xref: /openbmc/linux/drivers/net/geneve.c (revision f3dfffb3)
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 nh, 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 	/* Save offset of outer header relative to skb->head,
276 	 * because we are going to reset the network header to the inner header
277 	 * and might change skb->head.
278 	 */
279 	nh = skb_network_header(skb) - skb->head;
280 
281 	skb_reset_network_header(skb);
282 
283 	if (!pskb_inet_may_pull(skb)) {
284 		DEV_STATS_INC(geneve->dev, rx_length_errors);
285 		DEV_STATS_INC(geneve->dev, rx_errors);
286 		goto drop;
287 	}
288 
289 	/* Get the outer header. */
290 	oiph = skb->head + nh;
291 
292 	if (geneve_get_sk_family(gs) == AF_INET)
293 		err = IP_ECN_decapsulate(oiph, skb);
294 #if IS_ENABLED(CONFIG_IPV6)
295 	else
296 		err = IP6_ECN_decapsulate(oiph, skb);
297 #endif
298 
299 	if (unlikely(err)) {
300 		if (log_ecn_error) {
301 			if (geneve_get_sk_family(gs) == AF_INET)
302 				net_info_ratelimited("non-ECT from %pI4 "
303 						     "with TOS=%#x\n",
304 						     &((struct iphdr *)oiph)->saddr,
305 						     ((struct iphdr *)oiph)->tos);
306 #if IS_ENABLED(CONFIG_IPV6)
307 			else
308 				net_info_ratelimited("non-ECT from %pI6\n",
309 						     &((struct ipv6hdr *)oiph)->saddr);
310 #endif
311 		}
312 		if (err > 1) {
313 			++geneve->dev->stats.rx_frame_errors;
314 			++geneve->dev->stats.rx_errors;
315 			goto drop;
316 		}
317 	}
318 
319 	len = skb->len;
320 	err = gro_cells_receive(&geneve->gro_cells, skb);
321 	if (likely(err == NET_RX_SUCCESS))
322 		dev_sw_netstats_rx_add(geneve->dev, len);
323 
324 	return;
325 drop:
326 	/* Consume bad packet */
327 	kfree_skb(skb);
328 }
329 
330 /* Setup stats when device is created */
331 static int geneve_init(struct net_device *dev)
332 {
333 	struct geneve_dev *geneve = netdev_priv(dev);
334 	int err;
335 
336 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
337 	if (!dev->tstats)
338 		return -ENOMEM;
339 
340 	err = gro_cells_init(&geneve->gro_cells, dev);
341 	if (err) {
342 		free_percpu(dev->tstats);
343 		return err;
344 	}
345 
346 	err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
347 	if (err) {
348 		free_percpu(dev->tstats);
349 		gro_cells_destroy(&geneve->gro_cells);
350 		return err;
351 	}
352 	return 0;
353 }
354 
355 static void geneve_uninit(struct net_device *dev)
356 {
357 	struct geneve_dev *geneve = netdev_priv(dev);
358 
359 	dst_cache_destroy(&geneve->cfg.info.dst_cache);
360 	gro_cells_destroy(&geneve->gro_cells);
361 	free_percpu(dev->tstats);
362 }
363 
364 /* Callback from net/ipv4/udp.c to receive packets */
365 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
366 {
367 	struct genevehdr *geneveh;
368 	struct geneve_dev *geneve;
369 	struct geneve_sock *gs;
370 	__be16 inner_proto;
371 	int opts_len;
372 
373 	/* Need UDP and Geneve header to be present */
374 	if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
375 		goto drop;
376 
377 	/* Return packets with reserved bits set */
378 	geneveh = geneve_hdr(skb);
379 	if (unlikely(geneveh->ver != GENEVE_VER))
380 		goto drop;
381 
382 	gs = rcu_dereference_sk_user_data(sk);
383 	if (!gs)
384 		goto drop;
385 
386 	geneve = geneve_lookup_skb(gs, skb);
387 	if (!geneve)
388 		goto drop;
389 
390 	inner_proto = geneveh->proto_type;
391 
392 	if (unlikely((!geneve->cfg.inner_proto_inherit &&
393 		      inner_proto != htons(ETH_P_TEB)))) {
394 		geneve->dev->stats.rx_dropped++;
395 		goto drop;
396 	}
397 
398 	opts_len = geneveh->opt_len * 4;
399 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
400 				 !net_eq(geneve->net, dev_net(geneve->dev)))) {
401 		geneve->dev->stats.rx_dropped++;
402 		goto drop;
403 	}
404 
405 	geneve_rx(geneve, gs, skb);
406 	return 0;
407 
408 drop:
409 	/* Consume bad packet */
410 	kfree_skb(skb);
411 	return 0;
412 }
413 
414 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
415 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
416 {
417 	struct genevehdr *geneveh;
418 	struct geneve_sock *gs;
419 	u8 zero_vni[3] = { 0 };
420 	u8 *vni = zero_vni;
421 
422 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
423 		return -EINVAL;
424 
425 	geneveh = geneve_hdr(skb);
426 	if (geneveh->ver != GENEVE_VER)
427 		return -EINVAL;
428 
429 	if (geneveh->proto_type != htons(ETH_P_TEB))
430 		return -EINVAL;
431 
432 	gs = rcu_dereference_sk_user_data(sk);
433 	if (!gs)
434 		return -ENOENT;
435 
436 	if (geneve_get_sk_family(gs) == AF_INET) {
437 		struct iphdr *iph = ip_hdr(skb);
438 		__be32 addr4 = 0;
439 
440 		if (!gs->collect_md) {
441 			vni = geneve_hdr(skb)->vni;
442 			addr4 = iph->daddr;
443 		}
444 
445 		return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
446 	}
447 
448 #if IS_ENABLED(CONFIG_IPV6)
449 	if (geneve_get_sk_family(gs) == AF_INET6) {
450 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
451 		struct in6_addr addr6;
452 
453 		memset(&addr6, 0, sizeof(struct in6_addr));
454 
455 		if (!gs->collect_md) {
456 			vni = geneve_hdr(skb)->vni;
457 			addr6 = ip6h->daddr;
458 		}
459 
460 		return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
461 	}
462 #endif
463 
464 	return -EPFNOSUPPORT;
465 }
466 
467 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
468 					 __be16 port, bool ipv6_rx_csum)
469 {
470 	struct socket *sock;
471 	struct udp_port_cfg udp_conf;
472 	int err;
473 
474 	memset(&udp_conf, 0, sizeof(udp_conf));
475 
476 	if (ipv6) {
477 		udp_conf.family = AF_INET6;
478 		udp_conf.ipv6_v6only = 1;
479 		udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
480 	} else {
481 		udp_conf.family = AF_INET;
482 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
483 	}
484 
485 	udp_conf.local_udp_port = port;
486 
487 	/* Open UDP socket */
488 	err = udp_sock_create(net, &udp_conf, &sock);
489 	if (err < 0)
490 		return ERR_PTR(err);
491 
492 	udp_allow_gso(sock->sk);
493 	return sock;
494 }
495 
496 static int geneve_hlen(struct genevehdr *gh)
497 {
498 	return sizeof(*gh) + gh->opt_len * 4;
499 }
500 
501 static struct sk_buff *geneve_gro_receive(struct sock *sk,
502 					  struct list_head *head,
503 					  struct sk_buff *skb)
504 {
505 	struct sk_buff *pp = NULL;
506 	struct sk_buff *p;
507 	struct genevehdr *gh, *gh2;
508 	unsigned int hlen, gh_len, off_gnv;
509 	const struct packet_offload *ptype;
510 	__be16 type;
511 	int flush = 1;
512 
