xref: /openbmc/linux/drivers/net/geneve.c (revision fbb6b31a)
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 			geneve->dev->stats.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 			geneve->dev->stats.rx_frame_errors++;
250 			geneve->dev->stats.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 			geneve->dev->stats.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 			++geneve->dev->stats.rx_frame_errors;
300 			++geneve->dev->stats.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 	inner_proto = geneveh->proto_type;
369 
370 	if (unlikely((inner_proto != htons(ETH_P_TEB) &&
371 		      inner_proto != htons(ETH_P_IP) &&
372 		      inner_proto != htons(ETH_P_IPV6))))
373 		goto drop;
374 
375 	gs = rcu_dereference_sk_user_data(sk);
376 	if (!gs)
377 		goto drop;
378 
379 	geneve = geneve_lookup_skb(gs, skb);
380 	if (!geneve)
381 		goto drop;
382 
383 	if (unlikely((!geneve->cfg.inner_proto_inherit &&
384 		      inner_proto != htons(ETH_P_TEB)))) {
385 		geneve->dev->stats.rx_dropped++;
386 		goto drop;
387 	}
388 
389 	opts_len = geneveh->opt_len * 4;
390 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
391 				 !net_eq(geneve->net, dev_net(geneve->dev)))) {
392 		geneve->dev->stats.rx_dropped++;
393 		goto drop;
394 	}
395 
396 	geneve_rx(geneve, gs, skb);
397 	return 0;
398 
399 drop:
400 	/* Consume bad packet */
401 	kfree_skb(skb);
402 	return 0;
403 }
404 
405 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
406 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
407 {
408 	struct genevehdr *geneveh;
409 	struct geneve_sock *gs;
410 	u8 zero_vni[3] = { 0 };
411 	u8 *vni = zero_vni;
412 
413 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
414 		return -EINVAL;
415 
416 	geneveh = geneve_hdr(skb);
417 	if (geneveh->ver != GENEVE_VER)
418 		return -EINVAL;
419 
420 	if (geneveh->proto_type != htons(ETH_P_TEB))
421 		return -EINVAL;
422 
423 	gs = rcu_dereference_sk_user_data(sk);
424 	if (!gs)
425 		return -ENOENT;
426 
427 	if (geneve_get_sk_family(gs) == AF_INET) {
428 		struct iphdr *iph = ip_hdr(skb);
429 		__be32 addr4 = 0;
430 
431 		if (!gs->collect_md) {
432 			vni = geneve_hdr(skb)->vni;
433 			addr4 = iph->daddr;
434 		}
435 
436 		return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
437 	}
438 
439 #if IS_ENABLED(CONFIG_IPV6)
440 	if (geneve_get_sk_family(gs) == AF_INET6) {
441 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
442 		struct in6_addr addr6;
443 
444 		memset(&addr6, 0, sizeof(struct in6_addr));
445 
446 		if (!gs->collect_md) {
447 			vni = geneve_hdr(skb)->vni;
448 			addr6 = ip6h->daddr;
449 		}
450 
451 		return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
452 	}
453 #endif
454 
455 	return -EPFNOSUPPORT;
456 }
457 
458 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
459 					 __be16 port, bool ipv6_rx_csum)
460 {
461 	struct socket *sock;
462 	struct udp_port_cfg udp_conf;
463 	int err;
464 
465 	memset(&udp_conf, 0, sizeof(udp_conf));
466 
467 	if (ipv6) {
468 		udp_conf.family = AF_INET6;
469 		udp_conf.ipv6_v6only = 1;
470 		udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
471 	} else {
472 		udp_conf.family = AF_INET;
473 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
474 	}
475 
476 	udp_conf.local_udp_port = port;
477 
478 	/* Open UDP socket */
479 	err = udp_sock_create(net, &udp_conf, &sock);
480 	if (err < 0)
481 		return ERR_PTR(err);
482 
483 	udp_allow_gso(sock->sk);
484 	return sock;
485 }
486 
487 static int geneve_hlen(struct genevehdr *gh)
488 {
489 	return sizeof(*gh) + gh->opt_len * 4;
490 }
491 
492 static struct sk_buff *geneve_gro_receive(struct sock *sk,
493 					  struct list_head *head,
494 					  struct sk_buff *skb)
495 {
496 	struct sk_buff *pp = NULL;
497 	struct sk_buff *p;
498 	struct genevehdr *gh, *gh2;
499 	unsigned int hlen, gh_len, off_gnv;
500 	const struct packet_offload *ptype;
501 	__be16 type;
502 	int flush = 1;
503 
504 	off_gnv = skb_gro_offset(skb);
505 	hlen = off_gnv + sizeof(*gh);
506 	gh = skb_gro_header_fast(skb, off_gnv);
507 	if (skb_gro_header_hard(skb, hlen)) {
508 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
509 		if (unlikely(!gh))
510 			goto out;
511 	}
512 
513 	if (gh->ver != GENEVE_VER || gh->oam)
514 		goto out;
515 	gh_len = geneve_hlen(gh);
516 
517 	hlen = off_gnv + gh_len;
518 	if (skb_gro_header_hard(skb, hlen)) {
519 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
520 		if (unlikely(!gh))
521 			goto out;
522 	}
523 
524 	list_for_each_entry(p, head, list) {
525 		if (!NAPI_GRO_CB(p)->same_flow)
526 			continue;
527 
528 		gh2 = (struct genevehdr *)(p->data + off_gnv);
529 		if (gh->opt_len != gh2->opt_len ||
530 		    memcmp(gh, gh2, gh_len)) {
531 			NAPI_GRO_CB(p)->same_flow = 0;
532 			continue;
533 		}
534 	}
535 
536 	type = gh->proto_type;
537 
538 	ptype = gro_find_receive_by_type(type);
539 	if (!ptype)
540 		goto out;
541 
542 	skb_gro_pull(skb, gh_len);
543 	skb_gro_postpull_rcsum(skb, gh, gh_len);
544 	pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
545 	flush = 0;
546 
547 out:
548 	skb_gro_flush_final(skb, pp, flush);
549 
550 	return pp;
551 }
552 
553 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
554 			       int nhoff)
555 {
556 	struct genevehdr *gh;
557 	struct packet_offload *ptype;
558 	__be16 type;
559 	int gh_len;
560 	int err = -ENOSYS;
561 
562 	gh = (struct genevehdr *)(skb->data + nhoff);
563 	gh_len = geneve_hlen(gh);
564 	type = gh->proto_type;
565 
566 	ptype = gro_find_complete_by_type(type);
567 	if (ptype)
568 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
569 
570 	skb_set_inner_mac_header(skb, nhoff + gh_len);
571 
572 	return err;
573 }
574 
575 /* Create new listen socket if needed */
576 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
577 						bool ipv6, bool ipv6_rx_csum)
578 {
579 	struct geneve_net *gn = net_generic(net, geneve_net_id);
580 	struct geneve_sock *gs;
581 	struct socket *sock;
582 	struct udp_tunnel_sock_cfg tunnel_cfg;
583 	int h;
584 
585 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
586 	if (!gs)
587 		return ERR_PTR(-ENOMEM);
588 
