xref: /linux/drivers/net/gtp.c (revision 07fdad3a93756b872da7b53647715c48d0f4a2d0)
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/flow.h>
25 #include <net/inet_dscp.h>
26 #include <net/net_namespace.h>
27 #include <net/protocol.h>
28 #include <net/inet_sock.h>
29 #include <net/ip.h>
30 #include <net/ipv6.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>
38 
39 /* An active session for the subscriber. */
40 struct pdp_ctx {
41 	struct hlist_node	hlist_tid;
42 	struct hlist_node	hlist_addr;
43 
44 	union {
45 		struct {
46 			u64	tid;
47 			u16	flow;
48 		} v0;
49 		struct {
50 			u32	i_tei;
51 			u32	o_tei;
52 		} v1;
53 	} u;
54 	u8			gtp_version;
55 	u16			af;
56 
57 	union {
58 		struct in_addr	addr;
59 		struct in6_addr	addr6;
60 	} ms;
61 	union {
62 		struct in_addr	addr;
63 		struct in6_addr	addr6;
64 	} peer;
65 
66 	struct sock		*sk;
67 	struct net_device       *dev;
68 
69 	atomic_t		tx_seq;
70 	struct rcu_head		rcu_head;
71 };
72 
73 /* One instance of the GTP device. */
74 struct gtp_dev {
75 	struct list_head	list;
76 
77 	struct sock		*sk0;
78 	struct sock		*sk1u;
79 	u8			sk_created;
80 
81 	struct net_device	*dev;
82 	struct net		*net;
83 
84 	unsigned int		role;
85 	unsigned int		hash_size;
86 	struct hlist_head	*tid_hash;
87 	struct hlist_head	*addr_hash;
88 
89 	u8			restart_count;
90 };
91 
92 struct echo_info {
93 	u16			af;
94 	u8			gtp_version;
95 
96 	union {
97 		struct in_addr	addr;
98 	} ms;
99 	union {
100 		struct in_addr	addr;
101 	} peer;
102 };
103 
104 static unsigned int gtp_net_id __read_mostly;
105 
106 struct gtp_net {
107 	struct list_head gtp_dev_list;
108 };
109 
110 static u32 gtp_h_initval;
111 
112 static struct genl_family gtp_genl_family;
113 
114 enum gtp_multicast_groups {
115 	GTP_GENL_MCGRP,
116 };
117 
118 static const struct genl_multicast_group gtp_genl_mcgrps[] = {
119 	[GTP_GENL_MCGRP] = { .name = GTP_GENL_MCGRP_NAME },
120 };
121 
122 static void pdp_context_delete(struct pdp_ctx *pctx);
123 
124 static inline u32 gtp0_hashfn(u64 tid)
125 {
126 	u32 *tid32 = (u32 *) &tid;
127 	return jhash_2words(tid32[0], tid32[1], gtp_h_initval);
128 }
129 
130 static inline u32 gtp1u_hashfn(u32 tid)
131 {
132 	return jhash_1word(tid, gtp_h_initval);
133 }
134 
135 static inline u32 ipv4_hashfn(__be32 ip)
136 {
137 	return jhash_1word((__force u32)ip, gtp_h_initval);
138 }
139 
140 static u32 ipv6_hashfn(const struct in6_addr *ip6)
141 {
142 	return jhash_2words((__force u32)ip6->s6_addr32[0],
143 			    (__force u32)ip6->s6_addr32[1], gtp_h_initval);
144 }
145 
146 /* Resolve a PDP context structure based on the 64bit TID. */
147 static struct pdp_ctx *gtp0_pdp_find(struct gtp_dev *gtp, u64 tid, u16 family)
148 {
149 	struct hlist_head *head;
150 	struct pdp_ctx *pdp;
151 
152 	head = &gtp->tid_hash[gtp0_hashfn(tid) % gtp->hash_size];
153 
154 	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
155 		if (pdp->af == family &&
156 		    pdp->gtp_version == GTP_V0 &&
157 		    pdp->u.v0.tid == tid)
158 			return pdp;
159 	}
160 	return NULL;
161 }
162 
163 /* Resolve a PDP context structure based on the 32bit TEI. */
164 static struct pdp_ctx *gtp1_pdp_find(struct gtp_dev *gtp, u32 tid, u16 family)
165 {
166 	struct hlist_head *head;
167 	struct pdp_ctx *pdp;
168 
169 	head = &gtp->tid_hash[gtp1u_hashfn(tid) % gtp->hash_size];
170 
171 	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
172 		if (pdp->af == family &&
173 		    pdp->gtp_version == GTP_V1 &&
174 		    pdp->u.v1.i_tei == tid)
175 			return pdp;
176 	}
177 	return NULL;
178 }
179 
180 /* Resolve a PDP context based on IPv4 address of MS. */
181 static struct pdp_ctx *ipv4_pdp_find(struct gtp_dev *gtp, __be32 ms_addr)
182 {
183 	struct hlist_head *head;
184 	struct pdp_ctx *pdp;
185 
186 	head = &gtp->addr_hash[ipv4_hashfn(ms_addr) % gtp->hash_size];
187 
188 	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
189 		if (pdp->af == AF_INET &&
190 		    pdp->ms.addr.s_addr == ms_addr)
191 			return pdp;
192 	}
193 
194 	return NULL;
195 }
196 
197 /* 3GPP TS 29.060: PDN Connection: the association between a MS represented by
198  * [...] one IPv6 *prefix* and a PDN represented by an APN.
199  *
200  * Then, 3GPP TS 29.061, Section 11.2.1.3 says: The size of the prefix shall be
201  * according to the maximum prefix length for a global IPv6 address as
202  * specified in the IPv6 Addressing Architecture, see RFC 4291.
203  *
204  * Finally, RFC 4291 section 2.5.4 states: All Global Unicast addresses other
205  * than those that start with binary 000 have a 64-bit interface ID field
206  * (i.e., n + m = 64).
207  */
208 static bool ipv6_pdp_addr_equal(const struct in6_addr *a,
209 				const struct in6_addr *b)
210 {
211 	return a->s6_addr32[0] == b->s6_addr32[0] &&
212 	       a->s6_addr32[1] == b->s6_addr32[1];
213 }
214 
215 static struct pdp_ctx *ipv6_pdp_find(struct gtp_dev *gtp,
216 				     const struct in6_addr *ms_addr)
217 {
218 	struct hlist_head *head;
219 	struct pdp_ctx *pdp;
220 
221 	head = &gtp->addr_hash[ipv6_hashfn(ms_addr) % gtp->hash_size];
222 
223 	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
224 		if (pdp->af == AF_INET6 &&
225 		    ipv6_pdp_addr_equal(&pdp->ms.addr6, ms_addr))
226 			return pdp;
227 	}
228 
229 	return NULL;
230 }
231 
232 static bool gtp_check_ms_ipv4(struct sk_buff *skb, struct pdp_ctx *pctx,
233 				  unsigned int hdrlen, unsigned int role)
234 {
235 	struct iphdr *iph;
236 
237 	if (!pskb_may_pull(skb, hdrlen + sizeof(struct iphdr)))
238 		return false;
239 
240 	iph = (struct iphdr *)(skb->data + hdrlen);
241 
242 	if (role == GTP_ROLE_SGSN)
243 		return iph->daddr == pctx->ms.addr.s_addr;
244 	else
245 		return iph->saddr == pctx->ms.addr.s_addr;
246 }
247 
248 static bool gtp_check_ms_ipv6(struct sk_buff *skb, struct pdp_ctx *pctx,
249 			      unsigned int hdrlen, unsigned int role)
250 {
251 	struct ipv6hdr *ip6h;
252 	int ret;
253 
254 	if (!pskb_may_pull(skb, hdrlen + sizeof(struct ipv6hdr)))
255 		return false;
256 
257 	ip6h = (struct ipv6hdr *)(skb->data + hdrlen);
258 
259 	if ((ipv6_addr_type(&ip6h->saddr) & IPV6_ADDR_LINKLOCAL) ||
260 	    (ipv6_addr_type(&ip6h->daddr) & IPV6_ADDR_LINKLOCAL))
261 		return false;
262 
263 	if (role == GTP_ROLE_SGSN) {
264 		ret = ipv6_pdp_addr_equal(&ip6h->daddr, &pctx->ms.addr6);
265 	} else {
266 		ret = ipv6_pdp_addr_equal(&ip6h->saddr, &pctx->ms.addr6);
267 	}
268 
269 	return ret;
270 }
271 
272 /* Check if the inner IP address in this packet is assigned to any
273  * existing mobile subscriber.
274  */
275 static bool gtp_check_ms(struct sk_buff *skb, struct pdp_ctx *pctx,
276 			 unsigned int hdrlen, unsigned int role,
277 			 __u16 inner_proto)
278 {
279 	switch (inner_proto) {
280 	case ETH_P_IP:
281 		return gtp_check_ms_ipv4(skb, pctx, hdrlen, role);
282 	case ETH_P_IPV6:
283 		return gtp_check_ms_ipv6(skb, pctx, hdrlen, role);
284 	}
285 	return false;
286 }
287 
288 static int gtp_inner_proto(struct sk_buff *skb, unsigned int hdrlen,
289 			   __u16 *inner_proto)
290 {
291 	__u8 *ip_version, _ip_version;
292 
293 	ip_version = skb_header_pointer(skb, hdrlen, sizeof(*ip_version),
294 					&_ip_version);
295 	if (!ip_version)
296 		return -1;
297 
298 	switch (*ip_version & 0xf0) {
299 	case 0x40:
300 		*inner_proto = ETH_P_IP;
301 		break;
302 	case 0x60:
303 		*inner_proto = ETH_P_IPV6;
304 		break;
305 	default:
306 		return -1;
307 	}
308 
309 	return 0;
310 }
311 
312 static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb,
313 		  unsigned int hdrlen, unsigned int role, __u16 inner_proto)
314 {
315 	if (!gtp_check_ms(skb, pctx, hdrlen, role, inner_proto)) {
316 		netdev_dbg(pctx->dev, "No PDP ctx for this MS\n");
317 		return 1;
318 	}
319 
320 	/* Get rid of the GTP + UDP headers. */
321 	if (iptunnel_pull_header(skb, hdrlen, htons(inner_proto),
322 			 !net_eq(sock_net(pctx->sk), dev_net(pctx->dev)))) {
323 		pctx->dev->stats.rx_length_errors++;
324 		goto err;
325 	}
326 
327 	netdev_dbg(pctx->dev, "forwarding packet from GGSN to uplink\n");
328 
329 	/* Now that the UDP and the GTP header have been removed, set up the
330 	 * new network header. This is required by the upper layer to
331 	 * calculate the transport header.
332 	 */
333 	skb_reset_network_header(skb);
334 	skb_reset_mac_header(skb);
335 
336 	skb->dev = pctx->dev;
337 
338 	dev_sw_netstats_rx_add(pctx->dev, skb->len);
339 
340 	__netif_rx(skb);
341 	return 0;
342 
343 err:
344 	pctx->dev->stats.rx_dropped++;
345 	return -1;
346 }
347 
348 static struct rtable *ip4_route_output_gtp(struct flowi4 *fl4,
349 					   const struct sock *sk,
350 					   __be32 daddr, __be32 saddr)
351 {
352 	memset(fl4, 0, sizeof(*fl4));
353 	fl4->flowi4_oif		= sk->sk_bound_dev_if;
354 	fl4->daddr		= daddr;
355 	fl4->saddr		= saddr;
356 	fl4->flowi4_dscp	= inet_sk_dscp(inet_sk(sk));
357 	fl4->flowi4_scope	= ip_sock_rt_scope(sk);
358 	fl4->flowi4_proto	= sk->sk_protocol;
359 
360 	return ip_route_output_key(sock_net(sk), fl4);
361 }
362 
363 static struct rt6_info *ip6_route_output_gtp(struct net *net,
364 					     struct flowi6 *fl6,
365 					     const struct sock *sk,
366 					     const struct in6_addr *daddr,
367 					     struct in6_addr *saddr)
368 {
369 	struct dst_entry *dst;
370 
371 	memset(fl6, 0, sizeof(*fl6));
372 	fl6->flowi6_oif		= sk->sk_bound_dev_if;
373 	fl6->daddr		= *daddr;
374 	fl6->saddr		= *saddr;
375 	fl6->flowi6_proto	= sk->sk_protocol;
376 
377 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sk, fl6, NULL);
378 	if (IS_ERR(dst))
379 		return ERR_PTR(-ENETUNREACH);
380 
381 	return (struct rt6_info *)dst;
382 }
383 
384 /* GSM TS 09.60. 7.3
385  * In all Path Management messages:
386  * - TID: is not used and shall be set to 0.
