xref: /openbmc/linux/drivers/net/geneve.c (revision 9cfc5c90)
1 /*
2  * GENEVE: Generic Network Virtualization Encapsulation
3  *
4  * Copyright (c) 2015 Red Hat, Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/netdevice.h>
16 #include <linux/etherdevice.h>
17 #include <linux/hash.h>
18 #include <net/dst_metadata.h>
19 #include <net/gro_cells.h>
20 #include <net/rtnetlink.h>
21 #include <net/geneve.h>
22 #include <net/protocol.h>
23 
24 #define GENEVE_NETDEV_VER	"0.6"
25 
26 #define GENEVE_UDP_PORT		6081
27 
28 #define GENEVE_N_VID		(1u << 24)
29 #define GENEVE_VID_MASK		(GENEVE_N_VID - 1)
30 
31 #define VNI_HASH_BITS		10
32 #define VNI_HASH_SIZE		(1<<VNI_HASH_BITS)
33 
34 static bool log_ecn_error = true;
35 module_param(log_ecn_error, bool, 0644);
36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
37 
38 #define GENEVE_VER 0
39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
40 
41 /* per-network namespace private data for this module */
42 struct geneve_net {
43 	struct list_head	geneve_list;
44 	struct list_head	sock_list;
45 };
46 
47 static int geneve_net_id;
48 
49 union geneve_addr {
50 	struct sockaddr_in sin;
51 	struct sockaddr_in6 sin6;
52 	struct sockaddr sa;
53 };
54 
55 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
56 
57 /* Pseudo network device */
58 struct geneve_dev {
59 	struct hlist_node  hlist;	/* vni hash table */
60 	struct net	   *net;	/* netns for packet i/o */
61 	struct net_device  *dev;	/* netdev for geneve tunnel */
62 	struct geneve_sock *sock4;	/* IPv4 socket used for geneve tunnel */
63 #if IS_ENABLED(CONFIG_IPV6)
64 	struct geneve_sock *sock6;	/* IPv6 socket used for geneve tunnel */
65 #endif
66 	u8                 vni[3];	/* virtual network ID for tunnel */
67 	u8                 ttl;		/* TTL override */
68 	u8                 tos;		/* TOS override */
69 	union geneve_addr  remote;	/* IP address for link partner */
70 	struct list_head   next;	/* geneve's per namespace list */
71 	__be16		   dst_port;
72 	bool		   collect_md;
73 	struct gro_cells   gro_cells;
74 };
75 
76 struct geneve_sock {
77 	bool			collect_md;
78 	struct list_head	list;
79 	struct socket		*sock;
80 	struct rcu_head		rcu;
81 	int			refcnt;
82 	struct udp_offload	udp_offloads;
83 	struct hlist_head	vni_list[VNI_HASH_SIZE];
84 };
85 
86 static inline __u32 geneve_net_vni_hash(u8 vni[3])
87 {
88 	__u32 vnid;
89 
90 	vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
91 	return hash_32(vnid, VNI_HASH_BITS);
92 }
93 
94 static __be64 vni_to_tunnel_id(const __u8 *vni)
95 {
96 #ifdef __BIG_ENDIAN
97 	return (vni[0] << 16) | (vni[1] << 8) | vni[2];
98 #else
99 	return (__force __be64)(((__force u64)vni[0] << 40) |
100 				((__force u64)vni[1] << 48) |
101 				((__force u64)vni[2] << 56));
102 #endif
103 }
104 
105 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
106 					__be32 addr, u8 vni[])
107 {
108 	struct hlist_head *vni_list_head;
109 	struct geneve_dev *geneve;
110 	__u32 hash;
111 
112 	/* Find the device for this VNI */
113 	hash = geneve_net_vni_hash(vni);
114 	vni_list_head = &gs->vni_list[hash];
115 	hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
116 		if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
117 		    addr == geneve->remote.sin.sin_addr.s_addr)
118 			return geneve;
119 	}
120 	return NULL;
121 }
122 
123 #if IS_ENABLED(CONFIG_IPV6)
124 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
125 					 struct in6_addr addr6, u8 vni[])
126 {
127 	struct hlist_head *vni_list_head;
128 	struct geneve_dev *geneve;
129 	__u32 hash;
130 
131 	/* Find the device for this VNI */
132 	hash = geneve_net_vni_hash(vni);
133 	vni_list_head = &gs->vni_list[hash];
134 	hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
135 		if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
136 		    ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
137 			return geneve;
138 	}
139 	return NULL;
140 }
141 #endif
142 
143 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
144 {
145 	return (struct genevehdr *)(udp_hdr(skb) + 1);
146 }
147 
148 /* geneve receive/decap routine */
149 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
150 {
151 	struct genevehdr *gnvh = geneve_hdr(skb);
152 	struct metadata_dst *tun_dst = NULL;
153 	struct geneve_dev *geneve = NULL;
154 	struct pcpu_sw_netstats *stats;
155 	struct iphdr *iph = NULL;
156 	__be32 addr;
157 	static u8 zero_vni[3];
158 	u8 *vni;
159 	int err = 0;
160 	sa_family_t sa_family;
161 #if IS_ENABLED(CONFIG_IPV6)
162 	struct ipv6hdr *ip6h = NULL;
163 	struct in6_addr addr6;
164 	static struct in6_addr zero_addr6;
165 #endif
166 
167 	sa_family = gs->sock->sk->sk_family;
168 
169 	if (sa_family == AF_INET) {
170 		iph = ip_hdr(skb); /* outer IP header... */
171 
172 		if (gs->collect_md) {
173 			vni = zero_vni;
174 			addr = 0;
175 		} else {
176 			vni = gnvh->vni;
177 
178 			addr = iph->saddr;
179 		}
180 
181 		geneve = geneve_lookup(gs, addr, vni);
182 #if IS_ENABLED(CONFIG_IPV6)
183 	} else if (sa_family == AF_INET6) {
184 		ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
185 
186 		if (gs->collect_md) {
187 			vni = zero_vni;
188 			addr6 = zero_addr6;
189 		} else {
190 			vni = gnvh->vni;
191 
192 			addr6 = ip6h->saddr;
193 		}
194 
195 		geneve = geneve6_lookup(gs, addr6, vni);
196 #endif
197 	}
198 	if (!geneve)
199 		goto drop;
200 
201 	if (ip_tunnel_collect_metadata() || gs->collect_md) {
202 		__be16 flags;
203 
204 		flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
205 			(gnvh->oam ? TUNNEL_OAM : 0) |
206 			(gnvh->critical ? TUNNEL_CRIT_OPT : 0);
207 
208 		tun_dst = udp_tun_rx_dst(skb, sa_family, flags,
209 					 vni_to_tunnel_id(gnvh->vni),
210 					 gnvh->opt_len * 4);
211 		if (!tun_dst)
212 			goto drop;
213 		/* Update tunnel dst according to Geneve options. */
214 		ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
215 					gnvh->options, gnvh->opt_len * 4);
216 	} else {
217 		/* Drop packets w/ critical options,
218 		 * since we don't support any...
