• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * GENEVE: Generic Network Virtualization Encapsulation
3  *
4  * Copyright (c) 2015 Red Hat, Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/etherdevice.h>
16 #include <linux/hash.h>
17 #include <net/dst_metadata.h>
18 #include <net/gro_cells.h>
19 #include <net/rtnetlink.h>
20 #include <net/geneve.h>
21 #include <net/protocol.h>
22 
23 #define GENEVE_NETDEV_VER	"0.6"
24 
25 #define GENEVE_UDP_PORT		6081
26 
27 #define GENEVE_N_VID		(1u << 24)
28 #define GENEVE_VID_MASK		(GENEVE_N_VID - 1)
29 
30 #define VNI_HASH_BITS		10
31 #define VNI_HASH_SIZE		(1<<VNI_HASH_BITS)
32 
33 static bool log_ecn_error = true;
34 module_param(log_ecn_error, bool, 0644);
35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
36 
37 #define GENEVE_VER 0
38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
39 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
40 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
41 
42 /* per-network namespace private data for this module */
43 struct geneve_net {
44 	struct list_head	geneve_list;
45 	struct list_head	sock_list;
46 };
47 
48 static unsigned int geneve_net_id;
49 
50 struct geneve_dev_node {
51 	struct hlist_node hlist;
52 	struct geneve_dev *geneve;
53 };
54 
55 /* Pseudo network device */
56 struct geneve_dev {
57 	struct geneve_dev_node hlist4;	/* vni hash table for IPv4 socket */
58 #if IS_ENABLED(CONFIG_IPV6)
59 	struct geneve_dev_node hlist6;	/* vni hash table for IPv6 socket */
60 #endif
61 	struct net	   *net;	/* netns for packet i/o */
62 	struct net_device  *dev;	/* netdev for geneve tunnel */
63 	struct ip_tunnel_info info;
64 	struct geneve_sock __rcu *sock4;	/* IPv4 socket used for geneve tunnel */
65 #if IS_ENABLED(CONFIG_IPV6)
66 	struct geneve_sock __rcu *sock6;	/* IPv6 socket used for geneve tunnel */
67 #endif
68 	struct list_head   next;	/* geneve's per namespace list */
69 	struct gro_cells   gro_cells;
70 	bool		   collect_md;
71 	bool		   use_udp6_rx_checksums;
72 };
73 
74 struct geneve_sock {
75 	bool			collect_md;
76 	struct list_head	list;
77 	struct socket		*sock;
78 	struct rcu_head		rcu;
79 	int			refcnt;
80 	struct hlist_head	vni_list[VNI_HASH_SIZE];
81 };
82 
geneve_net_vni_hash(u8 vni[3])83 static inline __u32 geneve_net_vni_hash(u8 vni[3])
84 {
85 	__u32 vnid;
86 
87 	vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
88 	return hash_32(vnid, VNI_HASH_BITS);
89 }
90 
vni_to_tunnel_id(const __u8 * vni)91 static __be64 vni_to_tunnel_id(const __u8 *vni)
92 {
93 #ifdef __BIG_ENDIAN
94 	return (vni[0] << 16) | (vni[1] << 8) | vni[2];
95 #else
96 	return (__force __be64)(((__force u64)vni[0] << 40) |
97 				((__force u64)vni[1] << 48) |
98 				((__force u64)vni[2] << 56));
99 #endif
100 }
101 
102 /* Convert 64 bit tunnel ID to 24 bit VNI. */
tunnel_id_to_vni(__be64 tun_id,__u8 * vni)103 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
104 {
105 #ifdef __BIG_ENDIAN
106 	vni[0] = (__force __u8)(tun_id >> 16);
107 	vni[1] = (__force __u8)(tun_id >> 8);
108 	vni[2] = (__force __u8)tun_id;
109 #else
110 	vni[0] = (__force __u8)((__force u64)tun_id >> 40);
111 	vni[1] = (__force __u8)((__force u64)tun_id >> 48);
112 	vni[2] = (__force __u8)((__force u64)tun_id >> 56);
113 #endif
114 }
115 
eq_tun_id_and_vni(u8 * tun_id,u8 * vni)116 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
117 {
118 	return !memcmp(vni, &tun_id[5], 3);
119 }
120 
geneve_get_sk_family(struct geneve_sock * gs)121 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
122 {
123 	return gs->sock->sk->sk_family;
124 }
125 
geneve_lookup(struct geneve_sock * gs,__be32 addr,u8 vni[])126 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
127 					__be32 addr, u8 vni[])
