xref: /linux/drivers/net/geneve.c (revision be239684b18e1cdcafcf8c7face4a2f562c745ad)
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 	bool			inner_proto_inherit;
60 };
61 
62 /* Pseudo network device */
63 struct geneve_dev {
64 	struct geneve_dev_node hlist4;	/* vni hash table for IPv4 socket */
65 #if IS_ENABLED(CONFIG_IPV6)
66 	struct geneve_dev_node hlist6;	/* vni hash table for IPv6 socket */
67 #endif
68 	struct net	   *net;	/* netns for packet i/o */
69 	struct net_device  *dev;	/* netdev for geneve tunnel */
70 	struct geneve_sock __rcu *sock4;	/* IPv4 socket used for geneve tunnel */
71 #if IS_ENABLED(CONFIG_IPV6)
72 	struct geneve_sock __rcu *sock6;	/* IPv6 socket used for geneve tunnel */
73 #endif
74 	struct list_head   next;	/* geneve's per namespace list */
75 	struct gro_cells   gro_cells;
76 	struct geneve_config cfg;
77 };
78 
79 struct geneve_sock {
80 	bool			collect_md;
81 	struct list_head	list;
82 	struct socket		*sock;
83 	struct rcu_head		rcu;
84 	int			refcnt;
85 	struct hlist_head	vni_list[VNI_HASH_SIZE];
86 };
87 
88 static inline __u32 geneve_net_vni_hash(u8 vni[3])
89 {
90 	__u32 vnid;
91 
92 	vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
93 	return hash_32(vnid, VNI_HASH_BITS);
94 }
95 
96 static __be64 vni_to_tunnel_id(const __u8 *vni)
97 {
98 #ifdef __BIG_ENDIAN
99 	return (vni[0] << 16) | (vni[1] << 8) | vni[2];
100 #else
101 	return (__force __be64)(((__force u64)vni[0] << 40) |
102 				((__force u64)vni[1] << 48) |
103 				((__force u64)vni[2] << 56));
104 #endif
105 }
106 
107 /* Convert 64 bit tunnel ID to 24 bit VNI. */
108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
109 {
110 #ifdef __BIG_ENDIAN
111 	vni[0] = (__force __u8)(tun_id >> 16);
112 	vni[1] = (__force __u8)(tun_id >> 8);
113 	vni[2] = (__force __u8)tun_id;
114 #else
115 	vni[0] = (__force __u8)((__force u64)tun_id >> 40);
116 	vni[1] = (__force __u8)((__force u64)tun_id >> 48);
117 	vni[2] = (__force __u8)((__force u64)tun_id >> 56);
118 #endif
119 }
120 
121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
122 {
123 	return !memcmp(vni, &tun_id[5], 3);
124 }
125 
126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
127 {
128 	return gs->sock->sk->sk_family;
129 }
130 
131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
132 					__be32 addr, u8 vni[])
133 {
134 	struct hlist_head *vni_list_head;
135 	struct geneve_dev_node *node;
136 	__u32 hash;
137 
138 	/* Find the device for this VNI */
139 	hash = geneve_net_vni_hash(vni);
140 	vni_list_head = &gs->vni_list[hash];
141 	hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
142 		if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
143 		    addr == node->geneve->cfg.info.key.u.ipv4.dst)
144 			return node->geneve;
145 	}
146 	return NULL;
147 }
148 
149 #if IS_ENABLED(CONFIG_IPV6)
150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
151 					 struct in6_addr addr6, u8 vni[])
152 {
153 	struct hlist_head *vni_list_head;
154 	struct geneve_dev_node *node;
155 	__u32 hash;
156 
157 	/* Find the device for this VNI */
158 	hash = geneve_net_vni_hash(vni);
159 	vni_list_head = &gs->vni_list[hash];
160 	hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
161 		if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
162 		    ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
163 			return node->geneve;
164 	}
165 	return NULL;
166 }
167 #endif
168 
169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
170 {
171 	return (struct genevehdr *)(udp_hdr(skb) + 1);
172 }
173 
174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
175 					    struct sk_buff *skb)
176 {
177 	static u8 zero_vni[3];
178 	u8 *vni;
179 
180 	if (geneve_get_sk_family(gs) == AF_INET) {
181 		struct iphdr *iph;
182 		__be32 addr;
183 
184 		iph = ip_hdr(skb); /* outer IP header... */
185 
186 		if (gs->collect_md) {
187 			vni = zero_vni;
188 			addr = 0;
189 		} else {
190 			vni = geneve_hdr(skb)->vni;
191 			addr = iph->saddr;
192 		}
193 
194 		return geneve_lookup(gs, addr, vni);
195 #if IS_ENABLED(CONFIG_IPV6)
196 	} else if (geneve_get_sk_family(gs) == AF_INET6) {
197 		static struct in6_addr zero_addr6;
198 		struct ipv6hdr *ip6h;
199 		struct in6_addr addr6;
200 
201 		ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
202 
203 		if (gs->collect_md) {
204 			vni = zero_vni;
205 			addr6 = zero_addr6;
206 		} else {
207 			vni = geneve_hdr(skb)->vni;
208 			addr6 = ip6h->saddr;
209 		}
210 
211 		return geneve6_lookup(gs, addr6, vni);
212 #endif
213 	}
214 	return NULL;
215 }
216 
217 /* geneve receive/decap routine */
218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
219 		      struct sk_buff *skb)
220 {
221 	struct genevehdr *gnvh = geneve_hdr(skb);
222 	struct metadata_dst *tun_dst = NULL;
223 	unsigned int len;
224 	int err = 0;
225 	void *oiph;
226 
227 	if (ip_tunnel_collect_metadata() || gs->collect_md) {
228 		__be16 flags;
229 
230 		flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
231 			(gnvh->critical ? TUNNEL_CRIT_OPT : 0);
232 
233 		tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
234 					 vni_to_tunnel_id(gnvh->vni),
235 					 gnvh->opt_len * 4);
236 		if (!tun_dst) {
237 			DEV_STATS_INC(geneve->dev, rx_dropped);
238 			goto drop;
239 		}
240 		/* Update tunnel dst according to Geneve options. */
241 		ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
242 					gnvh->options, gnvh->opt_len * 4,
243 					TUNNEL_GENEVE_OPT);
244 	} else {
245 		/* Drop packets w/ critical options,
246 		 * since we don't support any...
