xref: /openbmc/linux/drivers/net/gtp.c (revision b4a6aaea)
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/udp.h>
28 #include <net/udp_tunnel.h>
29 #include <net/icmp.h>
30 #include <net/xfrm.h>
31 #include <net/genetlink.h>
32 #include <net/netns/generic.h>
33 #include <net/gtp.h>
34 
35 /* An active session for the subscriber. */
36 struct pdp_ctx {
37 	struct hlist_node	hlist_tid;
38 	struct hlist_node	hlist_addr;
39 
40 	union {
41 		struct {
42 			u64	tid;
43 			u16	flow;
44 		} v0;
45 		struct {
46 			u32	i_tei;
47 			u32	o_tei;
48 		} v1;
49 	} u;
50 	u8			gtp_version;
51 	u16			af;
52 
53 	struct in_addr		ms_addr_ip4;
54 	struct in_addr		peer_addr_ip4;
55 
56 	struct sock		*sk;
57 	struct net_device       *dev;
58 
59 	atomic_t		tx_seq;
60 	struct rcu_head		rcu_head;
61 };
62 
63 /* One instance of the GTP device. */
64 struct gtp_dev {
65 	struct list_head	list;
66 
67 	struct sock		*sk0;
68 	struct sock		*sk1u;
69 
70 	struct net_device	*dev;
71 
72 	unsigned int		role;
73 	unsigned int		hash_size;
74 	struct hlist_head	*tid_hash;
75 	struct hlist_head	*addr_hash;
76 };
77 
78 static unsigned int gtp_net_id __read_mostly;
79 
80 struct gtp_net {
81 	struct list_head gtp_dev_list;
82 };
83 
84 static u32 gtp_h_initval;
85 
86 static void pdp_context_delete(struct pdp_ctx *pctx);
87 
88 static inline u32 gtp0_hashfn(u64 tid)
89 {
90 	u32 *tid32 = (u32 *) &tid;
91 	return jhash_2words(tid32[0], tid32[1], gtp_h_initval);
92 }
93 
94 static inline u32 gtp1u_hashfn(u32 tid)
95 {
96 	return jhash_1word(tid, gtp_h_initval);
97 }
98 
99 static inline u32 ipv4_hashfn(__be32 ip)
100 {
101 	return jhash_1word((__force u32)ip, gtp_h_initval);
102 }
103 
104 /* Resolve a PDP context structure based on the 64bit TID. */
105 static struct pdp_ctx *gtp0_pdp_find(struct gtp_dev *gtp, u64 tid)
106 {
107 	struct hlist_head *head;
108 	struct pdp_ctx *pdp;
109 
110 	head = &gtp->tid_hash[gtp0_hashfn(tid) % gtp->hash_size];
111 
112 	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
113 		if (pdp->gtp_version == GTP_V0 &&
114 		    pdp->u.v0.tid == tid)
115 			return pdp;
116 	}
117 	return NULL;
118 }
119 
120 /* Resolve a PDP context structure based on the 32bit TEI. */
121 static struct pdp_ctx *gtp1_pdp_find(struct gtp_dev *gtp, u32 tid)
122 {
123 	struct hlist_head *head;
124 	struct pdp_ctx *pdp;
125 
126 	head = &gtp->tid_hash[gtp1u_hashfn(tid) % gtp->hash_size];
127 
128 	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
129 		if (pdp->gtp_version == GTP_V1 &&
130 		    pdp->u.v1.i_tei == tid)
131 			return pdp;
132 	}
133 	return NULL;
134 }
135 
136 /* Resolve a PDP context based on IPv4 address of MS. */
137 static struct pdp_ctx *ipv4_pdp_find(struct gtp_dev *gtp, __be32 ms_addr)
138 {
139 	struct hlist_head *head;
140 	struct pdp_ctx *pdp;
141 
142 	head = &gtp->addr_hash[ipv4_hashfn(ms_addr) % gtp->hash_size];
143 
144 	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
145 		if (pdp->af == AF_INET &&
146 		    pdp->ms_addr_ip4.s_addr == ms_addr)
147 			return pdp;
148 	}
149 
150 	return NULL;
151 }
152 
153 static bool gtp_check_ms_ipv4(struct sk_buff *skb, struct pdp_ctx *pctx,
154 				  unsigned int hdrlen, unsigned int role)
155 {
156 	struct iphdr *iph;
157 
158 	if (!pskb_may_pull(skb, hdrlen + sizeof(struct iphdr)))
159 		return false;
160 
161 	iph = (struct iphdr *)(skb->data + hdrlen);
162 
163 	if (role == GTP_ROLE_SGSN)
164 		return iph->daddr == pctx->ms_addr_ip4.s_addr;
165 	else
166 		return iph->saddr == pctx->ms_addr_ip4.s_addr;
167 }
168 
169 /* Check if the inner IP address in this packet is assigned to any
170  * existing mobile subscriber.
171  */
172 static bool gtp_check_ms(struct sk_buff *skb, struct pdp_ctx *pctx,
173 			     unsigned int hdrlen, unsigned int role)
174 {
175 	switch (ntohs(skb->protocol)) {
176 	case ETH_P_IP:
177 		return gtp_check_ms_ipv4(skb, pctx, hdrlen, role);
178 	}
179 	return false;
180 }
181 
182 static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb,
183 			unsigned int hdrlen, unsigned int role)
184 {
185 	if (!gtp_check_ms(skb, pctx, hdrlen, role)) {
186 		netdev_dbg(pctx->dev, "No PDP ctx for this MS\n");
187 		return 1;
188 	}
189 
190 	/* Get rid of the GTP + UDP headers. */
191 	if (iptunnel_pull_header(skb, hdrlen, skb->protocol,
192 			 !net_eq(sock_net(pctx->sk), dev_net(pctx->dev)))) {
193 		pctx->dev->stats.rx_length_errors++;
194 		goto err;
195 	}
196 
197 	netdev_dbg(pctx->dev, "forwarding packet from GGSN to uplink\n");
198 
199 	/* Now that the UDP and the GTP header have been removed, set up the
200 	 * new network header. This is required by the upper layer to
201 	 * calculate the transport header.