589 	sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
590 	if (IS_ERR(sock)) {
591 		kfree(gs);
592 		return ERR_CAST(sock);
593 	}
594 
595 	gs->sock = sock;
596 	gs->refcnt = 1;
597 	for (h = 0; h < VNI_HASH_SIZE; ++h)
598 		INIT_HLIST_HEAD(&gs->vni_list[h]);
599 
600 	/* Initialize the geneve udp offloads structure */
601 	udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
602 
603 	/* Mark socket as an encapsulation socket */
604 	memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
605 	tunnel_cfg.sk_user_data = gs;
606 	tunnel_cfg.encap_type = 1;
607 	tunnel_cfg.gro_receive = geneve_gro_receive;
608 	tunnel_cfg.gro_complete = geneve_gro_complete;
609 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
610 	tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
611 	tunnel_cfg.encap_destroy = NULL;
612 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
613 	list_add(&gs->list, &gn->sock_list);
614 	return gs;
615 }
616 
617 static void __geneve_sock_release(struct geneve_sock *gs)
618 {
619 	if (!gs || --gs->refcnt)
620 		return;
621 
622 	list_del(&gs->list);
623 	udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
624 	udp_tunnel_sock_release(gs->sock);
625 	kfree_rcu(gs, rcu);
626 }
627 
628 static void geneve_sock_release(struct geneve_dev *geneve)
629 {
630 	struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
631 #if IS_ENABLED(CONFIG_IPV6)
632 	struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
633 
634 	rcu_assign_pointer(geneve->sock6, NULL);
635 #endif
636 
637 	rcu_assign_pointer(geneve->sock4, NULL);
638 	synchronize_net();
639 
640 	__geneve_sock_release(gs4);
641 #if IS_ENABLED(CONFIG_IPV6)
642 	__geneve_sock_release(gs6);
643 #endif
644 }
645 
646 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
647 					    sa_family_t family,
648 					    __be16 dst_port)
649 {
650 	struct geneve_sock *gs;
651 
652 	list_for_each_entry(gs, &gn->sock_list, list) {
653 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
654 		    geneve_get_sk_family(gs) == family) {
655 			return gs;
656 		}
657 	}
658 	return NULL;
659 }
660 
661 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
662 {
663 	struct net *net = geneve->net;
664 	struct geneve_net *gn = net_generic(net, geneve_net_id);
665 	struct geneve_dev_node *node;
666 	struct geneve_sock *gs;
667 	__u8 vni[3];
668 	__u32 hash;
669 
670 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
671 	if (gs) {
672 		gs->refcnt++;
673 		goto out;
674 	}
675 
676 	gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
677 				  geneve->cfg.use_udp6_rx_checksums);
678 	if (IS_ERR(gs))
679 		return PTR_ERR(gs);
680 
681 out:
682 	gs->collect_md = geneve->cfg.collect_md;
683 #if IS_ENABLED(CONFIG_IPV6)
684 	if (ipv6) {
685 		rcu_assign_pointer(geneve->sock6, gs);
686 		node = &geneve->hlist6;
687 	} else
688 #endif
689 	{
690 		rcu_assign_pointer(geneve->sock4, gs);
691 		node = &geneve->hlist4;
692 	}
693 	node->geneve = geneve;
694 
695 	tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
696 	hash = geneve_net_vni_hash(vni);
697 	hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
698 	return 0;
699 }
700 
701 static int geneve_open(struct net_device *dev)
702 {
703 	struct geneve_dev *geneve = netdev_priv(dev);
704 	bool metadata = geneve->cfg.collect_md;
705 	bool ipv4, ipv6;
706 	int ret = 0;
707 
708 	ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
709 	ipv4 = !ipv6 || metadata;
710 #if IS_ENABLED(CONFIG_IPV6)
711 	if (ipv6) {
712 		ret = geneve_sock_add(geneve, true);
713 		if (ret < 0 && ret != -EAFNOSUPPORT)
714 			ipv4 = false;
715 	}
716 #endif
717 	if (ipv4)
718 		ret = geneve_sock_add(geneve, false);
719 	if (ret < 0)
720 		geneve_sock_release(geneve);
721 
722 	return ret;
723 }
724 
725 static int geneve_stop(struct net_device *dev)
726 {
727 	struct geneve_dev *geneve = netdev_priv(dev);
728 
729 	hlist_del_init_rcu(&geneve->hlist4.hlist);
730 #if IS_ENABLED(CONFIG_IPV6)
731 	hlist_del_init_rcu(&geneve->hlist6.hlist);
732 #endif
733 	geneve_sock_release(geneve);
734 	return 0;
735 }
736 
737 static void geneve_build_header(struct genevehdr *geneveh,
738 				const struct ip_tunnel_info *info,
739 				__be16 inner_proto)
740 {
741 	geneveh->ver = GENEVE_VER;
742 	geneveh->opt_len = info->options_len / 4;
743 	geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
744 	geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
745 	geneveh->rsvd1 = 0;
746 	tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
747 	geneveh->proto_type = inner_proto;
748 	geneveh->rsvd2 = 0;
749 
750 	if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
751 		ip_tunnel_info_opts_get(geneveh->options, info);
752 }
753 
754 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
755 			    const struct ip_tunnel_info *info,
756 			    bool xnet, int ip_hdr_len,
757 			    bool inner_proto_inherit)
758 {
759 	bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
760 	struct genevehdr *gnvh;
761 	__be16 inner_proto;
762 	int min_headroom;
763 	int err;
764 
765 	skb_reset_mac_header(skb);
766 	skb_scrub_packet(skb, xnet);
767 
768 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
769 		       GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
770 	err = skb_cow_head(skb, min_headroom);
771 	if (unlikely(err))
772 		goto free_dst;
773 
774 	err = udp_tunnel_handle_offloads(skb, udp_sum);
775 	if (err)
776 		goto free_dst;
777 
778 	gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
779 	inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
780 	geneve_build_header(gnvh, info, inner_proto);
781 	skb_set_inner_protocol(skb, inner_proto);
782 	return 0;
783 
784 free_dst:
785 	dst_release(dst);
786 	return err;
787 }
788 
789 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
790 				       struct net_device *dev,
791 				       struct geneve_sock *gs4,
792 				       struct flowi4 *fl4,
793 				       const struct ip_tunnel_info *info,
794 				       __be16 dport, __be16 sport)
795 {
796 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
797 	struct geneve_dev *geneve = netdev_priv(dev);
798 	struct dst_cache *dst_cache;
799 	struct rtable *rt = NULL;