219 		 */
220 		if (gnvh->critical)
221 			goto drop;
222 	}
223 
224 	skb_reset_mac_header(skb);
225 	skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
226 	skb->protocol = eth_type_trans(skb, geneve->dev);
227 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
228 
229 	if (tun_dst)
230 		skb_dst_set(skb, &tun_dst->dst);
231 
232 	/* Ignore packet loops (and multicast echo) */
233 	if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
234 		goto drop;
235 
236 	skb_reset_network_header(skb);
237 
238 	if (iph)
239 		err = IP_ECN_decapsulate(iph, skb);
240 #if IS_ENABLED(CONFIG_IPV6)
241 	if (ip6h)
242 		err = IP6_ECN_decapsulate(ip6h, skb);
243 #endif
244 
245 	if (unlikely(err)) {
246 		if (log_ecn_error) {
247 			if (iph)
248 				net_info_ratelimited("non-ECT from %pI4 "
249 						     "with TOS=%#x\n",
250 						     &iph->saddr, iph->tos);
251 #if IS_ENABLED(CONFIG_IPV6)
252 			if (ip6h)
253 				net_info_ratelimited("non-ECT from %pI6\n",
254 						     &ip6h->saddr);
255 #endif
256 		}
257 		if (err > 1) {
258 			++geneve->dev->stats.rx_frame_errors;
259 			++geneve->dev->stats.rx_errors;
260 			goto drop;
261 		}
262 	}
263 
264 	stats = this_cpu_ptr(geneve->dev->tstats);
265 	u64_stats_update_begin(&stats->syncp);
266 	stats->rx_packets++;
267 	stats->rx_bytes += skb->len;
268 	u64_stats_update_end(&stats->syncp);
269 
270 	gro_cells_receive(&geneve->gro_cells, skb);
271 	return;
272 drop:
273 	/* Consume bad packet */
274 	kfree_skb(skb);
275 }
276 
277 /* Setup stats when device is created */
278 static int geneve_init(struct net_device *dev)
279 {
280 	struct geneve_dev *geneve = netdev_priv(dev);
281 	int err;
282 
283 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
284 	if (!dev->tstats)
285 		return -ENOMEM;
286 
287 	err = gro_cells_init(&geneve->gro_cells, dev);
288 	if (err) {
289 		free_percpu(dev->tstats);
290 		return err;
291 	}
292 
293 	return 0;
294 }
295 
296 static void geneve_uninit(struct net_device *dev)
297 {
298 	struct geneve_dev *geneve = netdev_priv(dev);
299 
300 	gro_cells_destroy(&geneve->gro_cells);
301 	free_percpu(dev->tstats);
302 }
303 
304 /* Callback from net/ipv4/udp.c to receive packets */
305 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
306 {
307 	struct genevehdr *geneveh;
308 	struct geneve_sock *gs;
309 	int opts_len;
310 
311 	/* Need Geneve and inner Ethernet header to be present */
312 	if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
313 		goto error;
314 
315 	/* Return packets with reserved bits set */
316 	geneveh = geneve_hdr(skb);
317 	if (unlikely(geneveh->ver != GENEVE_VER))
318 		goto error;
319 
320 	if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
321 		goto error;
322 
323 	opts_len = geneveh->opt_len * 4;
324 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
325 				 htons(ETH_P_TEB)))
326 		goto drop;
327 
328 	gs = rcu_dereference_sk_user_data(sk);
329 	if (!gs)
330 		goto drop;
331 
332 	geneve_rx(gs, skb);
333 	return 0;
334 
335 drop:
336 	/* Consume bad packet */
337 	kfree_skb(skb);
338 	return 0;
339 
340 error:
341 	/* Let the UDP layer deal with the skb */
342 	return 1;
343 }
344 
345 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
346 					 __be16 port)
347 {
348 	struct socket *sock;
349 	struct udp_port_cfg udp_conf;
350 	int err;
351 
352 	memset(&udp_conf, 0, sizeof(udp_conf));
353 
354 	if (ipv6) {
355 		udp_conf.family = AF_INET6;
356 		udp_conf.ipv6_v6only = 1;
357 	} else {
358 		udp_conf.family = AF_INET;
359 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
360 	}
361 
362 	udp_conf.local_udp_port = port;
363 
364 	/* Open UDP socket */
365 	err = udp_sock_create(net, &udp_conf, &sock);
366 	if (err < 0)
367 		return ERR_PTR(err);
368 
369 	return sock;
370 }
371 
372 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
373 {
374 	struct sock *sk = gs->sock->sk;
375 	sa_family_t sa_family = sk->sk_family;
376 	int err;
377 
378 	if (sa_family == AF_INET) {
379 		err = udp_add_offload(&gs->udp_offloads);
380 		if (err)
381 			pr_warn("geneve: udp_add_offload failed with status %d\n",
382 				err);
383 	}
384 }
385 
386 static int geneve_hlen(struct genevehdr *gh)
387 {
388 	return sizeof(*gh) + gh->opt_len * 4;
389 }
390 
391 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
392 					   struct sk_buff *skb,
393 					   struct udp_offload *uoff)
394 {
395 	struct sk_buff *p, **pp = NULL;
396 	struct genevehdr *gh, *gh2;
397 	unsigned int hlen, gh_len, off_gnv;
398 	const struct packet_offload *ptype;
399 	__be16 type;
400 	int flush = 1;
401 
402 	off_gnv = skb_gro_offset(skb);
403 	hlen = off_gnv + sizeof(*gh);
404 	gh = skb_gro_header_fast(skb, off_gnv);
405 	if (skb_gro_header_hard(skb, hlen)) {
406 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
407 		if (unlikely(!gh))
408 			goto out;
409 	}
410 
411 	if (gh->ver != GENEVE_VER || gh->oam)
412 		goto out;