513 	off_gnv = skb_gro_offset(skb);
514 	hlen = off_gnv + sizeof(*gh);
515 	gh = skb_gro_header(skb, hlen, off_gnv);
516 	if (unlikely(!gh))
517 		goto out;
518 
519 	if (gh->ver != GENEVE_VER || gh->oam)
520 		goto out;
521 	gh_len = geneve_hlen(gh);
522 
523 	hlen = off_gnv + gh_len;
524 	if (skb_gro_header_hard(skb, hlen)) {
525 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
526 		if (unlikely(!gh))
527 			goto out;
528 	}
529 
530 	list_for_each_entry(p, head, list) {
531 		if (!NAPI_GRO_CB(p)->same_flow)
532 			continue;
533 
534 		gh2 = (struct genevehdr *)(p->data + off_gnv);
535 		if (gh->opt_len != gh2->opt_len ||
536 		    memcmp(gh, gh2, gh_len)) {
537 			NAPI_GRO_CB(p)->same_flow = 0;
538 			continue;
539 		}
540 	}
541 
542 	skb_gro_pull(skb, gh_len);
543 	skb_gro_postpull_rcsum(skb, gh, gh_len);
544 	type = gh->proto_type;
545 	if (likely(type == htons(ETH_P_TEB)))
546 		return call_gro_receive(eth_gro_receive, head, skb);
547 
548 	ptype = gro_find_receive_by_type(type);
549 	if (!ptype)
550 		goto out;
551 
552 	pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
553 	flush = 0;
554 
555 out:
556 	skb_gro_flush_final(skb, pp, flush);
557 
558 	return pp;
559 }
560 
561 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
562 			       int nhoff)
563 {
564 	struct genevehdr *gh;
565 	struct packet_offload *ptype;
566 	__be16 type;
567 	int gh_len;
568 	int err = -ENOSYS;
569 
570 	gh = (struct genevehdr *)(skb->data + nhoff);
571 	gh_len = geneve_hlen(gh);
572 	type = gh->proto_type;
573 
574 	/* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
575 	if (likely(type == htons(ETH_P_TEB)))
576 		return eth_gro_complete(skb, nhoff + gh_len);
577 
578 	ptype = gro_find_complete_by_type(type);
579 	if (ptype)
580 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
581 
582 	skb_set_inner_mac_header(skb, nhoff + gh_len);
583 
584 	return err;
585 }
586 
587 /* Create new listen socket if needed */
588 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
589 						bool ipv6, bool ipv6_rx_csum)
590 {
591 	struct geneve_net *gn = net_generic(net, geneve_net_id);
592 	struct geneve_sock *gs;
593 	struct socket *sock;
594 	struct udp_tunnel_sock_cfg tunnel_cfg;
595 	int h;
596 
597 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
598 	if (!gs)
599 		return ERR_PTR(-ENOMEM);
600 
601 	sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
602 	if (IS_ERR(sock)) {
603 		kfree(gs);
604 		return ERR_CAST(sock);
605 	}
606 
607 	gs->sock = sock;
608 	gs->refcnt = 1;
609 	for (h = 0; h < VNI_HASH_SIZE; ++h)
610 		INIT_HLIST_HEAD(&gs->vni_list[h]);
611 
612 	/* Initialize the geneve udp offloads structure */
613 	udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
614 
615 	/* Mark socket as an encapsulation socket */
616 	memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
617 	tunnel_cfg.sk_user_data = gs;
618 	tunnel_cfg.encap_type = 1;
619 	tunnel_cfg.gro_receive = geneve_gro_receive;
620 	tunnel_cfg.gro_complete = geneve_gro_complete;
621 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
622 	tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
623 	tunnel_cfg.encap_destroy = NULL;
624 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
625 	list_add(&gs->list, &gn->sock_list);
626 	return gs;
627 }
628 
629 static void __geneve_sock_release(struct geneve_sock *gs)
630 {
631 	if (!gs || --gs->refcnt)
632 		return;
633 
634 	list_del(&gs->list);
635 	udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
636 	udp_tunnel_sock_release(gs->sock);
637 	kfree_rcu(gs, rcu);
638 }
639 
640 static void geneve_sock_release(struct geneve_dev *geneve)
641 {
642 	struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
643 #if IS_ENABLED(CONFIG_IPV6)
644 	struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
645 
646 	rcu_assign_pointer(geneve->sock6, NULL);
647 #endif
648 
649 	rcu_assign_pointer(geneve->sock4, NULL);
650 	synchronize_net();
651 
652 	__geneve_sock_release(gs4);
653 #if IS_ENABLED(CONFIG_IPV6)
654 	__geneve_sock_release(gs6);
655 #endif
656 }
657 
658 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
659 					    sa_family_t family,
660 					    __be16 dst_port)
661 {
662 	struct geneve_sock *gs;
663 
664 	list_for_each_entry(gs, &gn->sock_list, list) {
665 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
666 		    geneve_get_sk_family(gs) == family) {
667 			return gs;
668 		}
669 	}
670 	return NULL;
671 }
672 
673 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
674 {
675 	struct net *net = geneve->net;
676 	struct geneve_net *gn = net_generic(net, geneve_net_id);
677 	struct geneve_dev_node *node;
678 	struct geneve_sock *gs;
679 	__u8 vni[3];
680 	__u32 hash;
681 
682 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
683 	if (gs) {
684 		gs->refcnt++;
685 		goto out;
686 	}
687 
688 	gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
689 				  geneve->cfg.use_udp6_rx_checksums);
690 	if (IS_ERR(gs))
691 		return PTR_ERR(gs);
692 
693 out:
694 	gs->collect_md = geneve->cfg.collect_md;
695 #if IS_ENABLED(CONFIG_IPV6)
696 	if (ipv6) {
697 		rcu_assign_pointer(geneve->sock6, gs);
698 		node = &geneve->hlist6;
699 	} else
700 #endif
701 	{
702 		rcu_assign_pointer(geneve->sock4, gs);
703 		node = &geneve->hlist4;
704 	}
705 	node->geneve = geneve;
706 
707 	tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
708 	hash = geneve_net_vni_hash(vni);
709 	hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
710 	return 0;
711 }
712 
713 static int geneve_open(struct net_device *dev)
714 {
715 	struct geneve_dev *geneve = netdev_priv(dev);
716 	bool metadata = geneve->cfg.collect_md;
717 	bool ipv4, ipv6;
718 	int ret = 0;
719 
720 	ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
721 	ipv4 = !ipv6 || metadata;
722 #if IS_ENABLED(CONFIG_IPV6)
723 	if (ipv6) {
724 		ret = geneve_sock_add(geneve, true);
725 		if (ret < 0 && ret != -EAFNOSUPPORT)
726 			ipv4 = false;
727 	}
728 #endif
729 	if (ipv4)
730 		ret = geneve_sock_add(geneve, false);
731 	if (ret < 0)
732 		geneve_sock_release(geneve);
733 
734 	return ret;
735 }
736 
737 static int geneve_stop(struct net_device *dev)
738 {
739 	struct geneve_dev *geneve = netdev_priv(dev);
740 
741 	hlist_del_init_rcu(&geneve->hlist4.hlist);
742 #if IS_ENABLED(CONFIG_IPV6)
743 	hlist_del_init_rcu(&geneve->hlist6.hlist);
744 #endif
745 	geneve_sock_release(geneve);