202 	 */
203 	skb_reset_network_header(skb);
204 	skb_reset_mac_header(skb);
205 
206 	skb->dev = pctx->dev;
207 
208 	dev_sw_netstats_rx_add(pctx->dev, skb->len);
209 
210 	netif_rx(skb);
211 	return 0;
212 
213 err:
214 	pctx->dev->stats.rx_dropped++;
215 	return -1;
216 }
217 
218 /* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */
219 static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
220 {
221 	unsigned int hdrlen = sizeof(struct udphdr) +
222 			      sizeof(struct gtp0_header);
223 	struct gtp0_header *gtp0;
224 	struct pdp_ctx *pctx;
225 
226 	if (!pskb_may_pull(skb, hdrlen))
227 		return -1;
228 
229 	gtp0 = (struct gtp0_header *)(skb->data + sizeof(struct udphdr));
230 
231 	if ((gtp0->flags >> 5) != GTP_V0)
232 		return 1;
233 
234 	if (gtp0->type != GTP_TPDU)
235 		return 1;
236 
237 	pctx = gtp0_pdp_find(gtp, be64_to_cpu(gtp0->tid));
238 	if (!pctx) {
239 		netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
240 		return 1;
241 	}
242 
243 	return gtp_rx(pctx, skb, hdrlen, gtp->role);
244 }
245 
246 static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
247 {
248 	unsigned int hdrlen = sizeof(struct udphdr) +
249 			      sizeof(struct gtp1_header);
250 	struct gtp1_header *gtp1;
251 	struct pdp_ctx *pctx;
252 
253 	if (!pskb_may_pull(skb, hdrlen))
254 		return -1;
255 
256 	gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
257 
258 	if ((gtp1->flags >> 5) != GTP_V1)
259 		return 1;
260 
261 	if (gtp1->type != GTP_TPDU)
262 		return 1;
263 
264 	/* From 29.060: "This field shall be present if and only if any one or
265 	 * more of the S, PN and E flags are set.".
266 	 *
267 	 * If any of the bit is set, then the remaining ones also have to be
268 	 * set.
269 	 */
270 	if (gtp1->flags & GTP1_F_MASK)
271 		hdrlen += 4;
272 
273 	/* Make sure the header is larger enough, including extensions. */
274 	if (!pskb_may_pull(skb, hdrlen))
275 		return -1;
276 
277 	gtp1 = (struct gtp1_header *)(skb->data + sizeof(struct udphdr));
278 
279 	pctx = gtp1_pdp_find(gtp, ntohl(gtp1->tid));
280 	if (!pctx) {
281 		netdev_dbg(gtp->dev, "No PDP ctx to decap skb=%p\n", skb);
282 		return 1;
283 	}
284 
285 	return gtp_rx(pctx, skb, hdrlen, gtp->role);
286 }
287 
288 static void __gtp_encap_destroy(struct sock *sk)
289 {
290 	struct gtp_dev *gtp;
291 
292 	lock_sock(sk);
293 	gtp = sk->sk_user_data;
294 	if (gtp) {
295 		if (gtp->sk0 == sk)
296 			gtp->sk0 = NULL;
297 		else
298 			gtp->sk1u = NULL;
299 		udp_sk(sk)->encap_type = 0;
300 		rcu_assign_sk_user_data(sk, NULL);
301 		sock_put(sk);
302 	}
303 	release_sock(sk);
304 }
305 
306 static void gtp_encap_destroy(struct sock *sk)
307 {
308 	rtnl_lock();
309 	__gtp_encap_destroy(sk);
310 	rtnl_unlock();
311 }
312 
313 static void gtp_encap_disable_sock(struct sock *sk)
314 {
315 	if (!sk)
316 		return;
317 
318 	__gtp_encap_destroy(sk);
319 }
320 
321 static void gtp_encap_disable(struct gtp_dev *gtp)
322 {
323 	gtp_encap_disable_sock(gtp->sk0);
324 	gtp_encap_disable_sock(gtp->sk1u);
325 }
326 
327 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
328  * Return codes: 0: success, <0: error, >0: pass up to userspace UDP socket.
329  */
330 static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb)
331 {
332 	struct gtp_dev *gtp;
333 	int ret = 0;
334 
335 	gtp = rcu_dereference_sk_user_data(sk);
336 	if (!gtp)
337 		return 1;
338 
339 	netdev_dbg(gtp->dev, "encap_recv sk=%p\n", sk);
340 
341 	switch (udp_sk(sk)->encap_type) {
342 	case UDP_ENCAP_GTP0:
343 		netdev_dbg(gtp->dev, "received GTP0 packet\n");
344 		ret = gtp0_udp_encap_recv(gtp, skb);
345 		break;
346 	case UDP_ENCAP_GTP1U:
347 		netdev_dbg(gtp->dev, "received GTP1U packet\n");
348 		ret = gtp1u_udp_encap_recv(gtp, skb);
349 		break;
350 	default:
351 		ret = -1; /* Shouldn't happen. */
352 	}
353 
354 	switch (ret) {
355 	case 1:
356 		netdev_dbg(gtp->dev, "pass up to the process\n");
357 		break;
358 	case 0:
359 		break;
360 	case -1:
361 		netdev_dbg(gtp->dev, "GTP packet has been dropped\n");
362 		kfree_skb(skb);
363 		ret = 0;
364 		break;
365 	}
366 
367 	return ret;
368 }
369 
370 static int gtp_dev_init(struct net_device *dev)
371 {
372 	struct gtp_dev *gtp = netdev_priv(dev);
373 
374 	gtp->dev = dev;
375 
376 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
377 	if (!dev->tstats)
378 		return -ENOMEM;
379 
380 	return 0;
381 }
382 
383 static void gtp_dev_uninit(struct net_device *dev)
384 {
385 	struct gtp_dev *gtp = netdev_priv(dev);
386 
387 	gtp_encap_disable(gtp);
388 	free_percpu(dev->tstats);
389 }
390 
391 static struct rtable *ip4_route_output_gtp(struct flowi4 *fl4,
392 					   const struct sock *sk,
393 					   __be32 daddr)
394 {
395 	memset(fl4, 0, sizeof(*fl4));
396 	fl4->flowi4_oif		= sk->sk_bound_dev_if;
397 	fl4->daddr		= daddr;
398 	fl4->saddr		= inet_sk(sk)->inet_saddr;
399 	fl4->flowi4_tos		= RT_CONN_FLAGS(sk);
400 	fl4->flowi4_proto	= sk->sk_protocol;
401 
402 	return ip_route_output_key(sock_net(sk), fl4);
403 }
404 
405 static inline void gtp0_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
406 {
407 	int payload_len = skb->len;
408 	struct gtp0_header *gtp0;
409 
410 	gtp0 = skb_push(skb, sizeof(*gtp0));
411 
412 	gtp0->flags	= 0x1e; /* v0, GTP-non-prime. */
413 	gtp0->type	= GTP_TPDU;
414 	gtp0->length	= htons(payload_len);
415 	gtp0->seq	= htons((atomic_inc_return(&pctx->tx_seq) - 1) % 0xffff);
416 	gtp0->flow	= htons(pctx->u.v0.flow);
417 	gtp0->number	= 0xff;
418 	gtp0->spare[0]	= gtp0->spare[1] = gtp0->spare[2] = 0xff;
419 	gtp0->tid	= cpu_to_be64(pctx->u.v0.tid);
420 }
421 
422 static inline void gtp1_push_header(struct sk_buff *skb, struct pdp_ctx *pctx)
423 {
424 	int payload_len = skb->len;
425 	struct gtp1_header *gtp1;
426 
427 	gtp1 = skb_push(skb, sizeof(*gtp1));
428 
429 	/* Bits    8  7  6  5  4  3  2	1
430 	 *	  +--+--+--+--+--+--+--+--+
431 	 *	  |version |PT| 0| E| S|PN|
432 	 *	  +--+--+--+--+--+--+--+--+
433 	 *	    0  0  1  1	1  0  0  0
434 	 */
435 	gtp1->flags	= 0x30; /* v1, GTP-non-prime. */
436 	gtp1->type	= GTP_TPDU;
437 	gtp1->length	= htons(payload_len);
438 	gtp1->tid	= htonl(pctx->u.v1.o_tei);
439 
440 	/* TODO: Support for extension header, sequence number and N-PDU.
