xref: /openbmc/linux/drivers/net/geneve.c (revision 8e74a48d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * GENEVE: Generic Network Virtualization Encapsulation
4  *
5  * Copyright (c) 2015 Red Hat, Inc.
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/ethtool.h>
11 #include <linux/kernel.h>
12 #include <linux/module.h>
13 #include <linux/etherdevice.h>
14 #include <linux/hash.h>
15 #include <net/ipv6_stubs.h>
16 #include <net/dst_metadata.h>
17 #include <net/gro_cells.h>
18 #include <net/rtnetlink.h>
19 #include <net/geneve.h>
20 #include <net/gro.h>
21 #include <net/protocol.h>
22 
23 #define GENEVE_NETDEV_VER	"0.6"
24 
25 #define GENEVE_N_VID		(1u << 24)
26 #define GENEVE_VID_MASK		(GENEVE_N_VID - 1)
27 
28 #define VNI_HASH_BITS		10
29 #define VNI_HASH_SIZE		(1<<VNI_HASH_BITS)
30 
31 static bool log_ecn_error = true;
32 module_param(log_ecn_error, bool, 0644);
33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
34 
35 #define GENEVE_VER 0
36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
39 
40 /* per-network namespace private data for this module */
41 struct geneve_net {
42 	struct list_head	geneve_list;
43 	struct list_head	sock_list;
44 };
45 
46 static unsigned int geneve_net_id;
47 
48 struct geneve_dev_node {
49 	struct hlist_node hlist;
50 	struct geneve_dev *geneve;
51 };
52 
53 struct geneve_config {
54 	struct ip_tunnel_info	info;
55 	bool			collect_md;
56 	bool			use_udp6_rx_checksums;
57 	bool			ttl_inherit;
58 	enum ifla_geneve_df	df;
59 };
60 
61 /* Pseudo network device */
62 struct geneve_dev {
63 	struct geneve_dev_node hlist4;	/* vni hash table for IPv4 socket */
64 #if IS_ENABLED(CONFIG_IPV6)
65 	struct geneve_dev_node hlist6;	/* vni hash table for IPv6 socket */
66 #endif
67 	struct net	   *net;	/* netns for packet i/o */
68 	struct net_device  *dev;	/* netdev for geneve tunnel */
69 	struct geneve_sock __rcu *sock4;	/* IPv4 socket used for geneve tunnel */
70 #if IS_ENABLED(CONFIG_IPV6)
71 	struct geneve_sock __rcu *sock6;	/* IPv6 socket used for geneve tunnel */
72 #endif
73 	struct list_head   next;	/* geneve's per namespace list */
74 	struct gro_cells   gro_cells;
75 	struct geneve_config cfg;
76 };
77 
78 struct geneve_sock {
79 	bool			collect_md;
80 	struct list_head	list;
81 	struct socket		*sock;
82 	struct rcu_head		rcu;
83 	int			refcnt;
84 	struct hlist_head	vni_list[VNI_HASH_SIZE];
85 };
86 
87 static inline __u32 geneve_net_vni_hash(u8 vni[3])
88 {
89 	__u32 vnid;
90 
91 	vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
92 	return hash_32(vnid, VNI_HASH_BITS);
93 }
94 
95 static __be64 vni_to_tunnel_id(const __u8 *vni)
96 {
97 #ifdef __BIG_ENDIAN
98 	return (vni[0] << 16) | (vni[1] << 8) | vni[2];
99 #else
100 	return (__force __be64)(((__force u64)vni[0] << 40) |
101 				((__force u64)vni[1] << 48) |
102 				((__force u64)vni[2] << 56));
103 #endif
104 }
105 
106 /* Convert 64 bit tunnel ID to 24 bit VNI. */
107 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
108 {
109 #ifdef __BIG_ENDIAN
110 	vni[0] = (__force __u8)(tun_id >> 16);
111 	vni[1] = (__force __u8)(tun_id >> 8);
112 	vni[2] = (__force __u8)tun_id;
113 #else
114 	vni[0] = (__force __u8)((__force u64)tun_id >> 40);
115 	vni[1] = (__force __u8)((__force u64)tun_id >> 48);
116 	vni[2] = (__force __u8)((__force u64)tun_id >> 56);
117 #endif
118 }
119 
120 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
121 {
122 	return !memcmp(vni, &tun_id[5], 3);
123 }
124 
125 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
126 {
127 	return gs->sock->sk->sk_family;
128 }
129 
130 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
131 					__be32 addr, u8 vni[])
132 {
133 	struct hlist_head *vni_list_head;
134 	struct geneve_dev_node *node;
135 	__u32 hash;
136 
137 	/* Find the device for this VNI */
138 	hash = geneve_net_vni_hash(vni);
139 	vni_list_head = &gs->vni_list[hash];
140 	hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
141 		if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
142 		    addr == node->geneve->cfg.info.key.u.ipv4.dst)
143 			return node->geneve;
144 	}
145 	return NULL;
146 }
147 
148 #if IS_ENABLED(CONFIG_IPV6)
149 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
150 					 struct in6_addr addr6, u8 vni[])
151 {
152 	struct hlist_head *vni_list_head;
153 	struct geneve_dev_node *node;
154 	__u32 hash;
155 
156 	/* Find the device for this VNI */
157 	hash = geneve_net_vni_hash(vni);
158 	vni_list_head = &gs->vni_list[hash];
159 	hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
160 		if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
161 		    ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
162 			return node->geneve;
163 	}
164 	return NULL;
165 }
166 #endif
167 
168 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
169 {
170 	return (struct genevehdr *)(udp_hdr(skb) + 1);
171 }
172 
173 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
174 					    struct sk_buff *skb)
175 {
176 	static u8 zero_vni[3];
177 	u8 *vni;
178 
179 	if (geneve_get_sk_family(gs) == AF_INET) {
180 		struct iphdr *iph;
181 		__be32 addr;
182 
183 		iph = ip_hdr(skb); /* outer IP header... */
184 
185 		if (gs->collect_md) {
186 			vni = zero_vni;
187 			addr = 0;
188 		} else {
189 			vni = geneve_hdr(skb)->vni;
190 			addr = iph->saddr;
191 		}
192 
193 		return geneve_lookup(gs, addr, vni);
194 #if IS_ENABLED(CONFIG_IPV6)
195 	} else if (geneve_get_sk_family(gs) == AF_INET6) {
196 		static struct in6_addr zero_addr6;
197 		struct ipv6hdr *ip6h;
198 		struct in6_addr addr6;
199 
200 		ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
201 
202 		if (gs->collect_md) {
203 			vni = zero_vni;
204 			addr6 = zero_addr6;
205 		} else {
206 			vni = geneve_hdr(skb)->vni;
207 			addr6 = ip6h->saddr;
208 		}
209 
210 		return geneve6_lookup(gs, addr6, vni);
211 #endif
212 	}
213 	return NULL;
214 }
215 
216 /* geneve receive/decap routine */
217 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
218 		      struct sk_buff *skb)
219 {
220 	struct genevehdr *gnvh = geneve_hdr(skb);
221 	struct metadata_dst *tun_dst = NULL;
222 	unsigned int len;
223 	int err = 0;
224 	void *oiph;
225 
226 	if (ip_tunnel_collect_metadata() || gs->collect_md) {
227 		__be16 flags;
228 
229 		flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
230 			(gnvh->critical ? TUNNEL_CRIT_OPT : 0);
231 
232 		tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
233 					 vni_to_tunnel_id(gnvh->vni),
234 					 gnvh->opt_len * 4);
235 		if (!tun_dst) {
236 			geneve->dev->stats.rx_dropped++;
237 			goto drop;
238 		}
239 		/* Update tunnel dst according to Geneve options. */
240 		ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
241 					gnvh->options, gnvh->opt_len * 4,
242 					TUNNEL_GENEVE_OPT);
243 	} else {
244 		/* Drop packets w/ critical options,
245 		 * since we don't support any...