387  * - Flow Label is not used and shall be set to 0
388  * In signalling messages:
389  * - number: this field is not yet used in signalling messages.
390  *   It shall be set to 255 by the sender and shall be ignored
391  *   by the receiver
392  * Returns true if the echo req was correct, false otherwise.
393  */
394 static bool gtp0_validate_echo_hdr(struct gtp0_header *gtp0)
395 {
396 	return !(gtp0->tid || (gtp0->flags ^ 0x1e) ||
397 		gtp0->number != 0xff || gtp0->flow);
398 }
399 
400 /* msg_type has to be GTP_ECHO_REQ or GTP_ECHO_RSP */
401 static void gtp0_build_echo_msg(struct gtp0_header *hdr, __u8 msg_type)
402 {
403 	int len_pkt, len_hdr;
404 
405 	hdr->flags = 0x1e; /* v0, GTP-non-prime. */
406 	hdr->type = msg_type;
407 	/* GSM TS 09.60. 7.3 In all Path Management Flow Label and TID
408 	 * are not used and shall be set to 0.
409 	 */
410 	hdr->flow = 0;
411 	hdr->tid = 0;
412 	hdr->number = 0xff;
413 	hdr->spare[0] = 0xff;
414 	hdr->spare[1] = 0xff;
415 	hdr->spare[2] = 0xff;
416 
417 	len_pkt = sizeof(struct gtp0_packet);
418 	len_hdr = sizeof(struct gtp0_header);
419 
420 	if (msg_type == GTP_ECHO_RSP)
421 		hdr->length = htons(len_pkt - len_hdr);
422 	else
423 		hdr->length = 0;
424 }
425 
426 static int gtp0_send_echo_resp_ip(struct gtp_dev *gtp, struct sk_buff *skb)
427 {
428 	struct iphdr *iph = ip_hdr(skb);
429 	struct flowi4 fl4;
430 	struct rtable *rt;
431 
432 	/* find route to the sender,
433 	 * src address becomes dst address and vice versa.
434 	 */
435 	rt = ip4_route_output_gtp(&fl4, gtp->sk0, iph->saddr, iph->daddr);
436 	if (IS_ERR(rt)) {
437 		netdev_dbg(gtp->dev, "no route for echo response from %pI4\n",
438 			   &iph->saddr);
439 		return -1;
440 	}
441 
442 	udp_tunnel_xmit_skb(rt, gtp->sk0, skb,
443 			    fl4.saddr, fl4.daddr,
444 			    iph->tos,
445 			    ip4_dst_hoplimit(&rt->dst),
446 			    0,
447 			    htons(GTP0_PORT), htons(GTP0_PORT),
448 			    !net_eq(sock_net(gtp->sk1u),
449 				    dev_net(gtp->dev)),
450 			    false,
451 			    0);
452 
453 	return 0;
454 }
455 
456 static int gtp0_send_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
457 {
458 	struct gtp0_packet *gtp_pkt;
459 	struct gtp0_header *gtp0;
460 	__be16 seq;
461 
462 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
463 
464 	if (!gtp0_validate_echo_hdr(gtp0))
465 		return -1;
466 
467 	seq = gtp0->seq;
468 
469 	/* pull GTP and UDP headers */
470 	skb_pull_data(skb, sizeof(struct gtp0_header) + sizeof(struct udphdr));
471 
472 	gtp_pkt = skb_push(skb, sizeof(struct gtp0_packet));
473 	memset(gtp_pkt, 0, sizeof(struct gtp0_packet));
474 
475 	gtp0_build_echo_msg(&gtp_pkt->gtp0_h, GTP_ECHO_RSP);
476 
477 	/* GSM TS 09.60. 7.3 The Sequence Number in a signalling response
478 	 * message shall be copied from the signalling request message
479 	 * that the GSN is replying to.
480 	 */
481 	gtp_pkt->gtp0_h.seq = seq;
482 
483 	gtp_pkt->ie.tag = GTPIE_RECOVERY;
484 	gtp_pkt->ie.val = gtp->restart_count;
485 
486 	switch (gtp->sk0->sk_family) {
487 	case AF_INET:
488 		if (gtp0_send_echo_resp_ip(gtp, skb) < 0)
489 			return -1;
490 		break;
491 	case AF_INET6:
492 		return -1;
493 	}
494 
495 	return 0;
496 }
497 
498 static int gtp_genl_fill_echo(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
499 			      int flags, u32 type, struct echo_info echo)
500 {
501 	void *genlh;
502 
503 	genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, flags,
504 			    type);
505 	if (!genlh)
506 		goto failure;
507 
508 	if (nla_put_u32(skb, GTPA_VERSION, echo.gtp_version) ||
509 	    nla_put_be32(skb, GTPA_PEER_ADDRESS, echo.peer.addr.s_addr) ||
510 	    nla_put_be32(skb, GTPA_MS_ADDRESS, echo.ms.addr.s_addr))
511 		goto failure;
512 
513 	genlmsg_end(skb, genlh);
514 	return 0;
515 
516 failure:
517 	genlmsg_cancel(skb, genlh);
518 	return -EMSGSIZE;
519 }
520 
521 static void gtp0_handle_echo_resp_ip(struct sk_buff *skb, struct echo_info *echo)
522 {
523 	struct iphdr *iph = ip_hdr(skb);
524 
525 	echo->ms.addr.s_addr = iph->daddr;
526 	echo->peer.addr.s_addr = iph->saddr;
527 	echo->gtp_version = GTP_V0;
528 }
529 
530 static int gtp0_handle_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
531 {
532 	struct gtp0_header *gtp0;
533 	struct echo_info echo;
534 	struct sk_buff *msg;
535 	int ret;
536 
537 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
538 
539 	if (!gtp0_validate_echo_hdr(gtp0))
540 		return -1;
541 
542 	switch (gtp->sk0->sk_family) {
543 	case AF_INET:
544 		gtp0_handle_echo_resp_ip(skb, &echo);
545 		break;
546 	case AF_INET6:
547 		return -1;
548 	}
549 
550 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
551 	if (!msg)
552 		return -ENOMEM;
553 
554 	ret = gtp_genl_fill_echo(msg, 0, 0, 0, GTP_CMD_ECHOREQ, echo);
555 	if (ret < 0) {
556 		nlmsg_free(msg);
557 		return ret;
558 	}
559 
560 	return genlmsg_multicast_netns(&gtp_genl_family, dev_net(gtp->dev),
561 				       msg, 0, GTP_GENL_MCGRP, GFP_ATOMIC);
562 }
563 
564 static int gtp_proto_to_family(__u16 proto)
565 {
566 	switch (proto) {
567 	case ETH_P_IP:
568 		return AF_INET;
569 	case ETH_P_IPV6:
570 		return AF_INET6;
571 	default:
572 		WARN_ON_ONCE(1);
573 		break;
574 	}
575 
576 	return AF_UNSPEC;
577 }
578 
579 /* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */
580 static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
581 {
582 	unsigned int hdrlen = sizeof(struct udphdr) +
583 			      sizeof(struct gtp0_header);
584 	struct gtp0_header *gtp0;
585 	struct pdp_ctx *pctx;
586 	__u16 inner_proto;
587 
588 	if (!pskb_may_pull(skb, hdrlen))
589 		return -1;
590 
591 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
592 
593 	if ((gtp0->flags >> 5) != GTP_V0)
594 		return 1;
595 
596 	/* If the sockets were created in kernel, it means that
597 	 * there is no daemon running in userspace which would
598 	 * handle echo request.
599 	 */
600 	if (gtp0->type == GTP_ECHO_REQ && gtp->sk_created)
601 		return gtp0_send_echo_resp(gtp, skb);
602 
603 	if (gtp0->type == GTP_ECHO_RSP && gtp->sk_created)
604 		return gtp0_handle_echo_resp(gtp, skb);
605 
606 	if (gtp0->type != GTP_TPDU)
607 		return 1;
608 
609 	if (gtp_inner_proto(skb, hdrlen, &inner_proto) < 0) {
610 		netdev_dbg(gtp->dev, "GTP packet does not encapsulate an IP packet\n");
611 		return -1;
612 	}
613 
614 	pctx = gtp0_pdp_find(gtp, be64_to_cpu(gtp0->tid),
615 			     gtp_proto_to_family(inner_proto));
616 	if (!pctx) {
617 		netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
618 		return 1;
619 	}
620 
621 	return gtp_rx(pctx, skb, hdrlen, gtp->role, inner_proto);
622 }
623 
624 /* msg_type has to be GTP_ECHO_REQ or GTP_ECHO_RSP */
625 static void gtp1u_build_echo_msg(struct gtp1_header_long *hdr, __u8 msg_type)
626 {
627 	int len_pkt, len_hdr;
628 
629 	/* S flag must be set to 1 */
630 	hdr->flags = 0x32; /* v1, GTP-non-prime. */
631 	hdr->type = msg_type;
632 	/* 3GPP TS 29.281 5.1 - TEID has to be set to 0 */
633 	hdr->tid = 0;
634 
635 	/* seq, npdu and next should be counted to the length of the GTP packet
636 	 * that's why szie of gtp1_header should be subtracted,
637 	 * not size of gtp1_header_long.
638 	 */
639 
640 	len_hdr = sizeof(struct gtp1_header);
641 
642 	if (msg_type == GTP_ECHO_RSP) {
643 		len_pkt = sizeof(struct gtp1u_packet);
644 		hdr->length = htons(len_pkt - len_hdr);
645 	} else {
646 		/* GTP_ECHO_REQ does not carry GTP Information Element,
647 		 * the why gtp1_header_long is used here.
648 		 */
649 		len_pkt = sizeof(struct gtp1_header_long);
650 		hdr->length = htons(len_pkt - len_hdr);
651 	}
652 }
653 
654 static int gtp1u_send_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
655 {
656 	struct gtp1_header_long *gtp1u;
657 	struct gtp1u_packet *gtp_pkt;
658 	struct rtable *rt;
659 	struct flowi4 fl4;
660 	struct iphdr *iph;
661 
662 	gtp1u = (struct gtp1_header_long *)(skb->data + sizeof(struct udphdr));
663 
664 	/* 3GPP TS 29.281 5.1 - For the Echo Request, Echo Response,
665 	 * Error Indication and Supported Extension Headers Notification
666 	 * messages, the S flag shall be set to 1 and TEID shall be set to 0.
667 	 */
668 	if (!(gtp1u->flags & GTP1_F_SEQ) || gtp1u->tid)
669 		return -1;
670 
671 	/* pull GTP and UDP headers */
672 	skb_pull_data(skb,
673 		      sizeof(struct gtp1_header_long) + sizeof(struct udphdr));
674 
675 	gtp_pkt = skb_push(skb, sizeof(struct gtp1u_packet));
676 	memset(gtp_pkt, 0, sizeof(struct gtp1u_packet));
677 
678 	gtp1u_build_echo_msg(&gtp_pkt->gtp1u_h, GTP_ECHO_RSP);
679 
680 	/* 3GPP TS 29.281 7.7.2 - The Restart Counter value in the
681 	 * Recovery information element shall not be used, i.e. it shall
682 	 * be set to zero by the sender and shall be ignored by the receiver.
683 	 * The Recovery information element is mandatory due to backwards
684 	 * compatibility reasons.
685 	 */
686 	gtp_pkt->ie.tag = GTPIE_RECOVERY;
687 	gtp_pkt->ie.val = 0;
688 
689 	iph = ip_hdr(skb);
690 
691 	/* find route to the sender,
692 	 * src address becomes dst address and vice versa.