441 	 *	 Update the length field if any of them is available.
442 	 */
443 }
444 
445 struct gtp_pktinfo {
446 	struct sock		*sk;
447 	struct iphdr		*iph;
448 	struct flowi4		fl4;
449 	struct rtable		*rt;
450 	struct pdp_ctx		*pctx;
451 	struct net_device	*dev;
452 	__be16			gtph_port;
453 };
454 
455 static void gtp_push_header(struct sk_buff *skb, struct gtp_pktinfo *pktinfo)
456 {
457 	switch (pktinfo->pctx->gtp_version) {
458 	case GTP_V0:
459 		pktinfo->gtph_port = htons(GTP0_PORT);
460 		gtp0_push_header(skb, pktinfo->pctx);
461 		break;
462 	case GTP_V1:
463 		pktinfo->gtph_port = htons(GTP1U_PORT);
464 		gtp1_push_header(skb, pktinfo->pctx);
465 		break;
466 	}
467 }
468 
469 static inline void gtp_set_pktinfo_ipv4(struct gtp_pktinfo *pktinfo,
470 					struct sock *sk, struct iphdr *iph,
471 					struct pdp_ctx *pctx, struct rtable *rt,
472 					struct flowi4 *fl4,
473 					struct net_device *dev)
474 {
475 	pktinfo->sk	= sk;
476 	pktinfo->iph	= iph;
477 	pktinfo->pctx	= pctx;
478 	pktinfo->rt	= rt;
479 	pktinfo->fl4	= *fl4;
480 	pktinfo->dev	= dev;
481 }
482 
483 static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev,
484 			     struct gtp_pktinfo *pktinfo)
485 {
486 	struct gtp_dev *gtp = netdev_priv(dev);
487 	struct pdp_ctx *pctx;
488 	struct rtable *rt;
489 	struct flowi4 fl4;
490 	struct iphdr *iph;
491 	__be16 df;
492 	int mtu;
493 
494 	/* Read the IP destination address and resolve the PDP context.
495 	 * Prepend PDP header with TEI/TID from PDP ctx.
496 	 */
497 	iph = ip_hdr(skb);
498 	if (gtp->role == GTP_ROLE_SGSN)
499 		pctx = ipv4_pdp_find(gtp, iph->saddr);
500 	else
501 		pctx = ipv4_pdp_find(gtp, iph->daddr);
502 
503 	if (!pctx) {
504 		netdev_dbg(dev, "no PDP ctx found for %pI4, skip\n",
505 			   &iph->daddr);
506 		return -ENOENT;
507 	}
508 	netdev_dbg(dev, "found PDP context %p\n", pctx);
509 
510 	rt = ip4_route_output_gtp(&fl4, pctx->sk, pctx->peer_addr_ip4.s_addr);
511 	if (IS_ERR(rt)) {
512 		netdev_dbg(dev, "no route to SSGN %pI4\n",
513 			   &pctx->peer_addr_ip4.s_addr);
514 		dev->stats.tx_carrier_errors++;
515 		goto err;
516 	}
517 
518 	if (rt->dst.dev == dev) {
519 		netdev_dbg(dev, "circular route to SSGN %pI4\n",
520 			   &pctx->peer_addr_ip4.s_addr);
521 		dev->stats.collisions++;
522 		goto err_rt;
523 	}
524 
525 	/* This is similar to tnl_update_pmtu(). */
526 	df = iph->frag_off;
527 	if (df) {
528 		mtu = dst_mtu(&rt->dst) - dev->hard_header_len -
529 			sizeof(struct iphdr) - sizeof(struct udphdr);
530 		switch (pctx->gtp_version) {
531 		case GTP_V0:
532 			mtu -= sizeof(struct gtp0_header);
533 			break;
534 		case GTP_V1:
535 			mtu -= sizeof(struct gtp1_header);
536 			break;
537 		}
538 	} else {
539 		mtu = dst_mtu(&rt->dst);
540 	}
541 
542 	skb_dst_update_pmtu_no_confirm(skb, mtu);
543 
544 	if (!skb_is_gso(skb) && (iph->frag_off & htons(IP_DF)) &&
545 	    mtu < ntohs(iph->tot_len)) {
546 		netdev_dbg(dev, "packet too big, fragmentation needed\n");
547 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
548 			      htonl(mtu));
549 		goto err_rt;
550 	}
551 
552 	gtp_set_pktinfo_ipv4(pktinfo, pctx->sk, iph, pctx, rt, &fl4, dev);
553 	gtp_push_header(skb, pktinfo);
554 
555 	return 0;
556 err_rt:
557 	ip_rt_put(rt);
558 err:
559 	return -EBADMSG;
560 }
561 
562 static netdev_tx_t gtp_dev_xmit(struct sk_buff *skb, struct net_device *dev)
563 {
564 	unsigned int proto = ntohs(skb->protocol);
565 	struct gtp_pktinfo pktinfo;
566 	int err;
567 
568 	/* Ensure there is sufficient headroom. */
569 	if (skb_cow_head(skb, dev->needed_headroom))
570 		goto tx_err;
571 
572 	skb_reset_inner_headers(skb);
573 
574 	/* PDP context lookups in gtp_build_skb_*() need rcu read-side lock. */
575 	rcu_read_lock();
576 	switch (proto) {
577 	case ETH_P_IP:
578 		err = gtp_build_skb_ip4(skb, dev, &pktinfo);
579 		break;
580 	default:
581 		err = -EOPNOTSUPP;
582 		break;
583 	}
584 	rcu_read_unlock();
585 
586 	if (err < 0)
587 		goto tx_err;
588 
589 	switch (proto) {
590 	case ETH_P_IP:
591 		netdev_dbg(pktinfo.dev, "gtp -> IP src: %pI4 dst: %pI4\n",
592 			   &pktinfo.iph->saddr, &pktinfo.iph->daddr);
593 		udp_tunnel_xmit_skb(pktinfo.rt, pktinfo.sk, skb,
594 				    pktinfo.fl4.saddr, pktinfo.fl4.daddr,
595 				    pktinfo.iph->tos,
596 				    ip4_dst_hoplimit(&pktinfo.rt->dst),
597 				    0,
598 				    pktinfo.gtph_port, pktinfo.gtph_port,
599 				    !net_eq(sock_net(pktinfo.pctx->sk),