246 		 */
247 		if (gnvh->critical) {
248 			geneve->dev->stats.rx_frame_errors++;
249 			geneve->dev->stats.rx_errors++;
250 			goto drop;
251 		}
252 	}
253 
254 	skb_reset_mac_header(skb);
255 	skb->protocol = eth_type_trans(skb, geneve->dev);
256 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
257 
258 	if (tun_dst)
259 		skb_dst_set(skb, &tun_dst->dst);
260 
261 	/* Ignore packet loops (and multicast echo) */
262 	if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
263 		geneve->dev->stats.rx_errors++;
264 		goto drop;
265 	}
266 
267 	oiph = skb_network_header(skb);
268 	skb_reset_network_header(skb);
269 
270 	if (geneve_get_sk_family(gs) == AF_INET)
271 		err = IP_ECN_decapsulate(oiph, skb);
272 #if IS_ENABLED(CONFIG_IPV6)
273 	else
274 		err = IP6_ECN_decapsulate(oiph, skb);
275 #endif
276 
277 	if (unlikely(err)) {
278 		if (log_ecn_error) {
279 			if (geneve_get_sk_family(gs) == AF_INET)
280 				net_info_ratelimited("non-ECT from %pI4 "
281 						     "with TOS=%#x\n",
282 						     &((struct iphdr *)oiph)->saddr,
283 						     ((struct iphdr *)oiph)->tos);
284 #if IS_ENABLED(CONFIG_IPV6)
285 			else
286 				net_info_ratelimited("non-ECT from %pI6\n",
287 						     &((struct ipv6hdr *)oiph)->saddr);
288 #endif
289 		}
290 		if (err > 1) {
291 			++geneve->dev->stats.rx_frame_errors;
292 			++geneve->dev->stats.rx_errors;
293 			goto drop;
294 		}
295 	}
296 
297 	len = skb->len;
298 	err = gro_cells_receive(&geneve->gro_cells, skb);
299 	if (likely(err == NET_RX_SUCCESS))
300 		dev_sw_netstats_rx_add(geneve->dev, len);
301 
302 	return;
303 drop:
304 	/* Consume bad packet */
305 	kfree_skb(skb);
306 }
307 
308 /* Setup stats when device is created */
309 static int geneve_init(struct net_device *dev)
310 {
311 	struct geneve_dev *geneve = netdev_priv(dev);
312 	int err;
313 
314 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
315 	if (!dev->tstats)
316 		return -ENOMEM;
317 
318 	err = gro_cells_init(&geneve->gro_cells, dev);
319 	if (err) {
320 		free_percpu(dev->tstats);
321 		return err;
322 	}
323 
324 	err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
325 	if (err) {
326 		free_percpu(dev->tstats);
327 		gro_cells_destroy(&geneve->gro_cells);
328 		return err;
329 	}
330 	return 0;
331 }
332 
333 static void geneve_uninit(struct net_device *dev)
334 {
335 	struct geneve_dev *geneve = netdev_priv(dev);
336 
337 	dst_cache_destroy(&geneve->cfg.info.dst_cache);
338 	gro_cells_destroy(&geneve->gro_cells);
339 	free_percpu(dev->tstats);
340 }
341 
342 /* Callback from net/ipv4/udp.c to receive packets */
343 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
344 {
345 	struct genevehdr *geneveh;
346 	struct geneve_dev *geneve;
347 	struct geneve_sock *gs;
348 	int opts_len;
349 
350 	/* Need UDP and Geneve header to be present */
351 	if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
352 		goto drop;
353 
354 	/* Return packets with reserved bits set */
355 	geneveh = geneve_hdr(skb);
356 	if (unlikely(geneveh->ver != GENEVE_VER))
357 		goto drop;
358 
359 	if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
360 		goto drop;
361 
362 	gs = rcu_dereference_sk_user_data(sk);
363 	if (!gs)
364 		goto drop;
365 
366 	geneve = geneve_lookup_skb(gs, skb);
367 	if (!geneve)
368 		goto drop;
369 
370 	opts_len = geneveh->opt_len * 4;
371 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
372 				 htons(ETH_P_TEB),
373 				 !net_eq(geneve->net, dev_net(geneve->dev)))) {
374 		geneve->dev->stats.rx_dropped++;
375 		goto drop;
376 	}
377 
378 	geneve_rx(geneve, gs, skb);
379 	return 0;
380 
381 drop:
382 	/* Consume bad packet */
383 	kfree_skb(skb);
384 	return 0;
385 }
386 
387 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
388 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
389 {
390 	struct genevehdr *geneveh;
391 	struct geneve_sock *gs;
392 	u8 zero_vni[3] = { 0 };
393 	u8 *vni = zero_vni;
394 
395 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
396 		return -EINVAL;
397 
398 	geneveh = geneve_hdr(skb);
399 	if (geneveh->ver != GENEVE_VER)
400 		return -EINVAL;
401 
402 	if (geneveh->proto_type != htons(ETH_P_TEB))
403 		return -EINVAL;
404 
405 	gs = rcu_dereference_sk_user_data(sk);
406 	if (!gs)
407 		return -ENOENT;
408 
409 	if (geneve_get_sk_family(gs) == AF_INET) {
410 		struct iphdr *iph = ip_hdr(skb);
411 		__be32 addr4 = 0;
412 
413 		if (!gs->collect_md) {
414 			vni = geneve_hdr(skb)->vni;
415 			addr4 = iph->daddr;
416 		}
417 
418 		return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
419 	}
420 
421 #if IS_ENABLED(CONFIG_IPV6)
422 	if (geneve_get_sk_family(gs) == AF_INET6) {
423 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
424 		struct in6_addr addr6;
425 
426 		memset(&addr6, 0, sizeof(struct in6_addr));
427 
428 		if (!gs->collect_md) {
429 			vni = geneve_hdr(skb)->vni;
430 			addr6 = ip6h->daddr;
431 		}
432 
433 		return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
434 	}
435 #endif
436 
437 	return -EPFNOSUPPORT;
438 }
439 
440 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
441 					 __be16 port, bool ipv6_rx_csum)
442 {
443 	struct socket *sock;
444 	struct udp_port_cfg udp_conf;
445 	int err;
446 
447 	memset(&udp_conf, 0, sizeof(udp_conf));
448 
449 	if (ipv6) {
450 		udp_conf.family = AF_INET6;
451 		udp_conf.ipv6_v6only = 1;
452 		udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
453 	} else {
454 		udp_conf.family = AF_INET;
455 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
456 	}
457 
458 	udp_conf.local_udp_port = port;
459 
460 	/* Open UDP socket */
461 	err = udp_sock_create(net, &udp_conf, &sock);
462 	if (err < 0)
463 		return ERR_PTR(err);
464 
465 	udp_allow_gso(sock->sk);
466 	return sock;
467 }
468 
469 static int geneve_hlen(struct genevehdr *gh)
470 {
471 	return sizeof(*gh) + gh->opt_len * 4;
472 }
473 
474 static struct sk_buff *geneve_gro_receive(struct sock *sk,
475 					  struct list_head *head,
476 					  struct sk_buff *skb)
477 {
478 	struct sk_buff *pp = NULL;
479 	struct sk_buff *p;
480 	struct genevehdr *gh, *gh2;
481 	unsigned int hlen, gh_len, off_gnv;
482 	const struct packet_offload *ptype;
483 	__be16 type;
484 	int flush = 1;
485 
486 	off_gnv = skb_gro_offset(skb);
487 	hlen = off_gnv + sizeof(*gh);
488 	gh = skb_gro_header_fast(skb, off_gnv);
489 	if (skb_gro_header_hard(skb, hlen)) {
490 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
491 		if (unlikely(!gh))
492 			goto out;
493 	}
494 
495 	if (gh->ver != GENEVE_VER || gh->oam)
496 		goto out;
497 	gh_len = geneve_hlen(gh);
498 
499 	hlen = off_gnv + gh_len;
500 	if (skb_gro_header_hard(skb, hlen)) {
501 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
502 		if (unlikely(!gh))
503 			goto out;
504 	}
505 
506 	list_for_each_entry(p, head, list) {
507 		if (!NAPI_GRO_CB(p)->same_flow)
508 			continue;
509 
510 		gh2 = (struct genevehdr *)(p->data + off_gnv);
511 		if (gh->opt_len != gh2->opt_len ||
512 		    memcmp(gh, gh2, gh_len)) {
513 			NAPI_GRO_CB(p)->same_flow = 0;