800 	__u8 tos;
801 
802 	if (!gs4)
803 		return ERR_PTR(-EIO);
804 
805 	memset(fl4, 0, sizeof(*fl4));
806 	fl4->flowi4_mark = skb->mark;
807 	fl4->flowi4_proto = IPPROTO_UDP;
808 	fl4->daddr = info->key.u.ipv4.dst;
809 	fl4->saddr = info->key.u.ipv4.src;
810 	fl4->fl4_dport = dport;
811 	fl4->fl4_sport = sport;
812 
813 	tos = info->key.tos;
814 	if ((tos == 1) && !geneve->cfg.collect_md) {
815 		tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
816 		use_cache = false;
817 	}
818 	fl4->flowi4_tos = RT_TOS(tos);
819 
820 	dst_cache = (struct dst_cache *)&info->dst_cache;
821 	if (use_cache) {
822 		rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
823 		if (rt)
824 			return rt;
825 	}
826 	rt = ip_route_output_key(geneve->net, fl4);
827 	if (IS_ERR(rt)) {
828 		netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
829 		return ERR_PTR(-ENETUNREACH);
830 	}
831 	if (rt->dst.dev == dev) { /* is this necessary? */
832 		netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
833 		ip_rt_put(rt);
834 		return ERR_PTR(-ELOOP);
835 	}
836 	if (use_cache)
837 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
838 	return rt;
839 }
840 
841 #if IS_ENABLED(CONFIG_IPV6)
842 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
843 					   struct net_device *dev,
844 					   struct geneve_sock *gs6,
845 					   struct flowi6 *fl6,
846 					   const struct ip_tunnel_info *info,
847 					   __be16 dport, __be16 sport)
848 {
849 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
850 	struct geneve_dev *geneve = netdev_priv(dev);
851 	struct dst_entry *dst = NULL;
852 	struct dst_cache *dst_cache;
853 	__u8 prio;
854 
855 	if (!gs6)
856 		return ERR_PTR(-EIO);
857 
858 	memset(fl6, 0, sizeof(*fl6));
859 	fl6->flowi6_mark = skb->mark;
860 	fl6->flowi6_proto = IPPROTO_UDP;
861 	fl6->daddr = info->key.u.ipv6.dst;
862 	fl6->saddr = info->key.u.ipv6.src;
863 	fl6->fl6_dport = dport;
864 	fl6->fl6_sport = sport;
865 
866 	prio = info->key.tos;
867 	if ((prio == 1) && !geneve->cfg.collect_md) {
868 		prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
869 		use_cache = false;
870 	}
871 
872 	fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
873 					   info->key.label);
874 	dst_cache = (struct dst_cache *)&info->dst_cache;
875 	if (use_cache) {
876 		dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
877 		if (dst)
878 			return dst;
879 	}
880 	dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
881 					      NULL);
882 	if (IS_ERR(dst)) {
883 		netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
884 		return ERR_PTR(-ENETUNREACH);
885 	}
886 	if (dst->dev == dev) { /* is this necessary? */
887 		netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
888 		dst_release(dst);
889 		return ERR_PTR(-ELOOP);
890 	}
891 
892 	if (use_cache)
893 		dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
894 	return dst;
895 }
896 #endif
897 
898 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
899 			   struct geneve_dev *geneve,
900 			   const struct ip_tunnel_info *info)
901 {
902 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
903 	struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
904 	const struct ip_tunnel_key *key = &info->key;
905 	struct rtable *rt;
906 	struct flowi4 fl4;
907 	__u8 tos, ttl;
908 	__be16 df = 0;
909 	__be16 sport;
910 	int err;
911 
912 	if (!pskb_inet_may_pull(skb))
913 		return -EINVAL;
914 
915 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
916 	rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
917 			      geneve->cfg.info.key.tp_dst, sport);
918 	if (IS_ERR(rt))
919 		return PTR_ERR(rt);
920 
921 	err = skb_tunnel_check_pmtu(skb, &rt->dst,
922 				    GENEVE_IPV4_HLEN + info->options_len,
923 				    netif_is_any_bridge_port(dev));
924 	if (err < 0) {
925 		dst_release(&rt->dst);
926 		return err;
927 	} else if (err) {
928 		struct ip_tunnel_info *info;
929 
930 		info = skb_tunnel_info(skb);
931 		if (info) {
932 			struct ip_tunnel_info *unclone;
933 
934 			unclone = skb_tunnel_info_unclone(skb);
935 			if (unlikely(!unclone)) {
936 				dst_release(&rt->dst);
937 				return -ENOMEM;
938 			}
939 
940 			unclone->key.u.ipv4.dst = fl4.saddr;
941 			unclone->key.u.ipv4.src = fl4.daddr;
942 		}
943 
944 		if (!pskb_may_pull(skb, ETH_HLEN)) {
945 			dst_release(&rt->dst);
946 			return -EINVAL;
947 		}
948 
949 		skb->protocol = eth_type_trans(skb, geneve->dev);
950 		__netif_rx(skb);
951 		dst_release(&rt->dst);
952 		return -EMSGSIZE;
953 	}
954 
955 	if (geneve->cfg.collect_md) {
956 		tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
957 		ttl = key->ttl;
958 
959 		df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
960 	} else {
961 		tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
962 		if (geneve->cfg.ttl_inherit)
963 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
964 		else
965 			ttl = key->ttl;
966 		ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
967 
968 		if (geneve->cfg.df == GENEVE_DF_SET) {
969 			df = htons(IP_DF);
970 		} else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
971 			struct ethhdr *eth = eth_hdr(skb);
972 
973 			if (ntohs(eth->h_proto) == ETH_P_IPV6) {
974 				df = htons(IP_DF);
975 			} else if (ntohs(eth->h_proto) == ETH_P_IP) {
976 				struct iphdr *iph = ip_hdr(skb);
977 
978 				if (iph->frag_off & htons(IP_DF))
979 					df = htons(IP_DF);
980 			}
981 		}
982 	}
983 
984 	err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
985 			       geneve->cfg.inner_proto_inherit);
986 	if (unlikely(err))
987 		return err;
988 
989 	udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
990 			    tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
991 			    !net_eq(geneve->net, dev_net(geneve->dev)),
992 			    !(info->key.tun_flags & TUNNEL_CSUM));
993 	return 0;
994 }
995 
996 #if IS_ENABLED(CONFIG_IPV6)
997 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
998 			    struct geneve_dev *geneve,
999 			    const struct ip_tunnel_info *info)
1000 {
1001 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1002 	struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1003 	const struct ip_tunnel_key *key = &info->key;