600 					    dev_net(dev)),
601 				    false);
602 		break;
603 	}
604 
605 	return NETDEV_TX_OK;
606 tx_err:
607 	dev->stats.tx_errors++;
608 	dev_kfree_skb(skb);
609 	return NETDEV_TX_OK;
610 }
611 
612 static const struct net_device_ops gtp_netdev_ops = {
613 	.ndo_init		= gtp_dev_init,
614 	.ndo_uninit		= gtp_dev_uninit,
615 	.ndo_start_xmit		= gtp_dev_xmit,
616 	.ndo_get_stats64	= dev_get_tstats64,
617 };
618 
619 static const struct device_type gtp_type = {
620 	.name = "gtp",
621 };
622 
623 static void gtp_link_setup(struct net_device *dev)
624 {
625 	unsigned int max_gtp_header_len = sizeof(struct iphdr) +
626 					  sizeof(struct udphdr) +
627 					  sizeof(struct gtp0_header);
628 
629 	dev->netdev_ops		= &gtp_netdev_ops;
630 	dev->needs_free_netdev	= true;
631 	SET_NETDEV_DEVTYPE(dev, &gtp_type);
632 
633 	dev->hard_header_len = 0;
634 	dev->addr_len = 0;
635 	dev->mtu = ETH_DATA_LEN - max_gtp_header_len;
636 
637 	/* Zero header length. */
638 	dev->type = ARPHRD_NONE;
639 	dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
640 
641 	dev->priv_flags	|= IFF_NO_QUEUE;
642 	dev->features	|= NETIF_F_LLTX;
643 	netif_keep_dst(dev);
644 
645 	dev->needed_headroom	= LL_MAX_HEADER + max_gtp_header_len;
646 }
647 
648 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize);
649 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]);
650 
651 static void gtp_destructor(struct net_device *dev)
652 {
653 	struct gtp_dev *gtp = netdev_priv(dev);
654 
655 	kfree(gtp->addr_hash);
656 	kfree(gtp->tid_hash);
657 }
658 
659 static int gtp_newlink(struct net *src_net, struct net_device *dev,
660 		       struct nlattr *tb[], struct nlattr *data[],
661 		       struct netlink_ext_ack *extack)
662 {
663 	struct gtp_dev *gtp;
664 	struct gtp_net *gn;
665 	int hashsize, err;
666 
667 	if (!data[IFLA_GTP_FD0] && !data[IFLA_GTP_FD1])
668 		return -EINVAL;
669 
670 	gtp = netdev_priv(dev);
671 
672 	if (!data[IFLA_GTP_PDP_HASHSIZE]) {
673 		hashsize = 1024;
674 	} else {
675 		hashsize = nla_get_u32(data[IFLA_GTP_PDP_HASHSIZE]);
676 		if (!hashsize)
677 			hashsize = 1024;
678 	}
679 
680 	err = gtp_hashtable_new(gtp, hashsize);
681 	if (err < 0)
682 		return err;
683 
684 	err = gtp_encap_enable(gtp, data);
685 	if (err < 0)
686 		goto out_hashtable;
687 
688 	err = register_netdevice(dev);
689 	if (err < 0) {
690 		netdev_dbg(dev, "failed to register new netdev %d\n", err);
691 		goto out_encap;
692 	}
693 
694 	gn = net_generic(dev_net(dev), gtp_net_id);
695 	list_add_rcu(&gtp->list, &gn->gtp_dev_list);
696 	dev->priv_destructor = gtp_destructor;
697 
698 	netdev_dbg(dev, "registered new GTP interface\n");
699 
700 	return 0;
701 
702 out_encap:
703 	gtp_encap_disable(gtp);
704 out_hashtable:
705 	kfree(gtp->addr_hash);
706 	kfree(gtp->tid_hash);
707 	return err;
708 }
709 
710 static void gtp_dellink(struct net_device *dev, struct list_head *head)
711 {
712 	struct gtp_dev *gtp = netdev_priv(dev);
713 	struct pdp_ctx *pctx;
714 	int i;
715 
716 	for (i = 0; i < gtp->hash_size; i++)
717 		hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i], hlist_tid)
718 			pdp_context_delete(pctx);
719 
720 	list_del_rcu(&gtp->list);
721 	unregister_netdevice_queue(dev, head);
722 }
723 
724 static const struct nla_policy gtp_policy[IFLA_GTP_MAX + 1] = {
725 	[IFLA_GTP_FD0]			= { .type = NLA_U32 },
726 	[IFLA_GTP_FD1]			= { .type = NLA_U32 },
727 	[IFLA_GTP_PDP_HASHSIZE]		= { .type = NLA_U32 },
728 	[IFLA_GTP_ROLE]			= { .type = NLA_U32 },
729 };
730 
731 static int gtp_validate(struct nlattr *tb[], struct nlattr *data[],
732 			struct netlink_ext_ack *extack)
733 {
734 	if (!data)
735 		return -EINVAL;
736 
737 	return 0;
738 }
739 
740 static size_t gtp_get_size(const struct net_device *dev)
741 {
742 	return nla_total_size(sizeof(__u32)) + /* IFLA_GTP_PDP_HASHSIZE */
743 		nla_total_size(sizeof(__u32)); /* IFLA_GTP_ROLE */
744 }
745 
746 static int gtp_fill_info(struct sk_buff *skb, const struct net_device *dev)
747 {
748 	struct gtp_dev *gtp = netdev_priv(dev);
749 
750 	if (nla_put_u32(skb, IFLA_GTP_PDP_HASHSIZE, gtp->hash_size))
751 		goto nla_put_failure;
752 	if (nla_put_u32(skb, IFLA_GTP_ROLE, gtp->role))
753 		goto nla_put_failure;
754 
755 	return 0;
756 
757 nla_put_failure:
758 	return -EMSGSIZE;
759 }
760 
761 static struct rtnl_link_ops gtp_link_ops __read_mostly = {
762 	.kind		= "gtp",
763 	.maxtype	= IFLA_GTP_MAX,