693 	 */
694 	rt = ip4_route_output_gtp(&fl4, gtp->sk1u, iph->saddr, iph->daddr);
695 	if (IS_ERR(rt)) {
696 		netdev_dbg(gtp->dev, "no route for echo response from %pI4\n",
697 			   &iph->saddr);
698 		return -1;
699 	}
700 
701 	udp_tunnel_xmit_skb(rt, gtp->sk1u, skb,
702 			    fl4.saddr, fl4.daddr,
703 			    iph->tos,
704 			    ip4_dst_hoplimit(&rt->dst),
705 			    0,
706 			    htons(GTP1U_PORT), htons(GTP1U_PORT),
707 			    !net_eq(sock_net(gtp->sk1u),
708 				    dev_net(gtp->dev)),
709 			    false,
710 			    0);
711 	return 0;
712 }
713 
714 static int gtp1u_handle_echo_resp(struct gtp_dev *gtp, struct sk_buff *skb)
715 {
716 	struct gtp1_header_long *gtp1u;
717 	struct echo_info echo;
718 	struct sk_buff *msg;
719 	struct iphdr *iph;
720 	int ret;
721 
722 	gtp1u = (struct gtp1_header_long *)(skb->data + sizeof(struct udphdr));
723 
724 	/* 3GPP TS 29.281 5.1 - For the Echo Request, Echo Response,
725 	 * Error Indication and Supported Extension Headers Notification
726 	 * messages, the S flag shall be set to 1 and TEID shall be set to 0.
727 	 */
728 	if (!(gtp1u->flags & GTP1_F_SEQ) || gtp1u->tid)
729 		return -1;
730 
731 	iph = ip_hdr(skb);
732 	echo.ms.addr.s_addr = iph->daddr;
733 	echo.peer.addr.s_addr = iph->saddr;
734 	echo.gtp_version = GTP_V1;
735 
736 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
737 	if (!msg)
738 		return -ENOMEM;
739 
740 	ret = gtp_genl_fill_echo(msg, 0, 0, 0, GTP_CMD_ECHOREQ, echo);
741 	if (ret < 0) {
742 		nlmsg_free(msg);
743 		return ret;
744 	}
745 
746 	return genlmsg_multicast_netns(&gtp_genl_family, dev_net(gtp->dev),
747 				       msg, 0, GTP_GENL_MCGRP, GFP_ATOMIC);
748 }
749 
750 static int gtp_parse_exthdrs(struct sk_buff *skb, unsigned int *hdrlen)
751 {
752 	struct gtp_ext_hdr *gtp_exthdr, _gtp_exthdr;
753 	unsigned int offset = *hdrlen;
754 	__u8 *next_type, _next_type;
755 
756 	/* From 29.060: "The Extension Header Length field specifies the length
757 	 * of the particular Extension header in 4 octets units."
758 	 *
759 	 * This length field includes length field size itself (1 byte),
760 	 * payload (variable length) and next type (1 byte). The extension
761 	 * header is aligned to to 4 bytes.
762 	 */
763 
764 	do {
765 		gtp_exthdr = skb_header_pointer(skb, offset, sizeof(*gtp_exthdr),
766 						&_gtp_exthdr);
767 		if (!gtp_exthdr || !gtp_exthdr->len)
768 			return -1;
769 
770 		offset += gtp_exthdr->len * 4;
771 
772 		/* From 29.060: "If no such Header follows, then the value of
773 		 * the Next Extension Header Type shall be 0."
774 		 */
775 		next_type = skb_header_pointer(skb, offset - 1,
776 					       sizeof(_next_type), &_next_type);
777 		if (!next_type)
778 			return -1;
779 
780 	} while (*next_type != 0);
781 
782 	*hdrlen = offset;
783 
784 	return 0;
785 }
786 
787 static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
788 {
789 	unsigned int hdrlen = sizeof(struct udphdr) +
790 			      sizeof(struct gtp1_header);
791 	struct gtp1_header *gtp1;
792 	struct pdp_ctx *pctx;
793 	__u16 inner_proto;
794 
795 	if (!pskb_may_pull(skb, hdrlen))
796 		return -1;
797 
798 	gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
799 
800 	if ((gtp1->flags >> 5) != GTP_V1)
801 		return 1;
802 
803 	/* If the sockets were created in kernel, it means that
804 	 * there is no daemon running in userspace which would
805 	 * handle echo request.
806 	 */
807 	if (gtp1->type == GTP_ECHO_REQ && gtp->sk_created)
808 		return gtp1u_send_echo_resp(gtp, skb);
809 
810 	if (gtp1->type == GTP_ECHO_RSP && gtp->sk_created)
811 		return gtp1u_handle_echo_resp(gtp, skb);
812 
813 	if (gtp1->type != GTP_TPDU)
814 		return 1;
815 
816 	/* From 29.060: "This field shall be present if and only if any one or
817 	 * more of the S, PN and E flags are set.".
818 	 *
819 	 * If any of the bit is set, then the remaining ones also have to be
820 	 * set.
821 	 */
822 	if (gtp1->flags & GTP1_F_MASK)
823 		hdrlen += 4;
824 
825 	/* Make sure the header is larger enough, including extensions. */
826 	if (!pskb_may_pull(skb, hdrlen))
827 		return -1;
828 
829 	if (gtp_inner_proto(skb, hdrlen, &inner_proto) < 0) {
830 		netdev_dbg(gtp->dev, "GTP packet does not encapsulate an IP packet\n");
831 		return -1;
832 	}
833 
834 	gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
835 
836 	pctx = gtp1_pdp_find(gtp, ntohl(gtp1->tid),
837 			     gtp_proto_to_family(inner_proto));
838 	if (!pctx) {
839 		netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
840 		return 1;
841 	}
842 
843 	if (gtp1->flags & GTP1_F_EXTHDR &&
844 	    gtp_parse_exthdrs(skb, &hdrlen) < 0)
845 		return -1;
846 
847 	return gtp_rx(pctx, skb, hdrlen, gtp->role, inner_proto);
848 }
849 
850 static void __gtp_encap_destroy(struct sock *sk)
851 {
852 	struct gtp_dev *gtp;
853 
854 	lock_sock(sk);
855 	gtp = sk->sk_user_data;
856 	if (gtp) {
857 		if (gtp->sk0 == sk)
858 			gtp->sk0 = NULL;
859 		else
860 			gtp->sk1u = NULL;
861 		WRITE_ONCE(udp_sk(sk)->encap_type, 0);
862 		rcu_assign_sk_user_data(sk, NULL);
863 		release_sock(sk);
864 		sock_put(sk);
865 		return;
866 	}
867 	release_sock(sk);
868 }
869 
870 static void gtp_encap_destroy(struct sock *sk)
871 {
872 	rtnl_lock();
873 	__gtp_encap_destroy(sk);
874 	rtnl_unlock();
875 }
876 
877 static void gtp_encap_disable_sock(struct sock *sk)
878 {
879 	if (!sk)
880 		return;
881 
882 	__gtp_encap_destroy(sk);
883 }
884 
885 static void gtp_encap_disable(struct gtp_dev *gtp)
886 {
887 	if (gtp->sk_created) {
888 		udp_tunnel_sock_release(gtp->sk0->sk_socket);
889 		udp_tunnel_sock_release(gtp->sk1u->sk_socket);
890 		gtp->sk_created = false;
891 		gtp->sk0 = NULL;
892 		gtp->sk1u = NULL;
893 	} else {
894 		gtp_encap_disable_sock(gtp->sk0);
895 		gtp_encap_disable_sock(gtp->sk1u);
896 	}
897 }
898 
899 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
900  * Return codes: 0: success, <0: error, >0: pass up to userspace UDP socket.
901  */
902 static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb)
903 {
904 	struct gtp_dev *gtp;
905 	int ret = 0;
906 
907 	gtp = rcu_dereference_sk_user_data(sk);
908 	if (!gtp)
909 		return 1;
910 
911 	netdev_dbg(gtp->dev, "encap_recv sk=%p\n", sk);
912 
913 	switch (READ_ONCE(udp_sk(sk)->encap_type)) {
914 	case UDP_ENCAP_GTP0:
915 		netdev_dbg(gtp->dev, "received GTP0 packet\n");
916 		ret = gtp0_udp_encap_recv(gtp, skb);
917 		break;
918 	case UDP_ENCAP_GTP1U:
919 		netdev_dbg(gtp->dev, "received GTP1U packet\n");
920 		ret = gtp1u_udp_encap_recv(gtp, skb);
921 		break;
922 	default:
923 		ret = -1; /* Shouldn't happen. */
924 	}
925 
926 	switch (ret) {
927 	case 1:
928 		netdev_dbg(gtp->dev, "pass up to the process\n");
929 		break;
930 	case 0:
931 		break;
932 	case -1:
933 		netdev_dbg(gtp->dev, "GTP packet has been dropped\n");
934 		kfree_skb(skb);
935 		ret = 0;
936 		break;
937 	}
938 
939 	return ret;
940 }
941 
942 static void gtp_dev_uninit(struct net_device *dev)
943 {
944 	struct gtp_dev *gtp = netdev_priv(dev);
945 
946 	gtp_encap_disable(gtp);
947 }
948 
949 static inline void gtp0_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
950 {
951 	int payload_len = skb->len;
952 	struct gtp0_header *gtp0;
953 
954 	gtp0 = skb_push(skb, sizeof(*gtp0));
955 
956 	gtp0->flags	= 0x1e; /* v0, GTP-non-prime. */
957 	gtp0->type	= GTP_TPDU;
958 	gtp0->length	= htons(payload_len);
959 	gtp0->seq	= htons((atomic_inc_return(&pctx->tx_seq) - 1) % 0xffff);
960 	gtp0->flow	= htons(pctx->u.v0.flow);
961 	gtp0->number	= 0xff;
962 	gtp0->spare[0]	= gtp0->spare[1] = gtp0->spare[2] = 0xff;
963 	gtp0->tid	= cpu_to_be64(pctx->u.v0.tid);
964 }
965 
966 static inline void gtp1_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
967 {
968 	int payload_len = skb->len;
969 	struct gtp1_header *gtp1;
970 
971 	gtp1 = skb_push(skb, sizeof(*gtp1));
972 
973 	/* Bits    8  7  6  5  4  3  2	1
974 	 *	  +--+--+--+--+--+--+--+--+
975 	 *	  |version |PT| 0| E| S|PN|
976 	 *	  +--+--+--+--+--+--+--+--+
977 	 *	    0  0  1  1	1  0  0  0
978 	 */
979 	gtp1->flags	= 0x30; /* v1, GTP-non-prime. */
980 	gtp1->type	= GTP_TPDU;
981 	gtp1->length	= htons(payload_len);
982 	gtp1->tid	= htonl(pctx->u.v1.o_tei);
983 
984 	/* TODO: Support for extension header, sequence number and N-PDU.
985 	 *	 Update the length field if any of them is available.