413 	gh_len = geneve_hlen(gh);
414 
415 	hlen = off_gnv + gh_len;
416 	if (skb_gro_header_hard(skb, hlen)) {
417 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
418 		if (unlikely(!gh))
419 			goto out;
420 	}
421 
422 	flush = 0;
423 
424 	for (p = *head; p; p = p->next) {
425 		if (!NAPI_GRO_CB(p)->same_flow)
426 			continue;
427 
428 		gh2 = (struct genevehdr *)(p->data + off_gnv);
429 		if (gh->opt_len != gh2->opt_len ||
430 		    memcmp(gh, gh2, gh_len)) {
431 			NAPI_GRO_CB(p)->same_flow = 0;
432 			continue;
433 		}
434 	}
435 
436 	type = gh->proto_type;
437 
438 	rcu_read_lock();
439 	ptype = gro_find_receive_by_type(type);
440 	if (!ptype) {
441 		flush = 1;
442 		goto out_unlock;
443 	}
444 
445 	skb_gro_pull(skb, gh_len);
446 	skb_gro_postpull_rcsum(skb, gh, gh_len);
447 	pp = ptype->callbacks.gro_receive(head, skb);
448 
449 out_unlock:
450 	rcu_read_unlock();
451 out:
452 	NAPI_GRO_CB(skb)->flush |= flush;
453 
454 	return pp;
455 }
456 
457 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
458 			       struct udp_offload *uoff)
459 {
460 	struct genevehdr *gh;
461 	struct packet_offload *ptype;
462 	__be16 type;
463 	int gh_len;
464 	int err = -ENOSYS;
465 
466 	udp_tunnel_gro_complete(skb, nhoff);
467 
468 	gh = (struct genevehdr *)(skb->data + nhoff);
469 	gh_len = geneve_hlen(gh);
470 	type = gh->proto_type;
471 
472 	rcu_read_lock();
473 	ptype = gro_find_complete_by_type(type);
474 	if (ptype)
475 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
476 
477 	rcu_read_unlock();
478 	return err;
479 }
480 
481 /* Create new listen socket if needed */
482 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
483 						bool ipv6)
484 {
485 	struct geneve_net *gn = net_generic(net, geneve_net_id);
486 	struct geneve_sock *gs;
487 	struct socket *sock;
488 	struct udp_tunnel_sock_cfg tunnel_cfg;
489 	int h;
490 
491 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
492 	if (!gs)
493 		return ERR_PTR(-ENOMEM);
494 
495 	sock = geneve_create_sock(net, ipv6, port);
496 	if (IS_ERR(sock)) {
497 		kfree(gs);
498 		return ERR_CAST(sock);
499 	}
500 
501 	gs->sock = sock;
502 	gs->refcnt = 1;
503 	for (h = 0; h < VNI_HASH_SIZE; ++h)
504 		INIT_HLIST_HEAD(&gs->vni_list[h]);
505 
506 	/* Initialize the geneve udp offloads structure */
507 	gs->udp_offloads.port = port;
508 	gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
509 	gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
510 	geneve_notify_add_rx_port(gs);
511 
512 	/* Mark socket as an encapsulation socket */
513 	tunnel_cfg.sk_user_data = gs;
514 	tunnel_cfg.encap_type = 1;
515 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
516 	tunnel_cfg.encap_destroy = NULL;
517 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
518 	list_add(&gs->list, &gn->sock_list);
519 	return gs;
520 }
521 
522 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
523 {
524 	struct sock *sk = gs->sock->sk;
525 	sa_family_t sa_family = sk->sk_family;
526 
527 	if (sa_family == AF_INET)
528 		udp_del_offload(&gs->udp_offloads);
529 }
530 
531 static void __geneve_sock_release(struct geneve_sock *gs)
532 {
533 	if (!gs || --gs->refcnt)
534 		return;
535 
536 	list_del(&gs->list);
537 	geneve_notify_del_rx_port(gs);
538 	udp_tunnel_sock_release(gs->sock);
539 	kfree_rcu(gs, rcu);
540 }
541 
542 static void geneve_sock_release(struct geneve_dev *geneve)
543 {
544 	__geneve_sock_release(geneve->sock4);
545 #if IS_ENABLED(CONFIG_IPV6)
546 	__geneve_sock_release(geneve->sock6);
547 #endif
548 }
549 
550 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
551 					    sa_family_t family,
552 					    __be16 dst_port)
553 {
554 	struct geneve_sock *gs;
555 
556 	list_for_each_entry(gs, &gn->sock_list, list) {
557 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
558 		    inet_sk(gs->sock->sk)->sk.sk_family == family) {
559 			return gs;
560 		}
561 	}
562 	return NULL;
563 }
564 
565 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
566 {
567 	struct net *net = geneve->net;
568 	struct geneve_net *gn = net_generic(net, geneve_net_id);
569 	struct geneve_sock *gs;
570 	__u32 hash;
571 
572 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
573 	if (gs) {
574 		gs->refcnt++;
575 		goto out;
576 	}
577 
578 	gs = geneve_socket_create(net, geneve->dst_port, ipv6);
579 	if (IS_ERR(gs))
580 		return PTR_ERR(gs);
581 
582 out:
583 	gs->collect_md = geneve->collect_md;
584 #if IS_ENABLED(CONFIG_IPV6)
585 	if (ipv6)
586 		geneve->sock6 = gs;
587 	else
588 #endif
589 		geneve->sock4 = gs;
590 
591 	hash = geneve_net_vni_hash(geneve->vni);
592 	hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
593 	return 0;
594 }
595 
596 static int geneve_open(struct net_device *dev)
597 {
598 	struct geneve_dev *geneve = netdev_priv(dev);
599 	bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
600 	bool metadata = geneve->collect_md;
601 	int ret = 0;
602 
