender,
430 	 * src address becomes dst address and vice versa.
431 	 */
432 	rt = ip4_route_output_gtp(&fl4, gtp->sk0, iph->saddr, iph->daddr);
433 	if (IS_ERR(rt)) {
434 		netdev_dbg(gtp->dev, "no route for echo response from %pI4\n",
435 			   &iph->saddr);
436 		return -1;
437 	}
438 
439 	udp_tunnel_xmit_skb(rt, gtp->sk0, skb,
440 			    fl4.saddr, fl4.daddr,
441 			    iph->tos,
442 			    ip4_dst_hoplimit(&rt->dst),
443 			    0,
444 			    htons(GTP0_PORT), htons(GTP0_PORT),
445 			    !net_eq(sock_net(gtp->sk1u),
446 				    dev_net(gtp->dev)),
447 			    false);
448 
449 	return 0;
450 }
451 
452 static int gtp0_send_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
453 {
454 	struct gtp0_packet *gtp_pkt;
455 	struct gtp0_header *gtp0;
456 	__be16 seq;
457 
458 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
459 
460 	if (!gtp0_validate_echo_hdr(gtp0))
461 		return -1;
462 
463 	seq = gtp0->seq;
464 
465 	/* pull GTP and UDP headers */
466 	skb_pull_data(skb, sizeof(struct gtp0_header) + sizeof(struct udphdr));
467 
468 	gtp_pkt = skb_push(skb, sizeof(struct gtp0_packet));
469 	memset(gtp_pkt, 0, sizeof(struct gtp0_packet));
470 
471 	gtp0_build_echo_msg(&gtp_pkt->gtp0_h, GTP_ECHO_RSP);
472 
473 	/* GSM TS 09.60. 7.3 The Sequence Number in a signalling response
474 	 * message shall be copied from the signalling request message
475 	 * that the GSN is replying to.
476 	 */
477 	gtp_pkt->gtp0_h.seq = seq;
478 
479 	gtp_pkt->ie.tag = GTPIE_RECOVERY;
480 	gtp_pkt->ie.val = gtp->restart_count;
481 
482 	switch (gtp->sk0->sk_family) {
483 	case AF_INET:
484 		if (gtp0_send_echo_resp_ip(gtp, skb) < 0)
485 			return -1;
486 		break;
487 	case AF_INET6:
488 		return -1;
489 	}
490 
491 	return 0;
492 }
493 
494 static int gtp_genl_fill_echo(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
495 			      int flags, u32 type, struct echo_info echo)
496 {
497 	void *genlh;
498 
499 	genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, flags,
500 			    type);
501 	if (!genlh)
502 		goto failure;
503 
504 	if (nla_put_u32(skb, GTPA_VERSION, echo.gtp_version) ||
505 	    nla_put_be32(skb, GTPA_PEER_ADDRESS, echo.peer.addr.s_addr) ||
506 	    nla_put_be32(skb, GTPA_MS_ADDRESS, echo.ms.addr.s_addr))
507 		goto failure;
508 
509 	genlmsg_end(skb, genlh);
510 	return 0;
511 
512 failure:
513 	genlmsg_cancel(skb, genlh);
514 	return -EMSGSIZE;
515 }
516 
517 static void gtp0_handle_echo_resp_ip(struct sk_buff *skb, struct echo_info *echo)
518 {
519 	struct iphdr *iph = ip_hdr(skb);
520 
521 	echo->ms.addr.s_addr = iph->daddr;
522 	echo->peer.addr.s_addr = iph->saddr;
523 	echo->gtp_version = GTP_V0;
524 }
525 
526 static int gtp0_handle_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
527 {
528 	struct gtp0_header *gtp0;
529 	struct echo_info echo;
530 	struct sk_buff *msg;
531 	int ret;
532 
533 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
534 
535 	if (!gtp0_validate_echo_hdr(gtp0))
536 		return -1;
537 
538 	switch (gtp->sk0->sk_family) {
539 	case AF_INET:
540 		gtp0_handle_echo_resp_ip(skb, &echo);
541 		break;
542 	case AF_INET6:
543 		return -1;
544 	}
545 
546 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
547 	if (!msg)
548 		return -ENOMEM;
549 
550 	ret = gtp_genl_fill_echo(msg, 0, 0, 0, GTP_CMD_ECHOREQ, echo);
551 	if (ret < 0) {
552 		nlmsg_free(msg);
553 		return ret;
554 	}
555 
556 	return genlmsg_multicast_netns(&gtp_genl_family, dev_net(gtp->dev),
557 				       msg, 0, GTP_GENL_MCGRP, GFP_ATOMIC);
558 }
559 
560 static int gtp_proto_to_family(__u16 proto)
561 {
562 	switch (proto) {
563 	case ETH_P_IP:
564 		return AF_INET;
565 	case ETH_P_IPV6:
566 		return AF_INET6;
567 	default:
568 		WARN_ON_ONCE(1);
569 		break;
570 	}
571 
572 	return AF_UNSPEC;
573 }
574 
575 /* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */
576 static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
577 {
578 	unsigned int hdrlen = sizeof(struct udphdr) +
579 			      sizeof(struct gtp0_header);
580 	struct gtp0_header *gtp0;
581 	struct pdp_ctx *pctx;
582 	__u16 inner_proto;
583 
584 	if (!pskb_may_pull(skb, hdrlen))
585 		return -1;
586 
587 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
588 
589 	if ((gtp0->flags >> 5) != GTP_V0)
590 		return 1;
591 
592 	/* If the sockets were created in kernel, it means that
593 	 * there is no daemon running in userspace which would
594 	 * handle echo request.
595 	 */
596 	if (gtp0->type == GTP_ECHO_REQ && gtp->sk_created)
597 		return gtp0_send_echo_resp(gtp, skb);
598 
599 	if (gtp0->type == GTP_ECHO_RSP && gtp->sk_created)
600 		return gtp0_handle_echo_resp(gtp, skb);
601 
602 	if (gtp0->type != GTP_TPDU)
603 		return 1;
604 
605 	if (gtp_inner_proto(skb, hdrlen, &inner_proto) < 0) {
606 		netdev_dbg(gtp->dev, "GTP packet does not encapsulate an IP packet\n");
607 		return -1;
608 	}
609 
610 	pctx = gtp0_pdp_find(gtp, be64_to_cpu(gtp0->tid),
611 			     gtp_proto_to_family(inner_proto));
612 	if (!pctx) {
613 		netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
614 		return 1;
615 	}
616 
617 	return gtp_rx(pctx, skb, hdrlen, gtp->role, inner_proto);
618 }
619 
620 /* msg_type has to be GTP_ECHO_REQ or GTP_ECHO_RSP */
621 static void gtp1u_build_echo_msg(struct gtp1_header_long *hdr, __u8 msg_type)
622 {
623 	int len_pkt, len_hdr;
624 
625 	/* S flag must be set to 1 */
626 	hdr->flags = 0x32; /* v1, GTP-non-prime. */
627 	hdr->type = msg_type;
628 	/* 3GPP TS 29.281 5.1 - TEID has to be set to 0 */
629 	hdr->tid = 0;
630 
631 	/* seq, npdu and next should be counted to the length of the GTP packet
632 	 * that's why szie of gtp1_header should be subtracted,
633 	 * not size of gtp1_header_long.
634 	 */
635 
636 	len_hdr = sizeof(struct gtp1_header);
637 
638 	if (msg_type == GTP_ECHO_RSP) {
639 		len_pkt = sizeof(struct gtp1u_packet);
640 		hdr->length = htons(len_pkt - len_hdr);
641 	} else {
642 		/* GTP_ECHO_REQ does not carry GTP Information Element,
643 		 * the why gtp1_header_long is used here.
644 		 */
645 		len_pkt = sizeof(struct gtp1_header_long);
646 		hdr->length = htons(len_pkt - len_hdr);
647 	}
648 }
649 
650 static int gtp1u_send_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
651 {
652 	struct gtp1_header_long *gtp1u;
653 	struct gtp1u_packet *gtp_pkt;
654 	struct rtable *rt;
655 	struct flowi4 fl4;
656 	struct iphdr *iph;
657 
658 	gtp1u = (struct gtp1_header_long *)(skb->data + sizeof(struct udphdr));
659 
660 	/* 3GPP TS 29.281 5.1 - For the Echo Request, Echo Response,
661 	 * Error Indication and Supported Extension Headers Notification
662 	 * messages, the S flag shall be set to 1 and TEID shall be set to 0.
663 	 */
664 	if (!(gtp1u->flags & GTP1_F_SEQ) || gtp1u->tid)
665 		return -1;
666 
667 	/* pull GTP and UDP headers */
668 	skb_pull_data(skb,
669 		      sizeof(struct gtp1_header_long) + sizeof(struct udphdr));
670 
671 	gtp_pkt = skb_push(skb, sizeof(struct gtp1u_packet));
672 	memset(gtp_pkt, 0, sizeof(struct gtp1u_packet));
673 
674 	gtp1u_build_echo_msg(&gtp_pkt->gtp1u_h, GTP_ECHO_RSP);
675 
676 	/* 3GPP TS 29.281 7.7.2 - The Restart Counter value in the
677 	 * Recovery information element shall not be used, i.e. it shall
678 	 * be set to zero by the sender and shall be ignored by the receiver.
679 	 * The Recovery information element is mandatory due to backwards
680 	 * compatibility reasons.
681 	 */
682 	gtp_pkt->ie.tag = GTPIE_RECOVERY;
683 	gtp_pkt->ie.val = 0;
684 
685 	iph = ip_hdr(skb);
686 
687 	/* find route to the sender,
688 	 * src address becomes dst address and vice versa.
689 	 */
690 	rt = ip4_route_output_gtp(&fl4, gtp->sk1u, iph->saddr, iph->daddr);
691 	if (IS_ERR(rt)) {
692 		netdev_dbg(gtp->dev, "no route for echo response from %pI4\n",
693 			   &iph->saddr);
694 		return -1;
695 	}
696 
697 	udp_tunnel_xmit_skb(rt, gtp->sk1u, skb,
698 			    fl4.saddr, fl4.daddr,
699 			    iph->tos,
700 			    ip4_dst_hoplimit(&rt->dst),
701 			    0,
702 			    htons(GTP1U_PORT), htons(GTP1U_PORT),
703 			    !net_eq(sock_net(gtp->sk1u),
704 				    dev_net(gtp->dev)),
705 			    false);
706 	return 0;
707 }
708 
709 static int gtp1u_handle_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
710 {
711 	struct gtp1_header_long *gtp1u;
712 	struct echo_info echo;
713 	struct sk_buff *msg;
714 	struct iphdr *iph;
715 	int ret;
716 
717 	gtp1u = (struct gtp1_header_long *)(skb->data + sizeof(struct udphdr));
718 
719 	/* 3GPP TS 29.281 5.1 - For the Echo Request, Echo Response,
720 	 * Error Indication and Supported Extension Headers Notification
721 	 * messages, the S flag shall be set to 1 and TEID shall be set to 0.