986 	 */
987 }
988 
989 struct gtp_pktinfo {
990 	struct sock		*sk;
991 	union {
992 		struct flowi4	fl4;
993 		struct flowi6	fl6;
994 	};
995 	union {
996 		struct rtable	*rt;
997 		struct rt6_info	*rt6;
998 	};
999 	struct pdp_ctx		*pctx;
1000 	struct net_device	*dev;
1001 	__u8			tos;
1002 	__be16			gtph_port;
1003 };
1004 
1005 static void gtp_push_header(struct sk_buff *skb, struct gtp_pktinfo *pktinfo)
1006 {
1007 	switch (pktinfo->pctx->gtp_version) {
1008 	case GTP_V0:
1009 		pktinfo->gtph_port = htons(GTP0_PORT);
1010 		gtp0_push_header(skb, pktinfo->pctx);
1011 		break;
1012 	case GTP_V1:
1013 		pktinfo->gtph_port = htons(GTP1U_PORT);
1014 		gtp1_push_header(skb, pktinfo->pctx);
1015 		break;
1016 	}
1017 }
1018 
1019 static inline void gtp_set_pktinfo_ipv4(struct gtp_pktinfo *pktinfo,
1020 					struct sock *sk, __u8 tos,
1021 					struct pdp_ctx *pctx, struct rtable *rt,
1022 					struct flowi4 *fl4,
1023 					struct net_device *dev)
1024 {
1025 	pktinfo->sk	= sk;
1026 	pktinfo->tos	= tos;
1027 	pktinfo->pctx	= pctx;
1028 	pktinfo->rt	= rt;
1029 	pktinfo->fl4	= *fl4;
1030 	pktinfo->dev	= dev;
1031 }
1032 
1033 static void gtp_set_pktinfo_ipv6(struct gtp_pktinfo *pktinfo,
1034 				 struct sock *sk, __u8 tos,
1035 				 struct pdp_ctx *pctx, struct rt6_info *rt6,
1036 				 struct flowi6 *fl6,
1037 				 struct net_device *dev)
1038 {
1039 	pktinfo->sk	= sk;
1040 	pktinfo->tos	= tos;
1041 	pktinfo->pctx	= pctx;
1042 	pktinfo->rt6	= rt6;
1043 	pktinfo->fl6	= *fl6;
1044 	pktinfo->dev	= dev;
1045 }
1046 
1047 static int gtp_build_skb_outer_ip4(struct sk_buff *skb, struct net_device *dev,
1048 				   struct gtp_pktinfo *pktinfo,
1049 				   struct pdp_ctx *pctx, __u8 tos,
1050 				   __be16 frag_off)
1051 {
1052 	struct rtable *rt;
1053 	struct flowi4 fl4;
1054 	__be16 df;
1055 	int mtu;
1056 
1057 	rt = ip4_route_output_gtp(&fl4, pctx->sk, pctx->peer.addr.s_addr,
1058 				  inet_sk(pctx->sk)->inet_saddr);
1059 	if (IS_ERR(rt)) {
1060 		netdev_dbg(dev, "no route to SSGN %pI4\n",
1061 			   &pctx->peer.addr.s_addr);
1062 		dev->stats.tx_carrier_errors++;
1063 		goto err;
1064 	}
1065 
1066 	if (rt->dst.dev == dev) {
1067 		netdev_dbg(dev, "circular route to SSGN %pI4\n",
1068 			   &pctx->peer.addr.s_addr);
1069 		dev->stats.collisions++;
1070 		goto err_rt;
1071 	}
1072 
1073 	/* This is similar to tnl_update_pmtu(). */
1074 	df = frag_off;
1075 	if (df) {
1076 		mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
1077 			sizeof(struct iphdr) - sizeof(struct udphdr);
1078 		switch (pctx->gtp_version) {
1079 		case GTP_V0:
1080 			mtu -= sizeof(struct gtp0_header);
1081 			break;
1082 		case GTP_V1:
1083 			mtu -= sizeof(struct gtp1_header);
1084 			break;
1085 		}
1086 	} else {
1087 		mtu = dst_mtu(&rt->dst);
1088 	}
1089 
1090 	skb_dst_update_pmtu_no_confirm(skb, mtu);
1091 
1092 	if (frag_off & htons(IP_DF) &&
1093 	    ((!skb_is_gso(skb) && skb->len > mtu) ||
1094 	     (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu)))) {
1095 		netdev_dbg(dev, "packet too big, fragmentation needed\n");
1096 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
1097 			      htonl(mtu));
1098 		goto err_rt;
1099 	}
1100 
1101 	gtp_set_pktinfo_ipv4(pktinfo, pctx->sk, tos, pctx, rt, &fl4, dev);
1102 	gtp_push_header(skb, pktinfo);
1103 
1104 	return 0;
1105 err_rt:
1106 	ip_rt_put(rt);
1107 err:
1108 	return -EBADMSG;
1109 }
1110 
1111 static int gtp_build_skb_outer_ip6(struct net *net, struct sk_buff *skb,
1112 				   struct net_device *dev,
1113 				   struct gtp_pktinfo *pktinfo,
1114 				   struct pdp_ctx *pctx, __u8 tos)
1115 {
1116 	struct dst_entry *dst;
1117 	struct rt6_info *rt;
1118 	struct flowi6 fl6;
1119 	int mtu;
1120 
1121 	rt = ip6_route_output_gtp(net, &fl6, pctx->sk, &pctx->peer.addr6,
1122 				  &inet6_sk(pctx->sk)->saddr);
1123 	if (IS_ERR(rt)) {
1124 		netdev_dbg(dev, "no route to SSGN %pI6\n",
1125 			   &pctx->peer.addr6);
1126 		dev->stats.tx_carrier_errors++;
1127 		goto err;
1128 	}
1129 	dst = &rt->dst;
1130 
1131 	if (rt->dst.dev == dev) {
1132 		netdev_dbg(dev, "circular route to SSGN %pI6\n",
1133 			   &pctx->peer.addr6);
1134 		dev->stats.collisions++;
1135 		goto err_rt;
1136 	}
1137 
1138 	mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
1139 		sizeof(struct ipv6hdr) - sizeof(struct udphdr);
1140 	switch (pctx->gtp_version) {
1141 	case GTP_V0:
1142 		mtu -= sizeof(struct gtp0_header);
1143 		break;
1144 	case GTP_V1:
1145 		mtu -= sizeof(struct gtp1_header);
1146 		break;
1147 	}
1148 
1149 	skb_dst_update_pmtu_no_confirm(skb, mtu);
1150 
1151 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
1152 	    (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))) {
1153 		netdev_dbg(dev, "packet too big, fragmentation needed\n");
1154 		icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1155 		goto err_rt;
1156 	}
1157 
1158 	gtp_set_pktinfo_ipv6(pktinfo, pctx->sk, tos, pctx, rt, &fl6, dev);
1159 	gtp_push_header(skb, pktinfo);
1160 
1161 	return 0;
1162 err_rt:
1163 	dst_release(dst);
1164 err:
1165 	return -EBADMSG;
1166 }
1167 
1168 static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev,
1169 			     struct gtp_pktinfo *pktinfo)
1170 {
1171 	struct gtp_dev *gtp = netdev_priv(dev);
1172 	struct net *net = gtp->net;
1173 	struct pdp_ctx *pctx;
1174 	struct iphdr *iph;
1175 	int ret;
1176 
1177 	/* Read the IP destination address and resolve the PDP context.
1178 	 * Prepend PDP header with TEI/TID from PDP ctx.
1179 	 */
1180 	iph = ip_hdr(skb);
1181 	if (gtp->role == GTP_ROLE_SGSN)
1182 		pctx = ipv4_pdp_find(gtp, iph->saddr);
1183 	else
1184 		pctx = ipv4_pdp_find(gtp, iph->daddr);
1185 
1186 	if (!pctx) {
1187 		netdev_dbg(dev, "no PDP ctx found for %pI4, skip\n",
1188 			   &iph->daddr);
1189 		return -ENOENT;
1190 	}
1191 	netdev_dbg(dev, "found PDP context %p\n", pctx);
1192 
1193 	switch (pctx->sk->sk_family) {
1194 	case AF_INET:
1195 		ret = gtp_build_skb_outer_ip4(skb, dev, pktinfo, pctx,
1196 					      iph->tos, iph->frag_off);
1197 		break;
1198 	case AF_INET6:
1199 		ret = gtp_build_skb_outer_ip6(net, skb, dev, pktinfo, pctx,
1200 					      iph->tos);
1201 		break;
1202 	default:
1203 		ret = -1;
1204 		WARN_ON_ONCE(1);
1205 		break;
1206 	}
1207 
1208 	if (ret < 0)
1209 		return ret;
1210 
1211 	netdev_dbg(dev, "gtp -> IP src: %pI4 dst: %pI4\n",
1212 		   &iph->saddr, &iph->daddr);
1213 
1214 	return 0;
1215 }
1216 
1217 static int gtp_build_skb_ip6(struct sk_buff *skb, struct net_device *dev,
1218 			     struct gtp_pktinfo *pktinfo)
1219 {
1220 	struct gtp_dev *gtp = netdev_priv(dev);
1221 	struct net *net = gtp->net;
1222 	struct pdp_ctx *pctx;
1223 	struct ipv6hdr *ip6h;
1224 	__u8 tos;
1225 	int ret;
1226 
1227 	/* Read the IP destination address and resolve the PDP context.
1228 	 * Prepend PDP header with TEI/TID from PDP ctx.
1229 	 */
1230 	ip6h = ipv6_hdr(skb);
1231 	if (gtp->role == GTP_ROLE_SGSN)
1232 		pctx = ipv6_pdp_find(gtp, &ip6h->saddr);
1233 	else
1234 		pctx = ipv6_pdp_find(gtp, &ip6h->daddr);
1235 
1236 	if (!pctx) {
1237 		netdev_dbg(dev, "no PDP ctx found for %pI6, skip\n",
1238 			   &ip6h->daddr);
1239 		return -ENOENT;
1240 	}
1241 	netdev_dbg(dev, "found PDP context %p\n", pctx);
1242 
1243 	tos = ipv6_get_dsfield(ip6h);
1244 
1245 	switch (pctx->sk->sk_family) {
1246 	case AF_INET:
1247 		ret = gtp_build_skb_outer_ip4(skb, dev, pktinfo, pctx, tos, 0);
1248 		break;
1249 	case AF_INET6:
1250 		ret = gtp_build_skb_outer_ip6(net, skb, dev, pktinfo, pctx, tos);
1251 		break;
1252 	default:
1253 		ret = -1;
1254 		WARN_ON_ONCE(1);
1255 		break;
1256 	}
1257 
1258 	if (ret < 0)
1259 		return ret;
1260 
1261 	netdev_dbg(dev, "gtp -> IP src: %pI6 dst: %pI6\n",
1262 		   &ip6h->saddr, &ip6h->daddr);
1263 
1264 	return 0;
1265 }
1266 
1267 static netdev_tx_t gtp_dev_xmit(struct sk_buff *skb, struct net_device *dev)
1268 {
1269 	unsigned int proto = ntohs(skb->protocol);
1270 	struct gtp_pktinfo pktinfo;
1271 	int err;
1272 
1273 	/* Ensure there is sufficient headroom. */
1274 	if (skb_cow_head(skb, dev->needed_headroom))
1275 		goto tx_err;
1276 
1277 	if (!pskb_inet_may_pull(skb))
1278 		goto tx_err;
1279 
1280 	skb_reset_inner_headers(skb);
1281 
1282 	/* PDP context lookups in gtp_build_skb_*() need rcu read-side lock. */
1283 	rcu_read_lock();
1284 	switch (proto) {
1285 	case ETH_P_IP:
1286 		err = gtp_build_skb_ip4(skb, dev, &pktinfo);
1287 		break;
1288 	case ETH_P_IPV6:
1289 		err = gtp_build_skb_ip6(skb, dev, &pktinfo);
1290 		break;
1291 	default:
1292 		err = -EOPNOTSUPP;
1293 		break;
1294 	}
1295 	rcu_read_unlock();
1296 
1297 	if (err < 0)
1298 		goto tx_err;
1299 
1300 	switch (pktinfo.pctx->sk->sk_family) {
1301 	case AF_INET:
1302 		udp_tunnel_xmit_skb(pktinfo.rt, pktinfo.sk, skb,
1303 				    pktinfo.fl4.saddr, pktinfo.fl4.daddr,
1304 				    pktinfo.tos,
1305 				    ip4_dst_hoplimit(&pktinfo.rt->dst),
1306 				    0,
1307 				    pktinfo.gtph_port, pktinfo.gtph_port,
1308 				    !net_eq(sock_net(pktinfo.pctx->sk),
1309 					    dev_net(dev)),
1310 				    false, 0);
1311 		break;
1312 	case AF_INET6:
1313 #if IS_ENABLED(CONFIG_IPV6)
1314 		udp_tunnel6_xmit_skb(&pktinfo.rt6->dst, pktinfo.sk, skb, dev,
1315 				     &pktinfo.fl6.saddr, &pktinfo.fl6.daddr,