603 	geneve->sock4 = NULL;
604 #if IS_ENABLED(CONFIG_IPV6)
605 	geneve->sock6 = NULL;
606 	if (ipv6 || metadata)
607 		ret = geneve_sock_add(geneve, true);
608 #endif
609 	if (!ret && (!ipv6 || metadata))
610 		ret = geneve_sock_add(geneve, false);
611 	if (ret < 0)
612 		geneve_sock_release(geneve);
613 
614 	return ret;
615 }
616 
617 static int geneve_stop(struct net_device *dev)
618 {
619 	struct geneve_dev *geneve = netdev_priv(dev);
620 
621 	if (!hlist_unhashed(&geneve->hlist))
622 		hlist_del_rcu(&geneve->hlist);
623 	geneve_sock_release(geneve);
624 	return 0;
625 }
626 
627 static void geneve_build_header(struct genevehdr *geneveh,
628 				__be16 tun_flags, u8 vni[3],
629 				u8 options_len, u8 *options)
630 {
631 	geneveh->ver = GENEVE_VER;
632 	geneveh->opt_len = options_len / 4;
633 	geneveh->oam = !!(tun_flags & TUNNEL_OAM);
634 	geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
635 	geneveh->rsvd1 = 0;
636 	memcpy(geneveh->vni, vni, 3);
637 	geneveh->proto_type = htons(ETH_P_TEB);
638 	geneveh->rsvd2 = 0;
639 
640 	memcpy(geneveh->options, options, options_len);
641 }
642 
643 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
644 			    __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
645 			    bool csum, bool xnet)
646 {
647 	struct genevehdr *gnvh;
648 	int min_headroom;
649 	int err;
650 
651 	skb_scrub_packet(skb, xnet);
652 
653 	min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
654 			+ GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
655 	err = skb_cow_head(skb, min_headroom);
656 	if (unlikely(err)) {
657 		kfree_skb(skb);
658 		goto free_rt;
659 	}
660 
661 	skb = udp_tunnel_handle_offloads(skb, csum);
662 	if (IS_ERR(skb)) {
663 		err = PTR_ERR(skb);
664 		goto free_rt;
665 	}
666 
667 	gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
668 	geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
669 
670 	skb_set_inner_protocol(skb, htons(ETH_P_TEB));
671 	return 0;
672 
673 free_rt:
674 	ip_rt_put(rt);
675 	return err;
676 }
677 
678 #if IS_ENABLED(CONFIG_IPV6)
679 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
680 			     __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
681 			     bool csum, bool xnet)
682 {
683 	struct genevehdr *gnvh;
684 	int min_headroom;
685 	int err;
686 
687 	skb_scrub_packet(skb, xnet);
688 
689 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
690 			+ GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
691 	err = skb_cow_head(skb, min_headroom);
692 	if (unlikely(err)) {
693 		kfree_skb(skb);
694 		goto free_dst;
695 	}
696 
697 	skb = udp_tunnel_handle_offloads(skb, csum);
698 	if (IS_ERR(skb)) {
699 		err = PTR_ERR(skb);
700 		goto free_dst;
701 	}
702 
703 	gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
704 	geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
705 
706 	skb_set_inner_protocol(skb, htons(ETH_P_TEB));
707 	return 0;
708 
709 free_dst:
710 	dst_release(dst);
711 	return err;
712 }
713 #endif
714 
715 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
716 				       struct net_device *dev,
717 				       struct flowi4 *fl4,
718 				       struct ip_tunnel_info *info)
719 {
720 	struct geneve_dev *geneve = netdev_priv(dev);
721 	struct rtable *rt = NULL;
722 	__u8 tos;
723 
724 	memset(fl4, 0, sizeof(*fl4));
725 	fl4->flowi4_mark = skb->mark;
726 	fl4->flowi4_proto = IPPROTO_UDP;
727 
728 	if (info) {
729 		fl4->daddr = info->key.u.ipv4.dst;
730 		fl4->saddr = info->key.u.ipv4.src;
731 		fl4->flowi4_tos = RT_TOS(info->key.tos);
732 	} else {
733 		tos = geneve->tos;
734 		if (tos == 1) {
735 			const struct iphdr *iip = ip_hdr(skb);
736 
737 			tos = ip_tunnel_get_dsfield(iip, skb);
738 		}
739 
740 		fl4->flowi4_tos = RT_TOS(tos);
741 		fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
742 	}
743 
744 	rt = ip_route_output_key(geneve->net, fl4);
745 	if (IS_ERR(rt)) {
746 		netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
747 		return ERR_PTR(-ENETUNREACH);
748 	}
749 	if (rt->dst.dev == dev) { /* is this necessary? */
750 		netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
751 		ip_rt_put(rt);
752 		return ERR_PTR(-ELOOP);
753 	}
754 	return rt;
755 }
756 
757 #if IS_ENABLED(CONFIG_IPV6)
758 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
759 					   struct net_device *dev,
760 					   struct flowi6 *fl6,
761 					   struct ip_tunnel_info *info)
762 {
763 	struct geneve_dev *geneve = netdev_priv(dev);
764 	struct geneve_sock *gs6 = geneve->sock6;
765 	struct dst_entry *dst = NULL;
766 	__u8 prio;
767 
768 	memset(fl6, 0, sizeof(*fl6));
769 	fl6->flowi6_mark = skb->mark;
770 	fl6->flowi6_proto = IPPROTO_UDP;
771 
772 	if (info) {
773 		fl6->daddr = info->key.u.ipv6.dst;
774 		fl6->saddr = info->key.u.ipv6.src;
775 		fl6->flowi6_tos = RT_TOS(info->key.tos);
776 	} else {
777 		prio = geneve->tos;
778 		if (prio == 1) {
779 			const struct iphdr *iip = ip_hdr(skb);