247 		 */
248 		if (gnvh->critical) {
249 			DEV_STATS_INC(geneve->dev, rx_frame_errors);
250 			DEV_STATS_INC(geneve->dev, rx_errors);
251 			goto drop;
252 		}
253 	}
254 
255 	if (tun_dst)
256 		skb_dst_set(skb, &tun_dst->dst);
257 
258 	if (gnvh->proto_type == htons(ETH_P_TEB)) {
259 		skb_reset_mac_header(skb);
260 		skb->protocol = eth_type_trans(skb, geneve->dev);
261 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
262 
263 		/* Ignore packet loops (and multicast echo) */
264 		if (ether_addr_equal(eth_hdr(skb)->h_source,
265 				     geneve->dev->dev_addr)) {
266 			DEV_STATS_INC(geneve->dev, rx_errors);
267 			goto drop;
268 		}
269 	} else {
270 		skb_reset_mac_header(skb);
271 		skb->dev = geneve->dev;
272 		skb->pkt_type = PACKET_HOST;
273 	}
274 
275 	oiph = skb_network_header(skb);
276 	skb_reset_network_header(skb);
277 
278 	if (geneve_get_sk_family(gs) == AF_INET)
279 		err = IP_ECN_decapsulate(oiph, skb);
280 #if IS_ENABLED(CONFIG_IPV6)
281 	else
282 		err = IP6_ECN_decapsulate(oiph, skb);
283 #endif
284 
285 	if (unlikely(err)) {
286 		if (log_ecn_error) {
287 			if (geneve_get_sk_family(gs) == AF_INET)
288 				net_info_ratelimited("non-ECT from %pI4 "
289 						     "with TOS=%#x\n",
290 						     &((struct iphdr *)oiph)->saddr,
291 						     ((struct iphdr *)oiph)->tos);
292 #if IS_ENABLED(CONFIG_IPV6)
293 			else
294 				net_info_ratelimited("non-ECT from %pI6\n",
295 						     &((struct ipv6hdr *)oiph)->saddr);
296 #endif
297 		}
298 		if (err > 1) {
299 			DEV_STATS_INC(geneve->dev, rx_frame_errors);
300 			DEV_STATS_INC(geneve->dev, rx_errors);
301 			goto drop;
302 		}
303 	}
304 
305 	len = skb->len;
306 	err = gro_cells_receive(&geneve->gro_cells, skb);
307 	if (likely(err == NET_RX_SUCCESS))
308 		dev_sw_netstats_rx_add(geneve->dev, len);
309 
310 	return;
311 drop:
312 	/* Consume bad packet */
313 	kfree_skb(skb);
314 }
315 
316 /* Setup stats when device is created */
317 static int geneve_init(struct net_device *dev)
318 {
319 	struct geneve_dev *geneve = netdev_priv(dev);
320 	int err;
321 
322 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
323 	if (!dev->tstats)
324 		return -ENOMEM;
325 
326 	err = gro_cells_init(&geneve->gro_cells, dev);
327 	if (err) {
328 		free_percpu(dev->tstats);
329 		return err;
330 	}
331 
332 	err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
333 	if (err) {
334 		free_percpu(dev->tstats);
335 		gro_cells_destroy(&geneve->gro_cells);
336 		return err;
337 	}
338 	return 0;
339 }
340 
341 static void geneve_uninit(struct net_device *dev)
342 {
343 	struct geneve_dev *geneve = netdev_priv(dev);
344 
345 	dst_cache_destroy(&geneve->cfg.info.dst_cache);
346 	gro_cells_destroy(&geneve->gro_cells);
347 	free_percpu(dev->tstats);
348 }
349 
350 /* Callback from net/ipv4/udp.c to receive packets */
351 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
352 {
353 	struct genevehdr *geneveh;
354 	struct geneve_dev *geneve;
355 	struct geneve_sock *gs;
356 	__be16 inner_proto;
357 	int opts_len;
358 
359 	/* Need UDP and Geneve header to be present */
360 	if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
361 		goto drop;
362 
363 	/* Return packets with reserved bits set */
364 	geneveh = geneve_hdr(skb);
365 	if (unlikely(geneveh->ver != GENEVE_VER))
366 		goto drop;
367 
368 	gs = rcu_dereference_sk_user_data(sk);
369 	if (!gs)
370 		goto drop;
371 
372 	geneve = geneve_lookup_skb(gs, skb);
373 	if (!geneve)
374 		goto drop;
375 
376 	inner_proto = geneveh->proto_type;
377 
378 	if (unlikely((!geneve->cfg.inner_proto_inherit &&
379 		      inner_proto != htons(ETH_P_TEB)))) {
380 		DEV_STATS_INC(geneve->dev, rx_dropped);
381 		goto drop;
382 	}
383 
384 	opts_len = geneveh->opt_len * 4;
385 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
386 				 !net_eq(geneve->net, dev_net(geneve->dev)))) {
387 		DEV_STATS_INC(geneve->dev, rx_dropped);
388 		goto drop;
389 	}
390 
391 	geneve_rx(geneve, gs, skb);
392 	return 0;
393 
394 drop:
395 	/* Consume bad packet */
396 	kfree_skb(skb);
397 	return 0;
398 }
399 
400 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
401 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
402 {
403 	struct genevehdr *geneveh;
404 	struct geneve_sock *gs;
405 	u8 zero_vni[3] = { 0 };
406 	u8 *vni = zero_vni;
407 
408 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
409 		return -EINVAL;
410 
411 	geneveh = geneve_hdr(skb);
412 	if (geneveh->ver != GENEVE_VER)
413 		return -EINVAL;
414 
415 	if (geneveh->proto_type != htons(ETH_P_TEB))
416 		return -EINVAL;
417 
418 	gs = rcu_dereference_sk_user_data(sk);
419 	if (!gs)
420 		return -ENOENT;
421 
422 	if (geneve_get_sk_family(gs) == AF_INET) {
423 		struct iphdr *iph = ip_hdr(skb);
424 		__be32 addr4 = 0;
425 
426 		if (!gs->collect_md) {
427 			vni = geneve_hdr(skb)->vni;
428 			addr4 = iph->daddr;
429 		}
430 
431 		return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
432 	}
433 
434 #if IS_ENABLED(CONFIG_IPV6)
435 	if (geneve_get_sk_family(gs) == AF_INET6) {
436 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
437 		struct in6_addr addr6;
438 
439 		memset(&addr6, 0, sizeof(struct in6_addr));
440 
441 		if (!gs->collect_md) {
442 			vni = geneve_hdr(skb)->vni;
443 			addr6 = ip6h->daddr;
444 		}
445 
446 		return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
447 	}
448 #endif
449 
450 	return -EPFNOSUPPORT;
451 }
452 
453 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
454 					 __be16 port, bool ipv6_rx_csum)
455 {
456 	struct socket *sock;
457 	struct udp_port_cfg udp_conf;
458 	int err;
459 
460 	memset(&udp_conf, 0, sizeof(udp_conf));
461 
462 	if (ipv6) {
463 		udp_conf.family = AF_INET6;
464 		udp_conf.ipv6_v6only = 1;
465 		udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
466 	} else {
467 		udp_conf.family = AF_INET;
468 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
469 	}
470 
471 	udp_conf.local_udp_port = port;
472 
473 	/* Open UDP socket */
474 	err = udp_sock_create(net, &udp_conf, &sock);
475 	if (err < 0)
476 		return ERR_PTR(err);
477 
478 	udp_allow_gso(sock->sk);
479 	return sock;
480 }
481 
482 static int geneve_hlen(struct genevehdr *gh)
483 {
484 	return sizeof(*gh) + gh->opt_len * 4;
485 }
486 
487 static struct sk_buff *geneve_gro_receive(struct sock *sk,
488 					  struct list_head *head,
489 					  struct sk_buff *skb)
490 {
491 	struct sk_buff *pp = NULL;
492 	struct sk_buff *p;
493 	struct genevehdr *gh, *gh2;
494 	unsigned int hlen, gh_len, off_gnv;
495 	const struct packet_offload *ptype;
496 	__be16 type;
497 	int flush = 1;
498 
499 	off_gnv = skb_gro_offset(skb);
500 	hlen = off_gnv + sizeof(*gh);
501 	gh = skb_gro_header(skb, hlen, off_gnv);
502 	if (unlikely(!gh))
503 		goto out;
504 
505 	if (gh->ver != GENEVE_VER || gh->oam)
506 		goto out;
507 	gh_len = geneve_hlen(gh);
508 
509 	hlen = off_gnv + gh_len;
510 	if (skb_gro_header_hard(skb, hlen)) {
511 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
512 		if (unlikely(!gh))
513 			goto out;
514 	}
515 
516 	list_for_each_entry(p, head, list) {
517 		if (!NAPI_GRO_CB(p)->same_flow)
518 			continue;
519 
520 		gh2 = (struct genevehdr *)(p->data + off_gnv);
521 		if (gh->opt_len != gh2->opt_len ||
522 		    memcmp(gh, gh2, gh_len)) {
523 			NAPI_GRO_CB(p)->same_flow = 0;
524 			continue;
525 		}
526 	}
527 
528 	skb_gro_pull(skb, gh_len);
529 	skb_gro_postpull_rcsum(skb, gh, gh_len);
530 	type = gh->proto_type;
531 	if (likely(type == htons(ETH_P_TEB)))
532 		return call_gro_receive(eth_gro_receive, head, skb);
533 
534 	ptype = gro_find_receive_by_type(type);
535 	if (!ptype)
536 		goto out;
537 
538 	pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
539 	flush = 0;
540 
541 out:
542 	skb_gro_flush_final(skb, pp, flush);
543 
544 	return pp;
545 }
546 
547 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
548 			       int nhoff)
549 {
550 	struct genevehdr *gh;
551 	struct packet_offload *ptype;
552 	__be16 type;
553 	int gh_len;