514 			continue;
515 		}
516 	}
517 
518 	type = gh->proto_type;
519 
520 	ptype = gro_find_receive_by_type(type);
521 	if (!ptype)
522 		goto out;
523 
524 	skb_gro_pull(skb, gh_len);
525 	skb_gro_postpull_rcsum(skb, gh, gh_len);
526 	pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
527 	flush = 0;
528 
529 out:
530 	skb_gro_flush_final(skb, pp, flush);
531 
532 	return pp;
533 }
534 
535 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
536 			       int nhoff)
537 {
538 	struct genevehdr *gh;
539 	struct packet_offload *ptype;
540 	__be16 type;
541 	int gh_len;
542 	int err = -ENOSYS;
543 
544 	gh = (struct genevehdr *)(skb->data + nhoff);
545 	gh_len = geneve_hlen(gh);
546 	type = gh->proto_type;
547 
548 	ptype = gro_find_complete_by_type(type);
549 	if (ptype)
550 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
551 
552 	skb_set_inner_mac_header(skb, nhoff + gh_len);
553 
554 	return err;
555 }
556 
557 /* Create new listen socket if needed */
558 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
559 						bool ipv6, bool ipv6_rx_csum)
560 {
561 	struct geneve_net *gn = net_generic(net, geneve_net_id);
562 	struct geneve_sock *gs;
563 	struct socket *sock;
564 	struct udp_tunnel_sock_cfg tunnel_cfg;
565 	int h;
566 
567 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
568 	if (!gs)
569 		return ERR_PTR(-ENOMEM);
570 
571 	sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
572 	if (IS_ERR(sock)) {
573 		kfree(gs);
574 		return ERR_CAST(sock);
575 	}
576 
577 	gs->sock = sock;
578 	gs->refcnt = 1;
579 	for (h = 0; h < VNI_HASH_SIZE; ++h)
580 		INIT_HLIST_HEAD(&gs->vni_list[h]);
581 
582 	/* Initialize the geneve udp offloads structure */
583 	udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
584 
585 	/* Mark socket as an encapsulation socket */
586 	memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
587 	tunnel_cfg.sk_user_data = gs;
588 	tunnel_cfg.encap_type = 1;
589 	tunnel_cfg.gro_receive = geneve_gro_receive;
590 	tunnel_cfg.gro_complete = geneve_gro_complete;
591 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
592 	tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
593 	tunnel_cfg.encap_destroy = NULL;
594 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
595 	list_add(&gs->list, &gn->sock_list);
596 	return gs;
597 }
598 
599 static void __geneve_sock_release(struct geneve_sock *gs)
600 {
601 	if (!gs || --gs->refcnt)
602 		return;
603 
604 	list_del(&gs->list);
605 	udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
606 	udp_tunnel_sock_release(gs->sock);
607 	kfree_rcu(gs, rcu);
608 }
609 
610 static void geneve_sock_release(struct geneve_dev *geneve)
611 {
612 	struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
613 #if IS_ENABLED(CONFIG_IPV6)
614 	struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
615 
616 	rcu_assign_pointer(geneve->sock6, NULL);
617 #endif
618 
619 	rcu_assign_pointer(geneve->sock4, NULL);
620 	synchronize_net();
621 
622 	__geneve_sock_release(gs4);
623 #if IS_ENABLED(CONFIG_IPV6)
624 	__geneve_sock_release(gs6);
625 #endif
626 }
627 
628 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
629 					    sa_family_t family,
630 					    __be16 dst_port)
631 {
632 	struct geneve_sock *gs;
633 
634 	list_for_each_entry(gs, &gn->sock_list, list) {
635 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
636 		    geneve_get_sk_family(gs) == family) {
637 			return gs;
638 		}
639 	}
640 	return NULL;
641 }
642 
643 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
644 {
645 	struct net *net = geneve->net;
646 	struct geneve_net *gn = net_generic(net, geneve_net_id);
647 	struct geneve_dev_node *node;
648 	struct geneve_sock *gs;
649 	__u8 vni[3];
650 	__u32 hash;
651 
652 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
653 	if (gs) {
654 		gs->refcnt++;
655 		goto out;
656 	}
657 
658 	gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
659 				  geneve->cfg.use_udp6_rx_checksums);
660 	if (IS_ERR(gs))
661 		return PTR_ERR(gs);
662 
663 out:
664 	gs->collect_md = geneve->cfg.collect_md;
665 #if IS_ENABLED(CONFIG_IPV6)
666 	if (ipv6) {
667 		rcu_assign_pointer(geneve->sock6, gs);
668 		node = &geneve->hlist6;
669 	} else
670 #endif
671 	{
672 		rcu_assign_pointer(geneve->sock4, gs);
673 		node = &geneve->hlist4;
674 	}
675 	node->geneve = geneve;
676 
677 	tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
678 	hash = geneve_net_vni_hash(vni);
679 	hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
680 	return 0;
681 }
682 
683 static int geneve_open(struct net_device *dev)
684 {
685 	struct geneve_dev *geneve = netdev_priv(dev);
686 	bool metadata = geneve->cfg.collect_md;
687 	bool ipv4, ipv6;
688 	int ret = 0;
689 
690 	ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
691 	ipv4 = !ipv6 || metadata;
692 #if IS_ENABLED(CONFIG_IPV6)
693 	if (ipv6) {
694 		ret = geneve_sock_add(geneve, true);
695 		if (ret < 0 && ret != -EAFNOSUPPORT)
696 			ipv4 = false;
697 	}
698 #endif
699 	if (ipv4)
700 		ret = geneve_sock_add(geneve, false);
701 	if (ret < 0)
702 		geneve_sock_release(geneve);
703 
704 	return ret;
705 }
706 
707 static int geneve_stop(struct net_device *dev)
708 {
709 	struct geneve_dev *geneve = netdev_priv(dev);
710 
711 	hlist_del_init_rcu(&geneve->hlist4.hlist);
712 #if IS_ENABLED(CONFIG_IPV6)
713 	hlist_del_init_rcu(&geneve->hlist6.hlist);
714 #endif
715 	geneve_sock_release(geneve);
716 	return 0;
717 }
718 
719 static void geneve_build_header(struct genevehdr *geneveh,
720 				const struct ip_tunnel_info *info)
721 {
722 	geneveh->ver = GENEVE_VER;
723 	geneveh->opt_len = info->options_len / 4;
724 	geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
725 	geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
726 	geneveh->rsvd1 = 0;
727 	tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
728 	geneveh->proto_type = htons(ETH_P_TEB);
729 	geneveh->rsvd2 = 0;
730 
731 	if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
732 		ip_tunnel_info_opts_get(geneveh->options, info);
733 }
734 
735 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
736 			    const struct ip_tunnel_info *info,
737 			    bool xnet, int ip_hdr_len)
738 {
739 	bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
740 	struct genevehdr *gnvh;
741 	int min_headroom;
742 	int err;
743 
744 	skb_reset_mac_header(skb);
745 	skb_scrub_packet(skb, xnet);
746 
747 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
748 		       GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
749 	err = skb_cow_head(skb, min_headroom);
750 	if (unlikely(err))
751 		goto free_dst;
752 
753 	err = udp_tunnel_handle_offloads(skb, udp_sum);
754 	if (err)
755 		goto free_dst;
756 
757 	gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
758 	geneve_build_header(gnvh, info);
759 	skb_set_inner_protocol(skb, htons(ETH_P_TEB));
760 	return 0;
761 
762 free_dst:
763 	dst_release(dst);
764 	return err;
765 }
766 
767 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
768 				       struct net_device *dev,