780 
781 			prio = ip_tunnel_get_dsfield(iip, skb);
782 		}
783 
784 		fl6->flowi6_tos = RT_TOS(prio);
785 		fl6->daddr = geneve->remote.sin6.sin6_addr;
786 	}
787 
788 	if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
789 		netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
790 		return ERR_PTR(-ENETUNREACH);
791 	}
792 	if (dst->dev == dev) { /* is this necessary? */
793 		netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
794 		dst_release(dst);
795 		return ERR_PTR(-ELOOP);
796 	}
797 
798 	return dst;
799 }
800 #endif
801 
802 /* Convert 64 bit tunnel ID to 24 bit VNI. */
803 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
804 {
805 #ifdef __BIG_ENDIAN
806 	vni[0] = (__force __u8)(tun_id >> 16);
807 	vni[1] = (__force __u8)(tun_id >> 8);
808 	vni[2] = (__force __u8)tun_id;
809 #else
810 	vni[0] = (__force __u8)((__force u64)tun_id >> 40);
811 	vni[1] = (__force __u8)((__force u64)tun_id >> 48);
812 	vni[2] = (__force __u8)((__force u64)tun_id >> 56);
813 #endif
814 }
815 
816 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
817 				   struct ip_tunnel_info *info)
818 {
819 	struct geneve_dev *geneve = netdev_priv(dev);
820 	struct geneve_sock *gs4 = geneve->sock4;
821 	struct rtable *rt = NULL;
822 	const struct iphdr *iip; /* interior IP header */
823 	int err = -EINVAL;
824 	struct flowi4 fl4;
825 	__u8 tos, ttl;
826 	__be16 sport;
827 	bool udp_csum;
828 	__be16 df;
829 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
830 
831 	if (geneve->collect_md) {
832 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
833 			netdev_dbg(dev, "no tunnel metadata\n");
834 			goto tx_error;
835 		}
836 		if (info && ip_tunnel_info_af(info) != AF_INET)
837 			goto tx_error;
838 	}
839 
840 	rt = geneve_get_v4_rt(skb, dev, &fl4, info);
841 	if (IS_ERR(rt)) {
842 		err = PTR_ERR(rt);
843 		goto tx_error;
844 	}
845 
846 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
847 	skb_reset_mac_header(skb);
848 
849 	iip = ip_hdr(skb);
850 
851 	if (info) {
852 		const struct ip_tunnel_key *key = &info->key;
853 		u8 *opts = NULL;
854 		u8 vni[3];
855 
856 		tunnel_id_to_vni(key->tun_id, vni);
857 		if (key->tun_flags & TUNNEL_GENEVE_OPT)
858 			opts = ip_tunnel_info_opts(info);
859 
860 		udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
861 		err = geneve_build_skb(rt, skb, key->tun_flags, vni,
862 				       info->options_len, opts, udp_csum, xnet);
863 		if (unlikely(err))
864 			goto err;
865 
866 		tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
867 		ttl = key->ttl;
868 		df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
869 	} else {
870 		udp_csum = false;
871 		err = geneve_build_skb(rt, skb, 0, geneve->vni,
872 				       0, NULL, udp_csum, xnet);
873 		if (unlikely(err))
874 			goto err;
875 
876 		tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
877 		ttl = geneve->ttl;
878 		if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
879 			ttl = 1;
880 		ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
881 		df = 0;
882 	}
883 	err = udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
884 				  tos, ttl, df, sport, geneve->dst_port,
885 				  !net_eq(geneve->net, dev_net(geneve->dev)),
886 				  !udp_csum);
887 
888 	iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
889 	return NETDEV_TX_OK;
890 
891 tx_error:
892 	dev_kfree_skb(skb);
893 err:
894 	if (err == -ELOOP)
895 		dev->stats.collisions++;
896 	else if (err == -ENETUNREACH)
897 		dev->stats.tx_carrier_errors++;
898 	else
899 		dev->stats.tx_errors++;
900 	return NETDEV_TX_OK;
901 }
902 
903 #if IS_ENABLED(CONFIG_IPV6)
904 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
905 				    struct ip_tunnel_info *info)
906 {
907 	struct geneve_dev *geneve = netdev_priv(dev);
908 	struct geneve_sock *gs6 = geneve->sock6;
909 	struct dst_entry *dst = NULL;
910 	const struct iphdr *iip; /* interior IP header */
911 	int err = -EINVAL;
912 	struct flowi6 fl6;
913 	__u8 prio, ttl;
914 	__be16 sport;
915 	bool udp_csum;
916 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
917 
918 	if (geneve->collect_md) {
919 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
920 			netdev_dbg(dev, "no tunnel metadata\n");
921 			goto tx_error;
922 		}
923 	}
924 
925 	dst = geneve_get_v6_dst(skb, dev, &fl6, info);
926 	if (IS_ERR(dst)) {
927 		err = PTR_ERR(dst);
928 		goto tx_error;
929 	}
930 
931 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
932 	skb_reset_mac_header(skb);
933 
934 	iip = ip_hdr(skb);
935 
936 	if (info) {
937 		const struct ip_tunnel_key *key = &info->key;
938 		u8 *opts = NULL;
939 		u8 vni[3];
940 
941 		tunnel_id_to_vni(key->tun_id, vni);
942 		if (key->tun_flags & TUNNEL_GENEVE_OPT)
943 			opts = ip_tunnel_info_opts(info);
944 
945 		udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
946 		err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