722 	 */
723 	if (!(gtp1u->flags & GTP1_F_SEQ) || gtp1u->tid)
724 		return -1;
725 
726 	iph = ip_hdr(skb);
727 	echo.ms.addr.s_addr = iph->daddr;
728 	echo.peer.addr.s_addr = iph->saddr;
729 	echo.gtp_version = GTP_V1;
730 
731 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
732 	if (!msg)
733 		return -ENOMEM;
734 
735 	ret = gtp_genl_fill_echo(msg, 0, 0, 0, GTP_CMD_ECHOREQ, echo);
736 	if (ret < 0) {
737 		nlmsg_free(msg);
738 		return ret;
739 	}
740 
741 	return genlmsg_multicast_netns(&gtp_genl_family, dev_net(gtp->dev),
742 				       msg, 0, GTP_GENL_MCGRP, GFP_ATOMIC);
743 }
744 
745 static int gtp_parse_exthdrs(struct sk_buff *skb, unsigned int *hdrlen)
746 {
747 	struct gtp_ext_hdr *gtp_exthdr, _gtp_exthdr;
748 	unsigned int offset = *hdrlen;
749 	__u8 *next_type, _next_type;
750 
751 	/* From 29.060: "The Extension Header Length field specifies the length
752 	 * of the particular Extension header in 4 octets units."
753 	 *
754 	 * This length field includes length field size itself (1 byte),
755 	 * payload (variable length) and next type (1 byte). The extension
756 	 * header is aligned to to 4 bytes.
757 	 */
758 
759 	do {
760 		gtp_exthdr = skb_header_pointer(skb, offset, sizeof(*gtp_exthdr),
761 						&_gtp_exthdr);
762 		if (!gtp_exthdr || !gtp_exthdr->len)
763 			return -1;
764 
765 		offset += gtp_exthdr->len * 4;
766 
767 		/* From 29.060: "If no such Header follows, then the value of
768 		 * the Next Extension Header Type shall be 0."
769 		 */
770 		next_type = skb_header_pointer(skb, offset - 1,
771 					       sizeof(_next_type), &_next_type);
772 		if (!next_type)
773 			return -1;
774 
775 	} while (*next_type != 0);
776 
777 	*hdrlen = offset;
778 
779 	return 0;
780 }
781 
782 static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
783 {
784 	unsigned int hdrlen = sizeof(struct udphdr) +
785 			      sizeof(struct gtp1_header);
786 	struct gtp1_header *gtp1;
787 	struct pdp_ctx *pctx;
788 	__u16 inner_proto;
789 
790 	if (!pskb_may_pull(skb, hdrlen))
791 		return -1;
792 
793 	gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
794 
795 	if ((gtp1->flags >> 5) != GTP_V1)
796 		return 1;
797 
798 	/* If the sockets were created in kernel, it means that
799 	 * there is no daemon running in userspace which would
800 	 * handle echo request.
801 	 */
802 	if (gtp1->type == GTP_ECHO_REQ && gtp->sk_created)
803 		return gtp1u_send_echo_resp(gtp, skb);
804 
805 	if (gtp1->type == GTP_ECHO_RSP && gtp->sk_created)
806 		return gtp1u_handle_echo_resp(gtp, skb);
807 
808 	if (gtp1->type != GTP_TPDU)
809 		return 1;
810 
811 	/* From 29.060: "This field shall be present if and only if any one or
812 	 * more of the S, PN and E flags are set.".
813 	 *
814 	 * If any of the bit is set, then the remaining ones also have to be
815 	 * set.
816 	 */
817 	if (gtp1->flags & GTP1_F_MASK)
818 		hdrlen += 4;
819 
820 	/* Make sure the header is larger enough, including extensions. */
821 	if (!pskb_may_pull(skb, hdrlen))
822 		return -1;
823 
824 	if (gtp_inner_proto(skb, hdrlen, &inner_proto) < 0) {
825 		netdev_dbg(gtp->dev, "GTP packet does not encapsulate an IP packet\n");
826 		return -1;
827 	}
828 
829 	gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
830 
831 	pctx = gtp1_pdp_find(gtp, ntohl(gtp1->tid),
832 			     gtp_proto_to_family(inner_proto));
833 	if (!pctx) {
834 		netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
835 		return 1;
836 	}
837 
838 	if (gtp1->flags & GTP1_F_EXTHDR &&
839 	    gtp_parse_exthdrs(skb, &hdrlen) < 0)
840 		return -1;
841 
842 	return gtp_rx(pctx, skb, hdrlen, gtp->role, inner_proto);
843 }
844 
845 static void __gtp_encap_destroy(struct sock *sk)
846 {
847 	struct gtp_dev *gtp;
848 
849 	lock_sock(sk);
850 	gtp = sk->sk_user_data;
851 	if (gtp) {
852 		if (gtp->sk0 == sk)
853 			gtp->sk0 = NULL;
854 		else
855 			gtp->sk1u = NULL;
856 		WRITE_ONCE(udp_sk(sk)->encap_type, 0);
857 		rcu_assign_sk_user_data(sk, NULL);
858 		release_sock(sk);
859 		sock_put(sk);
860 		return;
861 	}
862 	release_sock(sk);
863 }
864 
865 static void gtp_encap_destroy(struct sock *sk)
866 {
867 	rtnl_lock();
868 	__gtp_encap_destroy(sk);
869 	rtnl_unlock();
870 }
871 
872 static void gtp_encap_disable_sock(struct sock *sk)
873 {
874 	if (!sk)
875 		return;
876 
877 	__gtp_encap_destroy(sk);
878 }
879 
880 static void gtp_encap_disable(struct gtp_dev *gtp)
881 {
882 	if (gtp->sk_created) {
883 		udp_tunnel_sock_release(gtp->sk0->sk_socket);
884 		udp_tunnel_sock_release(gtp->sk1u->sk_socket);
885 		gtp->sk_created = false;
886 		gtp->sk0 = NULL;
887 		gtp->sk1u = NULL;
888 	} else {
889 		gtp_encap_disable_sock(gtp->sk0);
890 		gtp_encap_disable_sock(gtp->sk1u);
891 	}
892 }
893 
894 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
895  * Return codes: 0: success, <0: error, >0: pass up to userspace UDP socket.
896  */
897 static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb)
898 {
899 	struct gtp_dev *gtp;
900 	int ret = 0;
901 
902 	gtp = rcu_dereference_sk_user_data(sk);
903 	if (!gtp)
904 		return 1;
905 
906 	netdev_dbg(gtp->dev, "encap_recv sk=%p\n", sk);
907 
908 	switch (READ_ONCE(udp_sk(sk)->encap_type)) {
909 	case UDP_ENCAP_GTP0:
910 		netdev_dbg(gtp->dev, "received GTP0 packet\n");
911 		ret = gtp0_udp_encap_recv(gtp, skb);
912 		break;
913 	case UDP_ENCAP_GTP1U:
914 		netdev_dbg(gtp->dev, "received GTP1U packet\n");
915 		ret = gtp1u_udp_encap_recv(gtp, skb);
916 		break;
917 	default:
918 		ret = -1; /* Shouldn't happen. */
919 	}
920 
921 	switch (ret) {
922 	case 1:
923 		netdev_dbg(gtp->dev, "pass up to the process\n");
924 		break;
925 	case 0:
926 		break;
927 	case -1:
928 		netdev_dbg(gtp->dev, "GTP packet has been dropped\n");
929 		kfree_skb(skb);
930 		ret = 0;
931 		break;
932 	}
933 
934 	return ret;
935 }
936 
937 static void gtp_dev_uninit(struct net_device *dev)
938 {
939 	struct gtp_dev *gtp = netdev_priv(dev);
940 
941 	gtp_encap_disable(gtp);
942 }
943 
944 static inline void gtp0_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
945 {
946 	int payload_len = skb->len;
947 	struct gtp0_header *gtp0;
948 
949 	gtp0 = skb_push(skb, sizeof(*gtp0));
950 
951 	gtp0->flags	= 0x1e; /* v0, GTP-non-prime. */
952 	gtp0->type	= GTP_TPDU;
953 	gtp0->length	= htons(payload_len);
954 	gtp0->seq	= htons((atomic_inc_return(&pctx->tx_seq) - 1) % 0xffff);
955 	gtp0->flow	= htons(pctx->u.v0.flow);
956 	gtp0->number	= 0xff;
957 	gtp0->spare[0]	= gtp0->spare[1] = gtp0->spare[2] = 0xff;
958 	gtp0->tid	= cpu_to_be64(pctx->u.v0.tid);
959 }
960 
961 static inline void gtp1_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
962 {
963 	int payload_len = skb->len;
964 	struct gtp1_header *gtp1;
965 
966 	gtp1 = skb_push(skb, sizeof(*gtp1));
967 
968 	/* Bits    8  7  6  5  4  3  2	1
969 	 *	  +--+--+--+--+--+--+--+--+
970 	 *	  |version |PT| 0| E| S|PN|
971 	 *	  +--+--+--+--+--+--+--+--+
972 	 *	    0  0  1  1	1  0  0  0
973 	 */
974 	gtp1->flags	= 0x30; /* v1, GTP-non-prime. */
975 	gtp1->type	= GTP_TPDU;
976 	gtp1->length	= htons(payload_len);
977 	gtp1->tid	= htonl(pctx->u.v1.o_tei);
978 
979 	/* TODO: Support for extension header, sequence number and N-PDU.
980 	 *	 Update the length field if any of them is available.