764 	.policy		= gtp_policy,
765 	.priv_size	= sizeof(struct gtp_dev),
766 	.setup		= gtp_link_setup,
767 	.validate	= gtp_validate,
768 	.newlink	= gtp_newlink,
769 	.dellink	= gtp_dellink,
770 	.get_size	= gtp_get_size,
771 	.fill_info	= gtp_fill_info,
772 };
773 
774 static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize)
775 {
776 	int i;
777 
778 	gtp->addr_hash = kmalloc_array(hsize, sizeof(struct hlist_head),
779 				       GFP_KERNEL | __GFP_NOWARN);
780 	if (gtp->addr_hash == NULL)
781 		return -ENOMEM;
782 
783 	gtp->tid_hash = kmalloc_array(hsize, sizeof(struct hlist_head),
784 				      GFP_KERNEL | __GFP_NOWARN);
785 	if (gtp->tid_hash == NULL)
786 		goto err1;
787 
788 	gtp->hash_size = hsize;
789 
790 	for (i = 0; i < hsize; i++) {
791 		INIT_HLIST_HEAD(&gtp->addr_hash[i]);
792 		INIT_HLIST_HEAD(&gtp->tid_hash[i]);
793 	}
794 	return 0;
795 err1:
796 	kfree(gtp->addr_hash);
797 	return -ENOMEM;
798 }
799 
800 static struct sock *gtp_encap_enable_socket(int fd, int type,
801 					    struct gtp_dev *gtp)
802 {
803 	struct udp_tunnel_sock_cfg tuncfg = {NULL};
804 	struct socket *sock;
805 	struct sock *sk;
806 	int err;
807 
808 	pr_debug("enable gtp on %d, %d\n", fd, type);
809 
810 	sock = sockfd_lookup(fd, &err);
811 	if (!sock) {
812 		pr_debug("gtp socket fd=%d not found\n", fd);
813 		return NULL;
814 	}
815 
816 	sk = sock->sk;
817 	if (sk->sk_protocol != IPPROTO_UDP ||
818 	    sk->sk_type != SOCK_DGRAM ||
819 	    (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) {
820 		pr_debug("socket fd=%d not UDP\n", fd);
821 		sk = ERR_PTR(-EINVAL);
822 		goto out_sock;
823 	}
824 
825 	lock_sock(sk);
826 	if (sk->sk_user_data) {
827 		sk = ERR_PTR(-EBUSY);
828 		goto out_rel_sock;
829 	}
830 
831 	sock_hold(sk);
832 
833 	tuncfg.sk_user_data = gtp;
834 	tuncfg.encap_type = type;
835 	tuncfg.encap_rcv = gtp_encap_recv;
836 	tuncfg.encap_destroy = gtp_encap_destroy;
837 
838 	setup_udp_tunnel_sock(sock_net(sock->sk), sock, &tuncfg);
839 
840 out_rel_sock:
841 	release_sock(sock->sk);
842 out_sock:
843 	sockfd_put(sock);
844 	return sk;
845 }
846 
847 static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[])
848 {
849 	struct sock *sk1u = NULL;
850 	struct sock *sk0 = NULL;
851 	unsigned int role = GTP_ROLE_GGSN;
852 
853 	if (data[IFLA_GTP_FD0]) {
854 		u32 fd0 = nla_get_u32(data[IFLA_GTP_FD0]);
855 
856 		sk0 = gtp_encap_enable_socket(fd0, UDP_ENCAP_GTP0, gtp);
857 		if (IS_ERR(sk0))
858 			return PTR_ERR(sk0);
859 	}
860 
861 	if (data[IFLA_GTP_FD1]) {
862 		u32 fd1 = nla_get_u32(data[IFLA_GTP_FD1]);
863 
864 		sk1u = gtp_encap_enable_socket(fd1, UDP_ENCAP_GTP1U, gtp);
865 		if (IS_ERR(sk1u)) {
866 			gtp_encap_disable_sock(sk0);
867 			return PTR_ERR(sk1u);
868 		}
869 	}
870 
871 	if (data[IFLA_GTP_ROLE]) {
872 		role = nla_get_u32(data[IFLA_GTP_ROLE]);
873 		if (role > GTP_ROLE_SGSN) {
874 			gtp_encap_disable_sock(sk0);
875 			gtp_encap_disable_sock(sk1u);
876 			return -EINVAL;
877 		}
878 	}
879 
880 	gtp->sk0 = sk0;
881 	gtp->sk1u = sk1u;
882 	gtp->role = role;
883 
884 	return 0;
885 }
886 
887 static struct gtp_dev *gtp_find_dev(struct net *src_net, struct nlattr *nla[])
888 {
889 	struct gtp_dev *gtp = NULL;
890 	struct net_device *dev;
891 	struct net *net;
892 
893 	/* Examine the link attributes and figure out which network namespace
894 	 * we are talking about.
895 	 */
896 	if (nla[GTPA_NET_NS_FD])
897 		net = get_net_ns_by_fd(nla_get_u32(nla[GTPA_NET_NS_FD]));
898 	else
899 		net = get_net(src_net);
900 
901 	if (IS_ERR(net))
902 		return NULL;
903 
904 	/* Check if there's an existing gtpX device to configure */
905 	dev = dev_get_by_index_rcu(net, nla_get_u32(nla[GTPA_LINK]));
906 	if (dev && dev->netdev_ops == &gtp_netdev_ops)
907 		gtp = netdev_priv(dev);
908 
909 	put_net(net);
910 	return gtp;
911 }
912 
913 static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info)
914 {
915 	pctx->gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]);
916 	pctx->af = AF_INET;
917 	pctx->peer_addr_ip4.s_addr =
918 		nla_get_be32(info->attrs[GTPA_PEER_ADDRESS]);
919 	pctx->ms_addr_ip4.s_addr =
920 		nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
921 
922 	switch (pctx->gtp_version) {
923 	case GTP_V0:
924 		/* According to TS 09.60, sections 7.5.1 and 7.5.2, the flow
925 		 * label needs to be the same for uplink and downlink packets,
926 		 * so let's annotate this.