947 					info->options_len, opts,
948 					udp_csum, xnet);
949 		if (unlikely(err))
950 			goto err;
951 
952 		prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
953 		ttl = key->ttl;
954 	} else {
955 		udp_csum = false;
956 		err = geneve6_build_skb(dst, skb, 0, geneve->vni,
957 					0, NULL, udp_csum, xnet);
958 		if (unlikely(err))
959 			goto err;
960 
961 		prio = ip_tunnel_ecn_encap(fl6.flowi6_tos, iip, skb);
962 		ttl = geneve->ttl;
963 		if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
964 			ttl = 1;
965 		ttl = ttl ? : ip6_dst_hoplimit(dst);
966 	}
967 	err = udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
968 				   &fl6.saddr, &fl6.daddr, prio, ttl,
969 				   sport, geneve->dst_port, !udp_csum);
970 
971 	iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
972 	return NETDEV_TX_OK;
973 
974 tx_error:
975 	dev_kfree_skb(skb);
976 err:
977 	if (err == -ELOOP)
978 		dev->stats.collisions++;
979 	else if (err == -ENETUNREACH)
980 		dev->stats.tx_carrier_errors++;
981 	else
982 		dev->stats.tx_errors++;
983 	return NETDEV_TX_OK;
984 }
985 #endif
986 
987 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
988 {
989 	struct geneve_dev *geneve = netdev_priv(dev);
990 	struct ip_tunnel_info *info = NULL;
991 
992 	if (geneve->collect_md)
993 		info = skb_tunnel_info(skb);
994 
995 #if IS_ENABLED(CONFIG_IPV6)
996 	if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
997 	    (!info && geneve->remote.sa.sa_family == AF_INET6))
998 		return geneve6_xmit_skb(skb, dev, info);
999 #endif
1000 	return geneve_xmit_skb(skb, dev, info);
1001 }
1002 
1003 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1004 {
1005 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
1006 	struct geneve_dev *geneve = netdev_priv(dev);
1007 	struct rtable *rt;
1008 	struct flowi4 fl4;
1009 #if IS_ENABLED(CONFIG_IPV6)
1010 	struct dst_entry *dst;
1011 	struct flowi6 fl6;
1012 #endif
1013 
1014 	if (ip_tunnel_info_af(info) == AF_INET) {
1015 		rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1016 		if (IS_ERR(rt))
1017 			return PTR_ERR(rt);
1018 
1019 		ip_rt_put(rt);
1020 		info->key.u.ipv4.src = fl4.saddr;
1021 #if IS_ENABLED(CONFIG_IPV6)
1022 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
1023 		dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1024 		if (IS_ERR(dst))
1025 			return PTR_ERR(dst);
1026 
1027 		dst_release(dst);
1028 		info->key.u.ipv6.src = fl6.saddr;
1029 #endif
1030 	} else {
1031 		return -EINVAL;
1032 	}
1033 
1034 	info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1035 					     1, USHRT_MAX, true);
1036 	info->key.tp_dst = geneve->dst_port;
1037 	return 0;
1038 }
1039 
1040 static const struct net_device_ops geneve_netdev_ops = {
1041 	.ndo_init		= geneve_init,
1042 	.ndo_uninit		= geneve_uninit,
1043 	.ndo_open		= geneve_open,
1044 	.ndo_stop		= geneve_stop,
1045 	.ndo_start_xmit		= geneve_xmit,
1046 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1047 	.ndo_change_mtu		= eth_change_mtu,
1048 	.ndo_validate_addr	= eth_validate_addr,
1049 	.ndo_set_mac_address	= eth_mac_addr,
1050 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1051 };
1052 
1053 static void geneve_get_drvinfo(struct net_device *dev,
1054 			       struct ethtool_drvinfo *drvinfo)
1055 {
1056 	strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1057 	strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1058 }
1059 
1060 static const struct ethtool_ops geneve_ethtool_ops = {
1061 	.get_drvinfo	= geneve_get_drvinfo,
1062 	.get_link	= ethtool_op_get_link,
1063 };
1064 
1065 /* Info for udev, that this is a virtual tunnel endpoint */
1066 static struct device_type geneve_type = {
1067 	.name = "geneve",
1068 };
1069 
1070 /* Initialize the device structure. */
1071 static void geneve_setup(struct net_device *dev)
1072 {
1073 	ether_setup(dev);
1074 
1075 	dev->netdev_ops = &geneve_netdev_ops;
1076 	dev->ethtool_ops = &geneve_ethtool_ops;
1077 	dev->destructor = free_netdev;
1078 
1079 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1080 
1081 	dev->features    |= NETIF_F_LLTX;
1082 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1083 	dev->features    |= NETIF_F_RXCSUM;
1084 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1085 
1086 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1087 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1088 
1089 	netif_keep_dst(dev);
1090 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1091 	eth_hw_addr_random(dev);
1092 }
1093 
1094 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1095 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1096 	[IFLA_GENEVE_REMOTE]		= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1097 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1098 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1099 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1100 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1101 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1102 };