981 	 */
982 }
983 
984 struct gtp_pktinfo {
985 	struct sock		*sk;
986 	union {
987 		struct flowi4	fl4;
988 		struct flowi6	fl6;
989 	};
990 	union {
991 		struct rtable	*rt;
992 		struct rt6_info	*rt6;
993 	};
994 	struct pdp_ctx		*pctx;
995 	struct net_device	*dev;
996 	__u8			tos;
997 	__be16			gtph_port;
998 };
999 
1000 static void gtp_push_header(struct sk_buff *skb, struct gtp_pktinfo *pktinfo)
1001 {
1002 	switch (pktinfo->pctx->gtp_version) {
1003 	case GTP_V0:
1004 		pktinfo->gtph_port = htons(GTP0_PORT);
1005 		gtp0_push_header(skb, pktinfo->pctx);
1006 		break;
1007 	case GTP_V1:
1008 		pktinfo->gtph_port = htons(GTP1U_PORT);
1009 		gtp1_push_header(skb, pktinfo->pctx);
1010 		break;
1011 	}
1012 }
1013 
1014 static inline void gtp_set_pktinfo_ipv4(struct gtp_pktinfo *pktinfo,
1015 					struct sock *sk, __u8 tos,
1016 					struct pdp_ctx *pctx, struct rtable *rt,
1017 					struct flowi4 *fl4,
1018 					struct net_device *dev)
1019 {
1020 	pktinfo->sk	= sk;
1021 	pktinfo->tos	= tos;
1022 	pktinfo->pctx	= pctx;
1023 	pktinfo->rt	= rt;
1024 	pktinfo->fl4	= *fl4;
1025 	pktinfo->dev	= dev;
1026 }
1027 
1028 static void gtp_set_pktinfo_ipv6(struct gtp_pktinfo *pktinfo,
1029 				 struct sock *sk, __u8 tos,
1030 				 struct pdp_ctx *pctx, struct rt6_info *rt6,
1031 				 struct flowi6 *fl6,
1032 				 struct net_device *dev)
1033 {
1034 	pktinfo->sk	= sk;
1035 	pktinfo->tos	= tos;
1036 	pktinfo->pctx	= pctx;
1037 	pktinfo->rt6	= rt6;
1038 	pktinfo->fl6	= *fl6;
1039 	pktinfo->dev	= dev;
1040 }
1041 
1042 static int gtp_build_skb_outer_ip4(struct sk_buff *skb, struct net_device *dev,
1043 				   struct gtp_pktinfo *pktinfo,
1044 				   struct pdp_ctx *pctx, __u8 tos,
1045 				   __be16 frag_off)
1046 {
1047 	struct rtable *rt;
1048 	struct flowi4 fl4;
1049 	__be16 df;
1050 	int mtu;
1051 
1052 	rt = ip4_route_output_gtp(&fl4, pctx->sk, pctx->peer.addr.s_addr,
1053 				  inet_sk(pctx->sk)->inet_saddr);
1054 	if (IS_ERR(rt)) {
1055 		netdev_dbg(dev, "no route to SSGN %pI4\n",
1056 			   &pctx->peer.addr.s_addr);
1057 		dev->stats.tx_carrier_errors++;
1058 		goto err;
1059 	}
1060 
1061 	if (rt->dst.dev == dev) {
1062 		netdev_dbg(dev, "circular route to SSGN %pI4\n",
1063 			   &pctx->peer.addr.s_addr);
1064 		dev->stats.collisions++;
1065 		goto err_rt;
1066 	}
1067 
1068 	/* This is similar to tnl_update_pmtu(). */
1069 	df = frag_off;
1070 	if (df) {
1071 		mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
1072 			sizeof(struct iphdr) - sizeof(struct udphdr);
1073 		switch (pctx->gtp_version) {
1074 		case GTP_V0:
1075 			mtu -= sizeof(struct gtp0_header);
1076 			break;
1077 		case GTP_V1:
1078 			mtu -= sizeof(struct gtp1_header);
1079 			break;
1080 		}
1081 	} else {
1082 		mtu = dst_mtu(&rt->dst);
1083 	}
1084 
1085 	skb_dst_update_pmtu_no_confirm(skb, mtu);
1086 
1087 	if (frag_off & htons(IP_DF) &&
1088 	    ((!skb_is_gso(skb) && skb->len > mtu) ||
1089 	     (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu)))) {
1090 		netdev_dbg(dev, "packet too big, fragmentation needed\n");
1091 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
1092 			      htonl(mtu));
1093 		goto err_rt;
1094 	}
1095 
1096 	gtp_set_pktinfo_ipv4(pktinfo, pctx->sk, tos, pctx, rt, &fl4, dev);
1097 	gtp_push_header(skb, pktinfo);
1098 
1099 	return 0;
1100 err_rt:
1101 	ip_rt_put(rt);
1102 err:
1103 	return -EBADMSG;
1104 }
1105 
1106 static int gtp_build_skb_outer_ip6(struct net *net, struct sk_buff *skb,
1107 				   struct net_device *dev,
1108 				   struct gtp_pktinfo *pktinfo,
1109 				   struct pdp_ctx *pctx, __u8 tos)
1110 {
1111 	struct dst_entry *dst;
1112 	struct rt6_info *rt;
1113 	struct flowi6 fl6;
1114 	int mtu;
1115 
1116 	rt = ip6_route_output_gtp(net, &fl6, pctx->sk, &pctx->peer.addr6,
1117 				  &inet6_sk(pctx->sk)->saddr);
1118 	if (IS_ERR(rt)) {
1119 		netdev_dbg(dev, "no route to SSGN %pI6\n",
1120 			   &pctx->peer.addr6);
1121 		dev->stats.tx_carrier_errors++;
1122 		goto err;
1123 	}
1124 	dst = &rt->dst;
1125 
1126 	if (rt->dst.dev == dev) {
1127 		netdev_dbg(dev, "circular route to SSGN %pI6\n",
1128 			   &pctx->peer.addr6);
1129 		dev->stats.collisions++;
1130 		goto err_rt;
1131 	}
1132 
1133 	mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
1134 		sizeof(struct ipv6hdr) - sizeof(struct udphdr);
1135 	switch (pctx->gtp_version) {
1136 	case GTP_V0:
1137 		mtu -= sizeof(struct gtp0_header);
1138 		break;
1139 	case GTP_V1:
1140 		mtu -= sizeof(struct gtp1_header);
1141 		break;
1142 	}
1143 
1144 	skb_dst_update_pmtu_no_confirm(skb, mtu);
1145 
1146 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
1147 	    (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))) {
1148 		netdev_dbg(dev, "packet too big, fragmentation needed\n");
1149 		icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1150 		goto err_rt;
1151 	}
1152 
1153 	gtp_set_pktinfo_ipv6(pktinfo, pctx->sk, tos, pctx, rt, &fl6, dev);
1154 	gtp_push_header(skb, pktinfo);
1155 
1156 	return 0;
1157 err_rt:
1158 	dst_release(dst);
1159 err:
1160 	return -EBADMSG;
1161 }
1162 
1163 static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev,
1164 			     struct gtp_pktinfo *pktinfo)
1165 {
1166 	struct gtp_dev *gtp = netdev_priv(dev);
1167 	struct net *net = gtp->net;
1168 	struct pdp_ctx *pctx;
1169 	struct iphdr *iph;
1170 	int ret;
1171 
1172 	/* Read the IP destination address and resolve the PDP context.
1173 	 * Prepend PDP header with TEI/TID from PDP ctx.
1174 	 */
1175 	iph = ip_hdr(skb);
1176 	if (gtp->role == GTP_ROLE_SGSN)
1177 		pctx = ipv4_pdp_find(gtp, iph->saddr);
1178 	else
1179 		pctx = ipv4_pdp_find(gtp, iph->daddr);
1180 
1181 	if (!pctx) {
1182 		netdev_dbg(dev, "no PDP ctx found for %pI4, skip\n",
1183 			   &iph->daddr);
1184 		return -ENOENT;
1185 	}
1186 	netdev_dbg(dev, "found PDP context %p\n", pctx);
1187 
1188 	switch (pctx->sk->sk_family) {
1189 	case AF_INET:
1190 		ret = gtp_build_skb_outer_ip4(skb, dev, pktinfo, pctx,
1191 					      iph->tos, iph->frag_off);
1192 		break;
1193 	case AF_INET6:
1194 		ret = gtp_build_skb_outer_ip6(net, skb, dev, pktinfo, pctx,
1195 					      iph->tos);
1196 		break;
1197 	default:
1198 		ret = -1;
1199 		WARN_ON_ONCE(1);
1200 		break;
1201 	}
1202 
1203 	if (ret < 0)
1204 		return ret;
1205 
1206 	netdev_dbg(dev, "gtp -> IP src: %pI4 dst: %pI4\n",
1207 		   &iph->saddr, &iph->daddr);
1208 
1209 	return 0;
1210 }
1211 
1212 static int gtp_build_skb_ip6(struct sk_buff *skb, struct net_device *dev,
1213 			     struct gtp_pktinfo *pktinfo)
1214 {
1215 	struct gtp_dev *gtp = netdev_priv(dev);
1216 	struct net *net = gtp->net;
1217 	struct pdp_ctx *pctx;
1218 	struct ipv6hdr *ip6h;
1219 	__u8 tos;
1220 	int ret;
1221 
1222 	/* Read the IP destination address and resolve the PDP context.
1223 	 * Prepend PDP header with TEI/TID from PDP ctx.