1004 	struct dst_entry *dst = NULL;
1005 	struct flowi6 fl6;
1006 	__u8 prio, ttl;
1007 	__be16 sport;
1008 	int err;
1009 
1010 	if (!pskb_inet_may_pull(skb))
1011 		return -EINVAL;
1012 
1013 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1014 	dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1015 				geneve->cfg.info.key.tp_dst, sport);
1016 	if (IS_ERR(dst))
1017 		return PTR_ERR(dst);
1018 
1019 	err = skb_tunnel_check_pmtu(skb, dst,
1020 				    GENEVE_IPV6_HLEN + info->options_len,
1021 				    netif_is_any_bridge_port(dev));
1022 	if (err < 0) {
1023 		dst_release(dst);
1024 		return err;
1025 	} else if (err) {
1026 		struct ip_tunnel_info *info = skb_tunnel_info(skb);
1027 
1028 		if (info) {
1029 			struct ip_tunnel_info *unclone;
1030 
1031 			unclone = skb_tunnel_info_unclone(skb);
1032 			if (unlikely(!unclone)) {
1033 				dst_release(dst);
1034 				return -ENOMEM;
1035 			}
1036 
1037 			unclone->key.u.ipv6.dst = fl6.saddr;
1038 			unclone->key.u.ipv6.src = fl6.daddr;
1039 		}
1040 
1041 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1042 			dst_release(dst);
1043 			return -EINVAL;
1044 		}
1045 
1046 		skb->protocol = eth_type_trans(skb, geneve->dev);
1047 		__netif_rx(skb);
1048 		dst_release(dst);
1049 		return -EMSGSIZE;
1050 	}
1051 
1052 	if (geneve->cfg.collect_md) {
1053 		prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1054 		ttl = key->ttl;
1055 	} else {
1056 		prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1057 					   ip_hdr(skb), skb);
1058 		if (geneve->cfg.ttl_inherit)
1059 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1060 		else
1061 			ttl = key->ttl;
1062 		ttl = ttl ? : ip6_dst_hoplimit(dst);
1063 	}
1064 	err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1065 			       geneve->cfg.inner_proto_inherit);
1066 	if (unlikely(err))
1067 		return err;
1068 
1069 	udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1070 			     &fl6.saddr, &fl6.daddr, prio, ttl,
1071 			     info->key.label, sport, geneve->cfg.info.key.tp_dst,
1072 			     !(info->key.tun_flags & TUNNEL_CSUM));
1073 	return 0;
1074 }
1075 #endif
1076 
1077 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1078 {
1079 	struct geneve_dev *geneve = netdev_priv(dev);
1080 	struct ip_tunnel_info *info = NULL;
1081 	int err;
1082 
1083 	if (geneve->cfg.collect_md) {
1084 		info = skb_tunnel_info(skb);
1085 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1086 			netdev_dbg(dev, "no tunnel metadata\n");
1087 			dev_kfree_skb(skb);
1088 			dev->stats.tx_dropped++;
1089 			return NETDEV_TX_OK;
1090 		}
1091 	} else {
1092 		info = &geneve->cfg.info;
1093 	}
1094 
1095 	rcu_read_lock();
1096 #if IS_ENABLED(CONFIG_IPV6)
1097 	if (info->mode & IP_TUNNEL_INFO_IPV6)
1098 		err = geneve6_xmit_skb(skb, dev, geneve, info);
1099 	else
1100 #endif
1101 		err = geneve_xmit_skb(skb, dev, geneve, info);
1102 	rcu_read_unlock();
1103 
1104 	if (likely(!err))
1105 		return NETDEV_TX_OK;
1106 
1107 	if (err != -EMSGSIZE)
1108 		dev_kfree_skb(skb);
1109 
1110 	if (err == -ELOOP)
1111 		dev->stats.collisions++;
1112 	else if (err == -ENETUNREACH)
1113 		dev->stats.tx_carrier_errors++;
1114 
1115 	dev->stats.tx_errors++;
1116 	return NETDEV_TX_OK;
1117 }
1118 
1119 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1120 {
1121 	if (new_mtu > dev->max_mtu)
1122 		new_mtu = dev->max_mtu;
1123 	else if (new_mtu < dev->min_mtu)
1124 		new_mtu = dev->min_mtu;
1125 
1126 	dev->mtu = new_mtu;
1127 	return 0;
1128 }
1129 
1130 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1131 {
1132 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
1133 	struct geneve_dev *geneve = netdev_priv(dev);
1134 	__be16 sport;
1135 
1136 	if (ip_tunnel_info_af(info) == AF_INET) {
1137 		struct rtable *rt;
1138 		struct flowi4 fl4;
1139 
1140 		struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1141 		sport = udp_flow_src_port(geneve->net, skb,
1142 					  1, USHRT_MAX, true);
1143 
1144 		rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1145 				      geneve->cfg.info.key.tp_dst, sport);
1146 		if (IS_ERR(rt))
1147 			return PTR_ERR(rt);
1148 
1149 		ip_rt_put(rt);
1150 		info->key.u.ipv4.src = fl4.saddr;
1151 #if IS_ENABLED(CONFIG_IPV6)
1152 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
1153 		struct dst_entry *dst;
1154 		struct flowi6 fl6;
1155 
1156 		struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1157 		sport = udp_flow_src_port(geneve->net, skb,
1158 					  1, USHRT_MAX, true);
1159 
1160 		dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1161 					geneve->cfg.info.key.tp_dst, sport);
1162 		if (IS_ERR(dst))
1163 			return PTR_ERR(dst);
1164 
1165 		dst_release(dst);
1166 		info->key.u.ipv6.src = fl6.saddr;
1167 #endif
1168 	} else {
1169 		return -EINVAL;
1170 	}
1171 
1172 	info->key.tp_src = sport;
1173 	info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1174 	return 0;
1175 }
1176 
1177 static const struct net_device_ops geneve_netdev_ops = {
1178 	.ndo_init		= geneve_init,
1179 	.ndo_uninit		= geneve_uninit,
1180 	.ndo_open		= geneve_open,
1181 	.ndo_stop		= geneve_stop,
1182 	.ndo_start_xmit		= geneve_xmit,
1183 	.ndo_get_stats64	= dev_get_tstats64,
1184 	.ndo_change_mtu		= geneve_change_mtu,
1185 	.ndo_validate_addr	= eth_validate_addr,
1186 	.ndo_set_mac_address	= eth_mac_addr,
1187 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1188 };
1189 
1190 static void geneve_get_drvinfo(struct net_device *dev,
1191 			       struct ethtool_drvinfo *drvinfo)
1192 {
1193 	strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1194 	strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1195 }
1196 
1197 static const struct ethtool_ops geneve_ethtool_ops = {
1198 	.get_drvinfo	= geneve_get_drvinfo,
1199 	.get_link	= ethtool_op_get_link,
1200 };
1201 
1202 /* Info for udev, that this is a virtual tunnel endpoint */
1203 static struct device_type geneve_type = {
1204 	.name = "geneve",
1205 };
1206 
1207 /* Calls the ndo_udp_tunnel_add of the caller in order to
1208  * supply the listening GENEVE udp ports. Callers are expected
1209  * to implement the ndo_udp_tunnel_add.