1103 
1104 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1105 {
1106 	if (tb[IFLA_ADDRESS]) {
1107 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1108 			return -EINVAL;
1109 
1110 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1111 			return -EADDRNOTAVAIL;
1112 	}
1113 
1114 	if (!data)
1115 		return -EINVAL;
1116 
1117 	if (data[IFLA_GENEVE_ID]) {
1118 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1119 
1120 		if (vni >= GENEVE_VID_MASK)
1121 			return -ERANGE;
1122 	}
1123 
1124 	return 0;
1125 }
1126 
1127 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1128 					  __be16 dst_port,
1129 					  union geneve_addr *remote,
1130 					  u8 vni[],
1131 					  bool *tun_on_same_port,
1132 					  bool *tun_collect_md)
1133 {
1134 	struct geneve_dev *geneve, *t;
1135 
1136 	*tun_on_same_port = false;
1137 	*tun_collect_md = false;
1138 	t = NULL;
1139 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1140 		if (geneve->dst_port == dst_port) {
1141 			*tun_collect_md = geneve->collect_md;
1142 			*tun_on_same_port = true;
1143 		}
1144 		if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1145 		    !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1146 		    dst_port == geneve->dst_port)
1147 			t = geneve;
1148 	}
1149 	return t;
1150 }
1151 
1152 static int geneve_configure(struct net *net, struct net_device *dev,
1153 			    union geneve_addr *remote,
1154 			    __u32 vni, __u8 ttl, __u8 tos, __be16 dst_port,
1155 			    bool metadata)
1156 {
1157 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1158 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1159 	bool tun_collect_md, tun_on_same_port;
1160 	int err;
1161 
1162 	if (!remote)
1163 		return -EINVAL;
1164 	if (metadata &&
1165 	    (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl))
1166 		return -EINVAL;
1167 
1168 	geneve->net = net;
1169 	geneve->dev = dev;
1170 
1171 	geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1172 	geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1173 	geneve->vni[2] =  vni & 0x000000ff;
1174 
1175 	if ((remote->sa.sa_family == AF_INET &&
1176 	     IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1177 	    (remote->sa.sa_family == AF_INET6 &&
1178 	     ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1179 		return -EINVAL;
1180 	geneve->remote = *remote;
1181 
1182 	geneve->ttl = ttl;
1183 	geneve->tos = tos;
1184 	geneve->dst_port = dst_port;
1185 	geneve->collect_md = metadata;
1186 
1187 	t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1188 			    &tun_on_same_port, &tun_collect_md);
1189 	if (t)
1190 		return -EBUSY;
1191 
1192 	if (metadata) {
1193 		if (tun_on_same_port)
1194 			return -EPERM;
1195 	} else {
1196 		if (tun_collect_md)
1197 			return -EPERM;
1198 	}
1199 
1200 	err = register_netdevice(dev);
1201 	if (err)
1202 		return err;
1203 
1204 	list_add(&geneve->next, &gn->geneve_list);
1205 	return 0;
1206 }
1207 
1208 static int geneve_newlink(struct net *net, struct net_device *dev,
1209 			  struct nlattr *tb[], struct nlattr *data[])
1210 {
1211 	__be16 dst_port = htons(GENEVE_UDP_PORT);
1212 	__u8 ttl = 0, tos = 0;
1213 	bool metadata = false;
1214 	union geneve_addr remote = geneve_remote_unspec;
1215 	__u32 vni = 0;
1216 
1217 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1218 		return -EINVAL;
1219 
1220 	if (data[IFLA_GENEVE_REMOTE]) {
1221 		remote.sa.sa_family = AF_INET;
1222 		remote.sin.sin_addr.s_addr =
1223 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1224 	}
1225 
1226 	if (data[IFLA_GENEVE_REMOTE6]) {
1227 		if (!IS_ENABLED(CONFIG_IPV6))
1228 			return -EPFNOSUPPORT;
1229 
1230 		remote.sa.sa_family = AF_INET6;
1231 		remote.sin6.sin6_addr =
1232 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1233 
1234 		if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1235 		    IPV6_ADDR_LINKLOCAL) {
1236 			netdev_dbg(dev, "link-local remote is unsupported\n");
1237 			return -EINVAL;
1238 		}
1239 	}
1240 
1241 	if (data[IFLA_GENEVE_ID])
1242 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1243 
1244 	if (data[IFLA_GENEVE_TTL])
1245 		ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1246 
1247 	if (data[IFLA_GENEVE_TOS])
1248 		tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1249 
1250 	if (data[IFLA_GENEVE_PORT])
1251 		dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1252 
1253 	if (data[IFLA_GENEVE_COLLECT_METADATA])
1254 		metadata = true;
1255 
1256 	return geneve_configure(net, dev, &remote, vni, ttl, tos, dst_port,
1257 				metadata);
1258 }
1259 
1260 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1261 {
1262 	struct geneve_dev *geneve = netdev_priv(dev);
1263 
1264 	list_del(&geneve->next);
1265 	unregister_netdevice_queue(dev, head);
1266 }
1267 