554 	int err = -ENOSYS;
555 
556 	gh = (struct genevehdr *)(skb->data + nhoff);
557 	gh_len = geneve_hlen(gh);
558 	type = gh->proto_type;
559 
560 	/* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
561 	if (likely(type == htons(ETH_P_TEB)))
562 		return eth_gro_complete(skb, nhoff + gh_len);
563 
564 	ptype = gro_find_complete_by_type(type);
565 	if (ptype)
566 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
567 
568 	skb_set_inner_mac_header(skb, nhoff + gh_len);
569 
570 	return err;
571 }
572 
573 /* Create new listen socket if needed */
574 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
575 						bool ipv6, bool ipv6_rx_csum)
576 {
577 	struct geneve_net *gn = net_generic(net, geneve_net_id);
578 	struct geneve_sock *gs;
579 	struct socket *sock;
580 	struct udp_tunnel_sock_cfg tunnel_cfg;
581 	int h;
582 
583 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
584 	if (!gs)
585 		return ERR_PTR(-ENOMEM);
586 
587 	sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
588 	if (IS_ERR(sock)) {
589 		kfree(gs);
590 		return ERR_CAST(sock);
591 	}
592 
593 	gs->sock = sock;
594 	gs->refcnt = 1;
595 	for (h = 0; h < VNI_HASH_SIZE; ++h)
596 		INIT_HLIST_HEAD(&gs->vni_list[h]);
597 
598 	/* Initialize the geneve udp offloads structure */
599 	udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
600 
601 	/* Mark socket as an encapsulation socket */
602 	memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
603 	tunnel_cfg.sk_user_data = gs;
604 	tunnel_cfg.encap_type = 1;
605 	tunnel_cfg.gro_receive = geneve_gro_receive;
606 	tunnel_cfg.gro_complete = geneve_gro_complete;
607 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
608 	tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
609 	tunnel_cfg.encap_destroy = NULL;
610 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
611 	list_add(&gs->list, &gn->sock_list);
612 	return gs;
613 }
614 
615 static void __geneve_sock_release(struct geneve_sock *gs)
616 {
617 	if (!gs || --gs->refcnt)
618 		return;
619 
620 	list_del(&gs->list);
621 	udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
622 	udp_tunnel_sock_release(gs->sock);
623 	kfree_rcu(gs, rcu);
624 }
625 
626 static void geneve_sock_release(struct geneve_dev *geneve)
627 {
628 	struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
629 #if IS_ENABLED(CONFIG_IPV6)
630 	struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
631 
632 	rcu_assign_pointer(geneve->sock6, NULL);
633 #endif
634 
635 	rcu_assign_pointer(geneve->sock4, NULL);
636 	synchronize_net();
637 
638 	__geneve_sock_release(gs4);
639 #if IS_ENABLED(CONFIG_IPV6)
640 	__geneve_sock_release(gs6);
641 #endif
642 }
643 
644 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
645 					    sa_family_t family,
646 					    __be16 dst_port)
647 {
648 	struct geneve_sock *gs;
649 
650 	list_for_each_entry(gs, &gn->sock_list, list) {
651 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
652 		    geneve_get_sk_family(gs) == family) {
653 			return gs;
654 		}
655 	}
656 	return NULL;
657 }
658 
659 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
660 {
661 	struct net *net = geneve->net;
662 	struct geneve_net *gn = net_generic(net, geneve_net_id);
663 	struct geneve_dev_node *node;
664 	struct geneve_sock *gs;
665 	__u8 vni[3];
666 	__u32 hash;
667 
668 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
669 	if (gs) {
670 		gs->refcnt++;
671 		goto out;
672 	}
673 
674 	gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
675 				  geneve->cfg.use_udp6_rx_checksums);
676 	if (IS_ERR(gs))
677 		return PTR_ERR(gs);
678 
679 out:
680 	gs->collect_md = geneve->cfg.collect_md;
681 #if IS_ENABLED(CONFIG_IPV6)
682 	if (ipv6) {
683 		rcu_assign_pointer(geneve->sock6, gs);
684 		node = &geneve->hlist6;
685 	} else
686 #endif
687 	{
688 		rcu_assign_pointer(geneve->sock4, gs);
689 		node = &geneve->hlist4;
690 	}
691 	node->geneve = geneve;
692 
693 	tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
694 	hash = geneve_net_vni_hash(vni);
695 	hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
696 	return 0;
697 }
698 
699 static int geneve_open(struct net_device *dev)
700 {
701 	struct geneve_dev *geneve = netdev_priv(dev);
702 	bool metadata = geneve->cfg.collect_md;
703 	bool ipv4, ipv6;
704 	int ret = 0;
705 
706 	ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
707 	ipv4 = !ipv6 || metadata;
708 #if IS_ENABLED(CONFIG_IPV6)
709 	if (ipv6) {
710 		ret = geneve_sock_add(geneve, true);
711 		if (ret < 0 && ret != -EAFNOSUPPORT)
712 			ipv4 = false;
713 	}
714 #endif
715 	if (ipv4)
716 		ret = geneve_sock_add(geneve, false);
717 	if (ret < 0)
718 		geneve_sock_release(geneve);
719 
720 	return ret;
721 }
722 
723 static int geneve_stop(struct net_device *dev)
724 {
725 	struct geneve_dev *geneve = netdev_priv(dev);
726 
727 	hlist_del_init_rcu(&geneve->hlist4.hlist);
728 #if IS_ENABLED(CONFIG_IPV6)
729 	hlist_del_init_rcu(&geneve->hlist6.hlist);
730 #endif
731 	geneve_sock_release(geneve);
732 	return 0;
733 }
734 
735 static void geneve_build_header(struct genevehdr *geneveh,
736 				const struct ip_tunnel_info *info,
737 				__be16 inner_proto)
738 {
739 	geneveh->ver = GENEVE_VER;
740 	geneveh->opt_len = info->options_len / 4;
741 	geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
742 	geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
743 	geneveh->rsvd1 = 0;
744 	tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
745 	geneveh->proto_type = inner_proto;
746 	geneveh->rsvd2 = 0;
747 
748 	if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
749 		ip_tunnel_info_opts_get(geneveh->options, info);
750 }
751 
752 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
753 			    const struct ip_tunnel_info *info,
754 			    bool xnet, int ip_hdr_len,
755 			    bool inner_proto_inherit)
756 {
757 	bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
758 	struct genevehdr *gnvh;
759 	__be16 inner_proto;
760 	int min_headroom;
761 	int err;
762 
763 	skb_reset_mac_header(skb);
764 	skb_scrub_packet(skb, xnet);
765 
766 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
767 		       GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
768 	err = skb_cow_head(skb, min_headroom);
769 	if (unlikely(err))
770 		goto free_dst;
771 
772 	err = udp_tunnel_handle_offloads(skb, udp_sum);
773 	if (err)
774 		goto free_dst;
775 
776 	gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
777 	inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
778 	geneve_build_header(gnvh, info, inner_proto);
779 	skb_set_inner_protocol(skb, inner_proto);
780 	return 0;
781 
782 free_dst:
783 	dst_release(dst);
784 	return err;
785 }
786 
787 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
788 			     const struct ip_tunnel_info *info,
789 			     bool *use_cache)
790 {
791 	struct geneve_dev *geneve = netdev_priv(dev);
792 	u8 dsfield;
793 
794 	dsfield = info->key.tos;
795 	if (dsfield == 1 && !geneve->cfg.collect_md) {
796 		dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
797 		*use_cache = false;
798 	}
799 
800 	return dsfield;
801 }
802 
803 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
804 			   struct geneve_dev *geneve,
805 			   const struct ip_tunnel_info *info)
806 {
807 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
808 	struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
809 	const struct ip_tunnel_key *key = &info->key;
810 	struct rtable *rt;
811 	bool use_cache;
812 	__u8 tos, ttl;
813 	__be16 df = 0;
814 	__be32 saddr;
815 	__be16 sport;
816 	int err;
817 
818 	if (!pskb_inet_may_pull(skb))
819 		return -EINVAL;
820 
821 	if (!gs4)
822 		return -EIO;
823 
824 	use_cache = ip_tunnel_dst_cache_usable(skb, info);
825 	tos = geneve_get_dsfield(skb, dev, info, &use_cache);
826 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
827 
828 	rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
829 				   &info->key,
830 				   sport, geneve->cfg.info.key.tp_dst, tos,
831 				   use_cache ?