1316 				     pktinfo.tos,
1317 				     ip6_dst_hoplimit(&pktinfo.rt->dst),
1318 				     0,
1319 				     pktinfo.gtph_port, pktinfo.gtph_port,
1320 				     false, 0);
1321 #else
1322 		goto tx_err;
1323 #endif
1324 		break;
1325 	}
1326 
1327 	return NETDEV_TX_OK;
1328 tx_err:
1329 	dev->stats.tx_errors++;
1330 	dev_kfree_skb(skb);
1331 	return NETDEV_TX_OK;
1332 }
1333 
1334 static const struct net_device_ops gtp_netdev_ops = {
1335 	.ndo_uninit		= gtp_dev_uninit,
1336 	.ndo_start_xmit		= gtp_dev_xmit,
1337 };
1338 
1339 static const struct device_type gtp_type = {
1340 	.name = "gtp",
1341 };
1342 
1343 #define GTP_TH_MAXLEN	(sizeof(struct udphdr) + sizeof(struct gtp0_header))
1344 #define GTP_IPV4_MAXLEN	(sizeof(struct iphdr) + GTP_TH_MAXLEN)
1345 
1346 static void gtp_link_setup(struct net_device *dev)
1347 {
1348 	struct gtp_dev *gtp = netdev_priv(dev);
1349 
1350 	dev->netdev_ops		= &gtp_netdev_ops;
1351 	dev->needs_free_netdev	= true;
1352 	SET_NETDEV_DEVTYPE(dev, &gtp_type);
1353 
1354 	dev->hard_header_len = 0;
1355 	dev->addr_len = 0;
1356 	dev->mtu = ETH_DATA_LEN - GTP_IPV4_MAXLEN;
1357 
1358 	/* Zero header length. */
1359 	dev->type = ARPHRD_NONE;
1360 	dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1361 
1362 	dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1363 	dev->priv_flags	|= IFF_NO_QUEUE;
1364 	dev->lltx = true;
1365 	netif_keep_dst(dev);
1366 
1367 	dev->needed_headroom	= LL_MAX_HEADER + GTP_IPV4_MAXLEN;
1368 	gtp->dev = dev;
1369 }
1370 
1371 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize);
1372 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]);
1373 
1374 static void gtp_destructor(struct net_device *dev)
1375 {
1376 	struct gtp_dev *gtp = netdev_priv(dev);
1377 
1378 	kfree(gtp->addr_hash);
1379 	kfree(gtp->tid_hash);
1380 }
1381 
1382 static int gtp_sock_udp_config(struct udp_port_cfg *udp_conf,
1383 			       const struct nlattr *nla, int family)
1384 {
1385 	udp_conf->family = family;
1386 
1387 	switch (udp_conf->family) {
1388 	case AF_INET:
1389 		udp_conf->local_ip.s_addr = nla_get_be32(nla);
1390 		break;
1391 #if IS_ENABLED(CONFIG_IPV6)
1392 	case AF_INET6:
1393 		udp_conf->local_ip6 = nla_get_in6_addr(nla);
1394 		break;
1395 #endif
1396 	default:
1397 		return -EOPNOTSUPP;
1398 	}
1399 
1400 	return 0;
1401 }
1402 
1403 static struct sock *gtp_create_sock(int type, struct gtp_dev *gtp,
1404 				    const struct nlattr *nla, int family)
1405 {
1406 	struct udp_tunnel_sock_cfg tuncfg = {};
1407 	struct udp_port_cfg udp_conf = {};
1408 	struct net *net = gtp->net;
1409 	struct socket *sock;
1410 	int err;
1411 
1412 	if (nla) {
1413 		err = gtp_sock_udp_config(&udp_conf, nla, family);
1414 		if (err < 0)
1415 			return ERR_PTR(err);
1416 	} else {
1417 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
1418 		udp_conf.family = AF_INET;
1419 	}
1420 
1421 	if (type == UDP_ENCAP_GTP0)
1422 		udp_conf.local_udp_port = htons(GTP0_PORT);
1423 	else if (type == UDP_ENCAP_GTP1U)
1424 		udp_conf.local_udp_port = htons(GTP1U_PORT);
1425 	else
1426 		return ERR_PTR(-EINVAL);
1427 
1428 	err = udp_sock_create(net, &udp_conf, &sock);
1429 	if (err)
1430 		return ERR_PTR(err);
1431 
1432 	tuncfg.sk_user_data = gtp;
1433 	tuncfg.encap_type = type;
1434 	tuncfg.encap_rcv = gtp_encap_recv;
1435 	tuncfg.encap_destroy = NULL;
1436 
1437 	setup_udp_tunnel_sock(net, sock, &tuncfg);
1438 
1439 	return sock->sk;
1440 }
1441 
1442 static int gtp_create_sockets(struct gtp_dev *gtp, const struct nlattr *nla,
1443 			      int family)
1444 {
1445 	struct sock *sk1u;
1446 	struct sock *sk0;
1447 
1448 	sk0 = gtp_create_sock(UDP_ENCAP_GTP0, gtp, nla, family);
1449 	if (IS_ERR(sk0))
1450 		return PTR_ERR(sk0);
1451 
1452 	sk1u = gtp_create_sock(UDP_ENCAP_GTP1U, gtp, nla, family);
1453 	if (IS_ERR(sk1u)) {
1454 		udp_tunnel_sock_release(sk0->sk_socket);
1455 		return PTR_ERR(sk1u);
1456 	}
1457 
1458 	gtp->sk_created = true;
1459 	gtp->sk0 = sk0;
1460 	gtp->sk1u = sk1u;
1461 
1462 	return 0;
1463 }
1464 
1465 #define GTP_TH_MAXLEN	(sizeof(struct udphdr) + sizeof(struct gtp0_header))
1466 #define GTP_IPV6_MAXLEN	(sizeof(struct ipv6hdr) + GTP_TH_MAXLEN)
1467 
1468 static int gtp_newlink(struct net_device *dev,
1469 		       struct rtnl_newlink_params *params,
1470 		       struct netlink_ext_ack *extack)
1471 {
1472 	struct net *link_net = rtnl_newlink_link_net(params);
1473 	struct nlattr **data = params->data;
1474 	unsigned int role = GTP_ROLE_GGSN;
1475 	struct gtp_dev *gtp;
1476 	struct gtp_net *gn;
1477 	int hashsize, err;
1478 
1479 #if !IS_ENABLED(CONFIG_IPV6)
1480 	if (data[IFLA_GTP_LOCAL6])
1481 		return -EAFNOSUPPORT;
1482 #endif
1483 
1484 	gtp = netdev_priv(dev);
1485 
1486 	if (!data[IFLA_GTP_PDP_HASHSIZE]) {
1487 		hashsize = 1024;
1488 	} else {
1489 		hashsize = nla_get_u32(data[IFLA_GTP_PDP_HASHSIZE]);
1490 		if (!hashsize)
1491 			hashsize = 1024;
1492 	}
1493 
1494 	if (data[IFLA_GTP_ROLE]) {
1495 		role = nla_get_u32(data[IFLA_GTP_ROLE]);
1496 		if (role > GTP_ROLE_SGSN)
1497 			return -EINVAL;
1498 	}
1499 	gtp->role = role;
1500 
1501 	gtp->restart_count = nla_get_u8_default(data[IFLA_GTP_RESTART_COUNT],
1502 						0);
1503 
1504 	gtp->net = link_net;
1505 
1506 	err = gtp_hashtable_new(gtp, hashsize);
1507 	if (err < 0)
1508 		return err;
1509 
1510 	if (data[IFLA_GTP_CREATE_SOCKETS]) {
1511 		if (data[IFLA_GTP_LOCAL6])
1512 			err = gtp_create_sockets(gtp, data[IFLA_GTP_LOCAL6], AF_INET6);
1513 		else
1514 			err = gtp_create_sockets(gtp, data[IFLA_GTP_LOCAL], AF_INET);
1515 	} else {
1516 		err = gtp_encap_enable(gtp, data);
1517 	}
1518 
1519 	if (err < 0)
1520 		goto out_hashtable;
1521 
1522 	if ((gtp->sk0 && gtp->sk0->sk_family == AF_INET6) ||
1523 	    (gtp->sk1u && gtp->sk1u->sk_family == AF_INET6)) {
1524 		dev->mtu = ETH_DATA_LEN - GTP_IPV6_MAXLEN;
1525 		dev->needed_headroom = LL_MAX_HEADER + GTP_IPV6_MAXLEN;
1526 	}
1527 
1528 	err = register_netdevice(dev);
1529 	if (err < 0) {
1530 		netdev_dbg(dev, "failed to register new netdev %d\n", err);
1531 		goto out_encap;
1532 	}
1533 
1534 	gn = net_generic(link_net, gtp_net_id);
1535 	list_add(&gtp->list, &gn->gtp_dev_list);
1536 	dev->priv_destructor = gtp_destructor;
1537 
1538 	netdev_dbg(dev, "registered new GTP interface\n");
1539 
1540 	return 0;
1541 
1542 out_encap:
1543 	gtp_encap_disable(gtp);
1544 out_hashtable:
1545 	kfree(gtp->addr_hash);
1546 	kfree(gtp->tid_hash);
1547 	return err;
1548 }
1549 
1550 static void gtp_dellink(struct net_device *dev, struct list_head *head)
1551 {
1552 	struct gtp_dev *gtp = netdev_priv(dev);
1553 	struct hlist_node *next;
1554 	struct pdp_ctx *pctx;
1555 	int i;
1556 
1557 	for (i = 0; i < gtp->hash_size; i++)
1558 		hlist_for_each_entry_safe(pctx, next, &gtp->tid_hash[i], hlist_tid)
1559 			pdp_context_delete(pctx);
1560 
1561 	list_del(&gtp->list);
1562 	unregister_netdevice_queue(dev, head);
1563 }
1564 
1565 static const struct nla_policy gtp_policy[IFLA_GTP_MAX + 1] = {
1566 	[IFLA_GTP_FD0]			= { .type = NLA_U32 },
1567 	[IFLA_GTP_FD1]			= { .type = NLA_U32 },
1568 	[IFLA_GTP_PDP_HASHSIZE]		= { .type = NLA_U32 },
1569 	[IFLA_GTP_ROLE]			= { .type = NLA_U32 },
1570 	[IFLA_GTP_CREATE_SOCKETS]	= { .type = NLA_U8 },
1571 	[IFLA_GTP_RESTART_COUNT]	= { .type = NLA_U8 },
1572 	[IFLA_GTP_LOCAL]		= { .type = NLA_U32 },
1573 	[IFLA_GTP_LOCAL6]		= { .len = sizeof(struct in6_addr) },
1574 };
1575 
1576 static int gtp_validate(struct nlattr *tb[], struct nlattr *data[],
1577 			struct netlink_ext_ack *extack)
1578 {
1579 	if (!data)
1580 		return -EINVAL;
1581 
1582 	return 0;
1583 }
1584 
1585 static size_t gtp_get_size(const struct net_device *dev)
1586 {
1587 	return nla_total_size(sizeof(__u32)) + /* IFLA_GTP_PDP_HASHSIZE */
1588 		nla_total_size(sizeof(__u32)) + /* IFLA_GTP_ROLE */
1589 		nla_total_size(sizeof(__u8)); /* IFLA_GTP_RESTART_COUNT */
1590 }
1591 
1592 static int gtp_fill_info(struct sk_buff *skb, const struct net_device *dev)
1593 {
1594 	struct gtp_dev *gtp = netdev_priv(dev);
1595 
1596 	if (nla_put_u32(skb, IFLA_GTP_PDP_HASHSIZE, gtp->hash_size))
1597 		goto nla_put_failure;
1598 	if (nla_put_u32(skb, IFLA_GTP_ROLE, gtp->role))
1599 		goto nla_put_failure;
1600 	if (nla_put_u8(skb, IFLA_GTP_RESTART_COUNT, gtp->restart_count))
1601 		goto nla_put_failure;
1602 
1603 	return 0;
1604 
1605 nla_put_failure:
1606 	return -EMSGSIZE;
1607 }
1608 
1609 static struct rtnl_link_ops gtp_link_ops __read_mostly = {
1610 	.kind		= "gtp",
1611 	.maxtype	= IFLA_GTP_MAX,
1612 	.policy		= gtp_policy,
1613 	.priv_size	= sizeof(struct gtp_dev),
1614 	.setup		= gtp_link_setup,
1615 	.validate	= gtp_validate,
1616 	.newlink	= gtp_newlink,
1617 	.dellink	= gtp_dellink,
1618 	.get_size	= gtp_get_size,
1619 	.fill_info	= gtp_fill_info,
1620 };
1621 
1622 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize)
1623 {
1624 	int i;
1625 
1626 	gtp->addr_hash = kmalloc_array(hsize, sizeof(struct hlist_head),
1627 				       GFP_KERNEL | __GFP_NOWARN);
1628 	if (gtp->addr_hash == NULL)
1629 		return -ENOMEM;
1630 
1631 	gtp->tid_hash = kmalloc_array(hsize, sizeof(struct hlist_head),
1632 				      GFP_KERNEL | __GFP_NOWARN);
1633 	if (gtp->tid_hash == NULL)
1634 		goto err1;