746 	return 0;
747 }
748 
749 static void geneve_build_header(struct genevehdr *geneveh,
750 				const struct ip_tunnel_info *info,
751 				__be16 inner_proto)
752 {
753 	geneveh->ver = GENEVE_VER;
754 	geneveh->opt_len = info->options_len / 4;
755 	geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
756 	geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
757 	geneveh->rsvd1 = 0;
758 	tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
759 	geneveh->proto_type = inner_proto;
760 	geneveh->rsvd2 = 0;
761 
762 	if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
763 		ip_tunnel_info_opts_get(geneveh->options, info);
764 }
765 
766 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
767 			    const struct ip_tunnel_info *info,
768 			    bool xnet, int ip_hdr_len,
769 			    bool inner_proto_inherit)
770 {
771 	bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
772 	struct genevehdr *gnvh;
773 	__be16 inner_proto;
774 	int min_headroom;
775 	int err;
776 
777 	skb_reset_mac_header(skb);
778 	skb_scrub_packet(skb, xnet);
779 
780 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
781 		       GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
782 	err = skb_cow_head(skb, min_headroom);
783 	if (unlikely(err))
784 		goto free_dst;
785 
786 	err = udp_tunnel_handle_offloads(skb, udp_sum);
787 	if (err)
788 		goto free_dst;
789 
790 	gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
791 	inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
792 	geneve_build_header(gnvh, info, inner_proto);
793 	skb_set_inner_protocol(skb, inner_proto);
794 	return 0;
795 
796 free_dst:
797 	dst_release(dst);
798 	return err;
799 }
800 
801 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
802 				       struct net_device *dev,
803 				       struct geneve_sock *gs4,
804 				       struct flowi4 *fl4,
805 				       const struct ip_tunnel_info *info,
806 				       __be16 dport, __be16 sport,
807 				       __u8 *full_tos)
808 {
809 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
810 	struct geneve_dev *geneve = netdev_priv(dev);
811 	struct dst_cache *dst_cache;
812 	struct rtable *rt = NULL;
813 	__u8 tos;
814 
815 	if (!gs4)
816 		return ERR_PTR(-EIO);
817 
818 	memset(fl4, 0, sizeof(*fl4));
819 	fl4->flowi4_mark = skb->mark;
820 	fl4->flowi4_proto = IPPROTO_UDP;
821 	fl4->daddr = info->key.u.ipv4.dst;
822 	fl4->saddr = info->key.u.ipv4.src;
823 	fl4->fl4_dport = dport;
824 	fl4->fl4_sport = sport;
825 	fl4->flowi4_flags = info->key.flow_flags;
826 
827 	tos = info->key.tos;
828 	if ((tos == 1) && !geneve->cfg.collect_md) {
829 		tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
830 		use_cache = false;
831 	}
832 	fl4->flowi4_tos = RT_TOS(tos);
833 	if (full_tos)
834 		*full_tos = tos;
835 
836 	dst_cache = (struct dst_cache *)&info->dst_cache;
837 	if (use_cache) {
838 		rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
839 		if (rt)
840 			return rt;
841 	}
842 	rt = ip_route_output_key(geneve->net, fl4);
843 	if (IS_ERR(rt)) {
844 		netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
845 		return ERR_PTR(-ENETUNREACH);
846 	}
847 	if (rt->dst.dev == dev) { /* is this necessary? */
848 		netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
849 		ip_rt_put(rt);
850 		return ERR_PTR(-ELOOP);
851 	}
852 	if (use_cache)
853 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
854 	return rt;
855 }
856 
857 #if IS_ENABLED(CONFIG_IPV6)
858 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
859 					   struct net_device *dev,
860 					   struct geneve_sock *gs6,
861 					   struct flowi6 *fl6,
862 					   const struct ip_tunnel_info *info,
863 					   __be16 dport, __be16 sport)
864 {
865 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
866 	struct geneve_dev *geneve = netdev_priv(dev);
867 	struct dst_entry *dst = NULL;
868 	struct dst_cache *dst_cache;
869 	__u8 prio;
870 
871 	if (!gs6)
872 		return ERR_PTR(-EIO);
873 
874 	memset(fl6, 0, sizeof(*fl6));
875 	fl6->flowi6_mark = skb->mark;
876 	fl6->flowi6_proto = IPPROTO_UDP;
877 	fl6->daddr = info->key.u.ipv6.dst;
878 	fl6->saddr = info->key.u.ipv6.src;
879 	fl6->fl6_dport = dport;
880 	fl6->fl6_sport = sport;
881 
882 	prio = info->key.tos;
883 	if ((prio == 1) && !geneve->cfg.collect_md) {
884 		prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
885 		use_cache = false;
886 	}
887 
888 	fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
889 	dst_cache = (struct dst_cache *)&info->dst_cache;
890 	if (use_cache) {
891 		dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
892 		if (dst)
893 			return dst;
894 	}
895 	dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
896 					      NULL);
897 	if (IS_ERR(dst)) {
898 		netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
899 		return ERR_PTR(-ENETUNREACH);
900 	}
901 	if (dst->dev == dev) { /* is this necessary? */
902 		netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
903 		dst_release(dst);
904 		return ERR_PTR(-ELOOP);
905 	}
906 
907 	if (use_cache)
908 		dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
909 	return dst;
910 }
911 #endif
912 
913 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
914 			   struct geneve_dev *geneve,
915 			   const struct ip_tunnel_info *info)
916 {
917 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
918 	struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
919 	const struct ip_tunnel_key *key = &info->key;
920 	struct rtable *rt;
921 	struct flowi4 fl4;
922 	__u8 full_tos;
923 	__u8 tos, ttl;
924 	__be16 df = 0;
925 	__be16 sport;
926 	int err;
927 
928 	if (!pskb_inet_may_pull(skb))
929 		return -EINVAL;
930 
931 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
932 	rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
933 			      geneve->cfg.info.key.tp_dst, sport, &full_tos);
934 	if (IS_ERR(rt))
935 		return PTR_ERR(rt);
936 
937 	err = skb_tunnel_check_pmtu(skb, &rt->dst,
938 				    GENEVE_IPV4_HLEN + info->options_len,
939 				    netif_is_any_bridge_port(dev));
940 	if (err < 0) {
941 		dst_release(&rt->dst);
942 		return err;
943 	} else if (err) {
944 		struct ip_tunnel_info *info;
945 
946 		info = skb_tunnel_info(skb);
947 		if (info) {
948 			struct ip_tunnel_info *unclone;
949 
950 			unclone = skb_tunnel_info_unclone(skb);
951 			if (unlikely(!unclone)) {
952 				dst_release(&rt->dst);
953 				return -ENOMEM;