832 				   (struct dst_cache *)&info->dst_cache : NULL);
833 	if (IS_ERR(rt))
834 		return PTR_ERR(rt);
835 
836 	err = skb_tunnel_check_pmtu(skb, &rt->dst,
837 				    GENEVE_IPV4_HLEN + info->options_len,
838 				    netif_is_any_bridge_port(dev));
839 	if (err < 0) {
840 		dst_release(&rt->dst);
841 		return err;
842 	} else if (err) {
843 		struct ip_tunnel_info *info;
844 
845 		info = skb_tunnel_info(skb);
846 		if (info) {
847 			struct ip_tunnel_info *unclone;
848 
849 			unclone = skb_tunnel_info_unclone(skb);
850 			if (unlikely(!unclone)) {
851 				dst_release(&rt->dst);
852 				return -ENOMEM;
853 			}
854 
855 			unclone->key.u.ipv4.dst = saddr;
856 			unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
857 		}
858 
859 		if (!pskb_may_pull(skb, ETH_HLEN)) {
860 			dst_release(&rt->dst);
861 			return -EINVAL;
862 		}
863 
864 		skb->protocol = eth_type_trans(skb, geneve->dev);
865 		__netif_rx(skb);
866 		dst_release(&rt->dst);
867 		return -EMSGSIZE;
868 	}
869 
870 	tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
871 	if (geneve->cfg.collect_md) {
872 		ttl = key->ttl;
873 
874 		df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
875 	} else {
876 		if (geneve->cfg.ttl_inherit)
877 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
878 		else
879 			ttl = key->ttl;
880 		ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
881 
882 		if (geneve->cfg.df == GENEVE_DF_SET) {
883 			df = htons(IP_DF);
884 		} else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
885 			struct ethhdr *eth = eth_hdr(skb);
886 
887 			if (ntohs(eth->h_proto) == ETH_P_IPV6) {
888 				df = htons(IP_DF);
889 			} else if (ntohs(eth->h_proto) == ETH_P_IP) {
890 				struct iphdr *iph = ip_hdr(skb);
891 
892 				if (iph->frag_off & htons(IP_DF))
893 					df = htons(IP_DF);
894 			}
895 		}
896 	}
897 
898 	err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
899 			       geneve->cfg.inner_proto_inherit);
900 	if (unlikely(err))
901 		return err;
902 
903 	udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
904 			    tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
905 			    !net_eq(geneve->net, dev_net(geneve->dev)),
906 			    !(info->key.tun_flags & TUNNEL_CSUM));
907 	return 0;
908 }
909 
910 #if IS_ENABLED(CONFIG_IPV6)
911 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
912 			    struct geneve_dev *geneve,
913 			    const struct ip_tunnel_info *info)
914 {
915 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
916 	struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
917 	const struct ip_tunnel_key *key = &info->key;
918 	struct dst_entry *dst = NULL;
919 	struct in6_addr saddr;
920 	bool use_cache;
921 	__u8 prio, ttl;
922 	__be16 sport;
923 	int err;
924 
925 	if (!pskb_inet_may_pull(skb))
926 		return -EINVAL;
927 
928 	if (!gs6)
929 		return -EIO;
930 
931 	use_cache = ip_tunnel_dst_cache_usable(skb, info);
932 	prio = geneve_get_dsfield(skb, dev, info, &use_cache);
933 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
934 
935 	dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
936 				     &saddr, key, sport,
937 				     geneve->cfg.info.key.tp_dst, prio,
938 				     use_cache ?
939 				     (struct dst_cache *)&info->dst_cache : NULL);
940 	if (IS_ERR(dst))
941 		return PTR_ERR(dst);
942 
943 	err = skb_tunnel_check_pmtu(skb, dst,
944 				    GENEVE_IPV6_HLEN + info->options_len,
945 				    netif_is_any_bridge_port(dev));
946 	if (err < 0) {
947 		dst_release(dst);
948 		return err;
949 	} else if (err) {
950 		struct ip_tunnel_info *info = skb_tunnel_info(skb);
951 
952 		if (info) {
953 			struct ip_tunnel_info *unclone;
954 
955 			unclone = skb_tunnel_info_unclone(skb);
956 			if (unlikely(!unclone)) {
957 				dst_release(dst);
958 				return -ENOMEM;
959 			}
960 
961 			unclone->key.u.ipv6.dst = saddr;
962 			unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
963 		}
964 
965 		if (!pskb_may_pull(skb, ETH_HLEN)) {
966 			dst_release(dst);
967 			return -EINVAL;
968 		}
969 
970 		skb->protocol = eth_type_trans(skb, geneve->dev);
971 		__netif_rx(skb);
972 		dst_release(dst);
973 		return -EMSGSIZE;
974 	}
975 
976 	prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
977 	if (geneve->cfg.collect_md) {
978 		ttl = key->ttl;
979 	} else {
980 		if (geneve->cfg.ttl_inherit)
981 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
982 		else
983 			ttl = key->ttl;
984 		ttl = ttl ? : ip6_dst_hoplimit(dst);
985 	}
986 	err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
987 			       geneve->cfg.inner_proto_inherit);
988 	if (unlikely(err))
989 		return err;
990 
991 	udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
992 			     &saddr, &key->u.ipv6.dst, prio, ttl,
993 			     info->key.label, sport, geneve->cfg.info.key.tp_dst,
994 			     !(info->key.tun_flags & TUNNEL_CSUM));
995 	return 0;
996 }
997 #endif
998 
999 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1000 {
1001 	struct geneve_dev *geneve = netdev_priv(dev);
1002 	struct ip_tunnel_info *info = NULL;
1003 	int err;
1004 
1005 	if (geneve->cfg.collect_md) {
1006 		info = skb_tunnel_info(skb);
1007 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1008 			netdev_dbg(dev, "no tunnel metadata\n");
1009 			dev_kfree_skb(skb);
1010 			DEV_STATS_INC(dev, tx_dropped);
1011 			return NETDEV_TX_OK;
1012 		}
1013 	} else {
1014 		info = &geneve->cfg.info;
1015 	}
1016 
1017 	rcu_read_lock();
1018 #if IS_ENABLED(CONFIG_IPV6)
1019 	if (info->mode & IP_TUNNEL_INFO_IPV6)