1635 
1636 	gtp->hash_size = hsize;
1637 
1638 	for (i = 0; i < hsize; i++) {
1639 		INIT_HLIST_HEAD(&gtp->addr_hash[i]);
1640 		INIT_HLIST_HEAD(&gtp->tid_hash[i]);
1641 	}
1642 	return 0;
1643 err1:
1644 	kfree(gtp->addr_hash);
1645 	return -ENOMEM;
1646 }
1647 
1648 static struct sock *gtp_encap_enable_socket(int fd, int type,
1649 					    struct gtp_dev *gtp)
1650 {
1651 	struct udp_tunnel_sock_cfg tuncfg = {NULL};
1652 	struct socket *sock;
1653 	struct sock *sk;
1654 	int err;
1655 
1656 	pr_debug("enable gtp on %d, %d\n", fd, type);
1657 
1658 	sock = sockfd_lookup(fd, &err);
1659 	if (!sock) {
1660 		pr_debug("gtp socket fd=%d not found\n", fd);
1661 		return ERR_PTR(err);
1662 	}
1663 
1664 	sk = sock->sk;
1665 	if (sk->sk_protocol != IPPROTO_UDP ||
1666 	    sk->sk_type != SOCK_DGRAM ||
1667 	    (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) {
1668 		pr_debug("socket fd=%d not UDP\n", fd);
1669 		sk = ERR_PTR(-EINVAL);
1670 		goto out_sock;
1671 	}
1672 
1673 	if (sk->sk_family == AF_INET6 &&
1674 	    !sk->sk_ipv6only) {
1675 		sk = ERR_PTR(-EADDRNOTAVAIL);
1676 		goto out_sock;
1677 	}
1678 
1679 	lock_sock(sk);
1680 	if (sk->sk_user_data) {
1681 		sk = ERR_PTR(-EBUSY);
1682 		goto out_rel_sock;
1683 	}
1684 
1685 	sock_hold(sk);
1686 
1687 	tuncfg.sk_user_data = gtp;
1688 	tuncfg.encap_type = type;
1689 	tuncfg.encap_rcv = gtp_encap_recv;
1690 	tuncfg.encap_destroy = gtp_encap_destroy;
1691 
1692 	setup_udp_tunnel_sock(sock_net(sock->sk), sock, &tuncfg);
1693 
1694 out_rel_sock:
1695 	release_sock(sock->sk);
1696 out_sock:
1697 	sockfd_put(sock);
1698 	return sk;
1699 }
1700 
1701 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[])
1702 {
1703 	struct sock *sk1u = NULL;
1704 	struct sock *sk0 = NULL;
1705 
1706 	if (!data[IFLA_GTP_FD0] && !data[IFLA_GTP_FD1])
1707 		return -EINVAL;
1708 
1709 	if (data[IFLA_GTP_FD0]) {
1710 		int fd0 = nla_get_u32(data[IFLA_GTP_FD0]);
1711 
1712 		if (fd0 >= 0) {
1713 			sk0 = gtp_encap_enable_socket(fd0, UDP_ENCAP_GTP0, gtp);
1714 			if (IS_ERR(sk0))
1715 				return PTR_ERR(sk0);
1716 		}
1717 	}
1718 
1719 	if (data[IFLA_GTP_FD1]) {
1720 		int fd1 = nla_get_u32(data[IFLA_GTP_FD1]);
1721 
1722 		if (fd1 >= 0) {
1723 			sk1u = gtp_encap_enable_socket(fd1, UDP_ENCAP_GTP1U, gtp);
1724 			if (IS_ERR(sk1u)) {
1725 				gtp_encap_disable_sock(sk0);
1726 				return PTR_ERR(sk1u);
1727 			}
1728 		}
1729 	}
1730 
1731 	gtp->sk0 = sk0;
1732 	gtp->sk1u = sk1u;
1733 
1734 	if (sk0 && sk1u &&
1735 	    sk0->sk_family != sk1u->sk_family) {
1736 		gtp_encap_disable_sock(sk0);
1737 		gtp_encap_disable_sock(sk1u);
1738 		return -EINVAL;
1739 	}
1740 
1741 	return 0;
1742 }
1743 
1744 static struct gtp_dev *gtp_find_dev(struct net *src_net, struct nlattr *nla[])
1745 {
1746 	struct gtp_dev *gtp = NULL;
1747 	struct net_device *dev;
1748 	struct net *net;
1749 
1750 	/* Examine the link attributes and figure out which network namespace
1751 	 * we are talking about.
1752 	 */
1753 	if (nla[GTPA_NET_NS_FD])
1754 		net = get_net_ns_by_fd(nla_get_u32(nla[GTPA_NET_NS_FD]));
1755 	else
1756 		net = get_net(src_net);
1757 
1758 	if (IS_ERR(net))
1759 		return NULL;
1760 
1761 	/* Check if there's an existing gtpX device to configure */
1762 	dev = dev_get_by_index_rcu(net, nla_get_u32(nla[GTPA_LINK]));
1763 	if (dev && dev->netdev_ops == &gtp_netdev_ops)
1764 		gtp = netdev_priv(dev);
1765 
1766 	put_net(net);
1767 	return gtp;
1768 }
1769 
1770 static void gtp_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
1771 {
1772 	pctx->gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]);
1773 
1774 	switch (pctx->gtp_version) {
1775 	case GTP_V0:
1776 		/* According to TS 09.60, sections 7.5.1 and 7.5.2, the flow
1777 		 * label needs to be the same for uplink and downlink packets,
1778 		 * so let's annotate this.
1779 		 */
1780 		pctx->u.v0.tid = nla_get_u64(info->attrs[GTPA_TID]);
1781 		pctx->u.v0.flow = nla_get_u16(info->attrs[GTPA_FLOW]);
1782 		break;
1783 	case GTP_V1:
1784 		pctx->u.v1.i_tei = nla_get_u32(info->attrs[GTPA_I_TEI]);
1785 		pctx->u.v1.o_tei = nla_get_u32(info->attrs[GTPA_O_TEI]);
1786 		break;
1787 	default:
1788 		break;
1789 	}
1790 }
1791 
1792 static void ip_pdp_peer_fill(struct pdp_ctx *pctx, struct genl_info *info)
1793 {
1794 	if (info->attrs[GTPA_PEER_ADDRESS]) {
1795 		pctx->peer.addr.s_addr =
1796 			nla_get_be32(info->attrs[GTPA_PEER_ADDRESS]);
1797 	} else if (info->attrs[GTPA_PEER_ADDR6]) {
1798 		pctx->peer.addr6 = nla_get_in6_addr(info->attrs[GTPA_PEER_ADDR6]);
1799 	}
1800 }
1801 
1802 static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
1803 {
1804 	ip_pdp_peer_fill(pctx, info);
1805 	pctx->ms.addr.s_addr =
1806 		nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
1807 	gtp_pdp_fill(pctx, info);
1808 }
1809 
1810 static bool ipv6_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
1811 {
1812 	ip_pdp_peer_fill(pctx, info);
1813 	pctx->ms.addr6 = nla_get_in6_addr(info->attrs[GTPA_MS_ADDR6]);
1814 	if (pctx->ms.addr6.s6_addr32[2] ||
1815 	    pctx->ms.addr6.s6_addr32[3])
1816 		return false;
1817 
1818 	gtp_pdp_fill(pctx, info);
1819 
1820 	return true;
1821 }
1822 
1823 static struct pdp_ctx *gtp_pdp_add(struct gtp_dev *gtp, struct sock *sk,
1824 				   struct genl_info *info)
1825 {
1826 	struct pdp_ctx *pctx, *pctx_tid = NULL;
1827 	struct net_device *dev = gtp->dev;
1828 	u32 hash_ms, hash_tid = 0;
1829 	struct in6_addr ms_addr6;
1830 	unsigned int version;
1831 	bool found = false;
1832 	__be32 ms_addr;
1833 	int family;
1834 
1835 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
1836 
1837 	family = nla_get_u8_default(info->attrs[GTPA_FAMILY], AF_INET);
1838 
1839 #if !IS_ENABLED(CONFIG_IPV6)
1840 	if (family == AF_INET6)
1841 		return ERR_PTR(-EAFNOSUPPORT);
1842 #endif
1843 	if (!info->attrs[GTPA_PEER_ADDRESS] &&
1844 	    !info->attrs[GTPA_PEER_ADDR6])
1845 		return ERR_PTR(-EINVAL);
1846 
1847 	if ((info->attrs[GTPA_PEER_ADDRESS] &&
1848 	     sk->sk_family == AF_INET6) ||
1849 	    (info->attrs[GTPA_PEER_ADDR6] &&
1850 	     sk->sk_family == AF_INET))
1851 		return ERR_PTR(-EAFNOSUPPORT);
1852 
1853 	switch (family) {
1854 	case AF_INET:
1855 		if (!info->attrs[GTPA_MS_ADDRESS] ||
1856 		    info->attrs[GTPA_MS_ADDR6])
1857 			return ERR_PTR(-EINVAL);
1858 
1859 		ms_addr = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
1860 		hash_ms = ipv4_hashfn(ms_addr) % gtp->hash_size;
1861 		pctx = ipv4_pdp_find(gtp, ms_addr);
1862 		break;
1863 	case AF_INET6:
1864 		if (!info->attrs[GTPA_MS_ADDR6] ||
1865 		    info->attrs[GTPA_MS_ADDRESS])
1866 			return ERR_PTR(-EINVAL);
1867 
1868 		ms_addr6 = nla_get_in6_addr(info->attrs[GTPA_MS_ADDR6]);
1869 		hash_ms = ipv6_hashfn(&ms_addr6) % gtp->hash_size;
1870 		pctx = ipv6_pdp_find(gtp, &ms_addr6);
1871 		break;
1872 	default:
1873 		return ERR_PTR(-EAFNOSUPPORT);
1874 	}
1875 	if (pctx)
1876 		found = true;
1877 	if (version == GTP_V0)
1878 		pctx_tid = gtp0_pdp_find(gtp,
1879 					 nla_get_u64(info->attrs[GTPA_TID]),
1880 					 family);
1881 	else if (version == GTP_V1)
1882 		pctx_tid = gtp1_pdp_find(gtp,
1883 					 nla_get_u32(info->attrs[GTPA_I_TEI]),
1884 					 family);
1885 	if (pctx_tid)
1886 		found = true;
1887 
1888 	if (found) {
1889 		if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
1890 			return ERR_PTR(-EEXIST);
1891 		if (info->nlhdr->nlmsg_flags & NLM_F_REPLACE)
1892 			return ERR_PTR(-EOPNOTSUPP);
1893 
1894 		if (pctx && pctx_tid)
1895 			return ERR_PTR(-EEXIST);
1896 		if (!pctx)
1897 			pctx = pctx_tid;
1898 
1899 		switch (pctx->af) {
1900 		case AF_INET:
1901 			ipv4_pdp_fill(pctx, info);
1902 			break;
1903 		case AF_INET6:
1904 			if (!ipv6_pdp_fill(pctx, info))
1905 				return ERR_PTR(-EADDRNOTAVAIL);
1906 			break;
1907 		}
1908 
1909 		if (pctx->gtp_version == GTP_V0)
1910 			netdev_dbg(dev, "GTPv0-U: update tunnel id = %llx (pdp %p)\n",
1911 				   pctx->u.v0.tid, pctx);
1912 		else if (pctx->gtp_version == GTP_V1)
1913 			netdev_dbg(dev, "GTPv1-U: update tunnel id = %x/%x (pdp %p)\n",
1914 				   pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
1915 
1916 		return pctx;
1917 
1918 	}
1919 
1920 	pctx = kmalloc(sizeof(*pctx), GFP_ATOMIC);
1921 	if (pctx == NULL)
1922 		return ERR_PTR(-ENOMEM);
1923 
1924 	sock_hold(sk);
1925 	pctx->sk = sk;
1926 	pctx->dev = gtp->dev;
1927 	pctx->af = family;
1928 
1929 	switch (pctx->af) {
1930 	case AF_INET:
1931 		if (!info->attrs[GTPA_MS_ADDRESS]) {
1932 			sock_put(sk);
1933 			kfree(pctx);
1934 			return ERR_PTR(-EINVAL);
1935 		}
1936 
1937 		ipv4_pdp_fill(pctx, info);
1938 		break;
1939 	case AF_INET6:
1940 		if (!info->attrs[GTPA_MS_ADDR6]) {
1941 			sock_put(sk);
1942 			kfree(pctx);
1943 			return ERR_PTR(-EINVAL);
1944 		}
1945 
1946 		if (!ipv6_pdp_fill(pctx, info)) {
1947 			sock_put(sk);
1948 			kfree(pctx);
1949 			return ERR_PTR(-EADDRNOTAVAIL);
1950 		}
1951 		break;
1952 	}
1953 	atomic_set(&pctx->tx_seq, 0);
1954 
1955 	switch (pctx->gtp_version) {
1956 	case GTP_V0:
1957 		/* TS 09.60: "The flow label identifies unambiguously a GTP
1958 		 * flow.". We use the tid for this instead, I cannot find a
1959 		 * situation in which this doesn't unambiguosly identify the
1960 		 * PDP context.