954 			}
955 
956 			unclone->key.u.ipv4.dst = fl4.saddr;
957 			unclone->key.u.ipv4.src = fl4.daddr;
958 		}
959 
960 		if (!pskb_may_pull(skb, ETH_HLEN)) {
961 			dst_release(&rt->dst);
962 			return -EINVAL;
963 		}
964 
965 		skb->protocol = eth_type_trans(skb, geneve->dev);
966 		__netif_rx(skb);
967 		dst_release(&rt->dst);
968 		return -EMSGSIZE;
969 	}
970 
971 	if (geneve->cfg.collect_md) {
972 		tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
973 		ttl = key->ttl;
974 
975 		df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
976 	} else {
977 		tos = ip_tunnel_ecn_encap(full_tos, ip_hdr(skb), skb);
978 		if (geneve->cfg.ttl_inherit)
979 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
980 		else
981 			ttl = key->ttl;
982 		ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
983 
984 		if (geneve->cfg.df == GENEVE_DF_SET) {
985 			df = htons(IP_DF);
986 		} else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
987 			struct ethhdr *eth = eth_hdr(skb);
988 
989 			if (ntohs(eth->h_proto) == ETH_P_IPV6) {
990 				df = htons(IP_DF);
991 			} else if (ntohs(eth->h_proto) == ETH_P_IP) {
992 				struct iphdr *iph = ip_hdr(skb);
993 
994 				if (iph->frag_off & htons(IP_DF))
995 					df = htons(IP_DF);
996 			}
997 		}
998 	}
999 
1000 	err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
1001 			       geneve->cfg.inner_proto_inherit);
1002 	if (unlikely(err))
1003 		return err;
1004 
1005 	udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
1006 			    tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
1007 			    !net_eq(geneve->net, dev_net(geneve->dev)),
1008 			    !(info->key.tun_flags & TUNNEL_CSUM));
1009 	return 0;
1010 }
1011 
1012 #if IS_ENABLED(CONFIG_IPV6)
1013 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
1014 			    struct geneve_dev *geneve,
1015 			    const struct ip_tunnel_info *info)
1016 {
1017 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1018 	struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1019 	const struct ip_tunnel_key *key = &info->key;
1020 	struct dst_entry *dst = NULL;
1021 	struct flowi6 fl6;
1022 	__u8 prio, ttl;
1023 	__be16 sport;
1024 	int err;
1025 
1026 	if (!pskb_inet_may_pull(skb))
1027 		return -EINVAL;
1028 
1029 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1030 	dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1031 				geneve->cfg.info.key.tp_dst, sport);
1032 	if (IS_ERR(dst))
1033 		return PTR_ERR(dst);
1034 
1035 	err = skb_tunnel_check_pmtu(skb, dst,
1036 				    GENEVE_IPV6_HLEN + info->options_len,
1037 				    netif_is_any_bridge_port(dev));
1038 	if (err < 0) {
1039 		dst_release(dst);
1040 		return err;
1041 	} else if (err) {
1042 		struct ip_tunnel_info *info = skb_tunnel_info(skb);
1043 
1044 		if (info) {
1045 			struct ip_tunnel_info *unclone;
1046 
1047 			unclone = skb_tunnel_info_unclone(skb);
1048 			if (unlikely(!unclone)) {
1049 				dst_release(dst);
1050 				return -ENOMEM;
1051 			}
1052 
1053 			unclone->key.u.ipv6.dst = fl6.saddr;
1054 			unclone->key.u.ipv6.src = fl6.daddr;
1055 		}
1056 
1057 		if (!pskb_may_pull(skb, ETH_HLEN)) {
1058 			dst_release(dst);
1059 			return -EINVAL;
1060 		}
1061 
1062 		skb->protocol = eth_type_trans(skb, geneve->dev);
1063 		__netif_rx(skb);
1064 		dst_release(dst);
1065 		return -EMSGSIZE;
1066 	}
1067 
1068 	if (geneve->cfg.collect_md) {
1069 		prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1070 		ttl = key->ttl;
1071 	} else {
1072 		prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1073 					   ip_hdr(skb), skb);
1074 		if (geneve->cfg.ttl_inherit)
1075 			ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1076 		else
1077 			ttl = key->ttl;
1078 		ttl = ttl ? : ip6_dst_hoplimit(dst);
1079 	}
1080 	err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1081 			       geneve->cfg.inner_proto_inherit);
1082 	if (unlikely(err))
1083 		return err;
1084 
1085 	udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1086 			     &fl6.saddr, &fl6.daddr, prio, ttl,
1087 			     info->key.label, sport, geneve->cfg.info.key.tp_dst,
1088 			     !(info->key.tun_flags & TUNNEL_CSUM));
1089 	return 0;
1090 }
1091 #endif
1092 
1093 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1094 {
1095 	struct geneve_dev *geneve = netdev_priv(dev);
1096 	struct ip_tunnel_info *info = NULL;
1097 	int err;
1098 
1099 	if (geneve->cfg.collect_md) {
1100 		info = skb_tunnel_info(skb);
1101 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1102 			netdev_dbg(dev, "no tunnel metadata\n");
1103 			dev_kfree_skb(skb);
1104 			dev->stats.tx_dropped++;
1105 			return NETDEV_TX_OK;
1106 		}
1107 	} else {
1108 		info = &geneve->cfg.info;
1109 	}
1110 
1111 	rcu_read_lock();
1112 #if IS_ENABLED(CONFIG_IPV6)
1113 	if (info->mode & IP_TUNNEL_INFO_IPV6)
1114 		err = geneve6_xmit_skb(skb, dev, geneve, info);
1115 	else
1116 #endif
1117 		err = geneve_xmit_skb(skb, dev, geneve, info);
1118 	rcu_read_unlock();
1119 
1120 	if (likely(!err))
1121 		return NETDEV_TX_OK;
1122 
1123 	if (err != -EMSGSIZE)
1124 		dev_kfree_skb(skb);
1125 
1126 	if (err == -ELOOP)
1127 		dev->stats.collisions++;
1128 	else if (err == -ENETUNREACH)
1129 		dev->stats.tx_carrier_errors++;
1130 
1131 	dev->stats.tx_errors++;
1132 	return NETDEV_TX_OK;
1133 }
1134 
1135 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1136 {
1137 	if (new_mtu > dev->max_mtu)
1138 		new_mtu = dev->max_mtu;
1139 	else if (new_mtu < dev->min_mtu)
1140 		new_mtu = dev->min_mtu;
1141 
1142 	dev->mtu = new_mtu;
1143 	return 0;
1144 }
1145 
1146 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1147 {
1148 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
1149 	struct geneve_dev *geneve = netdev_priv(dev);
1150 	__be16 sport;
1151 
1152 	if (ip_tunnel_info_af(info) == AF_INET) {
1153 		struct rtable *rt;
1154 		struct flowi4 fl4;
1155 
1156 		struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1157 		sport = udp_flow_src_port(geneve->net, skb,
1158 					  1, USHRT_MAX, true);
1159 
1160 		rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1161 				      geneve->cfg.info.key.tp_dst, sport, NULL);
1162 		if (IS_ERR(rt))
1163 			return PTR_ERR(rt);
1164 
1165 		ip_rt_put(rt);
1166 		info->key.u.ipv4.src = fl4.saddr;
1167 #if IS_ENABLED(CONFIG_IPV6)
1168 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
1169 		struct dst_entry *dst;