1224 	 */
1225 	ip6h = ipv6_hdr(skb);
1226 	if (gtp->role == GTP_ROLE_SGSN)
1227 		pctx = ipv6_pdp_find(gtp, &ip6h->saddr);
1228 	else
1229 		pctx = ipv6_pdp_find(gtp, &ip6h->daddr);
1230 
1231 	if (!pctx) {
1232 		netdev_dbg(dev, "no PDP ctx found for %pI6, skip\n",
1233 			   &ip6h->daddr);
1234 		return -ENOENT;
1235 	}
1236 	netdev_dbg(dev, "found PDP context %p\n", pctx);
1237 
1238 	tos = ipv6_get_dsfield(ip6h);
1239 
1240 	switch (pctx->sk->sk_family) {
1241 	case AF_INET:
1242 		ret = gtp_build_skb_outer_ip4(skb, dev, pktinfo, pctx, tos, 0);
1243 		break;
1244 	case AF_INET6:
1245 		ret = gtp_build_skb_outer_ip6(net, skb, dev, pktinfo, pctx, tos);
1246 		break;
1247 	default:
1248 		ret = -1;
1249 		WARN_ON_ONCE(1);
1250 		break;
1251 	}
1252 
1253 	if (ret < 0)
1254 		return ret;
1255 
1256 	netdev_dbg(dev, "gtp -> IP src: %pI6 dst: %pI6\n",
1257 		   &ip6h->saddr, &ip6h->daddr);
1258 
1259 	return 0;
1260 }
1261 
1262 static netdev_tx_t gtp_dev_xmit(struct sk_buff *skb, struct net_device *dev)
1263 {
1264 	unsigned int proto = ntohs(skb->protocol);
1265 	struct gtp_pktinfo pktinfo;
1266 	int err;
1267 
1268 	/* Ensure there is sufficient headroom. */
1269 	if (skb_cow_head(skb, dev->needed_headroom))
1270 		goto tx_err;
1271 
1272 	skb_reset_inner_headers(skb);
1273 
1274 	/* PDP context lookups in gtp_build_skb_*() need rcu read-side lock. */
1275 	rcu_read_lock();
1276 	switch (proto) {
1277 	case ETH_P_IP:
1278 		err = gtp_build_skb_ip4(skb, dev, &pktinfo);
1279 		break;
1280 	case ETH_P_IPV6:
1281 		err = gtp_build_skb_ip6(skb, dev, &pktinfo);
1282 		break;
1283 	default:
1284 		err = -EOPNOTSUPP;
1285 		break;
1286 	}
1287 	rcu_read_unlock();
1288 
1289 	if (err < 0)
1290 		goto tx_err;
1291 
1292 	switch (pktinfo.pctx->sk->sk_family) {
1293 	case AF_INET:
1294 		udp_tunnel_xmit_skb(pktinfo.rt, pktinfo.sk, skb,
1295 				    pktinfo.fl4.saddr, pktinfo.fl4.daddr,
1296 				    pktinfo.tos,
1297 				    ip4_dst_hoplimit(&pktinfo.rt->dst),
1298 				    0,
1299 				    pktinfo.gtph_port, pktinfo.gtph_port,
1300 				    !net_eq(sock_net(pktinfo.pctx->sk),
1301 					    dev_net(dev)),
1302 				    false);
1303 		break;
1304 	case AF_INET6:
1305 #if IS_ENABLED(CONFIG_IPV6)
1306 		udp_tunnel6_xmit_skb(&pktinfo.rt6->dst, pktinfo.sk, skb, dev,
1307 				     &pktinfo.fl6.saddr, &pktinfo.fl6.daddr,
1308 				     pktinfo.tos,
1309 				     ip6_dst_hoplimit(&pktinfo.rt->dst),
1310 				     0,
1311 				     pktinfo.gtph_port, pktinfo.gtph_port,
1312 				     false);
1313 #else
1314 		goto tx_err;
1315 #endif
1316 		break;
1317 	}
1318 
1319 	return NETDEV_TX_OK;
1320 tx_err:
1321 	dev->stats.tx_errors++;
1322 	dev_kfree_skb(skb);
1323 	return NETDEV_TX_OK;
1324 }
1325 
1326 static const struct net_device_ops gtp_netdev_ops = {
1327 	.ndo_uninit		= gtp_dev_uninit,
1328 	.ndo_start_xmit		= gtp_dev_xmit,
1329 };
1330 
1331 static const struct device_type gtp_type = {
1332 	.name = "gtp",
1333 };
1334 
1335 #define GTP_TH_MAXLEN	(sizeof(struct udphdr) + sizeof(struct gtp0_header))
1336 #define GTP_IPV4_MAXLEN	(sizeof(struct iphdr) + GTP_TH_MAXLEN)
1337 
1338 static void gtp_link_setup(struct net_device *dev)
1339 {
1340 	struct gtp_dev *gtp = netdev_priv(dev);
1341 
1342 	dev->netdev_ops		= &gtp_netdev_ops;
1343 	dev->needs_free_netdev	= true;
1344 	SET_NETDEV_DEVTYPE(dev, &gtp_type);
1345 
1346 	dev->hard_header_len = 0;
1347 	dev->addr_len = 0;
1348 	dev->mtu = ETH_DATA_LEN - GTP_IPV4_MAXLEN;
1349 
1350 	/* Zero header length. */
1351 	dev->type = ARPHRD_NONE;
1352 	dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1353 
1354 	dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1355 	dev->priv_flags	|= IFF_NO_QUEUE;
1356 	dev->features	|= NETIF_F_LLTX;
1357 	netif_keep_dst(dev);
1358 
1359 	dev->needed_headroom	= LL_MAX_HEADER + GTP_IPV4_MAXLEN;
1360 	gtp->dev = dev;
1361 }
1362 
1363 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize);
1364 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]);
1365 
1366 static void gtp_destructor(struct net_device *dev)
1367 {
1368 	struct gtp_dev *gtp = netdev_priv(dev);
1369 
1370 	kfree(gtp->addr_hash);
1371 	kfree(gtp->tid_hash);
1372 }
1373 
1374 static int gtp_sock_udp_config(struct udp_port_cfg *udp_conf,
1375 			       const struct nlattr *nla, int family)
1376 {
1377 	udp_conf->family = family;
1378 
1379 	switch (udp_conf->family) {
1380 	case AF_INET:
1381 		udp_conf->local_ip.s_addr = nla_get_be32(nla);
1382 		break;
1383 #if IS_ENABLED(CONFIG_IPV6)
1384 	case AF_INET6:
1385 		udp_conf->local_ip6 = nla_get_in6_addr(nla);
1386 		break;
1387 #endif
1388 	default:
1389 		return -EOPNOTSUPP;
1390 	}
1391 
1392 	return 0;
1393 }
1394 
1395 static struct sock *gtp_create_sock(int type, struct gtp_dev *gtp,
1396 				    const struct nlattr *nla, int family)
1397 {
1398 	struct udp_tunnel_sock_cfg tuncfg = {};
1399 	struct udp_port_cfg udp_conf = {};
1400 	struct net *net = gtp->net;
1401 	struct socket *sock;
1402 	int err;
1403 
1404 	if (nla) {
1405 		err = gtp_sock_udp_config(&udp_conf, nla, family);
1406 		if (err < 0)
1407 			return ERR_PTR(err);
1408 	} else {
1409 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
1410 		udp_conf.family = AF_INET;
1411 	}
1412 
1413 	if (type == UDP_ENCAP_GTP0)
1414 		udp_conf.local_udp_port = htons(GTP0_PORT);
1415 	else if (type == UDP_ENCAP_GTP1U)
1416 		udp_conf.local_udp_port = htons(GTP1U_PORT);
1417 	else
1418 		return ERR_PTR(-EINVAL);
1419 
1420 	err = udp_sock_create(net, &udp_conf, &sock);
1421 	if (err)
1422 		return ERR_PTR(err);
1423 
1424 	tuncfg.sk_user_data = gtp;
1425 	tuncfg.encap_type = type;
1426 	tuncfg.encap_rcv = gtp_encap_recv;
1427 	tuncfg.encap_destroy = NULL;
1428 
1429 	setup_udp_tunnel_sock(net, sock, &tuncfg);
1430 
1431 	return sock->sk;
1432 }
1433 
1434 static int gtp_create_sockets(struct gtp_dev *gtp, const struct nlattr *nla,
1435 			      int family)
1436 {
1437 	struct sock *sk1u;
1438 	struct sock *sk0;
1439 
1440 	sk0 = gtp_create_sock(UDP_ENCAP_GTP0, gtp, nla, family);
1441 	if (IS_ERR(sk0))
1442 		return PTR_ERR(sk0);
1443 
1444 	sk1u = gtp_create_sock(UDP_ENCAP_GTP1U, gtp, nla, family);
1445 	if (IS_ERR(sk1u)) {
1446 		udp_tunnel_sock_release(sk0->sk_socket);
1447 		return PTR_ERR(sk1u);
1448 	}
1449 
1450 	gtp->sk_created = true;
1451 	gtp->sk0 = sk0;
1452 	gtp->sk1u = sk1u;
1453 
1454 	return 0;
1455 }
1456 
1457 #define GTP_TH_MAXLEN	(sizeof(struct udphdr) + sizeof(struct gtp0_header))
1458 #define GTP_IPV6_MAXLEN	(sizeof(struct ipv6hdr) + GTP_TH_MAXLEN)
1459 
1460 static int gtp_newlink(struct net *src_net, struct net_device *dev,
1461 		       struct nlattr *tb[], struct nlattr *data[],
1462 		       struct netlink_ext_ack *extack)
1463 {
1464 	unsigned int role = GTP_ROLE_GGSN;
1465 	struct gtp_dev *gtp;
1466 	struct gtp_net *gn;
1467 	int hashsize, err;
1468 
1469 #if !IS_ENABLED(CONFIG_IPV6)
1470 	if (data[IFLA_GTP_LOCAL6])
1471 		return -EAFNOSUPPORT;
1472 #endif
1473 
1474 	gtp = netdev_priv(dev);
1475 
1476 	if (!data[IFLA_GTP_PDP_HASHSIZE]) {
1477 		hashsize = 1024;
1478 	} else {
1479 		hashsize = nla_get_u32(data[IFLA_GTP_PDP_HASHSIZE]);
1480 		if (!hashsize)
1481 			hashsize = 1024;
1482 	}
1483 
1484 	if (data[IFLA_GTP_ROLE]) {
1485 		role = nla_get_u32(data[IFLA_GTP_ROLE]);
1486 		if (role > GTP_ROLE_SGSN)
1487 			return -EINVAL;
1488 	}
1489 	gtp->role = role;
1490 
1491 	if (!data[IFLA_GTP_RESTART_COUNT])
1492 		gtp->restart_count = 0;
1493 	else
1494 		gtp->restart_count = nla_get_u8(data[IFLA_GTP_RESTART_COUNT]);
1495 
1496 	gtp->net = src_net;
1497 
1498 	err = gtp_hashtable_new(gtp, hashsize);
1499 	if (err < 0)
1500 		return err;
1501 