769 				       struct geneve_sock *gs4,
770 				       struct flowi4 *fl4,
771 				       const struct ip_tunnel_info *info,
772 				       __be16 dport, __be16 sport)
773 {
774 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
775 	struct geneve_dev *geneve = netdev_priv(dev);
776 	struct dst_cache *dst_cache;
777 	struct rtable *rt = NULL;
778 	__u8 tos;
779 
780 	if (!gs4)
781 		return ERR_PTR(-EIO);
782 
783 	memset(fl4, 0, sizeof(*fl4));
784 	fl4->flowi4_mark = skb->mark;
785 	fl4->flowi4_proto = IPPROTO_UDP;
786 	fl4->daddr = info->key.u.ipv4.dst;
787 	fl4->saddr = info->key.u.ipv4.src;
788 	fl4->fl4_dport = dport;
789 	fl4->fl4_sport = sport;
790 
791 	tos = info->key.tos;
792 	if ((tos == 1) && !geneve->cfg.collect_md) {
793 		tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
794 		use_cache = false;
795 	}
796 	fl4->flowi4_tos = RT_TOS(tos);
797 
798 	dst_cache = (struct dst_cache *)&info->dst_cache;
799 	if (use_cache) {
800 		rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
801 		if (rt)
802 			return rt;
803 	}
804 	rt = ip_route_output_key(geneve->net, fl4);
805 	if (IS_ERR(rt)) {
806 		netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
807 		return ERR_PTR(-ENETUNREACH);
808 	}
809 	if (rt->dst.dev == dev) { /* is this necessary? */
810 		netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
811 		ip_rt_put(rt);
812 		return ERR_PTR(-ELOOP);
813 	}
814 	if (use_cache)
815 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
816 	return rt;
817 }
818 
819 #if IS_ENABLED(CONFIG_IPV6)
820 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
821 					   struct net_device *dev,
822 					   struct geneve_sock *gs6,
823 					   struct flowi6 *fl6,
824 					   const struct ip_tunnel_info *info,
825 					   __be16 dport, __be16 sport)
826 {
827 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
828 	struct geneve_dev *geneve = netdev_priv(dev);
829 	struct dst_entry *dst = NULL;
830 	struct dst_cache *dst_cache;
831 	__u8 prio;
832 
833 	if (!gs6)
834 		return ERR_PTR(-EIO);
835 
836 	memset(fl6, 0, sizeof(*fl6));
837 	fl6->flowi6_mark = skb->mark;
838 	fl6->flowi6_proto = IPPROTO_UDP;
839 	fl6->daddr = info->key.u.ipv6.dst;
840 	fl6->saddr = info->key.u.ipv6.src;
841 	fl6->fl6_dport = dport;
842 	fl6->fl6_sport = sport;
843 
844 	prio = info->key.tos;
845 	if ((prio == 1) && !geneve->cfg.collect_md) {
846 		prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
847 		use_cache = false;
848 	}
849 
850 	fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
851 					   info->key.label);
852 	dst_cache = (struct dst_cache *)&info->dst_cache;
853 	if (use_cache) {
854 		dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
855 		if (dst)
856 			return dst;
857 	}
858 	dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
859 					      NULL);
860 	if (IS_ERR(dst)) {
861 		netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
862 		return ERR_PTR(-ENETUNREACH);
863 	}
864 	if (dst->dev == dev) { /* is this necessary? */
865 		netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
866 		dst_release(dst);
867 		return ERR_PTR(-ELOOP);
868 	}
869 
870 	if (use_cache)
871 		dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
872 	return dst;
873 }
874 #endif
875 
876 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
877 			   struct geneve_dev *geneve,
878 			   const struct ip_tunnel_info *info)
879 {
880 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
881 	struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
882 	const struct ip_tunnel_key *key = &info->key;
883 	struct rtable *rt;
884 	struct flowi4 fl4;
885 	__u8 tos, ttl;
886 	__be16 df = 0;
887 	__be16 sport;
888 	int err;
889 
890 	if (!pskb_inet_may_pull(skb))
891 		return -EINVAL;
892 
893 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
894 	rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
895 			      geneve->cfg.info.key.tp_dst, sport);
896 	if (IS_ERR(rt))
897 		return PTR_ERR(rt);
898 
899 	err = skb_tunnel_check_pmtu(skb, &rt->dst,
900 				    GENEVE_IPV4_HLEN + info->options_len,
901 				    netif_is_any_bridge_port(dev));
902 	if (err < 0) {
903 		dst_release(&rt->dst);
904 		return err;
905 	} else if (err) {
906 		struct ip_tunnel_info *info;
907 
908 		info = skb_tunnel_info(skb);
909 		if (info) {
910 			struct ip_tunnel_info *unclone;
911 
912 			unclone = skb_tunnel_info_unclone(skb);
913 			if (unlikely(!unclone)) {
914 				dst_release(&rt->dst);
915 				return -ENOMEM;
916 			}
917 
918 			unclone->key.u.ipv4.dst = fl4.saddr;
919 			unclone->key.u.ipv4.src = fl4.daddr;
920 		}
921 
922 		if (!pskb_may_pull(skb, ETH_HLEN)) {
923 			dst_release(&rt->dst);
924 			return -EINVAL;
925 		}
926 
927 		skb->protocol = eth_type_trans(skb, geneve->dev);
928 		netif_rx(skb);
929 		dst_release(&rt->dst);
930 		return -EMSGSIZE;
931 	}
932 
933 	if (geneve->cfg.collect_md) {
934 		tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
935 		ttl = key->ttl;
936 
937 		df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
938 	} else {
939 		tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
940 		if (geneve->cfg.ttl_inherit)
941 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
942 		else
943 			ttl = key->ttl;
944 		ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
945 
946 		if (geneve->cfg.df == GENEVE_DF_SET) {
947 			df = htons(IP_DF);
948 		} else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
949 			struct ethhdr *eth = eth_hdr(skb);
950 
951 			if (ntohs(eth->h_proto) == ETH_P_IPV6) {
952 				df = htons(IP_DF);
953 			} else if (ntohs(eth->h_proto) == ETH_P_IP) {
954 				struct iphdr *iph = ip_hdr(skb);
955 
956 				if (iph->frag_off & htons(IP_DF))
957 					df = htons(IP_DF);
958 			}
959 		}
960 	}
961 
962 	err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
963 	if (unlikely(err))
964 		return err;
965 
966 	udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
967 			    tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
968 			    !net_eq(geneve->net, dev_net(geneve->dev)),
969 			    !(info->key.tun_flags & TUNNEL_CSUM));
970 	return 0;
971 }
972 
973 #if IS_ENABLED(CONFIG_IPV6)
974 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
975 			    struct geneve_dev *geneve,
976 			    const struct ip_tunnel_info *info)
977 {
978 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
979 	struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
980 	const struct ip_tunnel_key *key = &info->key;
981 	struct dst_entry *dst = NULL;
982 	struct flowi6 fl6;
983 	__u8 prio, ttl;
984 	__be16 sport;
985 	int err;
986 
987 	if (!pskb_inet_may_pull(skb))
988 		return -EINVAL;
989 
990 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
991 	dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
992 				geneve->cfg.info.key.tp_dst, sport);
993 	if (IS_ERR(dst))
994 		return PTR_ERR(dst);
995 
996 	err = skb_tunnel_check_pmtu(skb, dst,
997 				    GENEVE_IPV6_HLEN + info->options_len,
998 				    netif_is_any_bridge_port(dev));
999 	if (err < 0) {