1170 		struct flowi6 fl6;
1171 
1172 		struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1173 		sport = udp_flow_src_port(geneve->net, skb,
1174 					  1, USHRT_MAX, true);
1175 
1176 		dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1177 					geneve->cfg.info.key.tp_dst, sport);
1178 		if (IS_ERR(dst))
1179 			return PTR_ERR(dst);
1180 
1181 		dst_release(dst);
1182 		info->key.u.ipv6.src = fl6.saddr;
1183 #endif
1184 	} else {
1185 		return -EINVAL;
1186 	}
1187 
1188 	info->key.tp_src = sport;
1189 	info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1190 	return 0;
1191 }
1192 
1193 static const struct net_device_ops geneve_netdev_ops = {
1194 	.ndo_init		= geneve_init,
1195 	.ndo_uninit		= geneve_uninit,
1196 	.ndo_open		= geneve_open,
1197 	.ndo_stop		= geneve_stop,
1198 	.ndo_start_xmit		= geneve_xmit,
1199 	.ndo_get_stats64	= dev_get_tstats64,
1200 	.ndo_change_mtu		= geneve_change_mtu,
1201 	.ndo_validate_addr	= eth_validate_addr,
1202 	.ndo_set_mac_address	= eth_mac_addr,
1203 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1204 };
1205 
1206 static void geneve_get_drvinfo(struct net_device *dev,
1207 			       struct ethtool_drvinfo *drvinfo)
1208 {
1209 	strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1210 	strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1211 }
1212 
1213 static const struct ethtool_ops geneve_ethtool_ops = {
1214 	.get_drvinfo	= geneve_get_drvinfo,
1215 	.get_link	= ethtool_op_get_link,
1216 };
1217 
1218 /* Info for udev, that this is a virtual tunnel endpoint */
1219 static struct device_type geneve_type = {
1220 	.name = "geneve",
1221 };
1222 
1223 /* Calls the ndo_udp_tunnel_add of the caller in order to
1224  * supply the listening GENEVE udp ports. Callers are expected
1225  * to implement the ndo_udp_tunnel_add.
1226  */
1227 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1228 {
1229 	struct net *net = dev_net(dev);
1230 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1231 	struct geneve_sock *gs;
1232 
1233 	rcu_read_lock();
1234 	list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1235 		if (push) {
1236 			udp_tunnel_push_rx_port(dev, gs->sock,
1237 						UDP_TUNNEL_TYPE_GENEVE);
1238 		} else {
1239 			udp_tunnel_drop_rx_port(dev, gs->sock,
1240 						UDP_TUNNEL_TYPE_GENEVE);
1241 		}
1242 	}
1243 	rcu_read_unlock();
1244 }
1245 
1246 /* Initialize the device structure. */
1247 static void geneve_setup(struct net_device *dev)
1248 {
1249 	ether_setup(dev);
1250 
1251 	dev->netdev_ops = &geneve_netdev_ops;
1252 	dev->ethtool_ops = &geneve_ethtool_ops;
1253 	dev->needs_free_netdev = true;
1254 
1255 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1256 
1257 	dev->features    |= NETIF_F_LLTX;
1258 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1259 	dev->features    |= NETIF_F_RXCSUM;
1260 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1261 
1262 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1263 	dev->hw_features |= NETIF_F_RXCSUM;
1264 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1265 
1266 	/* MTU range: 68 - (something less than 65535) */
1267 	dev->min_mtu = ETH_MIN_MTU;
1268 	/* The max_mtu calculation does not take account of GENEVE
1269 	 * options, to avoid excluding potentially valid
1270 	 * configurations. This will be further reduced by IPvX hdr size.
1271 	 */
1272 	dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1273 
1274 	netif_keep_dst(dev);
1275 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1276 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1277 	eth_hw_addr_random(dev);
1278 }
1279 
1280 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1281 	[IFLA_GENEVE_UNSPEC]		= { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1282 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1283 	[IFLA_GENEVE_REMOTE]		= { .len = sizeof_field(struct iphdr, daddr) },
1284 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1285 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1286 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1287 	[IFLA_GENEVE_LABEL]		= { .type = NLA_U32 },
1288 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1289 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1290 	[IFLA_GENEVE_UDP_CSUM]		= { .type = NLA_U8 },
1291 	[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]	= { .type = NLA_U8 },
1292 	[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]	= { .type = NLA_U8 },
1293 	[IFLA_GENEVE_TTL_INHERIT]	= { .type = NLA_U8 },
1294 	[IFLA_GENEVE_DF]		= { .type = NLA_U8 },
1295 	[IFLA_GENEVE_INNER_PROTO_INHERIT]	= { .type = NLA_FLAG },
1296 };
1297 
1298 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1299 			   struct netlink_ext_ack *extack)
1300 {
1301 	if (tb[IFLA_ADDRESS]) {
1302 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1303 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1304 					    "Provided link layer address is not Ethernet");
1305 			return -EINVAL;
1306 		}
1307 
1308 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1309 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1310 					    "Provided Ethernet address is not unicast");
1311 			return -EADDRNOTAVAIL;
1312 		}
1313 	}
1314 
1315 	if (!data) {
1316 		NL_SET_ERR_MSG(extack,
1317 			       "Not enough attributes provided to perform the operation");
1318 		return -EINVAL;
1319 	}
1320 
1321 	if (data[IFLA_GENEVE_ID]) {
1322 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1323 
1324 		if (vni >= GENEVE_N_VID) {
1325 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1326 					    "Geneve ID must be lower than 16777216");
1327 			return -ERANGE;
1328 		}
1329 	}
1330 
1331 	if (data[IFLA_GENEVE_DF]) {
1332 		enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1333 
1334 		if (df < 0 || df > GENEVE_DF_MAX) {
1335 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1336 					    "Invalid DF attribute");
1337 			return -EINVAL;
1338 		}
1339 	}
1340 
1341 	return 0;
1342 }
1343 
1344 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1345 					  const struct ip_tunnel_info *info,
1346 					  bool *tun_on_same_port,
1347 					  bool *tun_collect_md)
1348 {
1349 	struct geneve_dev *geneve, *t = NULL;
1350 
1351 	*tun_on_same_port = false;
1352 	*tun_collect_md = false;
1353 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1354 		if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1355 			*tun_collect_md = geneve->cfg.collect_md;