1502 	if (data[IFLA_GTP_CREATE_SOCKETS]) {
1503 		if (data[IFLA_GTP_LOCAL6])
1504 			err = gtp_create_sockets(gtp, data[IFLA_GTP_LOCAL6], AF_INET6);
1505 		else
1506 			err = gtp_create_sockets(gtp, data[IFLA_GTP_LOCAL], AF_INET);
1507 	} else {
1508 		err = gtp_encap_enable(gtp, data);
1509 	}
1510 
1511 	if (err < 0)
1512 		goto out_hashtable;
1513 
1514 	if ((gtp->sk0 && gtp->sk0->sk_family == AF_INET6) ||
1515 	    (gtp->sk1u && gtp->sk1u->sk_family == AF_INET6)) {
1516 		dev->mtu = ETH_DATA_LEN - GTP_IPV6_MAXLEN;
1517 		dev->needed_headroom = LL_MAX_HEADER + GTP_IPV6_MAXLEN;
1518 	}
1519 
1520 	err = register_netdevice(dev);
1521 	if (err < 0) {
1522 		netdev_dbg(dev, "failed to register new netdev %d\n", err);
1523 		goto out_encap;
1524 	}
1525 
1526 	gn = net_generic(dev_net(dev), gtp_net_id);
1527 	list_add_rcu(&gtp->list, &gn->gtp_dev_list);
1528 	dev->priv_destructor = gtp_destructor;
1529 
1530 	netdev_dbg(dev, "registered new GTP interface\n");
1531 
1532 	return 0;
1533 
1534 out_encap:
1535 	gtp_encap_disable(gtp);
1536 out_hashtable:
1537 	kfree(gtp->addr_hash);
1538 	kfree(gtp->tid_hash);
1539 	return err;
1540 }
1541 
1542 static void gtp_dellink(struct net_device *dev, struct list_head *head)
1543 {
1544 	struct gtp_dev *gtp = netdev_priv(dev);
1545 	struct hlist_node *next;
1546 	struct pdp_ctx *pctx;
1547 	int i;
1548 
1549 	for (i = 0; i < gtp->hash_size; i++)
1550 		hlist_for_each_entry_safe(pctx, next, &gtp->tid_hash[i], hlist_tid)
1551 			pdp_context_delete(pctx);
1552 
1553 	list_del_rcu(&gtp->list);
1554 	unregister_netdevice_queue(dev, head);
1555 }
1556 
1557 static const struct nla_policy gtp_policy[IFLA_GTP_MAX + 1] = {
1558 	[IFLA_GTP_FD0]			= { .type = NLA_U32 },
1559 	[IFLA_GTP_FD1]			= { .type = NLA_U32 },
1560 	[IFLA_GTP_PDP_HASHSIZE]		= { .type = NLA_U32 },
1561 	[IFLA_GTP_ROLE]			= { .type = NLA_U32 },
1562 	[IFLA_GTP_CREATE_SOCKETS]	= { .type = NLA_U8 },
1563 	[IFLA_GTP_RESTART_COUNT]	= { .type = NLA_U8 },
1564 	[IFLA_GTP_LOCAL]		= { .type = NLA_U32 },
1565 	[IFLA_GTP_LOCAL6]		= { .len = sizeof(struct in6_addr) },
1566 };
1567 
1568 static int gtp_validate(struct nlattr *tb[], struct nlattr *data[],
1569 			struct netlink_ext_ack *extack)
1570 {
1571 	if (!data)
1572 		return -EINVAL;
1573 
1574 	return 0;
1575 }
1576 
1577 static size_t gtp_get_size(const struct net_device *dev)
1578 {
1579 	return nla_total_size(sizeof(__u32)) + /* IFLA_GTP_PDP_HASHSIZE */
1580 		nla_total_size(sizeof(__u32)) + /* IFLA_GTP_ROLE */
1581 		nla_total_size(sizeof(__u8)); /* IFLA_GTP_RESTART_COUNT */
1582 }
1583 
1584 static int gtp_fill_info(struct sk_buff *skb, const struct net_device *dev)
1585 {
1586 	struct gtp_dev *gtp = netdev_priv(dev);
1587 
1588 	if (nla_put_u32(skb, IFLA_GTP_PDP_HASHSIZE, gtp->hash_size))
1589 		goto nla_put_failure;
1590 	if (nla_put_u32(skb, IFLA_GTP_ROLE, gtp->role))
1591 		goto nla_put_failure;
1592 	if (nla_put_u8(skb, IFLA_GTP_RESTART_COUNT, gtp->restart_count))
1593 		goto nla_put_failure;
1594 
1595 	return 0;
1596 
1597 nla_put_failure:
1598 	return -EMSGSIZE;
1599 }
1600 
1601 static struct rtnl_link_ops gtp_link_ops __read_mostly = {
1602 	.kind		= "gtp",
1603 	.maxtype	= IFLA_GTP_MAX,
1604 	.policy		= gtp_policy,
1605 	.priv_size	= sizeof(struct gtp_dev),
1606 	.setup		= gtp_link_setup,
1607 	.validate	= gtp_validate,
1608 	.newlink	= gtp_newlink,
1609 	.dellink	= gtp_dellink,
1610 	.get_size	= gtp_get_size,
1611 	.fill_info	= gtp_fill_info,
1612 };
1613 
1614 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize)
1615 {
1616 	int i;
1617 
1618 	gtp->addr_hash = kmalloc_array(hsize, sizeof(struct hlist_head),
1619 				       GFP_KERNEL | __GFP_NOWARN);
1620 	if (gtp->addr_hash == NULL)
1621 		return -ENOMEM;
1622 
1623 	gtp->tid_hash = kmalloc_array(hsize, sizeof(struct hlist_head),
1624 				      GFP_KERNEL | __GFP_NOWARN);
1625 	if (gtp->tid_hash == NULL)
1626 		goto err1;
1627 
1628 	gtp->hash_size = hsize;
1629 
1630 	for (i = 0; i < hsize; i++) {
1631 		INIT_HLIST_HEAD(&gtp->addr_hash[i]);
1632 		INIT_HLIST_HEAD(&gtp->tid_hash[i]);
1633 	}
1634 	return 0;
1635 err1:
1636 	kfree(gtp->addr_hash);
1637 	return -ENOMEM;
1638 }
1639 
1640 static struct sock *gtp_encap_enable_socket(int fd, int type,
1641 					    struct gtp_dev *gtp)
1642 {
1643 	struct udp_tunnel_sock_cfg tuncfg = {NULL};
1644 	struct socket *sock;
1645 	struct sock *sk;
1646 	int err;
1647 
1648 	pr_debug("enable gtp on %d, %d\n", fd, type);
1649 
1650 	sock = sockfd_lookup(fd, &err);
1651 	if (!sock) {
1652 		pr_debug("gtp socket fd=%d not found\n", fd);
1653 		return NULL;
1654 	}
1655 
1656 	sk = sock->sk;
1657 	if (sk->sk_protocol != IPPROTO_UDP ||
1658 	    sk->sk_type != SOCK_DGRAM ||
1659 	    (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) {
1660 		pr_debug("socket fd=%d not UDP\n", fd);
1661 		sk = ERR_PTR(-EINVAL);
1662 		goto out_sock;
1663 	}
1664 
1665 	if (sk->sk_family == AF_INET6 &&
1666 	    !sk->sk_ipv6only) {
1667 		sk = ERR_PTR(-EADDRNOTAVAIL);
1668 		goto out_sock;
1669 	}
1670 
1671 	lock_sock(sk);
1672 	if (sk->sk_user_data) {
1673 		sk = ERR_PTR(-EBUSY);
1674 		goto out_rel_sock;
1675 	}
1676 
1677 	sock_hold(sk);
1678 
1679 	tuncfg.sk_user_data = gtp;
1680 	tuncfg.encap_type = type;
1681 	tuncfg.encap_rcv = gtp_encap_recv;
1682 	tuncfg.encap_destroy = gtp_encap_destroy;
1683 
1684 	setup_udp_tunnel_sock(sock_net(sock->sk), sock, &tuncfg);
1685 
1686 out_rel_sock:
1687 	release_sock(sock->sk);
1688 out_sock:
1689 	sockfd_put(sock);
1690 	return sk;
1691 }
1692 
1693 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[])
1694 {
1695 	struct sock *sk1u = NULL;
1696 	struct sock *sk0 = NULL;
1697 
1698 	if (!data[IFLA_GTP_FD0] && !data[IFLA_GTP_FD1])
1699 		return -EINVAL;
1700 
1701 	if (data[IFLA_GTP_FD0]) {
1702 		u32 fd0 = nla_get_u32(data[IFLA_GTP_FD0]);
1703 
1704 		sk0 = gtp_encap_enable_socket(fd0, UDP_ENCAP_GTP0, gtp);
1705 		if (IS_ERR(sk0))
1706 			return PTR_ERR(sk0);
1707 	}
1708 
1709 	if (data[IFLA_GTP_FD1]) {
1710 		u32 fd1 = nla_get_u32(data[IFLA_GTP_FD1]);
1711 
1712 		sk1u = gtp_encap_enable_socket(fd1, UDP_ENCAP_GTP1U, gtp);
1713 		if (IS_ERR(sk1u)) {
1714 			gtp_encap_disable_sock(sk0);
1715 			return PTR_ERR(sk1u);
1716 		}
1717 	}
1718 
1719 	gtp->sk0 = sk0;
1720 	gtp->sk1u = sk1u;
1721 
1722 	if (sk0 && sk1u &&
1723 	    sk0->sk_family != sk1u->sk_family) {
1724 		gtp_encap_disable_sock(sk0);
1725 		gtp_encap_disable_sock(sk1u);
1726 		return -EINVAL;
1727 	}
1728 
1729 	return 0;
1730 }
1731 
1732 static struct gtp_dev *gtp_find_dev(struct net *src_net, struct nlattr *nla[])
1733 {
1734 	struct gtp_dev *gtp = NULL;
1735 	struct net_device *dev;
1736 	struct net *net;
1737 
1738 	/* Examine the link attributes and figure out which network namespace
1739 	 * we are talking about.
1740 	 */
1741 	if (nla[GTPA_NET_NS_FD])
1742 		net = get_net_ns_by_fd(nla_get_u32(nla[GTPA_NET_NS_FD]));
1743 	else
1744 		net = get_net(src_net);
1745 
1746 	if (IS_ERR(net))
1747 		return NULL;
1748 
1749 	/* Check if there's an existing gtpX device to configure */
1750 	dev = dev_get_by_index_rcu(net, nla_get_u32(nla[GTPA_LINK]));
1751 	if (dev && dev->netdev_ops == &gtp_netdev_ops)
1752 		gtp = netdev_priv(dev);
1753 
1754 	put_net(net);
1755 	return gtp;
1756 }
1757 
1758 static void gtp_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
1759 {
1760 	pctx->gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]);
1761 
1762 	switch (pctx->gtp_version) {
1763 	case GTP_V0:
1764 		/* According to TS 09.60, sections 7.5.1 and 7.5.2, the flow
1765 		 * label needs to be the same for uplink and downlink packets,
1766 		 * so let's annotate this.