1961 		 */
1962 		hash_tid = gtp0_hashfn(pctx->u.v0.tid) % gtp->hash_size;
1963 		break;
1964 	case GTP_V1:
1965 		hash_tid = gtp1u_hashfn(pctx->u.v1.i_tei) % gtp->hash_size;
1966 		break;
1967 	}
1968 
1969 	hlist_add_head_rcu(&pctx->hlist_addr, &gtp->addr_hash[hash_ms]);
1970 	hlist_add_head_rcu(&pctx->hlist_tid, &gtp->tid_hash[hash_tid]);
1971 
1972 	switch (pctx->gtp_version) {
1973 	case GTP_V0:
1974 		netdev_dbg(dev, "GTPv0-U: new PDP ctx id=%llx ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
1975 			   pctx->u.v0.tid, &pctx->peer.addr,
1976 			   &pctx->ms.addr, pctx);
1977 		break;
1978 	case GTP_V1:
1979 		netdev_dbg(dev, "GTPv1-U: new PDP ctx id=%x/%x ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
1980 			   pctx->u.v1.i_tei, pctx->u.v1.o_tei,
1981 			   &pctx->peer.addr, &pctx->ms.addr, pctx);
1982 		break;
1983 	}
1984 
1985 	return pctx;
1986 }
1987 
1988 static void pdp_context_free(struct rcu_head *head)
1989 {
1990 	struct pdp_ctx *pctx = container_of(head, struct pdp_ctx, rcu_head);
1991 
1992 	sock_put(pctx->sk);
1993 	kfree(pctx);
1994 }
1995 
1996 static void pdp_context_delete(struct pdp_ctx *pctx)
1997 {
1998 	hlist_del_rcu(&pctx->hlist_tid);
1999 	hlist_del_rcu(&pctx->hlist_addr);
2000 	call_rcu(&pctx->rcu_head, pdp_context_free);
2001 }
2002 
2003 static int gtp_tunnel_notify(struct pdp_ctx *pctx, u8 cmd, gfp_t allocation);
2004 
2005 static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
2006 {
2007 	unsigned int version;
2008 	struct pdp_ctx *pctx;
2009 	struct gtp_dev *gtp;
2010 	struct sock *sk;
2011 	int err;
2012 
2013 	if (!info->attrs[GTPA_VERSION] ||
2014 	    !info->attrs[GTPA_LINK])
2015 		return -EINVAL;
2016 
2017 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
2018 
2019 	switch (version) {
2020 	case GTP_V0:
2021 		if (!info->attrs[GTPA_TID] ||
2022 		    !info->attrs[GTPA_FLOW])
2023 			return -EINVAL;
2024 		break;
2025 	case GTP_V1:
2026 		if (!info->attrs[GTPA_I_TEI] ||
2027 		    !info->attrs[GTPA_O_TEI])
2028 			return -EINVAL;
2029 		break;
2030 
2031 	default:
2032 		return -EINVAL;
2033 	}
2034 
2035 	rtnl_lock();
2036 
2037 	gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
2038 	if (!gtp) {
2039 		err = -ENODEV;
2040 		goto out_unlock;
2041 	}
2042 
2043 	if (version == GTP_V0)
2044 		sk = gtp->sk0;
2045 	else if (version == GTP_V1)
2046 		sk = gtp->sk1u;
2047 	else
2048 		sk = NULL;
2049 
2050 	if (!sk) {
2051 		err = -ENODEV;
2052 		goto out_unlock;
2053 	}
2054 
2055 	pctx = gtp_pdp_add(gtp, sk, info);
2056 	if (IS_ERR(pctx)) {
2057 		err = PTR_ERR(pctx);
2058 	} else {
2059 		gtp_tunnel_notify(pctx, GTP_CMD_NEWPDP, GFP_KERNEL);
2060 		err = 0;
2061 	}
2062 
2063 out_unlock:
2064 	rtnl_unlock();
2065 	return err;
2066 }
2067 
2068 static struct pdp_ctx *gtp_find_pdp_by_link(struct net *net,
2069 					    struct nlattr *nla[])
2070 {
2071 	struct gtp_dev *gtp;
2072 	int family;
2073 
2074 	family = nla_get_u8_default(nla[GTPA_FAMILY], AF_INET);
2075 
2076 	gtp = gtp_find_dev(net, nla);
2077 	if (!gtp)
2078 		return ERR_PTR(-ENODEV);
2079 
2080 	if (nla[GTPA_MS_ADDRESS]) {
2081 		__be32 ip = nla_get_be32(nla[GTPA_MS_ADDRESS]);
2082 
2083 		if (family != AF_INET)
2084 			return ERR_PTR(-EINVAL);
2085 
2086 		return ipv4_pdp_find(gtp, ip);
2087 	} else if (nla[GTPA_MS_ADDR6]) {
2088 		struct in6_addr addr = nla_get_in6_addr(nla[GTPA_MS_ADDR6]);
2089 
2090 		if (family != AF_INET6)
2091 			return ERR_PTR(-EINVAL);
2092 
2093 		if (addr.s6_addr32[2] ||
2094 		    addr.s6_addr32[3])
2095 			return ERR_PTR(-EADDRNOTAVAIL);
2096 
2097 		return ipv6_pdp_find(gtp, &addr);
2098 	} else if (nla[GTPA_VERSION]) {
2099 		u32 gtp_version = nla_get_u32(nla[GTPA_VERSION]);
2100 
2101 		if (gtp_version == GTP_V0 && nla[GTPA_TID]) {
2102 			return gtp0_pdp_find(gtp, nla_get_u64(nla[GTPA_TID]),
2103 					     family);
2104 		} else if (gtp_version == GTP_V1 && nla[GTPA_I_TEI]) {
2105 			return gtp1_pdp_find(gtp, nla_get_u32(nla[GTPA_I_TEI]),
2106 					     family);
2107 		}
2108 	}
2109 
2110 	return ERR_PTR(-EINVAL);
2111 }
2112 
2113 static struct pdp_ctx *gtp_find_pdp(struct net *net, struct nlattr *nla[])
2114 {
2115 	struct pdp_ctx *pctx;
2116 
2117 	if (nla[GTPA_LINK])
2118 		pctx = gtp_find_pdp_by_link(net, nla);
2119 	else
2120 		pctx = ERR_PTR(-EINVAL);
2121 
2122 	if (!pctx)
2123 		pctx = ERR_PTR(-ENOENT);
2124 
2125 	return pctx;
2126 }
2127 
2128 static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
2129 {
2130 	struct pdp_ctx *pctx;
2131 	int err = 0;
2132 
2133 	if (!info->attrs[GTPA_VERSION])
2134 		return -EINVAL;
2135 
2136 	rcu_read_lock();
2137 
2138 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
2139 	if (IS_ERR(pctx)) {
2140 		err = PTR_ERR(pctx);
2141 		goto out_unlock;
2142 	}
2143 
2144 	if (pctx->gtp_version == GTP_V0)
2145 		netdev_dbg(pctx->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n",
2146 			   pctx->u.v0.tid, pctx);
2147 	else if (pctx->gtp_version == GTP_V1)
2148 		netdev_dbg(pctx->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n",
2149 			   pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
2150 
2151 	gtp_tunnel_notify(pctx, GTP_CMD_DELPDP, GFP_ATOMIC);
2152 	pdp_context_delete(pctx);
2153 
2154 out_unlock:
2155 	rcu_read_unlock();
2156 	return err;
2157 }
2158 
2159 static int gtp_genl_fill_info(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
2160 			      int flags, u32 type, struct pdp_ctx *pctx)
2161 {
2162 	void *genlh;
2163 
2164 	genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, flags,
2165 			    type);
2166 	if (genlh == NULL)
2167 		goto nlmsg_failure;
2168 
2169 	if (nla_put_u32(skb, GTPA_VERSION, pctx->gtp_version) ||
2170 	    nla_put_u32(skb, GTPA_LINK, pctx->dev->ifindex) ||
2171 	    nla_put_u8(skb, GTPA_FAMILY, pctx->af))
2172 		goto nla_put_failure;
2173 
2174 	switch (pctx->af) {
2175 	case AF_INET:
2176 		if (nla_put_be32(skb, GTPA_MS_ADDRESS, pctx->ms.addr.s_addr))
2177 			goto nla_put_failure;
2178 		break;
2179 	case AF_INET6:
2180 		if (nla_put_in6_addr(skb, GTPA_MS_ADDR6, &pctx->ms.addr6))
2181 			goto nla_put_failure;
2182 		break;
2183 	}
2184 
2185 	switch (pctx->sk->sk_family) {
2186 	case AF_INET:
2187 		if (nla_put_be32(skb, GTPA_PEER_ADDRESS, pctx->peer.addr.s_addr))
2188 			goto nla_put_failure;
2189 		break;
2190 	case AF_INET6:
2191 		if (nla_put_in6_addr(skb, GTPA_PEER_ADDR6, &pctx->peer.addr6))
2192 			goto nla_put_failure;
2193 		break;
2194 	}
2195 
2196 	switch (pctx->gtp_version) {
2197 	case GTP_V0:
2198 		if (nla_put_u64_64bit(skb, GTPA_TID, pctx->u.v0.tid, GTPA_PAD) ||
2199 		    nla_put_u16(skb, GTPA_FLOW, pctx->u.v0.flow))
2200 			goto nla_put_failure;
2201 		break;
2202 	case GTP_V1:
2203 		if (nla_put_u32(skb, GTPA_I_TEI, pctx->u.v1.i_tei) ||
2204 		    nla_put_u32(skb, GTPA_O_TEI, pctx->u.v1.o_tei))
2205 			goto nla_put_failure;
2206 		break;
2207 	}
2208 	genlmsg_end(skb, genlh);
2209 	return 0;
2210 
2211 nlmsg_failure:
2212 nla_put_failure:
2213 	genlmsg_cancel(skb, genlh);
2214 	return -EMSGSIZE;
2215 }
2216 
2217 static int gtp_tunnel_notify(struct pdp_ctx *pctx, u8 cmd, gfp_t allocation)
2218 {
2219 	struct sk_buff *msg;
2220 	int ret;
2221 
2222 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, allocation);
2223 	if (!msg)
2224 		return -ENOMEM;
2225 
2226 	ret = gtp_genl_fill_info(msg, 0, 0, 0, cmd, pctx);
2227 	if (ret < 0) {
2228 		nlmsg_free(msg);
2229 		return ret;
2230 	}
2231 
2232 	ret = genlmsg_multicast_netns(&gtp_genl_family, dev_net(pctx->dev), msg,
2233 				      0, GTP_GENL_MCGRP, GFP_ATOMIC);
2234 	return ret;
2235 }
2236 
2237 static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info)
2238 {
2239 	struct pdp_ctx *pctx = NULL;
2240 	struct sk_buff *skb2;
2241 	int err;
2242 
2243 	if (!info->attrs[GTPA_VERSION])
2244 		return -EINVAL;
2245 
2246 	rcu_read_lock();
2247 
2248 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
2249 	if (IS_ERR(pctx)) {
2250 		err = PTR_ERR(pctx);
2251 		goto err_unlock;
2252 	}
2253 
2254 	skb2 = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
2255 	if (skb2 == NULL) {
2256 		err = -ENOMEM;
2257 		goto err_unlock;
2258 	}
2259 
2260 	err = gtp_genl_fill_info(skb2, NETLINK_CB(skb).portid, info->snd_seq,
2261 				 0, info->nlhdr->nlmsg_type, pctx);
2262 	if (err < 0)