1356 			*tun_on_same_port = true;
1357 		}
1358 		if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1359 		    info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1360 		    !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1361 			t = geneve;
1362 	}
1363 	return t;
1364 }
1365 
1366 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1367 {
1368 	return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1369 		 info->key.ttl || info->key.label || info->key.tp_src ||
1370 		 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1371 }
1372 
1373 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1374 				  struct ip_tunnel_info *b)
1375 {
1376 	if (ip_tunnel_info_af(a) == AF_INET)
1377 		return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1378 	else
1379 		return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1380 }
1381 
1382 static int geneve_configure(struct net *net, struct net_device *dev,
1383 			    struct netlink_ext_ack *extack,
1384 			    const struct geneve_config *cfg)
1385 {
1386 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1387 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1388 	const struct ip_tunnel_info *info = &cfg->info;
1389 	bool tun_collect_md, tun_on_same_port;
1390 	int err, encap_len;
1391 
1392 	if (cfg->collect_md && !is_tnl_info_zero(info)) {
1393 		NL_SET_ERR_MSG(extack,
1394 			       "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1395 		return -EINVAL;
1396 	}
1397 
1398 	geneve->net = net;
1399 	geneve->dev = dev;
1400 
1401 	t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1402 	if (t)
1403 		return -EBUSY;
1404 
1405 	/* make enough headroom for basic scenario */
1406 	encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1407 	if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1408 		encap_len += sizeof(struct iphdr);
1409 		dev->max_mtu -= sizeof(struct iphdr);
1410 	} else {
1411 		encap_len += sizeof(struct ipv6hdr);
1412 		dev->max_mtu -= sizeof(struct ipv6hdr);
1413 	}
1414 	dev->needed_headroom = encap_len + ETH_HLEN;
1415 
1416 	if (cfg->collect_md) {
1417 		if (tun_on_same_port) {
1418 			NL_SET_ERR_MSG(extack,
1419 				       "There can be only one externally controlled device on a destination port");
1420 			return -EPERM;
1421 		}
1422 	} else {
1423 		if (tun_collect_md) {
1424 			NL_SET_ERR_MSG(extack,
1425 				       "There already exists an externally controlled device on this destination port");
1426 			return -EPERM;
1427 		}
1428 	}
1429 
1430 	dst_cache_reset(&geneve->cfg.info.dst_cache);
1431 	memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1432 
1433 	if (geneve->cfg.inner_proto_inherit) {
1434 		dev->header_ops = NULL;
1435 		dev->type = ARPHRD_NONE;
1436 		dev->hard_header_len = 0;
1437 		dev->addr_len = 0;
1438 		dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1439 	}
1440 
1441 	err = register_netdevice(dev);
1442 	if (err)
1443 		return err;
1444 
1445 	list_add(&geneve->next, &gn->geneve_list);
1446 	return 0;
1447 }
1448 
1449 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1450 {
1451 	memset(info, 0, sizeof(*info));
1452 	info->key.tp_dst = htons(dst_port);
1453 }
1454 
1455 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1456 			  struct netlink_ext_ack *extack,
1457 			  struct geneve_config *cfg, bool changelink)
1458 {
1459 	struct ip_tunnel_info *info = &cfg->info;
1460 	int attrtype;
1461 
1462 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1463 		NL_SET_ERR_MSG(extack,
1464 			       "Cannot specify both IPv4 and IPv6 Remote addresses");
1465 		return -EINVAL;
1466 	}
1467 
1468 	if (data[IFLA_GENEVE_REMOTE]) {
1469 		if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1470 			attrtype = IFLA_GENEVE_REMOTE;
1471 			goto change_notsup;
1472 		}
1473 
1474 		info->key.u.ipv4.dst =
1475 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1476 
1477 		if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1478 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1479 					    "Remote IPv4 address cannot be Multicast");
1480 			return -EINVAL;
1481 		}
1482 	}
1483 
1484 	if (data[IFLA_GENEVE_REMOTE6]) {
1485 #if IS_ENABLED(CONFIG_IPV6)
1486 		if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1487 			attrtype = IFLA_GENEVE_REMOTE6;
1488 			goto change_notsup;
1489 		}
1490 
1491 		info->mode = IP_TUNNEL_INFO_IPV6;
1492 		info->key.u.ipv6.dst =
1493 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1494 
1495 		if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1496 		    IPV6_ADDR_LINKLOCAL) {
1497 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1498 					    "Remote IPv6 address cannot be link-local");
1499 			return -EINVAL;
1500 		}
1501 		if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1502 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1503 					    "Remote IPv6 address cannot be Multicast");
1504 			return -EINVAL;
1505 		}
1506 		info->key.tun_flags |= TUNNEL_CSUM;
1507 		cfg->use_udp6_rx_checksums = true;
1508 #else
1509 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1510 				    "IPv6 support not enabled in the kernel");
1511 		return -EPFNOSUPPORT;
1512 #endif
1513 	}
1514 
1515 	if (data[IFLA_GENEVE_ID]) {
1516 		__u32 vni;
1517 		__u8 tvni[3];
1518 		__be64 tunid;
1519 
1520 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1521 		tvni[0] = (vni & 0x00ff0000) >> 16;
1522 		tvni[1] = (vni & 0x0000ff00) >> 8;
1523 		tvni[2] =  vni & 0x000000ff;
1524 
1525 		tunid = vni_to_tunnel_id(tvni);
1526 		if (changelink && (tunid != info->key.tun_id)) {
1527 			attrtype = IFLA_GENEVE_ID;
1528 			goto change_notsup;
1529 		}
1530 		info->key.tun_id = tunid;
1531 	}
1532 
1533 	if (data[IFLA_GENEVE_TTL_INHERIT]) {
1534 		if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1535 			cfg->ttl_inherit = true;
1536 		else
1537 			cfg->ttl_inherit = false;
1538 	} else if (data[IFLA_GENEVE_TTL]) {
1539 		info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1540 		cfg->ttl_inherit = false;
1541 	}
1542 
1543 	if (data[IFLA_GENEVE_TOS])
1544 		info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1545 
1546 	if (data[IFLA_GENEVE_DF])
1547 		cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1548 
1549 	if (data[IFLA_GENEVE_LABEL]) {
1550 		info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1551 				  IPV6_FLOWLABEL_MASK;