1000 		dst_release(dst);
1001 		return err;
1002 	} else if (err) {
1003 		struct ip_tunnel_info *info = skb_tunnel_info(skb);
1004 
1005 		if (info) {
1006 			struct ip_tunnel_info *unclone;
1007 
1008 			unclone = skb_tunnel_info_unclone(skb);
1009 			if (unlikely(!unclone)) {
1010 				dst_release(dst);
1011 				return -ENOMEM;
1012 			}
1013 
1014 			unclone->key.u.ipv6.dst = fl6.saddr;
1015 			unclone->key.u.ipv6.src = fl6.daddr;
1016 		}
1017 
1018 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1019 			dst_release(dst);
1020 			return -EINVAL;
1021 		}
1022 
1023 		skb->protocol = eth_type_trans(skb, geneve->dev);
1024 		netif_rx(skb);
1025 		dst_release(dst);
1026 		return -EMSGSIZE;
1027 	}
1028 
1029 	if (geneve->cfg.collect_md) {
1030 		prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1031 		ttl = key->ttl;
1032 	} else {
1033 		prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1034 					   ip_hdr(skb), skb);
1035 		if (geneve->cfg.ttl_inherit)
1036 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1037 		else
1038 			ttl = key->ttl;
1039 		ttl = ttl ? : ip6_dst_hoplimit(dst);
1040 	}
1041 	err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1042 	if (unlikely(err))
1043 		return err;
1044 
1045 	udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1046 			     &fl6.saddr, &fl6.daddr, prio, ttl,
1047 			     info->key.label, sport, geneve->cfg.info.key.tp_dst,
1048 			     !(info->key.tun_flags & TUNNEL_CSUM));
1049 	return 0;
1050 }
1051 #endif
1052 
1053 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1054 {
1055 	struct geneve_dev *geneve = netdev_priv(dev);
1056 	struct ip_tunnel_info *info = NULL;
1057 	int err;
1058 
1059 	if (geneve->cfg.collect_md) {
1060 		info = skb_tunnel_info(skb);
1061 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1062 			netdev_dbg(dev, "no tunnel metadata\n");
1063 			dev_kfree_skb(skb);
1064 			dev->stats.tx_dropped++;
1065 			return NETDEV_TX_OK;
1066 		}
1067 	} else {
1068 		info = &geneve->cfg.info;
1069 	}
1070 
1071 	rcu_read_lock();
1072 #if IS_ENABLED(CONFIG_IPV6)
1073 	if (info->mode & IP_TUNNEL_INFO_IPV6)
1074 		err = geneve6_xmit_skb(skb, dev, geneve, info);
1075 	else
1076 #endif
1077 		err = geneve_xmit_skb(skb, dev, geneve, info);
1078 	rcu_read_unlock();
1079 
1080 	if (likely(!err))
1081 		return NETDEV_TX_OK;
1082 
1083 	if (err != -EMSGSIZE)
1084 		dev_kfree_skb(skb);
1085 
1086 	if (err == -ELOOP)
1087 		dev->stats.collisions++;
1088 	else if (err == -ENETUNREACH)
1089 		dev->stats.tx_carrier_errors++;
1090 
1091 	dev->stats.tx_errors++;
1092 	return NETDEV_TX_OK;
1093 }
1094 
1095 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1096 {
1097 	if (new_mtu > dev->max_mtu)
1098 		new_mtu = dev->max_mtu;
1099 	else if (new_mtu < dev->min_mtu)
1100 		new_mtu = dev->min_mtu;
1101 
1102 	dev->mtu = new_mtu;
1103 	return 0;
1104 }
1105 
1106 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1107 {
1108 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
1109 	struct geneve_dev *geneve = netdev_priv(dev);
1110 	__be16 sport;
1111 
1112 	if (ip_tunnel_info_af(info) == AF_INET) {
1113 		struct rtable *rt;
1114 		struct flowi4 fl4;
1115 
1116 		struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1117 		sport = udp_flow_src_port(geneve->net, skb,
1118 					  1, USHRT_MAX, true);
1119 
1120 		rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1121 				      geneve->cfg.info.key.tp_dst, sport);
1122 		if (IS_ERR(rt))
1123 			return PTR_ERR(rt);
1124 
1125 		ip_rt_put(rt);
1126 		info->key.u.ipv4.src = fl4.saddr;
1127 #if IS_ENABLED(CONFIG_IPV6)
1128 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
1129 		struct dst_entry *dst;
1130 		struct flowi6 fl6;
1131 
1132 		struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1133 		sport = udp_flow_src_port(geneve->net, skb,
1134 					  1, USHRT_MAX, true);
1135 
1136 		dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1137 					geneve->cfg.info.key.tp_dst, sport);
1138 		if (IS_ERR(dst))
1139 			return PTR_ERR(dst);
1140 
1141 		dst_release(dst);
1142 		info->key.u.ipv6.src = fl6.saddr;
1143 #endif
1144 	} else {
1145 		return -EINVAL;
1146 	}
1147 
1148 	info->key.tp_src = sport;
1149 	info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1150 	return 0;
1151 }
1152 
1153 static const struct net_device_ops geneve_netdev_ops = {
1154 	.ndo_init		= geneve_init,
1155 	.ndo_uninit		= geneve_uninit,
1156 	.ndo_open		= geneve_open,
1157 	.ndo_stop		= geneve_stop,
1158 	.ndo_start_xmit		= geneve_xmit,
1159 	.ndo_get_stats64	= dev_get_tstats64,
1160 	.ndo_change_mtu		= geneve_change_mtu,
1161 	.ndo_validate_addr	= eth_validate_addr,
1162 	.ndo_set_mac_address	= eth_mac_addr,
1163 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1164 };
1165 
1166 static void geneve_get_drvinfo(struct net_device *dev,
1167 			       struct ethtool_drvinfo *drvinfo)
1168 {
1169 	strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1170 	strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1171 }
1172 
1173 static const struct ethtool_ops geneve_ethtool_ops = {
1174 	.get_drvinfo	= geneve_get_drvinfo,
1175 	.get_link	= ethtool_op_get_link,
1176 };
1177 
1178 /* Info for udev, that this is a virtual tunnel endpoint */
1179 static struct device_type geneve_type = {
1180 	.name = "geneve",
1181 };
1182 
1183 /* Calls the ndo_udp_tunnel_add of the caller in order to
1184  * supply the listening GENEVE udp ports. Callers are expected
1185  * to implement the ndo_udp_tunnel_add.
1186  */
1187 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1188 {
1189 	struct net *net = dev_net(dev);
1190 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1191 	struct geneve_sock *gs;
1192 
1193 	rcu_read_lock();
1194 	list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1195 		if (push) {
1196 			udp_tunnel_push_rx_port(dev, gs->sock,
1197 						UDP_TUNNEL_TYPE_GENEVE);
1198 		} else {
1199 			udp_tunnel_drop_rx_port(dev, gs->sock,
1200 						UDP_TUNNEL_TYPE_GENEVE);
1201 		}
1202 	}
1203 	rcu_read_unlock();
1204 }
1205 
1206 /* Initialize the device structure. */
1207 static void geneve_setup(struct net_device *dev)
1208 {
1209 	ether_setup(dev);
1210 
1211 	dev->netdev_ops = &geneve_netdev_ops;
1212 	dev->ethtool_ops = &geneve_ethtool_ops;
1213 	dev->needs_free_netdev = true;
1214 
1215 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1216 
1217 	dev->features    |= NETIF_F_LLTX;
1218 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1219 	dev->features    |= NETIF_F_RXCSUM;
1220 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1221 
1222 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1223 	dev->hw_features |= NETIF_F_RXCSUM;
1224 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1225 
1226 	/* MTU range: 68 - (something less than 65535) */
1227 	dev->min_mtu = ETH_MIN_MTU;
1228 	/* The max_mtu calculation does not take account of GENEVE
1229 	 * options, to avoid excluding potentially valid
1230 	 * configurations. This will be further reduced by IPvX hdr size.