2263 		goto err_unlock_free;
2264 
2265 	rcu_read_unlock();
2266 	return genlmsg_unicast(genl_info_net(info), skb2, info->snd_portid);
2267 
2268 err_unlock_free:
2269 	kfree_skb(skb2);
2270 err_unlock:
2271 	rcu_read_unlock();
2272 	return err;
2273 }
2274 
2275 static int gtp_genl_dump_pdp(struct sk_buff *skb,
2276 				struct netlink_callback *cb)
2277 {
2278 	struct gtp_dev *last_gtp = (struct gtp_dev *)cb->args[2], *gtp;
2279 	int i, j, bucket = cb->args[0], skip = cb->args[1];
2280 	struct net *net = sock_net(skb->sk);
2281 	struct net_device *dev;
2282 	struct pdp_ctx *pctx;
2283 
2284 	if (cb->args[4])
2285 		return 0;
2286 
2287 	rcu_read_lock();
2288 	for_each_netdev_rcu(net, dev) {
2289 		if (dev->rtnl_link_ops != &gtp_link_ops)
2290 			continue;
2291 
2292 		gtp = netdev_priv(dev);
2293 
2294 		if (last_gtp && last_gtp != gtp)
2295 			continue;
2296 		else
2297 			last_gtp = NULL;
2298 
2299 		for (i = bucket; i < gtp->hash_size; i++) {
2300 			j = 0;
2301 			hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i],
2302 						 hlist_tid) {
2303 				if (j >= skip &&
2304 				    gtp_genl_fill_info(skb,
2305 					    NETLINK_CB(cb->skb).portid,
2306 					    cb->nlh->nlmsg_seq,
2307 					    NLM_F_MULTI,
2308 					    cb->nlh->nlmsg_type, pctx)) {
2309 					cb->args[0] = i;
2310 					cb->args[1] = j;
2311 					cb->args[2] = (unsigned long)gtp;
2312 					goto out;
2313 				}
2314 				j++;
2315 			}
2316 			skip = 0;
2317 		}
2318 		bucket = 0;
2319 	}
2320 	cb->args[4] = 1;
2321 out:
2322 	rcu_read_unlock();
2323 	return skb->len;
2324 }
2325 
2326 static int gtp_genl_send_echo_req(struct sk_buff *skb, struct genl_info *info)
2327 {
2328 	struct sk_buff *skb_to_send;
2329 	__be32 src_ip, dst_ip;
2330 	unsigned int version;
2331 	struct gtp_dev *gtp;
2332 	struct flowi4 fl4;
2333 	struct rtable *rt;
2334 	struct sock *sk;
2335 	__be16 port;
2336 	int len;
2337 
2338 	if (!info->attrs[GTPA_VERSION] ||
2339 	    !info->attrs[GTPA_LINK] ||
2340 	    !info->attrs[GTPA_PEER_ADDRESS] ||
2341 	    !info->attrs[GTPA_MS_ADDRESS])
2342 		return -EINVAL;
2343 
2344 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
2345 	dst_ip = nla_get_be32(info->attrs[GTPA_PEER_ADDRESS]);
2346 	src_ip = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
2347 
2348 	gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
2349 	if (!gtp)
2350 		return -ENODEV;
2351 
2352 	if (!gtp->sk_created)
2353 		return -EOPNOTSUPP;
2354 	if (!(gtp->dev->flags & IFF_UP))
2355 		return -ENETDOWN;
2356 
2357 	if (version == GTP_V0) {
2358 		struct gtp0_header *gtp0_h;
2359 
2360 		len = LL_RESERVED_SPACE(gtp->dev) + sizeof(struct gtp0_header) +
2361 			sizeof(struct iphdr) + sizeof(struct udphdr);
2362 
2363 		skb_to_send = netdev_alloc_skb_ip_align(gtp->dev, len);
2364 		if (!skb_to_send)
2365 			return -ENOMEM;
2366 
2367 		sk = gtp->sk0;
2368 		port = htons(GTP0_PORT);
2369 
2370 		gtp0_h = skb_push(skb_to_send, sizeof(struct gtp0_header));
2371 		memset(gtp0_h, 0, sizeof(struct gtp0_header));
2372 		gtp0_build_echo_msg(gtp0_h, GTP_ECHO_REQ);
2373 	} else if (version == GTP_V1) {
2374 		struct gtp1_header_long *gtp1u_h;
2375 
2376 		len = LL_RESERVED_SPACE(gtp->dev) +
2377 			sizeof(struct gtp1_header_long) +
2378 			sizeof(struct iphdr) + sizeof(struct udphdr);
2379 
2380 		skb_to_send = netdev_alloc_skb_ip_align(gtp->dev, len);
2381 		if (!skb_to_send)
2382 			return -ENOMEM;
2383 
2384 		sk = gtp->sk1u;
2385 		port = htons(GTP1U_PORT);
2386 
2387 		gtp1u_h = skb_push(skb_to_send,
2388 				   sizeof(struct gtp1_header_long));
2389 		memset(gtp1u_h, 0, sizeof(struct gtp1_header_long));
2390 		gtp1u_build_echo_msg(gtp1u_h, GTP_ECHO_REQ);
2391 	} else {
2392 		return -ENODEV;
2393 	}
2394 
2395 	rt = ip4_route_output_gtp(&fl4, sk, dst_ip, src_ip);
2396 	if (IS_ERR(rt)) {
2397 		netdev_dbg(gtp->dev, "no route for echo request to %pI4\n",
2398 			   &dst_ip);
2399 		kfree_skb(skb_to_send);
2400 		return -ENODEV;
2401 	}
2402 
2403 	udp_tunnel_xmit_skb(rt, sk, skb_to_send,
2404 			    fl4.saddr, fl4.daddr,
2405 			    inet_dscp_to_dsfield(fl4.flowi4_dscp),
2406 			    ip4_dst_hoplimit(&rt->dst),
2407 			    0,
2408 			    port, port,
2409 			    !net_eq(sock_net(sk),
2410 				    dev_net(gtp->dev)),
2411 			    false, 0);
2412 	return 0;
2413 }
2414 
2415 static const struct nla_policy gtp_genl_policy[GTPA_MAX + 1] = {
2416 	[GTPA_LINK]		= { .type = NLA_U32, },
2417 	[GTPA_VERSION]		= { .type = NLA_U32, },
2418 	[GTPA_TID]		= { .type = NLA_U64, },
2419 	[GTPA_PEER_ADDRESS]	= { .type = NLA_U32, },
2420 	[GTPA_MS_ADDRESS]	= { .type = NLA_U32, },
2421 	[GTPA_FLOW]		= { .type = NLA_U16, },
2422 	[GTPA_NET_NS_FD]	= { .type = NLA_U32, },
2423 	[GTPA_I_TEI]		= { .type = NLA_U32, },
2424 	[GTPA_O_TEI]		= { .type = NLA_U32, },
2425 	[GTPA_PEER_ADDR6]	= { .len = sizeof(struct in6_addr), },
2426 	[GTPA_MS_ADDR6]		= { .len = sizeof(struct in6_addr), },
2427 	[GTPA_FAMILY]		= { .type = NLA_U8, },
2428 };
2429 
2430 static const struct genl_small_ops gtp_genl_ops[] = {
2431 	{
2432 		.cmd = GTP_CMD_NEWPDP,
2433 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2434 		.doit = gtp_genl_new_pdp,
2435 		.flags = GENL_ADMIN_PERM,
2436 	},
2437 	{
2438 		.cmd = GTP_CMD_DELPDP,
2439 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2440 		.doit = gtp_genl_del_pdp,
2441 		.flags = GENL_ADMIN_PERM,
2442 	},
2443 	{
2444 		.cmd = GTP_CMD_GETPDP,
2445 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2446 		.doit = gtp_genl_get_pdp,
2447 		.dumpit = gtp_genl_dump_pdp,
2448 		.flags = GENL_ADMIN_PERM,
2449 	},
2450 	{
2451 		.cmd = GTP_CMD_ECHOREQ,
2452 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2453 		.doit = gtp_genl_send_echo_req,
2454 		.flags = GENL_ADMIN_PERM,
2455 	},
2456 };
2457 
2458 static struct genl_family gtp_genl_family __ro_after_init = {
2459 	.name		= "gtp",
2460 	.version	= 0,
2461 	.hdrsize	= 0,
2462 	.maxattr	= GTPA_MAX,
2463 	.policy = gtp_genl_policy,
2464 	.netnsok	= true,
2465 	.module		= THIS_MODULE,
2466 	.small_ops	= gtp_genl_ops,
2467 	.n_small_ops	= ARRAY_SIZE(gtp_genl_ops),
2468 	.resv_start_op	= GTP_CMD_ECHOREQ + 1,
2469 	.mcgrps		= gtp_genl_mcgrps,
2470 	.n_mcgrps	= ARRAY_SIZE(gtp_genl_mcgrps),
2471 };
2472 
2473 static int __net_init gtp_net_init(struct net *net)
2474 {
2475 	struct gtp_net *gn = net_generic(net, gtp_net_id);
2476 
2477 	INIT_LIST_HEAD(&gn->gtp_dev_list);
2478 	return 0;
2479 }
2480 
2481 static void __net_exit gtp_net_exit_rtnl(struct net *net,
2482 					 struct list_head *dev_to_kill)
2483 {
2484 	struct gtp_net *gn = net_generic(net, gtp_net_id);
2485 	struct gtp_dev *gtp, *gtp_next;
2486 
2487 	list_for_each_entry_safe(gtp, gtp_next, &gn->gtp_dev_list, list)
2488 		gtp_dellink(gtp->dev, dev_to_kill);
2489 }
2490 
2491 static struct pernet_operations gtp_net_ops = {
2492 	.init	= gtp_net_init,
2493 	.exit_rtnl = gtp_net_exit_rtnl,
2494 	.id	= &gtp_net_id,
2495 	.size	= sizeof(struct gtp_net),
2496 };
2497 
2498 static int __init gtp_init(void)
2499 {
2500 	int err;
2501 
2502 	get_random_bytes(&gtp_h_initval, sizeof(gtp_h_initval));
2503 
2504 	err = register_pernet_subsys(&gtp_net_ops);
2505 	if (err < 0)
2506 		goto error_out;
2507 
2508 	err = rtnl_link_register(&gtp_link_ops);
2509 	if (err < 0)
2510 		goto unreg_pernet_subsys;
2511 
2512 	err = genl_register_family(&gtp_genl_family);
2513 	if (err < 0)
2514 		goto unreg_rtnl_link;
2515 
2516 	pr_info("GTP module loaded (pdp ctx size %zd bytes)\n",
2517 		sizeof(struct pdp_ctx));
2518 	return 0;
2519 
2520 unreg_rtnl_link:
2521 	rtnl_link_unregister(&gtp_link_ops);
2522 unreg_pernet_subsys:
2523 	unregister_pernet_subsys(&gtp_net_ops);
2524 error_out:
2525 	pr_err("error loading GTP module loaded\n");
2526 	return err;
2527 }
2528 late_initcall(gtp_init);
2529 
2530 static void __exit gtp_fini(void)
2531 {
2532 	genl_unregister_family(&gtp_genl_family);
2533 	rtnl_link_unregister(&gtp_link_ops);
2534 	unregister_pernet_subsys(&gtp_net_ops);
2535 
2536 	pr_info("GTP module unloaded\n");
2537 }
2538 module_exit(gtp_fini);
2539 
2540 MODULE_LICENSE("GPL");
2541 MODULE_AUTHOR("Harald Welte <hwelte@sysmocom.de>");
2542 MODULE_DESCRIPTION("Interface driver for GTP encapsulated traffic");
2543 MODULE_ALIAS_RTNL_LINK("gtp");
2544 MODULE_ALIAS_GENL_FAMILY("gtp");
2545