1552 		if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1553 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1554 					    "Label attribute only applies for IPv6 Geneve devices");
1555 			return -EINVAL;
1556 		}
1557 	}
1558 
1559 	if (data[IFLA_GENEVE_PORT]) {
1560 		if (changelink) {
1561 			attrtype = IFLA_GENEVE_PORT;
1562 			goto change_notsup;
1563 		}
1564 		info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1565 	}
1566 
1567 	if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1568 		if (changelink) {
1569 			attrtype = IFLA_GENEVE_COLLECT_METADATA;
1570 			goto change_notsup;
1571 		}
1572 		cfg->collect_md = true;
1573 	}
1574 
1575 	if (data[IFLA_GENEVE_UDP_CSUM]) {
1576 		if (changelink) {
1577 			attrtype = IFLA_GENEVE_UDP_CSUM;
1578 			goto change_notsup;
1579 		}
1580 		if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1581 			info->key.tun_flags |= TUNNEL_CSUM;
1582 	}
1583 
1584 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1585 #if IS_ENABLED(CONFIG_IPV6)
1586 		if (changelink) {
1587 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1588 			goto change_notsup;
1589 		}
1590 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1591 			info->key.tun_flags &= ~TUNNEL_CSUM;
1592 #else
1593 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1594 				    "IPv6 support not enabled in the kernel");
1595 		return -EPFNOSUPPORT;
1596 #endif
1597 	}
1598 
1599 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1600 #if IS_ENABLED(CONFIG_IPV6)
1601 		if (changelink) {
1602 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1603 			goto change_notsup;
1604 		}
1605 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1606 			cfg->use_udp6_rx_checksums = false;
1607 #else
1608 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1609 				    "IPv6 support not enabled in the kernel");
1610 		return -EPFNOSUPPORT;
1611 #endif
1612 	}
1613 
1614 	if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1615 		if (changelink) {
1616 			attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1617 			goto change_notsup;
1618 		}
1619 		cfg->inner_proto_inherit = true;
1620 	}
1621 
1622 	return 0;
1623 change_notsup:
1624 	NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1625 			    "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1626 	return -EOPNOTSUPP;
1627 }
1628 
1629 static void geneve_link_config(struct net_device *dev,
1630 			       struct ip_tunnel_info *info, struct nlattr *tb[])
1631 {
1632 	struct geneve_dev *geneve = netdev_priv(dev);
1633 	int ldev_mtu = 0;
1634 
1635 	if (tb[IFLA_MTU]) {
1636 		geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1637 		return;
1638 	}
1639 
1640 	switch (ip_tunnel_info_af(info)) {
1641 	case AF_INET: {
1642 		struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1643 		struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1644 
1645 		if (!IS_ERR(rt) && rt->dst.dev) {
1646 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1647 			ip_rt_put(rt);
1648 		}
1649 		break;
1650 	}
1651 #if IS_ENABLED(CONFIG_IPV6)
1652 	case AF_INET6: {
1653 		struct rt6_info *rt;
1654 
1655 		if (!__in6_dev_get(dev))
1656 			break;
1657 
1658 		rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1659 				NULL, 0);
1660 
1661 		if (rt && rt->dst.dev)
1662 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1663 		ip6_rt_put(rt);
1664 		break;
1665 	}
1666 #endif
1667 	}
1668 
1669 	if (ldev_mtu <= 0)
1670 		return;
1671 
1672 	geneve_change_mtu(dev, ldev_mtu - info->options_len);
1673 }
1674 
1675 static int geneve_newlink(struct net *net, struct net_device *dev,
1676 			  struct nlattr *tb[], struct nlattr *data[],
1677 			  struct netlink_ext_ack *extack)
1678 {
1679 	struct geneve_config cfg = {
1680 		.df = GENEVE_DF_UNSET,
1681 		.use_udp6_rx_checksums = false,
1682 		.ttl_inherit = false,
1683 		.collect_md = false,
1684 	};
1685 	int err;
1686 
1687 	init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1688 	err = geneve_nl2info(tb, data, extack, &cfg, false);
1689 	if (err)
1690 		return err;
1691 
1692 	err = geneve_configure(net, dev, extack, &cfg);
1693 	if (err)
1694 		return err;
1695 
1696 	geneve_link_config(dev, &cfg.info, tb);
1697 
1698 	return 0;
1699 }
1700 
1701 /* Quiesces the geneve device data path for both TX and RX.
1702  *
1703  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1704  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1705  * to complete for the existing set of in-flight packets to be transmitted,
1706  * then we would have quiesced the transmit data path. All the future packets
1707  * will get dropped until we unquiesce the data path.
1708  *
1709  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1710  * if we set that to NULL under RCU and wait for synchronize_net() to
1711  * complete, then we would have quiesced the receive data path.
1712  */
1713 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1714 			   struct geneve_sock **gs6)
1715 {
1716 	*gs4 = rtnl_dereference(geneve->sock4);
1717 	rcu_assign_pointer(geneve->sock4, NULL);
1718 	if (*gs4)
1719 		rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1720 #if IS_ENABLED(CONFIG_IPV6)
1721 	*gs6 = rtnl_dereference(geneve->sock6);
1722 	rcu_assign_pointer(geneve->sock6, NULL);
1723 	if (*gs6)
1724 		rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1725 #else
1726 	*gs6 = NULL;
1727 #endif
1728 	synchronize_net();
1729 }
1730 
1731 /* Resumes the geneve device data path for both TX and RX. */
1732 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1733 			     struct geneve_sock __maybe_unused *gs6)
1734 {
1735 	rcu_assign_pointer(geneve->sock4, gs4);
1736 	if (gs4)
1737 		rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1738 #if IS_ENABLED(CONFIG_IPV6)
1739 	rcu_assign_pointer(geneve->sock6, gs6);
1740 	if (gs6)
1741 		rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1742 #endif
1743 	synchronize_net();
1744 }
1745 
1746 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1747 			     struct nlattr *data[],
1748 			     struct netlink_ext_ack *extack)
1749 {
1750 	struct geneve_dev *geneve = netdev_priv(dev);
1751 	struct geneve_sock *gs4, *gs6;
1752 	struct geneve_config cfg;
1753 	int err;
1754 
1755 	/* If the geneve device is configured for metadata (or externally
1756 	 * controlled, for example, OVS), then nothing can be changed.