1020 		err = geneve6_xmit_skb(skb, dev, geneve, info);
1021 	else
1022 #endif
1023 		err = geneve_xmit_skb(skb, dev, geneve, info);
1024 	rcu_read_unlock();
1025 
1026 	if (likely(!err))
1027 		return NETDEV_TX_OK;
1028 
1029 	if (err != -EMSGSIZE)
1030 		dev_kfree_skb(skb);
1031 
1032 	if (err == -ELOOP)
1033 		DEV_STATS_INC(dev, collisions);
1034 	else if (err == -ENETUNREACH)
1035 		DEV_STATS_INC(dev, tx_carrier_errors);
1036 
1037 	DEV_STATS_INC(dev, tx_errors);
1038 	return NETDEV_TX_OK;
1039 }
1040 
1041 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1042 {
1043 	if (new_mtu > dev->max_mtu)
1044 		new_mtu = dev->max_mtu;
1045 	else if (new_mtu < dev->min_mtu)
1046 		new_mtu = dev->min_mtu;
1047 
1048 	dev->mtu = new_mtu;
1049 	return 0;
1050 }
1051 
1052 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1053 {
1054 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
1055 	struct geneve_dev *geneve = netdev_priv(dev);
1056 	__be16 sport;
1057 
1058 	if (ip_tunnel_info_af(info) == AF_INET) {
1059 		struct rtable *rt;
1060 		struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1061 		bool use_cache;
1062 		__be32 saddr;
1063 		u8 tos;
1064 
1065 		if (!gs4)
1066 			return -EIO;
1067 
1068 		use_cache = ip_tunnel_dst_cache_usable(skb, info);
1069 		tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1070 		sport = udp_flow_src_port(geneve->net, skb,
1071 					  1, USHRT_MAX, true);
1072 
1073 		rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1074 					   &info->key,
1075 					   sport, geneve->cfg.info.key.tp_dst,
1076 					   tos,
1077 					   use_cache ? &info->dst_cache : NULL);
1078 		if (IS_ERR(rt))
1079 			return PTR_ERR(rt);
1080 
1081 		ip_rt_put(rt);
1082 		info->key.u.ipv4.src = saddr;
1083 #if IS_ENABLED(CONFIG_IPV6)
1084 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
1085 		struct dst_entry *dst;
1086 		struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1087 		struct in6_addr saddr;
1088 		bool use_cache;
1089 		u8 prio;
1090 
1091 		if (!gs6)
1092 			return -EIO;
1093 
1094 		use_cache = ip_tunnel_dst_cache_usable(skb, info);
1095 		prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1096 		sport = udp_flow_src_port(geneve->net, skb,
1097 					  1, USHRT_MAX, true);
1098 
1099 		dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1100 					     &saddr, &info->key, sport,
1101 					     geneve->cfg.info.key.tp_dst, prio,
1102 					     use_cache ? &info->dst_cache : NULL);
1103 		if (IS_ERR(dst))
1104 			return PTR_ERR(dst);
1105 
1106 		dst_release(dst);
1107 		info->key.u.ipv6.src = saddr;
1108 #endif
1109 	} else {
1110 		return -EINVAL;
1111 	}
1112 
1113 	info->key.tp_src = sport;
1114 	info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1115 	return 0;
1116 }
1117 
1118 static const struct net_device_ops geneve_netdev_ops = {
1119 	.ndo_init		= geneve_init,
1120 	.ndo_uninit		= geneve_uninit,
1121 	.ndo_open		= geneve_open,
1122 	.ndo_stop		= geneve_stop,
1123 	.ndo_start_xmit		= geneve_xmit,
1124 	.ndo_get_stats64	= dev_get_tstats64,
1125 	.ndo_change_mtu		= geneve_change_mtu,
1126 	.ndo_validate_addr	= eth_validate_addr,
1127 	.ndo_set_mac_address	= eth_mac_addr,
1128 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1129 };
1130 
1131 static void geneve_get_drvinfo(struct net_device *dev,
1132 			       struct ethtool_drvinfo *drvinfo)
1133 {
1134 	strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1135 	strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1136 }
1137 
1138 static const struct ethtool_ops geneve_ethtool_ops = {
1139 	.get_drvinfo	= geneve_get_drvinfo,
1140 	.get_link	= ethtool_op_get_link,
1141 };
1142 
1143 /* Info for udev, that this is a virtual tunnel endpoint */
1144 static struct device_type geneve_type = {
1145 	.name = "geneve",
1146 };
1147 
1148 /* Calls the ndo_udp_tunnel_add of the caller in order to
1149  * supply the listening GENEVE udp ports. Callers are expected
1150  * to implement the ndo_udp_tunnel_add.
1151  */
1152 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1153 {
1154 	struct net *net = dev_net(dev);
1155 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1156 	struct geneve_sock *gs;
1157 
1158 	rcu_read_lock();
1159 	list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1160 		if (push) {
1161 			udp_tunnel_push_rx_port(dev, gs->sock,
1162 						UDP_TUNNEL_TYPE_GENEVE);
1163 		} else {
1164 			udp_tunnel_drop_rx_port(dev, gs->sock,
1165 						UDP_TUNNEL_TYPE_GENEVE);
1166 		}
1167 	}
1168 	rcu_read_unlock();
1169 }
1170 
1171 /* Initialize the device structure. */
1172 static void geneve_setup(struct net_device *dev)
1173 {
1174 	ether_setup(dev);
1175 
1176 	dev->netdev_ops = &geneve_netdev_ops;
1177 	dev->ethtool_ops = &geneve_ethtool_ops;
1178 	dev->needs_free_netdev = true;
1179 
1180 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1181 
1182 	dev->features    |= NETIF_F_LLTX;
1183 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1184 	dev->features    |= NETIF_F_RXCSUM;
1185 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1186 
1187 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1188 	dev->hw_features |= NETIF_F_RXCSUM;
1189 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1190 
1191 	/* MTU range: 68 - (something less than 65535) */
1192 	dev->min_mtu = ETH_MIN_MTU;
1193 	/* The max_mtu calculation does not take account of GENEVE
1194 	 * options, to avoid excluding potentially valid
1195 	 * configurations. This will be further reduced by IPvX hdr size.