1210  */
1211 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1212 {
1213 	struct net *net = dev_net(dev);
1214 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1215 	struct geneve_sock *gs;
1216 
1217 	rcu_read_lock();
1218 	list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1219 		if (push) {
1220 			udp_tunnel_push_rx_port(dev, gs->sock,
1221 						UDP_TUNNEL_TYPE_GENEVE);
1222 		} else {
1223 			udp_tunnel_drop_rx_port(dev, gs->sock,
1224 						UDP_TUNNEL_TYPE_GENEVE);
1225 		}
1226 	}
1227 	rcu_read_unlock();
1228 }
1229 
1230 /* Initialize the device structure. */
1231 static void geneve_setup(struct net_device *dev)
1232 {
1233 	ether_setup(dev);
1234 
1235 	dev->netdev_ops = &geneve_netdev_ops;
1236 	dev->ethtool_ops = &geneve_ethtool_ops;
1237 	dev->needs_free_netdev = true;
1238 
1239 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1240 
1241 	dev->features    |= NETIF_F_LLTX;
1242 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1243 	dev->features    |= NETIF_F_RXCSUM;
1244 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1245 
1246 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1247 	dev->hw_features |= NETIF_F_RXCSUM;
1248 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1249 
1250 	/* MTU range: 68 - (something less than 65535) */
1251 	dev->min_mtu = ETH_MIN_MTU;
1252 	/* The max_mtu calculation does not take account of GENEVE
1253 	 * options, to avoid excluding potentially valid
1254 	 * configurations. This will be further reduced by IPvX hdr size.
1255 	 */
1256 	dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1257 
1258 	netif_keep_dst(dev);
1259 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1260 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1261 	eth_hw_addr_random(dev);
1262 }
1263 
1264 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1265 	[IFLA_GENEVE_UNSPEC]		= { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1266 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1267 	[IFLA_GENEVE_REMOTE]		= { .len = sizeof_field(struct iphdr, daddr) },
1268 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1269 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1270 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1271 	[IFLA_GENEVE_LABEL]		= { .type = NLA_U32 },
1272 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1273 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1274 	[IFLA_GENEVE_UDP_CSUM]		= { .type = NLA_U8 },
1275 	[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]	= { .type = NLA_U8 },
1276 	[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]	= { .type = NLA_U8 },
1277 	[IFLA_GENEVE_TTL_INHERIT]	= { .type = NLA_U8 },
1278 	[IFLA_GENEVE_DF]		= { .type = NLA_U8 },
1279 	[IFLA_GENEVE_INNER_PROTO_INHERIT]	= { .type = NLA_FLAG },
1280 };
1281 
1282 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1283 			   struct netlink_ext_ack *extack)
1284 {
1285 	if (tb[IFLA_ADDRESS]) {
1286 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1287 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1288 					    "Provided link layer address is not Ethernet");
1289 			return -EINVAL;
1290 		}
1291 
1292 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1293 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1294 					    "Provided Ethernet address is not unicast");
1295 			return -EADDRNOTAVAIL;
1296 		}
1297 	}
1298 
1299 	if (!data) {
1300 		NL_SET_ERR_MSG(extack,
1301 			       "Not enough attributes provided to perform the operation");
1302 		return -EINVAL;
1303 	}
1304 
1305 	if (data[IFLA_GENEVE_ID]) {
1306 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1307 
1308 		if (vni >= GENEVE_N_VID) {
1309 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1310 					    "Geneve ID must be lower than 16777216");
1311 			return -ERANGE;
1312 		}
1313 	}
1314 
1315 	if (data[IFLA_GENEVE_DF]) {
1316 		enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1317 
1318 		if (df < 0 || df > GENEVE_DF_MAX) {
1319 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1320 					    "Invalid DF attribute");
1321 			return -EINVAL;
1322 		}
1323 	}
1324 
1325 	return 0;
1326 }
1327 
1328 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1329 					  const struct ip_tunnel_info *info,
1330 					  bool *tun_on_same_port,
1331 					  bool *tun_collect_md)
1332 {
1333 	struct geneve_dev *geneve, *t = NULL;
1334 
1335 	*tun_on_same_port = false;
1336 	*tun_collect_md = false;
1337 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1338 		if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1339 			*tun_collect_md = geneve->cfg.collect_md;
1340 			*tun_on_same_port = true;
1341 		}
1342 		if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1343 		    info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1344 		    !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1345 			t = geneve;
1346 	}
1347 	return t;
1348 }
1349 
1350 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1351 {
1352 	return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1353 		 info->key.ttl || info->key.label || info->key.tp_src ||
1354 		 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1355 }
1356 
1357 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1358 				  struct ip_tunnel_info *b)
1359 {
1360 	if (ip_tunnel_info_af(a) == AF_INET)
1361 		return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1362 	else
1363 		return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1364 }
1365 
1366 static int geneve_configure(struct net *net, struct net_device *dev,
1367 			    struct netlink_ext_ack *extack,
1368 			    const struct geneve_config *cfg)
1369 {
1370 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1371 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1372 	const struct ip_tunnel_info *info = &cfg->info;
1373 	bool tun_collect_md, tun_on_same_port;
1374 	int err, encap_len;
1375 
1376 	if (cfg->collect_md && !is_tnl_info_zero(info)) {
1377 		NL_SET_ERR_MSG(extack,
1378 			       "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1379 		return -EINVAL;
1380 	}
1381 
1382 	geneve->net = net;
1383 	geneve->dev = dev;
1384 
1385 	t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1386 	if (t)
1387 		return -EBUSY;
1388 
1389 	/* make enough headroom for basic scenario */
1390 	encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1391 	if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1392 		encap_len += sizeof(struct iphdr);
1393 		dev->max_mtu -= sizeof(struct iphdr);
1394 	} else {
1395 		encap_len += sizeof(struct ipv6hdr);
1396 		dev->max_mtu -= sizeof(struct ipv6hdr);