1757 	 */
1758 	if (geneve->cfg.collect_md)
1759 		return -EOPNOTSUPP;
1760 
1761 	/* Start with the existing info. */
1762 	memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1763 	err = geneve_nl2info(tb, data, extack, &cfg, true);
1764 	if (err)
1765 		return err;
1766 
1767 	if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1768 		dst_cache_reset(&cfg.info.dst_cache);
1769 		geneve_link_config(dev, &cfg.info, tb);
1770 	}
1771 
1772 	geneve_quiesce(geneve, &gs4, &gs6);
1773 	memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1774 	geneve_unquiesce(geneve, gs4, gs6);
1775 
1776 	return 0;
1777 }
1778 
1779 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1780 {
1781 	struct geneve_dev *geneve = netdev_priv(dev);
1782 
1783 	list_del(&geneve->next);
1784 	unregister_netdevice_queue(dev, head);
1785 }
1786 
1787 static size_t geneve_get_size(const struct net_device *dev)
1788 {
1789 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1790 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1791 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1792 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1793 		nla_total_size(sizeof(__u8)) +	/* IFLA_GENEVE_DF */
1794 		nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1795 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1796 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1797 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1798 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1799 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1800 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1801 		nla_total_size(0) +	 /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1802 		0;
1803 }
1804 
1805 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1806 {
1807 	struct geneve_dev *geneve = netdev_priv(dev);
1808 	struct ip_tunnel_info *info = &geneve->cfg.info;
1809 	bool ttl_inherit = geneve->cfg.ttl_inherit;
1810 	bool metadata = geneve->cfg.collect_md;
1811 	__u8 tmp_vni[3];
1812 	__u32 vni;
1813 
1814 	tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1815 	vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1816 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1817 		goto nla_put_failure;
1818 
1819 	if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1820 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1821 				    info->key.u.ipv4.dst))
1822 			goto nla_put_failure;
1823 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1824 			       !!(info->key.tun_flags & TUNNEL_CSUM)))
1825 			goto nla_put_failure;
1826 
1827 #if IS_ENABLED(CONFIG_IPV6)
1828 	} else if (!metadata) {
1829 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1830 				     &info->key.u.ipv6.dst))
1831 			goto nla_put_failure;
1832 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1833 			       !(info->key.tun_flags & TUNNEL_CSUM)))
1834 			goto nla_put_failure;
1835 #endif
1836 	}
1837 
1838 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1839 	    nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1840 	    nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1841 		goto nla_put_failure;
1842 
1843 	if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1844 		goto nla_put_failure;
1845 
1846 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1847 		goto nla_put_failure;
1848 
1849 	if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1850 		goto nla_put_failure;
1851 
1852 #if IS_ENABLED(CONFIG_IPV6)
1853 	if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1854 		       !geneve->cfg.use_udp6_rx_checksums))
1855 		goto nla_put_failure;
1856 #endif
1857 
1858 	if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1859 		goto nla_put_failure;
1860 
1861 	if (geneve->cfg.inner_proto_inherit &&
1862 	    nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1863 		goto nla_put_failure;
1864 
1865 	return 0;
1866 
1867 nla_put_failure:
1868 	return -EMSGSIZE;
1869 }
1870 
1871 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1872 	.kind		= "geneve",
1873 	.maxtype	= IFLA_GENEVE_MAX,
1874 	.policy		= geneve_policy,
1875 	.priv_size	= sizeof(struct geneve_dev),
1876 	.setup		= geneve_setup,
1877 	.validate	= geneve_validate,
1878 	.newlink	= geneve_newlink,
1879 	.changelink	= geneve_changelink,
1880 	.dellink	= geneve_dellink,
1881 	.get_size	= geneve_get_size,
1882 	.fill_info	= geneve_fill_info,
1883 };
1884 
1885 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1886 					u8 name_assign_type, u16 dst_port)
1887 {
1888 	struct nlattr *tb[IFLA_MAX + 1];
1889 	struct net_device *dev;
1890 	LIST_HEAD(list_kill);
1891 	int err;
1892 	struct geneve_config cfg = {
1893 		.df = GENEVE_DF_UNSET,
1894 		.use_udp6_rx_checksums = true,
1895 		.ttl_inherit = false,
1896 		.collect_md = true,
1897 	};
1898 
1899 	memset(tb, 0, sizeof(tb));
1900 	dev = rtnl_create_link(net, name, name_assign_type,
1901 			       &geneve_link_ops, tb, NULL);
1902 	if (IS_ERR(dev))
1903 		return dev;
1904 
1905 	init_tnl_info(&cfg.info, dst_port);
1906 	err = geneve_configure(net, dev, NULL, &cfg);
1907 	if (err) {
1908 		free_netdev(dev);
1909 		return ERR_PTR(err);
1910 	}
1911 
1912 	/* openvswitch users expect packet sizes to be unrestricted,
1913 	 * so set the largest MTU we can.
1914 	 */
1915 	err = geneve_change_mtu(dev, IP_MAX_MTU);
1916 	if (err)
1917 		goto err;
1918 
1919 	err = rtnl_configure_link(dev, NULL, 0, NULL);
1920 	if (err < 0)
1921 		goto err;
1922 
1923 	return dev;
1924 err:
1925 	geneve_dellink(dev, &list_kill);
1926 	unregister_netdevice_many(&list_kill);
1927 	return ERR_PTR(err);
1928 }
1929 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1930 
1931 static int geneve_netdevice_event(struct notifier_block *unused,
1932 				  unsigned long event, void *ptr)
1933 {
1934 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1935 
1936 	if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1937 		geneve_offload_rx_ports(dev, true);
1938 	else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1939 		geneve_offload_rx_ports(dev, false);
1940 
1941 	return NOTIFY_DONE;
1942 }
1943 
1944 static struct notifier_block geneve_notifier_block __read_mostly = {
1945 	.notifier_call = geneve_netdevice_event,
1946 };
1947 
1948 static __net_init int geneve_init_net(struct net *net)
1949 {
1950 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1951 
1952 	INIT_LIST_HEAD(&gn->geneve_list);
1953 	INIT_LIST_HEAD(&gn->sock_list);
1954 	return 0;
1955 }
1956 
1957 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1958 {
1959 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1960 	struct geneve_dev *geneve, *next;
1961 	struct net_device *dev, *aux;
1962 
1963 	/* gather any geneve devices that were moved into this ns */
1964 	for_each_netdev_safe(net, dev, aux)
1965 		if (dev->rtnl_link_ops == &geneve_link_ops)
1966 			unregister_netdevice_queue(dev, head);
1967 
1968 	/* now gather any other geneve devices that were created in this ns */
1969 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1970 		/* If geneve->dev is in the same netns, it was already added
1971 		 * to the list by the previous loop.
1972 		 */
1973 		if (!net_eq(dev_net(geneve->dev), net))
1974 			unregister_netdevice_queue(geneve->dev, head);
1975 	}
1976 }
1977 
1978 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1979 {
1980 	struct net *net;
1981 	LIST_HEAD(list);
1982 
1983 	rtnl_lock();
1984 	list_for_each_entry(net, net_list, exit_list)
1985 		geneve_destroy_tunnels(net, &list);
1986 
1987 	/* unregister the devices gathered above */
1988 	unregister_netdevice_many(&list);
1989 	rtnl_unlock();
1990 
1991 	list_for_each_entry(net, net_list, exit_list) {
1992 		const struct geneve_net *gn = net_generic(net, geneve_net_id);
1993 
1994 		WARN_ON_ONCE(!list_empty(&gn->sock_list));
1995 	}
1996 }
1997 
1998 static struct pernet_operations geneve_net_ops = {
1999 	.init = geneve_init_net,
2000 	.exit_batch = geneve_exit_batch_net,
2001 	.id   = &geneve_net_id,
2002 	.size = sizeof(struct geneve_net),
2003 };
2004 
2005 static int __init geneve_init_module(void)
2006 {
2007 	int rc;
2008 
2009 	rc = register_pernet_subsys(&geneve_net_ops);
2010 	if (rc)
2011 		goto out1;
2012 
2013 	rc = register_netdevice_notifier(&geneve_notifier_block);
2014 	if (rc)
2015 		goto out2;
2016 
2017 	rc = rtnl_link_register(&geneve_link_ops);
2018 	if (rc)
2019 		goto out3;
2020 
2021 	return 0;
2022 out3:
2023 	unregister_netdevice_notifier(&geneve_notifier_block);
2024 out2:
2025 	unregister_pernet_subsys(&geneve_net_ops);
2026 out1:
2027 	return rc;
2028 }
2029 late_initcall(geneve_init_module);
2030 
2031 static void __exit geneve_cleanup_module(void)
2032 {
2033 	rtnl_link_unregister(&geneve_link_ops);
2034 	unregister_netdevice_notifier(&geneve_notifier_block);
2035 	unregister_pernet_subsys(&geneve_net_ops);
2036 }
2037 module_exit(geneve_cleanup_module);
2038 
2039 MODULE_LICENSE("GPL");
2040 MODULE_VERSION(GENEVE_NETDEV_VER);
2041 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2042 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2043 MODULE_ALIAS_RTNL_LINK("geneve");
2044