1231 	 */
1232 	dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1233 
1234 	netif_keep_dst(dev);
1235 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1236 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1237 	eth_hw_addr_random(dev);
1238 }
1239 
1240 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1241 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1242 	[IFLA_GENEVE_REMOTE]		= { .len = sizeof_field(struct iphdr, daddr) },
1243 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1244 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1245 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1246 	[IFLA_GENEVE_LABEL]		= { .type = NLA_U32 },
1247 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1248 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1249 	[IFLA_GENEVE_UDP_CSUM]		= { .type = NLA_U8 },
1250 	[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]	= { .type = NLA_U8 },
1251 	[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]	= { .type = NLA_U8 },
1252 	[IFLA_GENEVE_TTL_INHERIT]	= { .type = NLA_U8 },
1253 	[IFLA_GENEVE_DF]		= { .type = NLA_U8 },
1254 };
1255 
1256 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1257 			   struct netlink_ext_ack *extack)
1258 {
1259 	if (tb[IFLA_ADDRESS]) {
1260 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1261 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1262 					    "Provided link layer address is not Ethernet");
1263 			return -EINVAL;
1264 		}
1265 
1266 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1267 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1268 					    "Provided Ethernet address is not unicast");
1269 			return -EADDRNOTAVAIL;
1270 		}
1271 	}
1272 
1273 	if (!data) {
1274 		NL_SET_ERR_MSG(extack,
1275 			       "Not enough attributes provided to perform the operation");
1276 		return -EINVAL;
1277 	}
1278 
1279 	if (data[IFLA_GENEVE_ID]) {
1280 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1281 
1282 		if (vni >= GENEVE_N_VID) {
1283 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1284 					    "Geneve ID must be lower than 16777216");
1285 			return -ERANGE;
1286 		}
1287 	}
1288 
1289 	if (data[IFLA_GENEVE_DF]) {
1290 		enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1291 
1292 		if (df < 0 || df > GENEVE_DF_MAX) {
1293 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1294 					    "Invalid DF attribute");
1295 			return -EINVAL;
1296 		}
1297 	}
1298 
1299 	return 0;
1300 }
1301 
1302 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1303 					  const struct ip_tunnel_info *info,
1304 					  bool *tun_on_same_port,
1305 					  bool *tun_collect_md)
1306 {
1307 	struct geneve_dev *geneve, *t = NULL;
1308 
1309 	*tun_on_same_port = false;
1310 	*tun_collect_md = false;
1311 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1312 		if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1313 			*tun_collect_md = geneve->cfg.collect_md;
1314 			*tun_on_same_port = true;
1315 		}
1316 		if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1317 		    info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1318 		    !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1319 			t = geneve;
1320 	}
1321 	return t;
1322 }
1323 
1324 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1325 {
1326 	return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1327 		 info->key.ttl || info->key.label || info->key.tp_src ||
1328 		 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1329 }
1330 
1331 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1332 				  struct ip_tunnel_info *b)
1333 {
1334 	if (ip_tunnel_info_af(a) == AF_INET)
1335 		return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1336 	else
1337 		return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1338 }
1339 
1340 static int geneve_configure(struct net *net, struct net_device *dev,
1341 			    struct netlink_ext_ack *extack,
1342 			    const struct geneve_config *cfg)
1343 {
1344 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1345 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1346 	const struct ip_tunnel_info *info = &cfg->info;
1347 	bool tun_collect_md, tun_on_same_port;
1348 	int err, encap_len;
1349 
1350 	if (cfg->collect_md && !is_tnl_info_zero(info)) {
1351 		NL_SET_ERR_MSG(extack,
1352 			       "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1353 		return -EINVAL;
1354 	}
1355 
1356 	geneve->net = net;
1357 	geneve->dev = dev;
1358 
1359 	t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1360 	if (t)
1361 		return -EBUSY;
1362 
1363 	/* make enough headroom for basic scenario */
1364 	encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1365 	if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1366 		encap_len += sizeof(struct iphdr);
1367 		dev->max_mtu -= sizeof(struct iphdr);
1368 	} else {
1369 		encap_len += sizeof(struct ipv6hdr);
1370 		dev->max_mtu -= sizeof(struct ipv6hdr);
1371 	}
1372 	dev->needed_headroom = encap_len + ETH_HLEN;
1373 
1374 	if (cfg->collect_md) {
1375 		if (tun_on_same_port) {
1376 			NL_SET_ERR_MSG(extack,
1377 				       "There can be only one externally controlled device on a destination port");
1378 			return -EPERM;
1379 		}
1380 	} else {
1381 		if (tun_collect_md) {
1382 			NL_SET_ERR_MSG(extack,
1383 				       "There already exists an externally controlled device on this destination port");
1384 			return -EPERM;
1385 		}
1386 	}
1387 
1388 	dst_cache_reset(&geneve->cfg.info.dst_cache);
1389 	memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1390 
1391 	err = register_netdevice(dev);
1392 	if (err)
1393 		return err;
1394 
1395 	list_add(&geneve->next, &gn->geneve_list);
1396 	return 0;
1397 }
1398 
1399 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1400 {
1401 	memset(info, 0, sizeof(*info));
1402 	info->key.tp_dst = htons(dst_port);
1403 }
1404 
1405 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1406 			  struct netlink_ext_ack *extack,
1407 			  struct geneve_config *cfg, bool changelink)
1408 {
1409 	struct ip_tunnel_info *info = &cfg->info;
1410 	int attrtype;
1411 
1412 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1413 		NL_SET_ERR_MSG(extack,
1414 			       "Cannot specify both IPv4 and IPv6 Remote addresses");
1415 		return -EINVAL;
1416 	}
1417 
1418 	if (data[IFLA_GENEVE_REMOTE]) {
1419 		if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1420 			attrtype = IFLA_GENEVE_REMOTE;
1421 			goto change_notsup;
1422 		}
1423 
1424 		info->key.u.ipv4.dst =
1425 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1426 
1427 		if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1428 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1429 					    "Remote IPv4 address cannot be Multicast");
1430 			return -EINVAL;
1431 		}
1432 	}
1433 
1434 	if (data[IFLA_GENEVE_REMOTE6]) {
1435 #if IS_ENABLED(CONFIG_IPV6)
1436 		if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1437 			attrtype = IFLA_GENEVE_REMOTE6;
1438 			goto change_notsup;
1439 		}
1440 
1441 		info->mode = IP_TUNNEL_INFO_IPV6;
1442 		info->key.u.ipv6.dst =
1443 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1444 
1445 		if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1446 		    IPV6_ADDR_LINKLOCAL) {
1447 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1448 					    "Remote IPv6 address cannot be link-local");
1449 			return -EINVAL;
1450 		}
1451 		if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1452 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1453 					    "Remote IPv6 address cannot be Multicast");