1268 static size_t geneve_get_size(const struct net_device *dev)
1269 {
1270 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1271 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1272 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1273 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1274 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1275 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1276 		0;
1277 }
1278 
1279 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1280 {
1281 	struct geneve_dev *geneve = netdev_priv(dev);
1282 	__u32 vni;
1283 
1284 	vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1285 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1286 		goto nla_put_failure;
1287 
1288 	if (geneve->remote.sa.sa_family == AF_INET) {
1289 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1290 				    geneve->remote.sin.sin_addr.s_addr))
1291 			goto nla_put_failure;
1292 #if IS_ENABLED(CONFIG_IPV6)
1293 	} else {
1294 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1295 				     &geneve->remote.sin6.sin6_addr))
1296 			goto nla_put_failure;
1297 #endif
1298 	}
1299 
1300 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1301 	    nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
1302 		goto nla_put_failure;
1303 
1304 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1305 		goto nla_put_failure;
1306 
1307 	if (geneve->collect_md) {
1308 		if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1309 			goto nla_put_failure;
1310 	}
1311 
1312 	return 0;
1313 
1314 nla_put_failure:
1315 	return -EMSGSIZE;
1316 }
1317 
1318 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1319 	.kind		= "geneve",
1320 	.maxtype	= IFLA_GENEVE_MAX,
1321 	.policy		= geneve_policy,
1322 	.priv_size	= sizeof(struct geneve_dev),
1323 	.setup		= geneve_setup,
1324 	.validate	= geneve_validate,
1325 	.newlink	= geneve_newlink,
1326 	.dellink	= geneve_dellink,
1327 	.get_size	= geneve_get_size,
1328 	.fill_info	= geneve_fill_info,
1329 };
1330 
1331 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1332 					u8 name_assign_type, u16 dst_port)
1333 {
1334 	struct nlattr *tb[IFLA_MAX + 1];
1335 	struct net_device *dev;
1336 	int err;
1337 
1338 	memset(tb, 0, sizeof(tb));
1339 	dev = rtnl_create_link(net, name, name_assign_type,
1340 			       &geneve_link_ops, tb);
1341 	if (IS_ERR(dev))
1342 		return dev;
1343 
1344 	err = geneve_configure(net, dev, &geneve_remote_unspec,
1345 			       0, 0, 0, htons(dst_port), true);
1346 	if (err) {
1347 		free_netdev(dev);
1348 		return ERR_PTR(err);
1349 	}
1350 	return dev;
1351 }
1352 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1353 
1354 static __net_init int geneve_init_net(struct net *net)
1355 {
1356 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1357 
1358 	INIT_LIST_HEAD(&gn->geneve_list);
1359 	INIT_LIST_HEAD(&gn->sock_list);
1360 	return 0;
1361 }
1362 
1363 static void __net_exit geneve_exit_net(struct net *net)
1364 {
1365 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1366 	struct geneve_dev *geneve, *next;
1367 	struct net_device *dev, *aux;
1368 	LIST_HEAD(list);
1369 
1370 	rtnl_lock();
1371 
1372 	/* gather any geneve devices that were moved into this ns */
1373 	for_each_netdev_safe(net, dev, aux)
1374 		if (dev->rtnl_link_ops == &geneve_link_ops)
1375 			unregister_netdevice_queue(dev, &list);
1376 
1377 	/* now gather any other geneve devices that were created in this ns */
1378 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1379 		/* If geneve->dev is in the same netns, it was already added
1380 		 * to the list by the previous loop.
1381 		 */
1382 		if (!net_eq(dev_net(geneve->dev), net))
1383 			unregister_netdevice_queue(geneve->dev, &list);
1384 	}
1385 
1386 	/* unregister the devices gathered above */
1387 	unregister_netdevice_many(&list);
1388 	rtnl_unlock();
1389 }
1390 
1391 static struct pernet_operations geneve_net_ops = {
1392 	.init = geneve_init_net,
1393 	.exit = geneve_exit_net,
1394 	.id   = &geneve_net_id,
1395 	.size = sizeof(struct geneve_net),
1396 };
1397 
1398 static int __init geneve_init_module(void)
1399 {
1400 	int rc;
1401 
1402 	rc = register_pernet_subsys(&geneve_net_ops);
1403 	if (rc)
1404 		goto out1;
1405 
1406 	rc = rtnl_link_register(&geneve_link_ops);
1407 	if (rc)
1408 		goto out2;
1409 
1410 	return 0;
1411 out2:
1412 	unregister_pernet_subsys(&geneve_net_ops);
1413 out1:
1414 	return rc;
1415 }
1416 late_initcall(geneve_init_module);
1417 
1418 static void __exit geneve_cleanup_module(void)
1419 {
1420 	rtnl_link_unregister(&geneve_link_ops);
1421 	unregister_pernet_subsys(&geneve_net_ops);
1422 }
1423 module_exit(geneve_cleanup_module);
1424 
1425 MODULE_LICENSE("GPL");
1426 MODULE_VERSION(GENEVE_NETDEV_VER);
1427 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1428 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1429 MODULE_ALIAS_RTNL_LINK("geneve");
1430