1397 	}
1398 	dev->needed_headroom = encap_len + ETH_HLEN;
1399 
1400 	if (cfg->collect_md) {
1401 		if (tun_on_same_port) {
1402 			NL_SET_ERR_MSG(extack,
1403 				       "There can be only one externally controlled device on a destination port");
1404 			return -EPERM;
1405 		}
1406 	} else {
1407 		if (tun_collect_md) {
1408 			NL_SET_ERR_MSG(extack,
1409 				       "There already exists an externally controlled device on this destination port");
1410 			return -EPERM;
1411 		}
1412 	}
1413 
1414 	dst_cache_reset(&geneve->cfg.info.dst_cache);
1415 	memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1416 
1417 	if (geneve->cfg.inner_proto_inherit) {
1418 		dev->header_ops = NULL;
1419 		dev->type = ARPHRD_NONE;
1420 		dev->hard_header_len = 0;
1421 		dev->addr_len = 0;
1422 		dev->flags = IFF_NOARP;
1423 	}
1424 
1425 	err = register_netdevice(dev);
1426 	if (err)
1427 		return err;
1428 
1429 	list_add(&geneve->next, &gn->geneve_list);
1430 	return 0;
1431 }
1432 
1433 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1434 {
1435 	memset(info, 0, sizeof(*info));
1436 	info->key.tp_dst = htons(dst_port);
1437 }
1438 
1439 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1440 			  struct netlink_ext_ack *extack,
1441 			  struct geneve_config *cfg, bool changelink)
1442 {
1443 	struct ip_tunnel_info *info = &cfg->info;
1444 	int attrtype;
1445 
1446 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1447 		NL_SET_ERR_MSG(extack,
1448 			       "Cannot specify both IPv4 and IPv6 Remote addresses");
1449 		return -EINVAL;
1450 	}
1451 
1452 	if (data[IFLA_GENEVE_REMOTE]) {
1453 		if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1454 			attrtype = IFLA_GENEVE_REMOTE;
1455 			goto change_notsup;
1456 		}
1457 
1458 		info->key.u.ipv4.dst =
1459 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1460 
1461 		if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1462 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1463 					    "Remote IPv4 address cannot be Multicast");
1464 			return -EINVAL;
1465 		}
1466 	}
1467 
1468 	if (data[IFLA_GENEVE_REMOTE6]) {
1469 #if IS_ENABLED(CONFIG_IPV6)
1470 		if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1471 			attrtype = IFLA_GENEVE_REMOTE6;
1472 			goto change_notsup;
1473 		}
1474 
1475 		info->mode = IP_TUNNEL_INFO_IPV6;
1476 		info->key.u.ipv6.dst =
1477 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1478 
1479 		if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1480 		    IPV6_ADDR_LINKLOCAL) {
1481 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1482 					    "Remote IPv6 address cannot be link-local");
1483 			return -EINVAL;
1484 		}
1485 		if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1486 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1487 					    "Remote IPv6 address cannot be Multicast");
1488 			return -EINVAL;
1489 		}
1490 		info->key.tun_flags |= TUNNEL_CSUM;
1491 		cfg->use_udp6_rx_checksums = true;
1492 #else
1493 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1494 				    "IPv6 support not enabled in the kernel");
1495 		return -EPFNOSUPPORT;
1496 #endif
1497 	}
1498 
1499 	if (data[IFLA_GENEVE_ID]) {
1500 		__u32 vni;
1501 		__u8 tvni[3];
1502 		__be64 tunid;
1503 
1504 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1505 		tvni[0] = (vni & 0x00ff0000) >> 16;
1506 		tvni[1] = (vni & 0x0000ff00) >> 8;
1507 		tvni[2] =  vni & 0x000000ff;
1508 
1509 		tunid = vni_to_tunnel_id(tvni);
1510 		if (changelink && (tunid != info->key.tun_id)) {
1511 			attrtype = IFLA_GENEVE_ID;
1512 			goto change_notsup;
1513 		}
1514 		info->key.tun_id = tunid;
1515 	}
1516 
1517 	if (data[IFLA_GENEVE_TTL_INHERIT]) {
1518 		if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1519 			cfg->ttl_inherit = true;
1520 		else
1521 			cfg->ttl_inherit = false;
1522 	} else if (data[IFLA_GENEVE_TTL]) {
1523 		info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1524 		cfg->ttl_inherit = false;
1525 	}
1526 
1527 	if (data[IFLA_GENEVE_TOS])
1528 		info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1529 
1530 	if (data[IFLA_GENEVE_DF])
1531 		cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1532 
1533 	if (data[IFLA_GENEVE_LABEL]) {
1534 		info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1535 				  IPV6_FLOWLABEL_MASK;
1536 		if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1537 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1538 					    "Label attribute only applies for IPv6 Geneve devices");
1539 			return -EINVAL;
1540 		}
1541 	}
1542 
1543 	if (data[IFLA_GENEVE_PORT]) {
1544 		if (changelink) {
1545 			attrtype = IFLA_GENEVE_PORT;
1546 			goto change_notsup;
1547 		}
1548 		info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1549 	}
1550 
1551 	if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1552 		if (changelink) {
1553 			attrtype = IFLA_GENEVE_COLLECT_METADATA;
1554 			goto change_notsup;
1555 		}
1556 		cfg->collect_md = true;
1557 	}
1558 
1559 	if (data[IFLA_GENEVE_UDP_CSUM]) {
1560 		if (changelink) {
1561 			attrtype = IFLA_GENEVE_UDP_CSUM;
1562 			goto change_notsup;
1563 		}
1564 		if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1565 			info->key.tun_flags |= TUNNEL_CSUM;
1566 	}
1567 
1568 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1569 #if IS_ENABLED(CONFIG_IPV6)
1570 		if (changelink) {
1571 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1572 			goto change_notsup;
1573 		}
1574 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1575 			info->key.tun_flags &= ~TUNNEL_CSUM;
1576 #else
1577 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1578 				    "IPv6 support not enabled in the kernel");
1579 		return -EPFNOSUPPORT;
1580 #endif
1581 	}
1582 
1583 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1584 #if IS_ENABLED(CONFIG_IPV6)
1585 		if (changelink) {
1586 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1587 			goto change_notsup;
1588 		}
1589 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1590 			cfg->use_udp6_rx_checksums = false;
1591 #else
1592 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1593 				    "IPv6 support not enabled in the kernel");
1594 		return -EPFNOSUPPORT;
1595 #endif
1596 	}
1597 
1598 	if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1599 		if (changelink) {
1600 			attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1601 			goto change_notsup;
1602 		}
1603 		cfg->inner_proto_inherit = true;
1604 	}
1605 
1606 	return 0;
1607 change_notsup:
1608 	NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1609 			    "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1610 	return -EOPNOTSUPP;