1767 		 */
1768 		pctx->u.v0.tid = nla_get_u64(info->attrs[GTPA_TID]);
1769 		pctx->u.v0.flow = nla_get_u16(info->attrs[GTPA_FLOW]);
1770 		break;
1771 	case GTP_V1:
1772 		pctx->u.v1.i_tei = nla_get_u32(info->attrs[GTPA_I_TEI]);
1773 		pctx->u.v1.o_tei = nla_get_u32(info->attrs[GTPA_O_TEI]);
1774 		break;
1775 	default:
1776 		break;
1777 	}
1778 }
1779 
1780 static void ip_pdp_peer_fill(struct pdp_ctx *pctx, struct genl_info *info)
1781 {
1782 	if (info->attrs[GTPA_PEER_ADDRESS]) {
1783 		pctx->peer.addr.s_addr =
1784 			nla_get_be32(info->attrs[GTPA_PEER_ADDRESS]);
1785 	} else if (info->attrs[GTPA_PEER_ADDR6]) {
1786 		pctx->peer.addr6 = nla_get_in6_addr(info->attrs[GTPA_PEER_ADDR6]);
1787 	}
1788 }
1789 
1790 static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
1791 {
1792 	ip_pdp_peer_fill(pctx, info);
1793 	pctx->ms.addr.s_addr =
1794 		nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
1795 	gtp_pdp_fill(pctx, info);
1796 }
1797 
1798 static bool ipv6_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
1799 {
1800 	ip_pdp_peer_fill(pctx, info);
1801 	pctx->ms.addr6 = nla_get_in6_addr(info->attrs[GTPA_MS_ADDR6]);
1802 	if (pctx->ms.addr6.s6_addr32[2] ||
1803 	    pctx->ms.addr6.s6_addr32[3])
1804 		return false;
1805 
1806 	gtp_pdp_fill(pctx, info);
1807 
1808 	return true;
1809 }
1810 
1811 static struct pdp_ctx *gtp_pdp_add(struct gtp_dev *gtp, struct sock *sk,
1812 				   struct genl_info *info)
1813 {
1814 	struct pdp_ctx *pctx, *pctx_tid = NULL;
1815 	struct net_device *dev = gtp->dev;
1816 	u32 hash_ms, hash_tid = 0;
1817 	struct in6_addr ms_addr6;
1818 	unsigned int version;
1819 	bool found = false;
1820 	__be32 ms_addr;
1821 	int family;
1822 
1823 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
1824 
1825 	if (info->attrs[GTPA_FAMILY])
1826 		family = nla_get_u8(info->attrs[GTPA_FAMILY]);
1827 	else
1828 		family = AF_INET;
1829 
1830 #if !IS_ENABLED(CONFIG_IPV6)
1831 	if (family == AF_INET6)
1832 		return ERR_PTR(-EAFNOSUPPORT);
1833 #endif
1834 	if (!info->attrs[GTPA_PEER_ADDRESS] &&
1835 	    !info->attrs[GTPA_PEER_ADDR6])
1836 		return ERR_PTR(-EINVAL);
1837 
1838 	if ((info->attrs[GTPA_PEER_ADDRESS] &&
1839 	     sk->sk_family == AF_INET6) ||
1840 	    (info->attrs[GTPA_PEER_ADDR6] &&
1841 	     sk->sk_family == AF_INET))
1842 		return ERR_PTR(-EAFNOSUPPORT);
1843 
1844 	switch (family) {
1845 	case AF_INET:
1846 		if (!info->attrs[GTPA_MS_ADDRESS] ||
1847 		    info->attrs[GTPA_MS_ADDR6])
1848 			return ERR_PTR(-EINVAL);
1849 
1850 		ms_addr = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
1851 		hash_ms = ipv4_hashfn(ms_addr) % gtp->hash_size;
1852 		pctx = ipv4_pdp_find(gtp, ms_addr);
1853 		break;
1854 	case AF_INET6:
1855 		if (!info->attrs[GTPA_MS_ADDR6] ||
1856 		    info->attrs[GTPA_MS_ADDRESS])
1857 			return ERR_PTR(-EINVAL);
1858 
1859 		ms_addr6 = nla_get_in6_addr(info->attrs[GTPA_MS_ADDR6]);
1860 		hash_ms = ipv6_hashfn(&ms_addr6) % gtp->hash_size;
1861 		pctx = ipv6_pdp_find(gtp, &ms_addr6);
1862 		break;
1863 	default:
1864 		return ERR_PTR(-EAFNOSUPPORT);
1865 	}
1866 	if (pctx)
1867 		found = true;
1868 	if (version == GTP_V0)
1869 		pctx_tid = gtp0_pdp_find(gtp,
1870 					 nla_get_u64(info->attrs[GTPA_TID]),
1871 					 family);
1872 	else if (version == GTP_V1)
1873 		pctx_tid = gtp1_pdp_find(gtp,
1874 					 nla_get_u32(info->attrs[GTPA_I_TEI]),
1875 					 family);
1876 	if (pctx_tid)
1877 		found = true;
1878 
1879 	if (found) {
1880 		if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
1881 			return ERR_PTR(-EEXIST);
1882 		if (info->nlhdr->nlmsg_flags & NLM_F_REPLACE)
1883 			return ERR_PTR(-EOPNOTSUPP);
1884 
1885 		if (pctx && pctx_tid)
1886 			return ERR_PTR(-EEXIST);
1887 		if (!pctx)
1888 			pctx = pctx_tid;
1889 
1890 		switch (pctx->af) {
1891 		case AF_INET:
1892 			ipv4_pdp_fill(pctx, info);
1893 			break;
1894 		case AF_INET6:
1895 			if (!ipv6_pdp_fill(pctx, info))
1896 				return ERR_PTR(-EADDRNOTAVAIL);
1897 			break;
1898 		}
1899 
1900 		if (pctx->gtp_version == GTP_V0)
1901 			netdev_dbg(dev, "GTPv0-U: update tunnel id = %llx (pdp %p)\n",
1902 				   pctx->u.v0.tid, pctx);
1903 		else if (pctx->gtp_version == GTP_V1)
1904 			netdev_dbg(dev, "GTPv1-U: update tunnel id = %x/%x (pdp %p)\n",
1905 				   pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
1906 
1907 		return pctx;
1908 
1909 	}
1910 
1911 	pctx = kmalloc(sizeof(*pctx), GFP_ATOMIC);
1912 	if (pctx == NULL)
1913 		return ERR_PTR(-ENOMEM);
1914 
1915 	sock_hold(sk);
1916 	pctx->sk = sk;
1917 	pctx->dev = gtp->dev;
1918 	pctx->af = family;
1919 
1920 	switch (pctx->af) {
1921 	case AF_INET:
1922 		if (!info->attrs[GTPA_MS_ADDRESS]) {
1923 			sock_put(sk);
1924 			kfree(pctx);
1925 			return ERR_PTR(-EINVAL);
1926 		}
1927 
1928 		ipv4_pdp_fill(pctx, info);
1929 		break;
1930 	case AF_INET6:
1931 		if (!info->attrs[GTPA_MS_ADDR6]) {
1932 			sock_put(sk);
1933 			kfree(pctx);
1934 			return ERR_PTR(-EINVAL);
1935 		}
1936 
1937 		if (!ipv6_pdp_fill(pctx, info)) {
1938 			sock_put(sk);
1939 			kfree(pctx);
1940 			return ERR_PTR(-EADDRNOTAVAIL);
1941 		}
1942 		break;
1943 	}
1944 	atomic_set(&pctx->tx_seq, 0);
1945 
1946 	switch (pctx->gtp_version) {
1947 	case GTP_V0:
1948 		/* TS 09.60: "The flow label identifies unambiguously a GTP
1949 		 * flow.". We use the tid for this instead, I cannot find a
1950 		 * situation in which this doesn't unambiguosly identify the
1951 		 * PDP context.