1196 	 */
1197 	dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1198 
1199 	netif_keep_dst(dev);
1200 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1201 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1202 	eth_hw_addr_random(dev);
1203 }
1204 
1205 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1206 	[IFLA_GENEVE_UNSPEC]		= { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1207 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1208 	[IFLA_GENEVE_REMOTE]		= { .len = sizeof_field(struct iphdr, daddr) },
1209 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1210 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1211 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1212 	[IFLA_GENEVE_LABEL]		= { .type = NLA_U32 },
1213 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1214 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1215 	[IFLA_GENEVE_UDP_CSUM]		= { .type = NLA_U8 },
1216 	[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]	= { .type = NLA_U8 },
1217 	[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]	= { .type = NLA_U8 },
1218 	[IFLA_GENEVE_TTL_INHERIT]	= { .type = NLA_U8 },
1219 	[IFLA_GENEVE_DF]		= { .type = NLA_U8 },
1220 	[IFLA_GENEVE_INNER_PROTO_INHERIT]	= { .type = NLA_FLAG },
1221 };
1222 
1223 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1224 			   struct netlink_ext_ack *extack)
1225 {
1226 	if (tb[IFLA_ADDRESS]) {
1227 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1228 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1229 					    "Provided link layer address is not Ethernet");
1230 			return -EINVAL;
1231 		}
1232 
1233 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1234 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1235 					    "Provided Ethernet address is not unicast");
1236 			return -EADDRNOTAVAIL;
1237 		}
1238 	}
1239 
1240 	if (!data) {
1241 		NL_SET_ERR_MSG(extack,
1242 			       "Not enough attributes provided to perform the operation");
1243 		return -EINVAL;
1244 	}
1245 
1246 	if (data[IFLA_GENEVE_ID]) {
1247 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1248 
1249 		if (vni >= GENEVE_N_VID) {
1250 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1251 					    "Geneve ID must be lower than 16777216");
1252 			return -ERANGE;
1253 		}
1254 	}
1255 
1256 	if (data[IFLA_GENEVE_DF]) {
1257 		enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1258 
1259 		if (df < 0 || df > GENEVE_DF_MAX) {
1260 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1261 					    "Invalid DF attribute");
1262 			return -EINVAL;
1263 		}
1264 	}
1265 
1266 	return 0;
1267 }
1268 
1269 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1270 					  const struct ip_tunnel_info *info,
1271 					  bool *tun_on_same_port,
1272 					  bool *tun_collect_md)
1273 {
1274 	struct geneve_dev *geneve, *t = NULL;
1275 
1276 	*tun_on_same_port = false;
1277 	*tun_collect_md = false;
1278 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1279 		if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1280 			*tun_collect_md = geneve->cfg.collect_md;
1281 			*tun_on_same_port = true;
1282 		}
1283 		if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1284 		    info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1285 		    !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1286 			t = geneve;
1287 	}
1288 	return t;
1289 }
1290 
1291 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1292 {
1293 	return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1294 		 info->key.ttl || info->key.label || info->key.tp_src ||
1295 		 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1296 }
1297 
1298 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1299 				  struct ip_tunnel_info *b)
1300 {
1301 	if (ip_tunnel_info_af(a) == AF_INET)
1302 		return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1303 	else
1304 		return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1305 }
1306 
1307 static int geneve_configure(struct net *net, struct net_device *dev,
1308 			    struct netlink_ext_ack *extack,
1309 			    const struct geneve_config *cfg)
1310 {
1311 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1312 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1313 	const struct ip_tunnel_info *info = &cfg->info;
1314 	bool tun_collect_md, tun_on_same_port;
1315 	int err, encap_len;
1316 
1317 	if (cfg->collect_md && !is_tnl_info_zero(info)) {
1318 		NL_SET_ERR_MSG(extack,
1319 			       "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1320 		return -EINVAL;
1321 	}
1322 
1323 	geneve->net = net;
1324 	geneve->dev = dev;
1325 
1326 	t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1327 	if (t)
1328 		return -EBUSY;
1329 
1330 	/* make enough headroom for basic scenario */
1331 	encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1332 	if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1333 		encap_len += sizeof(struct iphdr);
1334 		dev->max_mtu -= sizeof(struct iphdr);
1335 	} else {
1336 		encap_len += sizeof(struct ipv6hdr);
1337 		dev->max_mtu -= sizeof(struct ipv6hdr);
1338 	}
1339 	dev->needed_headroom = encap_len + ETH_HLEN;
1340 
1341 	if (cfg->collect_md) {
1342 		if (tun_on_same_port) {
1343 			NL_SET_ERR_MSG(extack,
1344 				       "There can be only one externally controlled device on a destination port");
1345 			return -EPERM;
1346 		}
1347 	} else {
1348 		if (tun_collect_md) {
1349 			NL_SET_ERR_MSG(extack,
1350 				       "There already exists an externally controlled device on this destination port");
1351 			return -EPERM;
1352 		}
1353 	}
1354 
1355 	dst_cache_reset(&geneve->cfg.info.dst_cache);
1356 	memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1357 
1358 	if (geneve->cfg.inner_proto_inherit) {
1359 		dev->header_ops = NULL;
1360 		dev->type = ARPHRD_NONE;
1361 		dev->hard_header_len = 0;
1362 		dev->addr_len = 0;
1363 		dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1364 	}
1365 
1366 	err = register_netdevice(dev);
1367 	if (err)
1368 		return err;
1369 
1370 	list_add(&geneve->next, &gn->geneve_list);
1371 	return 0;
1372 }
1373 
1374 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1375 {
1376 	memset(info, 0, sizeof(*info));
1377 	info->key.tp_dst = htons(dst_port);
1378 }
1379 
1380 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1381 			  struct netlink_ext_ack *extack,
1382 			  struct geneve_config *cfg, bool changelink)
1383 {
1384 	struct ip_tunnel_info *info = &cfg->info;
1385 	int attrtype;
1386 
1387 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1388 		NL_SET_ERR_MSG(extack,
1389 			       "Cannot specify both IPv4 and IPv6 Remote addresses");
1390 		return -EINVAL;
1391 	}
1392 
1393 	if (data[IFLA_GENEVE_REMOTE]) {
1394 		if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1395 			attrtype = IFLA_GENEVE_REMOTE;
1396 			goto change_notsup;
1397 		}
1398 
1399 		info->key.u.ipv4.dst =
1400 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1401 
1402 		if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1403 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1404 					    "Remote IPv4 address cannot be Multicast");
1405 			return -EINVAL;
1406 		}
1407 	}
1408 
1409 	if (data[IFLA_GENEVE_REMOTE6]) {
1410 #if IS_ENABLED(CONFIG_IPV6)
1411 		if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1412 			attrtype = IFLA_GENEVE_REMOTE6;
1413 			goto change_notsup;
1414 		}
1415 
1416 		info->mode = IP_TUNNEL_INFO_IPV6;
1417 		info->key.u.ipv6.dst =
1418 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1419 
1420 		if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1421 		    IPV6_ADDR_LINKLOCAL) {
1422 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1423 					    "Remote IPv6 address cannot be link-local");
1424 			return -EINVAL;
1425 		}
1426 		if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1427 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1428 					    "Remote IPv6 address cannot be Multicast");