1611 }
1612 
1613 static void geneve_link_config(struct net_device *dev,
1614 			       struct ip_tunnel_info *info, struct nlattr *tb[])
1615 {
1616 	struct geneve_dev *geneve = netdev_priv(dev);
1617 	int ldev_mtu = 0;
1618 
1619 	if (tb[IFLA_MTU]) {
1620 		geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1621 		return;
1622 	}
1623 
1624 	switch (ip_tunnel_info_af(info)) {
1625 	case AF_INET: {
1626 		struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1627 		struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1628 
1629 		if (!IS_ERR(rt) && rt->dst.dev) {
1630 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1631 			ip_rt_put(rt);
1632 		}
1633 		break;
1634 	}
1635 #if IS_ENABLED(CONFIG_IPV6)
1636 	case AF_INET6: {
1637 		struct rt6_info *rt;
1638 
1639 		if (!__in6_dev_get(dev))
1640 			break;
1641 
1642 		rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1643 				NULL, 0);
1644 
1645 		if (rt && rt->dst.dev)
1646 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1647 		ip6_rt_put(rt);
1648 		break;
1649 	}
1650 #endif
1651 	}
1652 
1653 	if (ldev_mtu <= 0)
1654 		return;
1655 
1656 	geneve_change_mtu(dev, ldev_mtu - info->options_len);
1657 }
1658 
1659 static int geneve_newlink(struct net *net, struct net_device *dev,
1660 			  struct nlattr *tb[], struct nlattr *data[],
1661 			  struct netlink_ext_ack *extack)
1662 {
1663 	struct geneve_config cfg = {
1664 		.df = GENEVE_DF_UNSET,
1665 		.use_udp6_rx_checksums = false,
1666 		.ttl_inherit = false,
1667 		.collect_md = false,
1668 	};
1669 	int err;
1670 
1671 	init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1672 	err = geneve_nl2info(tb, data, extack, &cfg, false);
1673 	if (err)
1674 		return err;
1675 
1676 	err = geneve_configure(net, dev, extack, &cfg);
1677 	if (err)
1678 		return err;
1679 
1680 	geneve_link_config(dev, &cfg.info, tb);
1681 
1682 	return 0;
1683 }
1684 
1685 /* Quiesces the geneve device data path for both TX and RX.
1686  *
1687  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1688  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1689  * to complete for the existing set of in-flight packets to be transmitted,
1690  * then we would have quiesced the transmit data path. All the future packets
1691  * will get dropped until we unquiesce the data path.
1692  *
1693  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1694  * if we set that to NULL under RCU and wait for synchronize_net() to
1695  * complete, then we would have quiesced the receive data path.
1696  */
1697 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1698 			   struct geneve_sock **gs6)
1699 {
1700 	*gs4 = rtnl_dereference(geneve->sock4);
1701 	rcu_assign_pointer(geneve->sock4, NULL);
1702 	if (*gs4)
1703 		rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1704 #if IS_ENABLED(CONFIG_IPV6)
1705 	*gs6 = rtnl_dereference(geneve->sock6);
1706 	rcu_assign_pointer(geneve->sock6, NULL);
1707 	if (*gs6)
1708 		rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1709 #else
1710 	*gs6 = NULL;
1711 #endif
1712 	synchronize_net();
1713 }
1714 
1715 /* Resumes the geneve device data path for both TX and RX. */
1716 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1717 			     struct geneve_sock __maybe_unused *gs6)
1718 {
1719 	rcu_assign_pointer(geneve->sock4, gs4);
1720 	if (gs4)
1721 		rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1722 #if IS_ENABLED(CONFIG_IPV6)
1723 	rcu_assign_pointer(geneve->sock6, gs6);
1724 	if (gs6)
1725 		rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1726 #endif
1727 	synchronize_net();
1728 }
1729 
1730 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1731 			     struct nlattr *data[],
1732 			     struct netlink_ext_ack *extack)
1733 {
1734 	struct geneve_dev *geneve = netdev_priv(dev);
1735 	struct geneve_sock *gs4, *gs6;
1736 	struct geneve_config cfg;
1737 	int err;
1738 
1739 	/* If the geneve device is configured for metadata (or externally
1740 	 * controlled, for example, OVS), then nothing can be changed.
1741 	 */
1742 	if (geneve->cfg.collect_md)
1743 		return -EOPNOTSUPP;
1744 
1745 	/* Start with the existing info. */
1746 	memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1747 	err = geneve_nl2info(tb, data, extack, &cfg, true);
1748 	if (err)
1749 		return err;
1750 
1751 	if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1752 		dst_cache_reset(&cfg.info.dst_cache);
1753 		geneve_link_config(dev, &cfg.info, tb);
1754 	}
1755 
1756 	geneve_quiesce(geneve, &gs4, &gs6);
1757 	memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1758 	geneve_unquiesce(geneve, gs4, gs6);
1759 
1760 	return 0;
1761 }
1762 
1763 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1764 {
1765 	struct geneve_dev *geneve = netdev_priv(dev);
1766 
1767 	list_del(&geneve->next);
1768 	unregister_netdevice_queue(dev, head);
1769 }
1770 
1771 static size_t geneve_get_size(const struct net_device *dev)
1772 {
1773 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1774 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1775 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1776 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1777 		nla_total_size(sizeof(__u8)) +	/* IFLA_GENEVE_DF */
1778 		nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1779 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1780 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1781 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1782 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1783 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1784 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1785 		nla_total_size(0) +	 /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1786 		0;
1787 }
1788 
1789 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1790 {
1791 	struct geneve_dev *geneve = netdev_priv(dev);
1792 	struct ip_tunnel_info *info = &geneve->cfg.info;
1793 	bool ttl_inherit = geneve->cfg.ttl_inherit;
1794 	bool metadata = geneve->cfg.collect_md;
1795 	__u8 tmp_vni[3];
1796 	__u32 vni;
1797 
1798 	tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1799 	vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1800 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1801 		goto nla_put_failure;
1802 
1803 	if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1804 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1805 				    info->key.u.ipv4.dst))
1806 			goto nla_put_failure;
1807 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1808 			       !!(info->key.tun_flags & TUNNEL_CSUM)))