1952 		 */
1953 		hash_tid = gtp0_hashfn(pctx->u.v0.tid) % gtp->hash_size;
1954 		break;
1955 	case GTP_V1:
1956 		hash_tid = gtp1u_hashfn(pctx->u.v1.i_tei) % gtp->hash_size;
1957 		break;
1958 	}
1959 
1960 	hlist_add_head_rcu(&pctx->hlist_addr, &gtp->addr_hash[hash_ms]);
1961 	hlist_add_head_rcu(&pctx->hlist_tid, &gtp->tid_hash[hash_tid]);
1962 
1963 	switch (pctx->gtp_version) {
1964 	case GTP_V0:
1965 		netdev_dbg(dev, "GTPv0-U: new PDP ctx id=%llx ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
1966 			   pctx->u.v0.tid, &pctx->peer.addr,
1967 			   &pctx->ms.addr, pctx);
1968 		break;
1969 	case GTP_V1:
1970 		netdev_dbg(dev, "GTPv1-U: new PDP ctx id=%x/%x ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
1971 			   pctx->u.v1.i_tei, pctx->u.v1.o_tei,
1972 			   &pctx->peer.addr, &pctx->ms.addr, pctx);
1973 		break;
1974 	}
1975 
1976 	return pctx;
1977 }
1978 
1979 static void pdp_context_free(struct rcu_head *head)
1980 {
1981 	struct pdp_ctx *pctx = container_of(head, struct pdp_ctx, rcu_head);
1982 
1983 	sock_put(pctx->sk);
1984 	kfree(pctx);
1985 }
1986 
1987 static void pdp_context_delete(struct pdp_ctx *pctx)
1988 {
1989 	hlist_del_rcu(&pctx->hlist_tid);
1990 	hlist_del_rcu(&pctx->hlist_addr);
1991 	call_rcu(&pctx->rcu_head, pdp_context_free);
1992 }
1993 
1994 static int gtp_tunnel_notify(struct pdp_ctx *pctx, u8 cmd, gfp_t allocation);
1995 
1996 static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
1997 {
1998 	unsigned int version;
1999 	struct pdp_ctx *pctx;
2000 	struct gtp_dev *gtp;
2001 	struct sock *sk;
2002 	int err;
2003 
2004 	if (!info->attrs[GTPA_VERSION] ||
2005 	    !info->attrs[GTPA_LINK])
2006 		return -EINVAL;
2007 
2008 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
2009 
2010 	switch (version) {
2011 	case GTP_V0:
2012 		if (!info->attrs[GTPA_TID] ||
2013 		    !info->attrs[GTPA_FLOW])
2014 			return -EINVAL;
2015 		break;
2016 	case GTP_V1:
2017 		if (!info->attrs[GTPA_I_TEI] ||
2018 		    !info->attrs[GTPA_O_TEI])
2019 			return -EINVAL;
2020 		break;
2021 
2022 	default:
2023 		return -EINVAL;
2024 	}
2025 
2026 	rtnl_lock();
2027 
2028 	gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
2029 	if (!gtp) {
2030 		err = -ENODEV;
2031 		goto out_unlock;
2032 	}
2033 
2034 	if (version == GTP_V0)
2035 		sk = gtp->sk0;
2036 	else if (version == GTP_V1)
2037 		sk = gtp->sk1u;
2038 	else
2039 		sk = NULL;
2040 
2041 	if (!sk) {
2042 		err = -ENODEV;
2043 		goto out_unlock;
2044 	}
2045 
2046 	pctx = gtp_pdp_add(gtp, sk, info);
2047 	if (IS_ERR(pctx)) {
2048 		err = PTR_ERR(pctx);
2049 	} else {
2050 		gtp_tunnel_notify(pctx, GTP_CMD_NEWPDP, GFP_KERNEL);
2051 		err = 0;
2052 	}
2053 
2054 out_unlock:
2055 	rtnl_unlock();
2056 	return err;
2057 }
2058 
2059 static struct pdp_ctx *gtp_find_pdp_by_link(struct net *net,
2060 					    struct nlattr *nla[])
2061 {
2062 	struct gtp_dev *gtp;
2063 	int family;
2064 
2065 	if (nla[GTPA_FAMILY])
2066 		family = nla_get_u8(nla[GTPA_FAMILY]);
2067 	else
2068 		family = AF_INET;
2069 
2070 	gtp = gtp_find_dev(net, nla);
2071 	if (!gtp)
2072 		return ERR_PTR(-ENODEV);
2073 
2074 	if (nla[GTPA_MS_ADDRESS]) {
2075 		__be32 ip = nla_get_be32(nla[GTPA_MS_ADDRESS]);
2076 
2077 		if (family != AF_INET)
2078 			return ERR_PTR(-EINVAL);
2079 
2080 		return ipv4_pdp_find(gtp, ip);
2081 	} else if (nla[GTPA_MS_ADDR6]) {
2082 		struct in6_addr addr = nla_get_in6_addr(nla[GTPA_MS_ADDR6]);
2083 
2084 		if (family != AF_INET6)
2085 			return ERR_PTR(-EINVAL);
2086 
2087 		if (addr.s6_addr32[2] ||
2088 		    addr.s6_addr32[3])
2089 			return ERR_PTR(-EADDRNOTAVAIL);
2090 
2091 		return ipv6_pdp_find(gtp, &addr);
2092 	} else if (nla[GTPA_VERSION]) {
2093 		u32 gtp_version = nla_get_u32(nla[GTPA_VERSION]);
2094 
2095 		if (gtp_version == GTP_V0 && nla[GTPA_TID]) {
2096 			return gtp0_pdp_find(gtp, nla_get_u64(nla[GTPA_TID]),
2097 					     family);
2098 		} else if (gtp_version == GTP_V1 && nla[GTPA_I_TEI]) {
2099 			return gtp1_pdp_find(gtp, nla_get_u32(nla[GTPA_I_TEI]),
2100 					     family);
2101 		}
2102 	}
2103 
2104 	return ERR_PTR(-EINVAL);
2105 }
2106 
2107 static struct pdp_ctx *gtp_find_pdp(struct net *net, struct nlattr *nla[])
2108 {
2109 	struct pdp_ctx *pctx;
2110 
2111 	if (nla[GTPA_LINK])
2112 		pctx = gtp_find_pdp_by_link(net, nla);
2113 	else
2114 		pctx = ERR_PTR(-EINVAL);
2115 
2116 	if (!pctx)
2117 		pctx = ERR_PTR(-ENOENT);
2118 
2119 	return pctx;
2120 }
2121 
2122 static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
2123 {
2124 	struct pdp_ctx *pctx;
2125 	int err = 0;
2126 
2127 	if (!info->attrs[GTPA_VERSION])
2128 		return -EINVAL;
2129 
2130 	rcu_read_lock();
2131 
2132 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
2133 	if (IS_ERR(pctx)) {
2134 		err = PTR_ERR(pctx);
2135 		goto out_unlock;
2136 	}
2137 
2138 	if (pctx->gtp_version == GTP_V0)
2139 		netdev_dbg(pctx->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n",
2140 			   pctx->u.v0.tid, pctx);
2141 	else if (pctx->gtp_version == GTP_V1)
2142 		netdev_dbg(pctx->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n",
2143 			   pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
2144 
2145 	gtp_tunnel_notify(pctx, GTP_CMD_DELPDP, GFP_ATOMIC);
2146 	pdp_context_delete(pctx);
2147 
2148 out_unlock:
2149 	rcu_read_unlock();
2150 	return err;
2151 }
2152 
2153 static int gtp_genl_fill_info(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
2154 			      int flags, u32 type, struct pdp_ctx *pctx)
2155 {
2156 	void *genlh;
2157 
2158 	genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, flags,
2159 			    type);
2160 	if (genlh == NULL)
2161 		goto nlmsg_failure;
2162 
2163 	if (nla_put_u32(skb, GTPA_VERSION, pctx->gtp_version) ||
2164 	    nla_put_u32(skb, GTPA_LINK, pctx->dev->ifindex) ||
2165 	    nla_put_u8(skb, GTPA_FAMILY, pctx->af))
2166 		goto nla_put_failure;
2167 
2168 	switch (pctx->af) {
2169 	case AF_INET:
2170 		if (nla_put_be32(skb, GTPA_MS_ADDRESS, pctx->ms.addr.s_addr))
2171 			goto nla_put_failure;
2172 		break;
2173 	case AF_INET6:
2174 		if (nla_put_in6_addr(skb, GTPA_MS_ADDR6, &pctx->ms.addr6))
2175 			goto nla_put_failure;
2176 		break;
2177 	}
2178 
2179 	switch (pctx->sk->sk_family) {
2180 	case AF_INET:
2181 		if (nla_put_be32(skb, GTPA_PEER_ADDRESS, pctx->peer.addr.s_addr))
2182 			goto nla_put_failure;
2183 		break;
2184 	case AF_INET6:
2185 		if (nla_put_in6_addr(skb, GTPA_PEER_ADDR6, &pctx->peer.addr6))
2186 			goto nla_put_failure;
2187 		break;
2188 	}
2189 
2190 	switch (pctx->gtp_version) {
2191 	case GTP_V0:
2192 		if (nla_put_u64_64bit(skb, GTPA_TID, pctx->u.v0.tid, GTPA_PAD) ||
2193 		    nla_put_u16(skb, GTPA_FLOW, pctx->u.v0.flow))
2194 			goto nla_put_failure;
2195 		break;
2196 	case GTP_V1:
2197 		if (nla_put_u32(skb, GTPA_I_TEI, pctx->u.v1.i_tei) ||
2198 		    nla_put_u32(skb, GTPA_O_TEI, pctx->u.v1.o_tei))
2199 			goto nla_put_failure;
2200 		break;
2201 	}
2202 	genlmsg_end(skb, genlh);
2203 	return 0;
2204 
2205 nlmsg_failure:
2206 nla_put_failure:
2207 	genlmsg_cancel(skb, genlh);
2208 	return -EMSGSIZE;
2209 }
2210 
2211 static int gtp_tunnel_notify(struct pdp_ctx *pctx, u8 cmd, gfp_t allocation)
2212 {
2213 	struct sk_buff *msg;
2214 	int ret;
2215 
2216 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, allocation);
2217 	if (!msg)
2218 		return -ENOMEM;
2219 
2220 	ret = gtp_genl_fill_info(msg, 0, 0, 0, cmd, pctx);
2221 	if (ret < 0) {
2222 		nlmsg_free(msg);
2223 		return ret;
2224 	}
2225 
2226 	ret = genlmsg_multicast_netns(&gtp_genl_family, dev_net(pctx->dev), msg,
2227 				      0, GTP_GENL_MCGRP, GFP_ATOMIC);
2228 	return ret;
2229 }
2230 
2231 static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info)
2232 {
2233 	struct pdp_ctx *pctx = NULL;
2234 	struct sk_buff *skb2;
2235 	int err;
2236 
2237 	if (!info->attrs[GTPA_VERSION])
2238 		return -EINVAL;
2239 
2240 	rcu_read_lock();
2241 
2242 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
2243 	if (IS_ERR(pctx)) {
2244 		err = PTR_ERR(pctx);
2245 		goto err_unlock;
2246 	}
2247 
2248 	skb2 = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
2249 	if (skb2 == NULL) {
2250 		err = -ENOMEM;
2251 		goto err_unlock;
2252 	}
2253 
2254 	err = gtp_genl_fill_info(skb2, NETLINK_CB(skb).portid, info->snd_seq,
2255 				 0, info->nlhdr->nlmsg_type, pctx);
2256 	if (err < 0)