1429 			return -EINVAL;
1430 		}
1431 		info->key.tun_flags |= TUNNEL_CSUM;
1432 		cfg->use_udp6_rx_checksums = true;
1433 #else
1434 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1435 				    "IPv6 support not enabled in the kernel");
1436 		return -EPFNOSUPPORT;
1437 #endif
1438 	}
1439 
1440 	if (data[IFLA_GENEVE_ID]) {
1441 		__u32 vni;
1442 		__u8 tvni[3];
1443 		__be64 tunid;
1444 
1445 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1446 		tvni[0] = (vni & 0x00ff0000) >> 16;
1447 		tvni[1] = (vni & 0x0000ff00) >> 8;
1448 		tvni[2] =  vni & 0x000000ff;
1449 
1450 		tunid = vni_to_tunnel_id(tvni);
1451 		if (changelink && (tunid != info->key.tun_id)) {
1452 			attrtype = IFLA_GENEVE_ID;
1453 			goto change_notsup;
1454 		}
1455 		info->key.tun_id = tunid;
1456 	}
1457 
1458 	if (data[IFLA_GENEVE_TTL_INHERIT]) {
1459 		if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1460 			cfg->ttl_inherit = true;
1461 		else
1462 			cfg->ttl_inherit = false;
1463 	} else if (data[IFLA_GENEVE_TTL]) {
1464 		info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1465 		cfg->ttl_inherit = false;
1466 	}
1467 
1468 	if (data[IFLA_GENEVE_TOS])
1469 		info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1470 
1471 	if (data[IFLA_GENEVE_DF])
1472 		cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1473 
1474 	if (data[IFLA_GENEVE_LABEL]) {
1475 		info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1476 				  IPV6_FLOWLABEL_MASK;
1477 		if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1478 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1479 					    "Label attribute only applies for IPv6 Geneve devices");
1480 			return -EINVAL;
1481 		}
1482 	}
1483 
1484 	if (data[IFLA_GENEVE_PORT]) {
1485 		if (changelink) {
1486 			attrtype = IFLA_GENEVE_PORT;
1487 			goto change_notsup;
1488 		}
1489 		info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1490 	}
1491 
1492 	if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1493 		if (changelink) {
1494 			attrtype = IFLA_GENEVE_COLLECT_METADATA;
1495 			goto change_notsup;
1496 		}
1497 		cfg->collect_md = true;
1498 	}
1499 
1500 	if (data[IFLA_GENEVE_UDP_CSUM]) {
1501 		if (changelink) {
1502 			attrtype = IFLA_GENEVE_UDP_CSUM;
1503 			goto change_notsup;
1504 		}
1505 		if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1506 			info->key.tun_flags |= TUNNEL_CSUM;
1507 	}
1508 
1509 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1510 #if IS_ENABLED(CONFIG_IPV6)
1511 		if (changelink) {
1512 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1513 			goto change_notsup;
1514 		}
1515 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1516 			info->key.tun_flags &= ~TUNNEL_CSUM;
1517 #else
1518 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1519 				    "IPv6 support not enabled in the kernel");
1520 		return -EPFNOSUPPORT;
1521 #endif
1522 	}
1523 
1524 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1525 #if IS_ENABLED(CONFIG_IPV6)
1526 		if (changelink) {
1527 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1528 			goto change_notsup;
1529 		}
1530 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1531 			cfg->use_udp6_rx_checksums = false;
1532 #else
1533 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1534 				    "IPv6 support not enabled in the kernel");
1535 		return -EPFNOSUPPORT;
1536 #endif
1537 	}
1538 
1539 	if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1540 		if (changelink) {
1541 			attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1542 			goto change_notsup;
1543 		}
1544 		cfg->inner_proto_inherit = true;
1545 	}
1546 
1547 	return 0;
1548 change_notsup:
1549 	NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1550 			    "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1551 	return -EOPNOTSUPP;
1552 }
1553 
1554 static void geneve_link_config(struct net_device *dev,
1555 			       struct ip_tunnel_info *info, struct nlattr *tb[])
1556 {
1557 	struct geneve_dev *geneve = netdev_priv(dev);
1558 	int ldev_mtu = 0;
1559 
1560 	if (tb[IFLA_MTU]) {
1561 		geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1562 		return;
1563 	}
1564 
1565 	switch (ip_tunnel_info_af(info)) {
1566 	case AF_INET: {
1567 		struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1568 		struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1569 
1570 		if (!IS_ERR(rt) && rt->dst.dev) {
1571 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1572 			ip_rt_put(rt);
1573 		}
1574 		break;
1575 	}
1576 #if IS_ENABLED(CONFIG_IPV6)
1577 	case AF_INET6: {
1578 		struct rt6_info *rt;
1579 
1580 		if (!__in6_dev_get(dev))
1581 			break;
1582 
1583 		rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1584 				NULL, 0);
1585 
1586 		if (rt && rt->dst.dev)
1587 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1588 		ip6_rt_put(rt);
1589 		break;
1590 	}
1591 #endif
1592 	}
1593 
1594 	if (ldev_mtu <= 0)
1595 		return;
1596 
1597 	geneve_change_mtu(dev, ldev_mtu - info->options_len);
1598 }
1599 
1600 static int geneve_newlink(struct net *net, struct net_device *dev,
1601 			  struct nlattr *tb[], struct nlattr *data[],
1602 			  struct netlink_ext_ack *extack)
1603 {
1604 	struct geneve_config cfg = {
1605 		.df = GENEVE_DF_UNSET,
1606 		.use_udp6_rx_checksums = false,
1607 		.ttl_inherit = false,
1608 		.collect_md = false,
1609 	};
1610 	int err;
1611 
1612 	init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1613 	err = geneve_nl2info(tb, data, extack, &cfg, false);
1614 	if (err)
1615 		return err;
1616 
1617 	err = geneve_configure(net, dev, extack, &cfg);
1618 	if (err)
1619 		return err;
1620 
1621 	geneve_link_config(dev, &cfg.info, tb);
1622 
1623 	return 0;
1624 }
1625 
1626 /* Quiesces the geneve device data path for both TX and RX.
1627  *
1628  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1629  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1630  * to complete for the existing set of in-flight packets to be transmitted,
1631  * then we would have quiesced the transmit data path. All the future packets
1632  * will get dropped until we unquiesce the data path.
1633  *
1634  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1635  * if we set that to NULL under RCU and wait for synchronize_net() to
1636  * complete, then we would have quiesced the receive data path.
1637  */
1638 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1639 			   struct geneve_sock **gs6)
1640 {
1641 	*gs4 = rtnl_dereference(geneve->sock4);
1642 	rcu_assign_pointer(geneve->sock4, NULL);
1643 	if (*gs4)
1644 		rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1645 #if IS_ENABLED(CONFIG_IPV6)
1646 	*gs6 = rtnl_dereference(geneve->sock6);
1647 	rcu_assign_pointer(geneve->sock6, NULL);
1648 	if (*gs6)
1649 		rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1650 #else
1651 	*gs6 = NULL;
1652 #endif
1653 	synchronize_net();
1654 }
1655 
1656 /* Resumes the geneve device data path for both TX and RX. */
1657 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1658 			     struct geneve_sock __maybe_unused *gs6)
1659 {
1660 	rcu_assign_pointer(geneve->sock4, gs4);
1661 	if (gs4)
1662 		rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1663 #if IS_ENABLED(CONFIG_IPV6)
1664 	rcu_assign_pointer(geneve->sock6, gs6);
1665 	if (gs6)
1666 		rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1667 #endif
1668 	synchronize_net();
1669 }
1670 
1671 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1672 			     struct nlattr *data[],
1673 			     struct netlink_ext_ack *extack)
1674 {
1675 	struct geneve_dev *geneve = netdev_priv(dev);
1676 	struct geneve_sock *gs4, *gs6;
1677 	struct geneve_config cfg;
1678 	int err;
1679 
1680 	/* If the geneve device is configured for metadata (or externally
1681 	 * controlled, for example, OVS), then nothing can be changed.