1809 			goto nla_put_failure;
1810 
1811 #if IS_ENABLED(CONFIG_IPV6)
1812 	} else if (!metadata) {
1813 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1814 				     &info->key.u.ipv6.dst))
1815 			goto nla_put_failure;
1816 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1817 			       !(info->key.tun_flags & TUNNEL_CSUM)))
1818 			goto nla_put_failure;
1819 #endif
1820 	}
1821 
1822 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1823 	    nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1824 	    nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1825 		goto nla_put_failure;
1826 
1827 	if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1828 		goto nla_put_failure;
1829 
1830 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1831 		goto nla_put_failure;
1832 
1833 	if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1834 		goto nla_put_failure;
1835 
1836 #if IS_ENABLED(CONFIG_IPV6)
1837 	if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1838 		       !geneve->cfg.use_udp6_rx_checksums))
1839 		goto nla_put_failure;
1840 #endif
1841 
1842 	if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1843 		goto nla_put_failure;
1844 
1845 	if (geneve->cfg.inner_proto_inherit &&
1846 	    nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1847 		goto nla_put_failure;
1848 
1849 	return 0;
1850 
1851 nla_put_failure:
1852 	return -EMSGSIZE;
1853 }
1854 
1855 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1856 	.kind		= "geneve",
1857 	.maxtype	= IFLA_GENEVE_MAX,
1858 	.policy		= geneve_policy,
1859 	.priv_size	= sizeof(struct geneve_dev),
1860 	.setup		= geneve_setup,
1861 	.validate	= geneve_validate,
1862 	.newlink	= geneve_newlink,
1863 	.changelink	= geneve_changelink,
1864 	.dellink	= geneve_dellink,
1865 	.get_size	= geneve_get_size,
1866 	.fill_info	= geneve_fill_info,
1867 };
1868 
1869 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1870 					u8 name_assign_type, u16 dst_port)
1871 {
1872 	struct nlattr *tb[IFLA_MAX + 1];
1873 	struct net_device *dev;
1874 	LIST_HEAD(list_kill);
1875 	int err;
1876 	struct geneve_config cfg = {
1877 		.df = GENEVE_DF_UNSET,
1878 		.use_udp6_rx_checksums = true,
1879 		.ttl_inherit = false,
1880 		.collect_md = true,
1881 	};
1882 
1883 	memset(tb, 0, sizeof(tb));
1884 	dev = rtnl_create_link(net, name, name_assign_type,
1885 			       &geneve_link_ops, tb, NULL);
1886 	if (IS_ERR(dev))
1887 		return dev;
1888 
1889 	init_tnl_info(&cfg.info, dst_port);
1890 	err = geneve_configure(net, dev, NULL, &cfg);
1891 	if (err) {
1892 		free_netdev(dev);
1893 		return ERR_PTR(err);
1894 	}
1895 
1896 	/* openvswitch users expect packet sizes to be unrestricted,
1897 	 * so set the largest MTU we can.
1898 	 */
1899 	err = geneve_change_mtu(dev, IP_MAX_MTU);
1900 	if (err)
1901 		goto err;
1902 
1903 	err = rtnl_configure_link(dev, NULL);
1904 	if (err < 0)
1905 		goto err;
1906 
1907 	return dev;
1908 err:
1909 	geneve_dellink(dev, &list_kill);
1910 	unregister_netdevice_many(&list_kill);
1911 	return ERR_PTR(err);
1912 }
1913 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1914 
1915 static int geneve_netdevice_event(struct notifier_block *unused,
1916 				  unsigned long event, void *ptr)
1917 {
1918 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1919 
1920 	if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1921 		geneve_offload_rx_ports(dev, true);
1922 	else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1923 		geneve_offload_rx_ports(dev, false);
1924 
1925 	return NOTIFY_DONE;
1926 }
1927 
1928 static struct notifier_block geneve_notifier_block __read_mostly = {
1929 	.notifier_call = geneve_netdevice_event,
1930 };
1931 
1932 static __net_init int geneve_init_net(struct net *net)
1933 {
1934 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1935 
1936 	INIT_LIST_HEAD(&gn->geneve_list);
1937 	INIT_LIST_HEAD(&gn->sock_list);
1938 	return 0;
1939 }
1940 
1941 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1942 {
1943 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1944 	struct geneve_dev *geneve, *next;
1945 	struct net_device *dev, *aux;
1946 
1947 	/* gather any geneve devices that were moved into this ns */
1948 	for_each_netdev_safe(net, dev, aux)
1949 		if (dev->rtnl_link_ops == &geneve_link_ops)
1950 			unregister_netdevice_queue(dev, head);
1951 
1952 	/* now gather any other geneve devices that were created in this ns */
1953 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1954 		/* If geneve->dev is in the same netns, it was already added
1955 		 * to the list by the previous loop.
1956 		 */
1957 		if (!net_eq(dev_net(geneve->dev), net))
1958 			unregister_netdevice_queue(geneve->dev, head);
1959 	}
1960 }
1961 
1962 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1963 {
1964 	struct net *net;
1965 	LIST_HEAD(list);
1966 
1967 	rtnl_lock();
1968 	list_for_each_entry(net, net_list, exit_list)
1969 		geneve_destroy_tunnels(net, &list);
1970 
1971 	/* unregister the devices gathered above */
1972 	unregister_netdevice_many(&list);
1973 	rtnl_unlock();
1974 
1975 	list_for_each_entry(net, net_list, exit_list) {
1976 		const struct geneve_net *gn = net_generic(net, geneve_net_id);
1977 
1978 		WARN_ON_ONCE(!list_empty(&gn->sock_list));
1979 	}
1980 }
1981 
1982 static struct pernet_operations geneve_net_ops = {
1983 	.init = geneve_init_net,
1984 	.exit_batch = geneve_exit_batch_net,
1985 	.id   = &geneve_net_id,
1986 	.size = sizeof(struct geneve_net),
1987 };
1988 
1989 static int __init geneve_init_module(void)
1990 {
1991 	int rc;
1992 
1993 	rc = register_pernet_subsys(&geneve_net_ops);
1994 	if (rc)
1995 		goto out1;
1996 
1997 	rc = register_netdevice_notifier(&geneve_notifier_block);
1998 	if (rc)
1999 		goto out2;
2000 
2001 	rc = rtnl_link_register(&geneve_link_ops);
2002 	if (rc)
2003 		goto out3;
2004 
2005 	return 0;
2006 out3:
2007 	unregister_netdevice_notifier(&geneve_notifier_block);
2008 out2:
2009 	unregister_pernet_subsys(&geneve_net_ops);
2010 out1:
2011 	return rc;
2012 }
2013 late_initcall(geneve_init_module);
2014 
2015 static void __exit geneve_cleanup_module(void)
2016 {
2017 	rtnl_link_unregister(&geneve_link_ops);
2018 	unregister_netdevice_notifier(&geneve_notifier_block);
2019 	unregister_pernet_subsys(&geneve_net_ops);
2020 }
2021 module_exit(geneve_cleanup_module);
2022 
2023 MODULE_LICENSE("GPL");
2024 MODULE_VERSION(GENEVE_NETDEV_VER);
2025 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2026 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2027 MODULE_ALIAS_RTNL_LINK("geneve");
2028