1454 			return -EINVAL;
1455 		}
1456 		info->key.tun_flags |= TUNNEL_CSUM;
1457 		cfg->use_udp6_rx_checksums = true;
1458 #else
1459 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1460 				    "IPv6 support not enabled in the kernel");
1461 		return -EPFNOSUPPORT;
1462 #endif
1463 	}
1464 
1465 	if (data[IFLA_GENEVE_ID]) {
1466 		__u32 vni;
1467 		__u8 tvni[3];
1468 		__be64 tunid;
1469 
1470 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1471 		tvni[0] = (vni & 0x00ff0000) >> 16;
1472 		tvni[1] = (vni & 0x0000ff00) >> 8;
1473 		tvni[2] =  vni & 0x000000ff;
1474 
1475 		tunid = vni_to_tunnel_id(tvni);
1476 		if (changelink && (tunid != info->key.tun_id)) {
1477 			attrtype = IFLA_GENEVE_ID;
1478 			goto change_notsup;
1479 		}
1480 		info->key.tun_id = tunid;
1481 	}
1482 
1483 	if (data[IFLA_GENEVE_TTL_INHERIT]) {
1484 		if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1485 			cfg->ttl_inherit = true;
1486 		else
1487 			cfg->ttl_inherit = false;
1488 	} else if (data[IFLA_GENEVE_TTL]) {
1489 		info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1490 		cfg->ttl_inherit = false;
1491 	}
1492 
1493 	if (data[IFLA_GENEVE_TOS])
1494 		info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1495 
1496 	if (data[IFLA_GENEVE_DF])
1497 		cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1498 
1499 	if (data[IFLA_GENEVE_LABEL]) {
1500 		info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1501 				  IPV6_FLOWLABEL_MASK;
1502 		if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1503 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1504 					    "Label attribute only applies for IPv6 Geneve devices");
1505 			return -EINVAL;
1506 		}
1507 	}
1508 
1509 	if (data[IFLA_GENEVE_PORT]) {
1510 		if (changelink) {
1511 			attrtype = IFLA_GENEVE_PORT;
1512 			goto change_notsup;
1513 		}
1514 		info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1515 	}
1516 
1517 	if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1518 		if (changelink) {
1519 			attrtype = IFLA_GENEVE_COLLECT_METADATA;
1520 			goto change_notsup;
1521 		}
1522 		cfg->collect_md = true;
1523 	}
1524 
1525 	if (data[IFLA_GENEVE_UDP_CSUM]) {
1526 		if (changelink) {
1527 			attrtype = IFLA_GENEVE_UDP_CSUM;
1528 			goto change_notsup;
1529 		}
1530 		if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1531 			info->key.tun_flags |= TUNNEL_CSUM;
1532 	}
1533 
1534 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1535 #if IS_ENABLED(CONFIG_IPV6)
1536 		if (changelink) {
1537 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1538 			goto change_notsup;
1539 		}
1540 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1541 			info->key.tun_flags &= ~TUNNEL_CSUM;
1542 #else
1543 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1544 				    "IPv6 support not enabled in the kernel");
1545 		return -EPFNOSUPPORT;
1546 #endif
1547 	}
1548 
1549 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1550 #if IS_ENABLED(CONFIG_IPV6)
1551 		if (changelink) {
1552 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1553 			goto change_notsup;
1554 		}
1555 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1556 			cfg->use_udp6_rx_checksums = false;
1557 #else
1558 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1559 				    "IPv6 support not enabled in the kernel");
1560 		return -EPFNOSUPPORT;
1561 #endif
1562 	}
1563 
1564 	return 0;
1565 change_notsup:
1566 	NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1567 			    "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1568 	return -EOPNOTSUPP;
1569 }
1570 
1571 static void geneve_link_config(struct net_device *dev,
1572 			       struct ip_tunnel_info *info, struct nlattr *tb[])
1573 {
1574 	struct geneve_dev *geneve = netdev_priv(dev);
1575 	int ldev_mtu = 0;
1576 
1577 	if (tb[IFLA_MTU]) {
1578 		geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1579 		return;
1580 	}
1581 
1582 	switch (ip_tunnel_info_af(info)) {
1583 	case AF_INET: {
1584 		struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1585 		struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1586 
1587 		if (!IS_ERR(rt) && rt->dst.dev) {
1588 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1589 			ip_rt_put(rt);
1590 		}
1591 		break;
1592 	}
1593 #if IS_ENABLED(CONFIG_IPV6)
1594 	case AF_INET6: {
1595 		struct rt6_info *rt;
1596 
1597 		if (!__in6_dev_get(dev))
1598 			break;
1599 
1600 		rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1601 				NULL, 0);
1602 
1603 		if (rt && rt->dst.dev)
1604 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1605 		ip6_rt_put(rt);
1606 		break;
1607 	}
1608 #endif
1609 	}
1610 
1611 	if (ldev_mtu <= 0)
1612 		return;
1613 
1614 	geneve_change_mtu(dev, ldev_mtu - info->options_len);
1615 }
1616 
1617 static int geneve_newlink(struct net *net, struct net_device *dev,
1618 			  struct nlattr *tb[], struct nlattr *data[],
1619 			  struct netlink_ext_ack *extack)
1620 {
1621 	struct geneve_config cfg = {
1622 		.df = GENEVE_DF_UNSET,
1623 		.use_udp6_rx_checksums = false,
1624 		.ttl_inherit = false,
1625 		.collect_md = false,
1626 	};
1627 	int err;
1628 
1629 	init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1630 	err = geneve_nl2info(tb, data, extack, &cfg, false);
1631 	if (err)
1632 		return err;
1633 
1634 	err = geneve_configure(net, dev, extack, &cfg);
1635 	if (err)
1636 		return err;
1637 
1638 	geneve_link_config(dev, &cfg.info, tb);
1639 
1640 	return 0;
1641 }
1642 
1643 /* Quiesces the geneve device data path for both TX and RX.
1644  *
1645  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1646  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1647  * to complete for the existing set of in-flight packets to be transmitted,
1648  * then we would have quiesced the transmit data path. All the future packets
1649  * will get dropped until we unquiesce the data path.
1650  *
1651  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1652  * if we set that to NULL under RCU and wait for synchronize_net() to
1653  * complete, then we would have quiesced the receive data path.
1654  */
1655 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1656 			   struct geneve_sock **gs6)
1657 {
1658 	*gs4 = rtnl_dereference(geneve->sock4);
1659 	rcu_assign_pointer(geneve->sock4, NULL);
1660 	if (*gs4)
1661 		rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1662 #if IS_ENABLED(CONFIG_IPV6)
1663 	*gs6 = rtnl_dereference(geneve->sock6);
1664 	rcu_assign_pointer(geneve->sock6, NULL);
1665 	if (*gs6)
1666 		rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1667 #else
1668 	*gs6 = NULL;
1669 #endif
1670 	synchronize_net();
1671 }
1672 
1673 /* Resumes the geneve device data path for both TX and RX. */
1674 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1675 			     struct geneve_sock __maybe_unused *gs6)
1676 {
1677 	rcu_assign_pointer(geneve->sock4, gs4);
1678 	if (gs4)
1679 		rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1680 #if IS_ENABLED(CONFIG_IPV6)
1681 	rcu_assign_pointer(geneve->sock6, gs6);
1682 	if (gs6)
1683 		rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1684 #endif
1685 	synchronize_net();
1686 }
1687 
1688 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1689 			     struct nlattr *data[],
1690 			     struct netlink_ext_ack *extack)
1691 {
1692 	struct geneve_dev *geneve = netdev_priv(dev);
1693 	struct geneve_sock *gs4, *gs6;
1694 	struct geneve_config cfg;
1695 	int err;
1696 
1697 	/* If the geneve device is configured for metadata (or externally
1698 	 * controlled, for example, OVS), then nothing can be changed.