1682 	 */
1683 	if (geneve->cfg.collect_md)
1684 		return -EOPNOTSUPP;
1685 
1686 	/* Start with the existing info. */
1687 	memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1688 	err = geneve_nl2info(tb, data, extack, &cfg, true);
1689 	if (err)
1690 		return err;
1691 
1692 	if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1693 		dst_cache_reset(&cfg.info.dst_cache);
1694 		geneve_link_config(dev, &cfg.info, tb);
1695 	}
1696 
1697 	geneve_quiesce(geneve, &gs4, &gs6);
1698 	memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1699 	geneve_unquiesce(geneve, gs4, gs6);
1700 
1701 	return 0;
1702 }
1703 
1704 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1705 {
1706 	struct geneve_dev *geneve = netdev_priv(dev);
1707 
1708 	list_del(&geneve->next);
1709 	unregister_netdevice_queue(dev, head);
1710 }
1711 
1712 static size_t geneve_get_size(const struct net_device *dev)
1713 {
1714 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1715 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1716 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1717 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1718 		nla_total_size(sizeof(__u8)) +	/* IFLA_GENEVE_DF */
1719 		nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1720 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1721 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1722 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1723 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1724 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1725 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1726 		nla_total_size(0) +	 /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1727 		0;
1728 }
1729 
1730 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1731 {
1732 	struct geneve_dev *geneve = netdev_priv(dev);
1733 	struct ip_tunnel_info *info = &geneve->cfg.info;
1734 	bool ttl_inherit = geneve->cfg.ttl_inherit;
1735 	bool metadata = geneve->cfg.collect_md;
1736 	__u8 tmp_vni[3];
1737 	__u32 vni;
1738 
1739 	tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1740 	vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1741 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1742 		goto nla_put_failure;
1743 
1744 	if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1745 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1746 				    info->key.u.ipv4.dst))
1747 			goto nla_put_failure;
1748 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1749 			       !!(info->key.tun_flags & TUNNEL_CSUM)))
1750 			goto nla_put_failure;
1751 
1752 #if IS_ENABLED(CONFIG_IPV6)
1753 	} else if (!metadata) {
1754 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1755 				     &info->key.u.ipv6.dst))
1756 			goto nla_put_failure;
1757 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1758 			       !(info->key.tun_flags & TUNNEL_CSUM)))
1759 			goto nla_put_failure;
1760 #endif
1761 	}
1762 
1763 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1764 	    nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1765 	    nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1766 		goto nla_put_failure;
1767 
1768 	if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1769 		goto nla_put_failure;
1770 
1771 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1772 		goto nla_put_failure;
1773 
1774 	if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1775 		goto nla_put_failure;
1776 
1777 #if IS_ENABLED(CONFIG_IPV6)
1778 	if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1779 		       !geneve->cfg.use_udp6_rx_checksums))
1780 		goto nla_put_failure;
1781 #endif
1782 
1783 	if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1784 		goto nla_put_failure;
1785 
1786 	if (geneve->cfg.inner_proto_inherit &&
1787 	    nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1788 		goto nla_put_failure;
1789 
1790 	return 0;
1791 
1792 nla_put_failure:
1793 	return -EMSGSIZE;
1794 }
1795 
1796 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1797 	.kind		= "geneve",
1798 	.maxtype	= IFLA_GENEVE_MAX,
1799 	.policy		= geneve_policy,
1800 	.priv_size	= sizeof(struct geneve_dev),
1801 	.setup		= geneve_setup,
1802 	.validate	= geneve_validate,
1803 	.newlink	= geneve_newlink,
1804 	.changelink	= geneve_changelink,
1805 	.dellink	= geneve_dellink,
1806 	.get_size	= geneve_get_size,
1807 	.fill_info	= geneve_fill_info,
1808 };
1809 
1810 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1811 					u8 name_assign_type, u16 dst_port)
1812 {
1813 	struct nlattr *tb[IFLA_MAX + 1];
1814 	struct net_device *dev;
1815 	LIST_HEAD(list_kill);
1816 	int err;
1817 	struct geneve_config cfg = {
1818 		.df = GENEVE_DF_UNSET,
1819 		.use_udp6_rx_checksums = true,
1820 		.ttl_inherit = false,
1821 		.collect_md = true,
1822 	};
1823 
1824 	memset(tb, 0, sizeof(tb));
1825 	dev = rtnl_create_link(net, name, name_assign_type,
1826 			       &geneve_link_ops, tb, NULL);
1827 	if (IS_ERR(dev))
1828 		return dev;
1829 
1830 	init_tnl_info(&cfg.info, dst_port);
1831 	err = geneve_configure(net, dev, NULL, &cfg);
1832 	if (err) {
1833 		free_netdev(dev);
1834 		return ERR_PTR(err);
1835 	}
1836 
1837 	/* openvswitch users expect packet sizes to be unrestricted,
1838 	 * so set the largest MTU we can.
1839 	 */
1840 	err = geneve_change_mtu(dev, IP_MAX_MTU);
1841 	if (err)
1842 		goto err;
1843 
1844 	err = rtnl_configure_link(dev, NULL, 0, NULL);
1845 	if (err < 0)
1846 		goto err;
1847 
1848 	return dev;
1849 err:
1850 	geneve_dellink(dev, &list_kill);
1851 	unregister_netdevice_many(&list_kill);
1852 	return ERR_PTR(err);
1853 }
1854 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1855 
1856 static int geneve_netdevice_event(struct notifier_block *unused,
1857 				  unsigned long event, void *ptr)
1858 {
1859 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1860 
1861 	if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1862 		geneve_offload_rx_ports(dev, true);
1863 	else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1864 		geneve_offload_rx_ports(dev, false);
1865 
1866 	return NOTIFY_DONE;
1867 }
1868 
1869 static struct notifier_block geneve_notifier_block __read_mostly = {
1870 	.notifier_call = geneve_netdevice_event,
1871 };
1872 
1873 static __net_init int geneve_init_net(struct net *net)
1874 {
1875 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1876 
1877 	INIT_LIST_HEAD(&gn->geneve_list);
1878 	INIT_LIST_HEAD(&gn->sock_list);
1879 	return 0;
1880 }
1881 
1882 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1883 {
1884 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1885 	struct geneve_dev *geneve, *next;
1886 	struct net_device *dev, *aux;
1887 
1888 	/* gather any geneve devices that were moved into this ns */
1889 	for_each_netdev_safe(net, dev, aux)
1890 		if (dev->rtnl_link_ops == &geneve_link_ops)
1891 			unregister_netdevice_queue(dev, head);
1892 
1893 	/* now gather any other geneve devices that were created in this ns */
1894 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1895 		/* If geneve->dev is in the same netns, it was already added
1896 		 * to the list by the previous loop.
1897 		 */
1898 		if (!net_eq(dev_net(geneve->dev), net))
1899 			unregister_netdevice_queue(geneve->dev, head);
1900 	}
1901 }
1902 
1903 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1904 {
1905 	struct net *net;
1906 	LIST_HEAD(list);
1907 
1908 	rtnl_lock();
1909 	list_for_each_entry(net, net_list, exit_list)
1910 		geneve_destroy_tunnels(net, &list);
1911 
1912 	/* unregister the devices gathered above */
1913 	unregister_netdevice_many(&list);
1914 	rtnl_unlock();
1915 
1916 	list_for_each_entry(net, net_list, exit_list) {
1917 		const struct geneve_net *gn = net_generic(net, geneve_net_id);
1918 
1919 		WARN_ON_ONCE(!list_empty(&gn->sock_list));
1920 	}
1921 }
1922 
1923 static struct pernet_operations geneve_net_ops = {
1924 	.init = geneve_init_net,
1925 	.exit_batch = geneve_exit_batch_net,
1926 	.id   = &geneve_net_id,
1927 	.size = sizeof(struct geneve_net),
1928 };
1929 
1930 static int __init geneve_init_module(void)
1931 {
1932 	int rc;
1933 
1934 	rc = register_pernet_subsys(&geneve_net_ops);
1935 	if (rc)
1936 		goto out1;
1937 
1938 	rc = register_netdevice_notifier(&geneve_notifier_block);
1939 	if (rc)
1940 		goto out2;
1941 
1942 	rc = rtnl_link_register(&geneve_link_ops);
1943 	if (rc)
1944 		goto out3;
1945 
1946 	return 0;
1947 out3:
1948 	unregister_netdevice_notifier(&geneve_notifier_block);
1949 out2:
1950 	unregister_pernet_subsys(&geneve_net_ops);
1951 out1:
1952 	return rc;
1953 }
1954 late_initcall(geneve_init_module);
1955 
1956 static void __exit geneve_cleanup_module(void)
1957 {
1958 	rtnl_link_unregister(&geneve_link_ops);
1959 	unregister_netdevice_notifier(&geneve_notifier_block);
1960 	unregister_pernet_subsys(&geneve_net_ops);
1961 }
1962 module_exit(geneve_cleanup_module);
1963 
1964 MODULE_LICENSE("GPL");
1965 MODULE_VERSION(GENEVE_NETDEV_VER);
1966 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1967 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1968 MODULE_ALIAS_RTNL_LINK("geneve");
1969