927 		 */
928 		pctx->u.v0.tid = nla_get_u64(info->attrs[GTPA_TID]);
929 		pctx->u.v0.flow = nla_get_u16(info->attrs[GTPA_FLOW]);
930 		break;
931 	case GTP_V1:
932 		pctx->u.v1.i_tei = nla_get_u32(info->attrs[GTPA_I_TEI]);
933 		pctx->u.v1.o_tei = nla_get_u32(info->attrs[GTPA_O_TEI]);
934 		break;
935 	default:
936 		break;
937 	}
938 }
939 
940 static struct pdp_ctx *gtp_pdp_add(struct gtp_dev *gtp, struct sock *sk,
941 				   struct genl_info *info)
942 {
943 	struct pdp_ctx *pctx, *pctx_tid = NULL;
944 	struct net_device *dev = gtp->dev;
945 	u32 hash_ms, hash_tid = 0;
946 	unsigned int version;
947 	bool found = false;
948 	__be32 ms_addr;
949 
950 	ms_addr = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]);
951 	hash_ms = ipv4_hashfn(ms_addr) % gtp->hash_size;
952 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
953 
954 	pctx = ipv4_pdp_find(gtp, ms_addr);
955 	if (pctx)
956 		found = true;
957 	if (version == GTP_V0)
958 		pctx_tid = gtp0_pdp_find(gtp,
959 					 nla_get_u64(info->attrs[GTPA_TID]));
960 	else if (version == GTP_V1)
961 		pctx_tid = gtp1_pdp_find(gtp,
962 					 nla_get_u32(info->attrs[GTPA_I_TEI]));
963 	if (pctx_tid)
964 		found = true;
965 
966 	if (found) {
967 		if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
968 			return ERR_PTR(-EEXIST);
969 		if (info->nlhdr->nlmsg_flags & NLM_F_REPLACE)
970 			return ERR_PTR(-EOPNOTSUPP);
971 
972 		if (pctx && pctx_tid)
973 			return ERR_PTR(-EEXIST);
974 		if (!pctx)
975 			pctx = pctx_tid;
976 
977 		ipv4_pdp_fill(pctx, info);
978 
979 		if (pctx->gtp_version == GTP_V0)
980 			netdev_dbg(dev, "GTPv0-U: update tunnel id = %llx (pdp %p)\n",
981 				   pctx->u.v0.tid, pctx);
982 		else if (pctx->gtp_version == GTP_V1)
983 			netdev_dbg(dev, "GTPv1-U: update tunnel id = %x/%x (pdp %p)\n",
984 				   pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
985 
986 		return pctx;
987 
988 	}
989 
990 	pctx = kmalloc(sizeof(*pctx), GFP_ATOMIC);
991 	if (pctx == NULL)
992 		return ERR_PTR(-ENOMEM);
993 
994 	sock_hold(sk);
995 	pctx->sk = sk;
996 	pctx->dev = gtp->dev;
997 	ipv4_pdp_fill(pctx, info);
998 	atomic_set(&pctx->tx_seq, 0);
999 
1000 	switch (pctx->gtp_version) {
1001 	case GTP_V0:
1002 		/* TS 09.60: "The flow label identifies unambiguously a GTP
1003 		 * flow.". We use the tid for this instead, I cannot find a
1004 		 * situation in which this doesn't unambiguosly identify the
1005 		 * PDP context.
1006 		 */
1007 		hash_tid = gtp0_hashfn(pctx->u.v0.tid) % gtp->hash_size;
1008 		break;
1009 	case GTP_V1:
1010 		hash_tid = gtp1u_hashfn(pctx->u.v1.i_tei) % gtp->hash_size;
1011 		break;
1012 	}
1013 
1014 	hlist_add_head_rcu(&pctx->hlist_addr, &gtp->addr_hash[hash_ms]);
1015 	hlist_add_head_rcu(&pctx->hlist_tid, &gtp->tid_hash[hash_tid]);
1016 
1017 	switch (pctx->gtp_version) {
1018 	case GTP_V0:
1019 		netdev_dbg(dev, "GTPv0-U: new PDP ctx id=%llx ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
1020 			   pctx->u.v0.tid, &pctx->peer_addr_ip4,
1021 			   &pctx->ms_addr_ip4, pctx);
1022 		break;
1023 	case GTP_V1:
1024 		netdev_dbg(dev, "GTPv1-U: new PDP ctx id=%x/%x ssgn=%pI4 ms=%pI4 (pdp=%p)\n",
1025 			   pctx->u.v1.i_tei, pctx->u.v1.o_tei,
1026 			   &pctx->peer_addr_ip4, &pctx->ms_addr_ip4, pctx);
1027 		break;
1028 	}
1029 
1030 	return pctx;
1031 }
1032 
1033 static void pdp_context_free(struct rcu_head *head)
1034 {
1035 	struct pdp_ctx *pctx = container_of(head, struct pdp_ctx, rcu_head);
1036 
1037 	sock_put(pctx->sk);
1038 	kfree(pctx);
1039 }
1040 
1041 static void pdp_context_delete(struct pdp_ctx *pctx)
1042 {
1043 	hlist_del_rcu(&pctx->hlist_tid);
1044 	hlist_del_rcu(&pctx->hlist_addr);
1045 	call_rcu(&pctx->rcu_head, pdp_context_free);
1046 }
1047 
1048 static int gtp_tunnel_notify(struct pdp_ctx *pctx, u8 cmd, gfp_t allocation);
1049 
1050 static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
1051 {
1052 	unsigned int version;
1053 	struct pdp_ctx *pctx;
1054 	struct gtp_dev *gtp;
1055 	struct sock *sk;
1056 	int err;
1057 
1058 	if (!info->attrs[GTPA_VERSION] ||
1059 	    !info->attrs[GTPA_LINK] ||
1060 	    !info->attrs[GTPA_PEER_ADDRESS] ||
1061 	    !info->attrs[GTPA_MS_ADDRESS])
1062 		return -EINVAL;
1063 
1064 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
1065 
1066 	switch (version) {
1067 	case GTP_V0:
1068 		if (!info->attrs[GTPA_TID] ||
1069 		    !info->attrs[GTPA_FLOW])
1070 			return -EINVAL;
1071 		break;
1072 	case GTP_V1:
1073 		if (!info->attrs[GTPA_I_TEI] ||
1074 		    !info->attrs[GTPA_O_TEI])
1075 			return -EINVAL;
1076 		break;
1077 
1078 	default:
1079 		return -EINVAL;
1080 	}
1081 
1082 	rtnl_lock();
1083 
1084 	gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
1085 	if (!gtp) {
1086 		err = -ENODEV;
1087 		goto out_unlock;
1088 	}
1089 
1090 	if (version == GTP_V0)
1091 		sk = gtp->sk0;
1092 	else if (version == GTP_V1)
1093 		sk = gtp->sk1u;
1094 	else
1095 		sk = NULL;
1096 
1097 	if (!sk) {
1098 		err = -ENODEV;
1099 		goto out_unlock;
1100 	}
1101 
1102 	pctx = gtp_pdp_add(gtp, sk, info);