1699 	 */
1700 	if (geneve->cfg.collect_md)
1701 		return -EOPNOTSUPP;
1702 
1703 	/* Start with the existing info. */
1704 	memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1705 	err = geneve_nl2info(tb, data, extack, &cfg, true);
1706 	if (err)
1707 		return err;
1708 
1709 	if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1710 		dst_cache_reset(&cfg.info.dst_cache);
1711 		geneve_link_config(dev, &cfg.info, tb);
1712 	}
1713 
1714 	geneve_quiesce(geneve, &gs4, &gs6);
1715 	memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1716 	geneve_unquiesce(geneve, gs4, gs6);
1717 
1718 	return 0;
1719 }
1720 
1721 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1722 {
1723 	struct geneve_dev *geneve = netdev_priv(dev);
1724 
1725 	list_del(&geneve->next);
1726 	unregister_netdevice_queue(dev, head);
1727 }
1728 
1729 static size_t geneve_get_size(const struct net_device *dev)
1730 {
1731 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1732 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1733 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1734 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1735 		nla_total_size(sizeof(__u8)) +	/* IFLA_GENEVE_DF */
1736 		nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1737 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1738 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1739 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1740 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1741 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1742 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1743 		0;
1744 }
1745 
1746 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1747 {
1748 	struct geneve_dev *geneve = netdev_priv(dev);
1749 	struct ip_tunnel_info *info = &geneve->cfg.info;
1750 	bool ttl_inherit = geneve->cfg.ttl_inherit;
1751 	bool metadata = geneve->cfg.collect_md;
1752 	__u8 tmp_vni[3];
1753 	__u32 vni;
1754 
1755 	tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1756 	vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1757 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1758 		goto nla_put_failure;
1759 
1760 	if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1761 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1762 				    info->key.u.ipv4.dst))
1763 			goto nla_put_failure;
1764 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1765 			       !!(info->key.tun_flags & TUNNEL_CSUM)))
1766 			goto nla_put_failure;
1767 
1768 #if IS_ENABLED(CONFIG_IPV6)
1769 	} else if (!metadata) {
1770 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1771 				     &info->key.u.ipv6.dst))
1772 			goto nla_put_failure;
1773 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1774 			       !(info->key.tun_flags & TUNNEL_CSUM)))
1775 			goto nla_put_failure;
1776 #endif
1777 	}
1778 
1779 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1780 	    nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1781 	    nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1782 		goto nla_put_failure;
1783 
1784 	if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1785 		goto nla_put_failure;
1786 
1787 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1788 		goto nla_put_failure;
1789 
1790 	if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1791 		goto nla_put_failure;
1792 
1793 #if IS_ENABLED(CONFIG_IPV6)
1794 	if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1795 		       !geneve->cfg.use_udp6_rx_checksums))
1796 		goto nla_put_failure;
1797 #endif
1798 
1799 	if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1800 		goto nla_put_failure;
1801 
1802 	return 0;
1803 
1804 nla_put_failure:
1805 	return -EMSGSIZE;
1806 }
1807 
1808 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1809 	.kind		= "geneve",
1810 	.maxtype	= IFLA_GENEVE_MAX,
1811 	.policy		= geneve_policy,
1812 	.priv_size	= sizeof(struct geneve_dev),
1813 	.setup		= geneve_setup,
1814 	.validate	= geneve_validate,
1815 	.newlink	= geneve_newlink,
1816 	.changelink	= geneve_changelink,
1817 	.dellink	= geneve_dellink,
1818 	.get_size	= geneve_get_size,
1819 	.fill_info	= geneve_fill_info,
1820 };
1821 
1822 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1823 					u8 name_assign_type, u16 dst_port)
1824 {
1825 	struct nlattr *tb[IFLA_MAX + 1];
1826 	struct net_device *dev;
1827 	LIST_HEAD(list_kill);
1828 	int err;
1829 	struct geneve_config cfg = {
1830 		.df = GENEVE_DF_UNSET,
1831 		.use_udp6_rx_checksums = true,
1832 		.ttl_inherit = false,
1833 		.collect_md = true,
1834 	};
1835 
1836 	memset(tb, 0, sizeof(tb));
1837 	dev = rtnl_create_link(net, name, name_assign_type,
1838 			       &geneve_link_ops, tb, NULL);
1839 	if (IS_ERR(dev))
1840 		return dev;
1841 
1842 	init_tnl_info(&cfg.info, dst_port);
1843 	err = geneve_configure(net, dev, NULL, &cfg);
1844 	if (err) {
1845 		free_netdev(dev);
1846 		return ERR_PTR(err);
1847 	}
1848 
1849 	/* openvswitch users expect packet sizes to be unrestricted,
1850 	 * so set the largest MTU we can.
1851 	 */
1852 	err = geneve_change_mtu(dev, IP_MAX_MTU);
1853 	if (err)
1854 		goto err;
1855 
1856 	err = rtnl_configure_link(dev, NULL);
1857 	if (err < 0)
1858 		goto err;
1859 
1860 	return dev;
1861 err:
1862 	geneve_dellink(dev, &list_kill);
1863 	unregister_netdevice_many(&list_kill);
1864 	return ERR_PTR(err);
1865 }
1866 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1867 
1868 static int geneve_netdevice_event(struct notifier_block *unused,
1869 				  unsigned long event, void *ptr)
1870 {
1871 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1872 
1873 	if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1874 		geneve_offload_rx_ports(dev, true);
1875 	else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1876 		geneve_offload_rx_ports(dev, false);
1877 
1878 	return NOTIFY_DONE;
1879 }
1880 
1881 static struct notifier_block geneve_notifier_block __read_mostly = {
1882 	.notifier_call = geneve_netdevice_event,
1883 };
1884 
1885 static __net_init int geneve_init_net(struct net *net)
1886 {
1887 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1888 
1889 	INIT_LIST_HEAD(&gn->geneve_list);
1890 	INIT_LIST_HEAD(&gn->sock_list);
1891 	return 0;
1892 }
1893 
1894 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1895 {
1896 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1897 	struct geneve_dev *geneve, *next;
1898 	struct net_device *dev, *aux;
1899 
1900 	/* gather any geneve devices that were moved into this ns */
1901 	for_each_netdev_safe(net, dev, aux)
1902 		if (dev->rtnl_link_ops == &geneve_link_ops)
1903 			unregister_netdevice_queue(dev, head);
1904 
1905 	/* now gather any other geneve devices that were created in this ns */
1906 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1907 		/* If geneve->dev is in the same netns, it was already added
1908 		 * to the list by the previous loop.
1909 		 */
1910 		if (!net_eq(dev_net(geneve->dev), net))
1911 			unregister_netdevice_queue(geneve->dev, head);
1912 	}
1913 }
1914 
1915 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1916 {
1917 	struct net *net;
1918 	LIST_HEAD(list);
1919 
1920 	rtnl_lock();
1921 	list_for_each_entry(net, net_list, exit_list)
1922 		geneve_destroy_tunnels(net, &list);
1923 
1924 	/* unregister the devices gathered above */
1925 	unregister_netdevice_many(&list);
1926 	rtnl_unlock();
1927 
1928 	list_for_each_entry(net, net_list, exit_list) {
1929 		const struct geneve_net *gn = net_generic(net, geneve_net_id);
1930 
1931 		WARN_ON_ONCE(!list_empty(&gn->sock_list));
1932 	}
1933 }
1934 
1935 static struct pernet_operations geneve_net_ops = {
1936 	.init = geneve_init_net,
1937 	.exit_batch = geneve_exit_batch_net,
1938 	.id   = &geneve_net_id,
1939 	.size = sizeof(struct geneve_net),
1940 };
1941 
1942 static int __init geneve_init_module(void)
1943 {
1944 	int rc;
1945 
1946 	rc = register_pernet_subsys(&geneve_net_ops);
1947 	if (rc)
1948 		goto out1;
1949 
1950 	rc = register_netdevice_notifier(&geneve_notifier_block);
1951 	if (rc)
1952 		goto out2;
1953 
1954 	rc = rtnl_link_register(&geneve_link_ops);
1955 	if (rc)
1956 		goto out3;
1957 
1958 	return 0;
1959 out3:
1960 	unregister_netdevice_notifier(&geneve_notifier_block);
1961 out2:
1962 	unregister_pernet_subsys(&geneve_net_ops);
1963 out1:
1964 	return rc;
1965 }
1966 late_initcall(geneve_init_module);
1967 
1968 static void __exit geneve_cleanup_module(void)
1969 {
1970 	rtnl_link_unregister(&geneve_link_ops);
1971 	unregister_netdevice_notifier(&geneve_notifier_block);
1972 	unregister_pernet_subsys(&geneve_net_ops);
1973 }
1974 module_exit(geneve_cleanup_module);
1975 
1976 MODULE_LICENSE("GPL");
1977 MODULE_VERSION(GENEVE_NETDEV_VER);
1978 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1979 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1980 MODULE_ALIAS_RTNL_LINK("geneve");
1981