2257 		goto err_unlock_free;
2258 
2259 	rcu_read_unlock();
2260 	return genlmsg_unicast(genl_info_net(info), skb2, info->snd_portid);
2261 
2262 err_unlock_free:
2263 	kfree_skb(skb2);
2264 err_unlock:
2265 	rcu_read_unlock();
2266 	return err;
2267 }
2268 
2269 static int gtp_genl_dump_pdp(struct sk_buff *skb,
2270 				struct netlink_callback *cb)
2271 {
2272 	struct gtp_dev *last_gtp = (struct gtp_dev *)cb->args[2], *gtp;
2273 	int i, j, bucket = cb->args[0], skip = cb->args[1];
2274 	struct net *net = sock_net(skb->sk);
2275 	struct pdp_ctx *pctx;
2276 	struct gtp_net *gn;
2277 
2278 	gn = net_generic(net, gtp_net_id);
2279 
2280 	if (cb->args[4])
2281 		return 0;
2282 
2283 	rcu_read_lock();
2284 	list_for_each_entry_rcu(gtp, &gn->gtp_dev_list, list) {
2285 		if (last_gtp && last_gtp != gtp)
2286 			continue;
2287 		else
2288 			last_gtp = NULL;
2289 
2290 		for (i = bucket; i < gtp->hash_size; i++) {
2291 			j = 0;
2292 			hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i],
2293 						 hlist_tid) {
2294 				if (j >= skip &&
2295 				    gtp_genl_fill_info(skb,
2296 					    NETLINK_CB(cb->skb).portid,
2297 					    cb->nlh->nlmsg_seq,
2298 					    NLM_F_MULTI,
2299 					    cb->nlh->nlmsg_type, pctx)) {
2300 					cb->args[0] = i;
2301 					cb->args[1] = j;
2302 					cb->args[2] = (unsigned long)gtp;
2303 					goto out;
2304 				}
2305 				j++;
2306 			}
2307 			skip = 0;
2308 		}
2309 		bucket = 0;
2310 	}
2311 	cb->args[4] = 1;
2312 out:
2313 	rcu_read_unlock();
2314 	return skb->len;
2315 }
2316 
2317 static int gtp_genl_send_echo_req(struct sk_buff *skb, struct genl_info *info)
2318 {
2319 	struct sk_buff *skb_to_send;
2320 	__be32 src_ip, dst_ip;
2321 	unsigned int version;
2322 	struct gtp_dev *gtp;
2323 	struct flowi4 fl4;
2324 	struct rtable *rt;
2325 	struct sock *sk;
2326 	__be16 port;
2327 	int len;
2328 
2329 	if (!info->attrs[GTPA_VERSION] ||
2330 	    !info->attrs[GTPA_LINK] ||
2331 	    !info->attrs[GTPA_PEER_ADDRESS] ||
2332 	    !info->attrs[GTPA_MS_ADDRESS])
2333 		return -EINVAL;
2334 
2335 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
2336 	dst_ip = nla_get_be32(info->attrs[GTPA_PEER_ADDRESS]);
2337 	src_ip = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
2338 
2339 	gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
2340 	if (!gtp)
2341 		return -ENODEV;
2342 
2343 	if (!gtp->sk_created)
2344 		return -EOPNOTSUPP;
2345 	if (!(gtp->dev->flags & IFF_UP))
2346 		return -ENETDOWN;
2347 
2348 	if (version == GTP_V0) {
2349 		struct gtp0_header *gtp0_h;
2350 
2351 		len = LL_RESERVED_SPACE(gtp->dev) + sizeof(struct gtp0_header) +
2352 			sizeof(struct iphdr) + sizeof(struct udphdr);
2353 
2354 		skb_to_send = netdev_alloc_skb_ip_align(gtp->dev, len);
2355 		if (!skb_to_send)
2356 			return -ENOMEM;
2357 
2358 		sk = gtp->sk0;
2359 		port = htons(GTP0_PORT);
2360 
2361 		gtp0_h = skb_push(skb_to_send, sizeof(struct gtp0_header));
2362 		memset(gtp0_h, 0, sizeof(struct gtp0_header));
2363 		gtp0_build_echo_msg(gtp0_h, GTP_ECHO_REQ);
2364 	} else if (version == GTP_V1) {
2365 		struct gtp1_header_long *gtp1u_h;
2366 
2367 		len = LL_RESERVED_SPACE(gtp->dev) +
2368 			sizeof(struct gtp1_header_long) +
2369 			sizeof(struct iphdr) + sizeof(struct udphdr);
2370 
2371 		skb_to_send = netdev_alloc_skb_ip_align(gtp->dev, len);
2372 		if (!skb_to_send)
2373 			return -ENOMEM;
2374 
2375 		sk = gtp->sk1u;
2376 		port = htons(GTP1U_PORT);
2377 
2378 		gtp1u_h = skb_push(skb_to_send,
2379 				   sizeof(struct gtp1_header_long));
2380 		memset(gtp1u_h, 0, sizeof(struct gtp1_header_long));
2381 		gtp1u_build_echo_msg(gtp1u_h, GTP_ECHO_REQ);
2382 	} else {
2383 		return -ENODEV;
2384 	}
2385 
2386 	rt = ip4_route_output_gtp(&fl4, sk, dst_ip, src_ip);
2387 	if (IS_ERR(rt)) {
2388 		netdev_dbg(gtp->dev, "no route for echo request to %pI4\n",
2389 			   &dst_ip);
2390 		kfree_skb(skb_to_send);
2391 		return -ENODEV;
2392 	}
2393 
2394 	udp_tunnel_xmit_skb(rt, sk, skb_to_send,
2395 			    fl4.saddr, fl4.daddr,
2396 			    fl4.flowi4_tos,
2397 			    ip4_dst_hoplimit(&rt->dst),
2398 			    0,
2399 			    port, port,
2400 			    !net_eq(sock_net(sk),
2401 				    dev_net(gtp->dev)),
2402 			    false);
2403 	return 0;
2404 }
2405 
2406 static const struct nla_policy gtp_genl_policy[GTPA_MAX + 1] = {
2407 	[GTPA_LINK]		= { .type = NLA_U32, },
2408 	[GTPA_VERSION]		= { .type = NLA_U32, },
2409 	[GTPA_TID]		= { .type = NLA_U64, },
2410 	[GTPA_PEER_ADDRESS]	= { .type = NLA_U32, },
2411 	[GTPA_MS_ADDRESS]	= { .type = NLA_U32, },
2412 	[GTPA_FLOW]		= { .type = NLA_U16, },
2413 	[GTPA_NET_NS_FD]	= { .type = NLA_U32, },
2414 	[GTPA_I_TEI]		= { .type = NLA_U32, },
2415 	[GTPA_O_TEI]		= { .type = NLA_U32, },
2416 	[GTPA_PEER_ADDR6]	= { .len = sizeof(struct in6_addr), },
2417 	[GTPA_MS_ADDR6]		= { .len = sizeof(struct in6_addr), },
2418 	[GTPA_FAMILY]		= { .type = NLA_U8, },
2419 };
2420 
2421 static const struct genl_small_ops gtp_genl_ops[] = {
2422 	{
2423 		.cmd = GTP_CMD_NEWPDP,
2424 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2425 		.doit = gtp_genl_new_pdp,
2426 		.flags = GENL_ADMIN_PERM,
2427 	},
2428 	{
2429 		.cmd = GTP_CMD_DELPDP,
2430 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2431 		.doit = gtp_genl_del_pdp,
2432 		.flags = GENL_ADMIN_PERM,
2433 	},
2434 	{
2435 		.cmd = GTP_CMD_GETPDP,
2436 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2437 		.doit = gtp_genl_get_pdp,
2438 		.dumpit = gtp_genl_dump_pdp,
2439 		.flags = GENL_ADMIN_PERM,
2440 	},
2441 	{
2442 		.cmd = GTP_CMD_ECHOREQ,
2443 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2444 		.doit = gtp_genl_send_echo_req,
2445 		.flags = GENL_ADMIN_PERM,
2446 	},
2447 };
2448 
2449 static struct genl_family gtp_genl_family __ro_after_init = {
2450 	.name		= "gtp",
2451 	.version	= 0,
2452 	.hdrsize	= 0,
2453 	.maxattr	= GTPA_MAX,
2454 	.policy = gtp_genl_policy,
2455 	.netnsok	= true,
2456 	.module		= THIS_MODULE,
2457 	.small_ops	= gtp_genl_ops,
2458 	.n_small_ops	= ARRAY_SIZE(gtp_genl_ops),
2459 	.resv_start_op	= GTP_CMD_ECHOREQ + 1,
2460 	.mcgrps		= gtp_genl_mcgrps,
2461 	.n_mcgrps	= ARRAY_SIZE(gtp_genl_mcgrps),
2462 };
2463 
2464 static int __net_init gtp_net_init(struct net *net)
2465 {
2466 	struct gtp_net *gn = net_generic(net, gtp_net_id);
2467 
2468 	INIT_LIST_HEAD(&gn->gtp_dev_list);
2469 	return 0;
2470 }
2471 
2472 static void __net_exit gtp_net_exit_batch_rtnl(struct list_head *net_list,
2473 					       struct list_head *dev_to_kill)
2474 {
2475 	struct net *net;
2476 
2477 	list_for_each_entry(net, net_list, exit_list) {
2478 		struct gtp_net *gn = net_generic(net, gtp_net_id);
2479 		struct gtp_dev *gtp;
2480 
2481 		list_for_each_entry(gtp, &gn->gtp_dev_list, list)
2482 			gtp_dellink(gtp->dev, dev_to_kill);
2483 	}
2484 }
2485 
2486 static struct pernet_operations gtp_net_ops = {
2487 	.init	= gtp_net_init,
2488 	.exit_batch_rtnl = gtp_net_exit_batch_rtnl,
2489 	.id	= &gtp_net_id,
2490 	.size	= sizeof(struct gtp_net),
2491 };
2492 
2493 static int __init gtp_init(void)
2494 {
2495 	int err;
2496 
2497 	get_random_bytes(&gtp_h_initval, sizeof(gtp_h_initval));
2498 
2499 	err = register_pernet_subsys(&gtp_net_ops);
2500 	if (err < 0)
2501 		goto error_out;
2502 
2503 	err = rtnl_link_register(&gtp_link_ops);
2504 	if (err < 0)
2505 		goto unreg_pernet_subsys;
2506 
2507 	err = genl_register_family(&gtp_genl_family);
2508 	if (err < 0)
2509 		goto unreg_rtnl_link;
2510 
2511 	pr_info("GTP module loaded (pdp ctx size %zd bytes)\n",
2512 		sizeof(struct pdp_ctx));
2513 	return 0;
2514 
2515 unreg_rtnl_link:
2516 	rtnl_link_unregister(&gtp_link_ops);
2517 unreg_pernet_subsys:
2518 	unregister_pernet_subsys(&gtp_net_ops);
2519 error_out:
2520 	pr_err("error loading GTP module loaded\n");
2521 	return err;
2522 }
2523 late_initcall(gtp_init);
2524 
2525 static void __exit gtp_fini(void)
2526 {
2527 	genl_unregister_family(&gtp_genl_family);
2528 	rtnl_link_unregister(&gtp_link_ops);
2529 	unregister_pernet_subsys(&gtp_net_ops);
2530 
2531 	pr_info("GTP module unloaded\n");
2532 }
2533 module_exit(gtp_fini);
2534 
2535 MODULE_LICENSE("GPL");
2536 MODULE_AUTHOR("Harald Welte <hwelte@sysmocom.de>");
2537 MODULE_DESCRIPTION("Interface driver for GTP encapsulated traffic");
2538 MODULE_ALIAS_RTNL_LINK("gtp");
2539 MODULE_ALIAS_GENL_FAMILY("gtp");
2540