1103 	if (IS_ERR(pctx)) {
1104 		err = PTR_ERR(pctx);
1105 	} else {
1106 		gtp_tunnel_notify(pctx, GTP_CMD_NEWPDP, GFP_KERNEL);
1107 		err = 0;
1108 	}
1109 
1110 out_unlock:
1111 	rtnl_unlock();
1112 	return err;
1113 }
1114 
1115 static struct pdp_ctx *gtp_find_pdp_by_link(struct net *net,
1116 					    struct nlattr *nla[])
1117 {
1118 	struct gtp_dev *gtp;
1119 
1120 	gtp = gtp_find_dev(net, nla);
1121 	if (!gtp)
1122 		return ERR_PTR(-ENODEV);
1123 
1124 	if (nla[GTPA_MS_ADDRESS]) {
1125 		__be32 ip = nla_get_be32(nla[GTPA_MS_ADDRESS]);
1126 
1127 		return ipv4_pdp_find(gtp, ip);
1128 	} else if (nla[GTPA_VERSION]) {
1129 		u32 gtp_version = nla_get_u32(nla[GTPA_VERSION]);
1130 
1131 		if (gtp_version == GTP_V0 && nla[GTPA_TID])
1132 			return gtp0_pdp_find(gtp, nla_get_u64(nla[GTPA_TID]));
1133 		else if (gtp_version == GTP_V1 && nla[GTPA_I_TEI])
1134 			return gtp1_pdp_find(gtp, nla_get_u32(nla[GTPA_I_TEI]));
1135 	}
1136 
1137 	return ERR_PTR(-EINVAL);
1138 }
1139 
1140 static struct pdp_ctx *gtp_find_pdp(struct net *net, struct nlattr *nla[])
1141 {
1142 	struct pdp_ctx *pctx;
1143 
1144 	if (nla[GTPA_LINK])
1145 		pctx = gtp_find_pdp_by_link(net, nla);
1146 	else
1147 		pctx = ERR_PTR(-EINVAL);
1148 
1149 	if (!pctx)
1150 		pctx = ERR_PTR(-ENOENT);
1151 
1152 	return pctx;
1153 }
1154 
1155 static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
1156 {
1157 	struct pdp_ctx *pctx;
1158 	int err = 0;
1159 
1160 	if (!info->attrs[GTPA_VERSION])
1161 		return -EINVAL;
1162 
1163 	rcu_read_lock();
1164 
1165 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
1166 	if (IS_ERR(pctx)) {
1167 		err = PTR_ERR(pctx);
1168 		goto out_unlock;
1169 	}
1170 
1171 	if (pctx->gtp_version == GTP_V0)
1172 		netdev_dbg(pctx->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n",
1173 			   pctx->u.v0.tid, pctx);
1174 	else if (pctx->gtp_version == GTP_V1)
1175 		netdev_dbg(pctx->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n",
1176 			   pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx);
1177 
1178 	gtp_tunnel_notify(pctx, GTP_CMD_DELPDP, GFP_ATOMIC);
1179 	pdp_context_delete(pctx);
1180 
1181 out_unlock:
1182 	rcu_read_unlock();
1183 	return err;
1184 }
1185 
1186 static struct genl_family gtp_genl_family;
1187 
1188 enum gtp_multicast_groups {
1189 	GTP_GENL_MCGRP,
1190 };
1191 
1192 static const struct genl_multicast_group gtp_genl_mcgrps[] = {
1193 	[GTP_GENL_MCGRP] = { .name = GTP_GENL_MCGRP_NAME },
1194 };
1195 
1196 static int gtp_genl_fill_info(struct sk_buff *skb, u32 snd_portid, u32 snd_seq,
1197 			      int flags, u32 type, struct pdp_ctx *pctx)
1198 {
1199 	void *genlh;
1200 
1201 	genlh = genlmsg_put(skb, snd_portid, snd_seq, &gtp_genl_family, flags,
1202 			    type);
1203 	if (genlh == NULL)
1204 		goto nlmsg_failure;
1205 
1206 	if (nla_put_u32(skb, GTPA_VERSION, pctx->gtp_version) ||
1207 	    nla_put_u32(skb, GTPA_LINK, pctx->dev->ifindex) ||
1208 	    nla_put_be32(skb, GTPA_PEER_ADDRESS, pctx->peer_addr_ip4.s_addr) ||
1209 	    nla_put_be32(skb, GTPA_MS_ADDRESS, pctx->ms_addr_ip4.s_addr))
1210 		goto nla_put_failure;
1211 
1212 	switch (pctx->gtp_version) {
1213 	case GTP_V0:
1214 		if (nla_put_u64_64bit(skb, GTPA_TID, pctx->u.v0.tid, GTPA_PAD) ||
1215 		    nla_put_u16(skb, GTPA_FLOW, pctx->u.v0.flow))
1216 			goto nla_put_failure;
1217 		break;
1218 	case GTP_V1:
1219 		if (nla_put_u32(skb, GTPA_I_TEI, pctx->u.v1.i_tei) ||
1220 		    nla_put_u32(skb, GTPA_O_TEI, pctx->u.v1.o_tei))
1221 			goto nla_put_failure;
1222 		break;
1223 	}
1224 	genlmsg_end(skb, genlh);
1225 	return 0;
1226 
1227 nlmsg_failure:
1228 nla_put_failure:
1229 	genlmsg_cancel(skb, genlh);
1230 	return -EMSGSIZE;
1231 }
1232 
1233 static int gtp_tunnel_notify(struct pdp_ctx *pctx, u8 cmd, gfp_t allocation)
1234 {
1235 	struct sk_buff *msg;
1236 	int ret;
1237 
1238 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, allocation);
1239 	if (!msg)
1240 		return -ENOMEM;
1241 
1242 	ret = gtp_genl_fill_info(msg, 0, 0, 0, cmd, pctx);
1243 	if (ret < 0) {
1244 		nlmsg_free(msg);
1245 		return ret;
1246 	}
1247 
1248 	ret = genlmsg_multicast_netns(&gtp_genl_family, dev_net(pctx->dev), msg,
1249 				      0, GTP_GENL_MCGRP, GFP_ATOMIC);
1250 	return ret;
1251 }
1252 
1253 static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info)
1254 {
1255 	struct pdp_ctx *pctx = NULL;
1256 	struct sk_buff *skb2;
1257 	int err;
1258 
1259 	if (!info->attrs[GTPA_VERSION])
1260 		return -EINVAL;
1261 
1262 	rcu_read_lock();
1263 
1264 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
1265 	if (IS_ERR(pctx)) {
1266 		err = PTR_ERR(pctx);
1267 		goto err_unlock;
1268 	}
1269 
1270 	skb2 = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
1271 	if (skb2 == NULL) {
1272 		err = -ENOMEM;
1273 		goto err_unlock;
1274 	}
1275 
1276 	err = gtp_genl_fill_info(skb2, NETLINK_CB(skb).portid, info->snd_seq,
1277 				 0, info->nlhdr->nlmsg_type, pctx);
1278 	if (err < 0)
1279 		goto err_unlock_free;
1280 
1281 	rcu_read_unlock();
1282 	return genlmsg_unicast(genl_info_net(info), skb2, info->snd_portid);
1283 
1284 err_unlock_free:
1285 	kfree_skb(skb2);
1286 err_unlock:
1287 	rcu_read_unlock();
1288 	return err;
1289 }
1290 
1291 static int gtp_genl_dump_pdp(struct sk_buff *skb,
1292 				struct netlink_callback *cb)
1293 {
1294 	struct gtp_dev *last_gtp = (struct gtp_dev *)cb->args[2], *gtp;
1295 	int i, j, bucket = cb->args[0], skip = cb->args[1];
1296 	struct net *net = sock_net(skb->sk);
1297 	struct pdp_ctx *pctx;
1298 	struct gtp_net *gn;
1299 
1300 	gn = net_generic(net, gtp_net_id);
1301 
1302 	if (cb->args[4])
1303 		return 0;
1304 
1305 	rcu_read_lock();
1306 	list_for_each_entry_rcu(gtp, &gn->gtp_dev_list, list) {
1307 		if (last_gtp && last_gtp != gtp)
1308 			continue;
1309 		else
1310 			last_gtp = NULL;
1311 
1312 		for (i = bucket; i < gtp->hash_size; i++) {
1313 			j = 0;
1314 			hlist_for_each_entry_rcu(pctx, &gtp->tid_hash[i],
1315 						 hlist_tid) {
1316 				if (j >= skip &&
1317 				    gtp_genl_fill_info(skb,
1318 					    NETLINK_CB(cb->skb).portid,
1319 					    cb->nlh->nlmsg_seq,
1320 					    NLM_F_MULTI,
1321 					    cb->nlh->nlmsg_type, pctx)) {
1322 					cb->args[0] = i;
1323 					cb->args[1] = j;
1324 					cb->args[2] = (unsigned long)gtp;
1325 					goto out;
1326 				}
1327 				j++;
1328 			}
1329 			skip = 0;
1330 		}
1331 		bucket = 0;
1332 	}
1333 	cb->args[4] = 1;
1334 out:
1335 	rcu_read_unlock();
1336 	return skb->len;
1337 }
1338 
1339 static const struct nla_policy gtp_genl_policy[GTPA_MAX + 1] = {
1340 	[GTPA_LINK]		= { .type = NLA_U32, },
1341 	[GTPA_VERSION]		= { .type = NLA_U32, },
1342 	[GTPA_TID]		= { .type = NLA_U64, },
1343 	[GTPA_PEER_ADDRESS]	= { .type = NLA_U32, },
1344 	[GTPA_MS_ADDRESS]	= { .type = NLA_U32, },
1345 	[GTPA_FLOW]		= { .type = NLA_U16, },
1346 	[GTPA_NET_NS_FD]	= { .type = NLA_U32, },
1347 	[GTPA_I_TEI]		= { .type = NLA_U32, },
1348 	[GTPA_O_TEI]		= { .type = NLA_U32, },
1349 };
1350 
1351 static const struct genl_small_ops gtp_genl_ops[] = {
1352 	{
1353 		.cmd = GTP_CMD_NEWPDP,
1354 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1355 		.doit = gtp_genl_new_pdp,
1356 		.flags = GENL_ADMIN_PERM,
1357 	},
1358 	{
1359 		.cmd = GTP_CMD_DELPDP,
1360 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1361 		.doit = gtp_genl_del_pdp,
1362 		.flags = GENL_ADMIN_PERM,
1363 	},
1364 	{
1365 		.cmd = GTP_CMD_GETPDP,
1366 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1367 		.doit = gtp_genl_get_pdp,
1368 		.dumpit = gtp_genl_dump_pdp,
1369 		.flags = GENL_ADMIN_PERM,
1370 	},
1371 };
1372 
1373 static struct genl_family gtp_genl_family __ro_after_init = {
1374 	.name		= "gtp",
1375 	.version	= 0,
1376 	.hdrsize	= 0,
1377 	.maxattr	= GTPA_MAX,
1378 	.policy = gtp_genl_policy,
1379 	.netnsok	= true,
1380 	.module		= THIS_MODULE,
1381 	.small_ops	= gtp_genl_ops,
1382 	.n_small_ops	= ARRAY_SIZE(gtp_genl_ops),
1383 	.mcgrps		= gtp_genl_mcgrps,
1384 	.n_mcgrps	= ARRAY_SIZE(gtp_genl_mcgrps),
1385 };
1386 
1387 static int __net_init gtp_net_init(struct net *net)
1388 {
1389 	struct gtp_net *gn = net_generic(net, gtp_net_id);
1390 
1391 	INIT_LIST_HEAD(&gn->gtp_dev_list);
1392 	return 0;
1393 }
1394 
1395 static void __net_exit gtp_net_exit(struct net *net)
1396 {
1397 	struct gtp_net *gn = net_generic(net, gtp_net_id);
1398 	struct gtp_dev *gtp;
1399 	LIST_HEAD(list);
1400 
1401 	rtnl_lock();
1402 	list_for_each_entry(gtp, &gn->gtp_dev_list, list)
1403 		gtp_dellink(gtp->dev, &list);
1404 
1405 	unregister_netdevice_many(&list);
1406 	rtnl_unlock();
1407 }
1408 
1409 static struct pernet_operations gtp_net_ops = {
1410 	.init	= gtp_net_init,
1411 	.exit	= gtp_net_exit,
1412 	.id	= &gtp_net_id,
1413 	.size	= sizeof(struct gtp_net),
1414 };
1415 
1416 static int __init gtp_init(void)
1417 {
1418 	int err;
1419 
1420 	get_random_bytes(&gtp_h_initval, sizeof(gtp_h_initval));
1421 
1422 	err = rtnl_link_register(&gtp_link_ops);
1423 	if (err < 0)
1424 		goto error_out;
1425 
1426 	err = genl_register_family(&gtp_genl_family);
1427 	if (err < 0)
1428 		goto unreg_rtnl_link;
1429 
1430 	err = register_pernet_subsys(&gtp_net_ops);
1431 	if (err < 0)
1432 		goto unreg_genl_family;
1433 
1434 	pr_info("GTP module loaded (pdp ctx size %zd bytes)\n",
1435 		sizeof(struct pdp_ctx));
1436 	return 0;
1437 
1438 unreg_genl_family:
1439 	genl_unregister_family(&gtp_genl_family);
1440 unreg_rtnl_link:
1441 	rtnl_link_unregister(&gtp_link_ops);
1442 error_out:
1443 	pr_err("error loading GTP module loaded\n");
1444 	return err;
1445 }
1446 late_initcall(gtp_init);
1447 
1448 static void __exit gtp_fini(void)
1449 {
1450 	genl_unregister_family(&gtp_genl_family);
1451 	rtnl_link_unregister(&gtp_link_ops);
1452 	unregister_pernet_subsys(&gtp_net_ops);
1453 
1454 	pr_info("GTP module unloaded\n");
1455 }
1456 module_exit(gtp_fini);
1457 
1458 MODULE_LICENSE("GPL");
1459 MODULE_AUTHOR("Harald Welte <hwelte@sysmocom.de>");
1460 MODULE_DESCRIPTION("Interface driver for GTP encapsulated traffic");
1461 MODULE_ALIAS_RTNL_LINK("gtp");
1462 MODULE_ALIAS_GENL_FAMILY("gtp");
1463