128 {
129 	struct hlist_head *vni_list_head;
130 	struct geneve_dev_node *node;
131 	__u32 hash;
132 
133 	/* Find the device for this VNI */
134 	hash = geneve_net_vni_hash(vni);
135 	vni_list_head = &gs->vni_list[hash];
136 	hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
137 		if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
138 		    addr == node->geneve->info.key.u.ipv4.dst)
139 			return node->geneve;
140 	}
141 	return NULL;
142 }
143 
144 #if IS_ENABLED(CONFIG_IPV6)
geneve6_lookup(struct geneve_sock * gs,struct in6_addr addr6,u8 vni[])145 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
146 					 struct in6_addr addr6, u8 vni[])
147 {
148 	struct hlist_head *vni_list_head;
149 	struct geneve_dev_node *node;
150 	__u32 hash;
151 
152 	/* Find the device for this VNI */
153 	hash = geneve_net_vni_hash(vni);
154 	vni_list_head = &gs->vni_list[hash];
155 	hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
156 		if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
157 		    ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
158 			return node->geneve;
159 	}
160 	return NULL;
161 }
162 #endif
163 
geneve_hdr(const struct sk_buff * skb)164 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
165 {
166 	return (struct genevehdr *)(udp_hdr(skb) + 1);
167 }
168 
geneve_lookup_skb(struct geneve_sock * gs,struct sk_buff * skb)169 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
170 					    struct sk_buff *skb)
171 {
172 	static u8 zero_vni[3];
173 	u8 *vni;
174 
175 	if (geneve_get_sk_family(gs) == AF_INET) {
176 		struct iphdr *iph;
177 		__be32 addr;
178 
179 		iph = ip_hdr(skb); /* outer IP header... */
180 
181 		if (gs->collect_md) {
182 			vni = zero_vni;
183 			addr = 0;
184 		} else {
185 			vni = geneve_hdr(skb)->vni;
186 			addr = iph->saddr;
187 		}
188 
189 		return geneve_lookup(gs, addr, vni);
190 #if IS_ENABLED(CONFIG_IPV6)
191 	} else if (geneve_get_sk_family(gs) == AF_INET6) {
192 		static struct in6_addr zero_addr6;
193 		struct ipv6hdr *ip6h;
194 		struct in6_addr addr6;
195 
196 		ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
197 
198 		if (gs->collect_md) {
199 			vni = zero_vni;
200 			addr6 = zero_addr6;
201 		} else {
202 			vni = geneve_hdr(skb)->vni;
203 			addr6 = ip6h->saddr;
204 		}
205 
206 		return geneve6_lookup(gs, addr6, vni);
207 #endif
208 	}
209 	return NULL;
210 }
211 
212 /* geneve receive/decap routine */
geneve_rx(struct geneve_dev * geneve,struct geneve_sock * gs,struct sk_buff * skb)213 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
214 		      struct sk_buff *skb)
215 {
216 	struct genevehdr *gnvh = geneve_hdr(skb);
217 	struct metadata_dst *tun_dst = NULL;
218 	struct pcpu_sw_netstats *stats;
219 	unsigned int len;
220 	int err = 0;
221 	void *oiph;
222 
223 	if (ip_tunnel_collect_metadata() || gs->collect_md) {
224 		__be16 flags;
225 
226 		flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
227 			(gnvh->oam ? TUNNEL_OAM : 0) |
228 			(gnvh->critical ? TUNNEL_CRIT_OPT : 0);
229 
230 		tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
231 					 vni_to_tunnel_id(gnvh->vni),
232 					 gnvh->opt_len * 4);
233 		if (!tun_dst) {
234 			geneve->dev->stats.rx_dropped++;
235 			goto drop;
236 		}
237 		/* Update tunnel dst according to Geneve options. */
238 		ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
239 					gnvh->options, gnvh->opt_len * 4,
240 					TUNNEL_GENEVE_OPT);
241 	} else {
242 		/* Drop packets w/ critical options,
243 		 * since we don't support any...
244 		 */
245 		if (gnvh->critical) {
246 			geneve->dev->stats.rx_frame_errors++;
247 			geneve->dev->stats.rx_errors++;
248 			goto drop;
249 		}
250 	}
251 
252 	skb_reset_mac_header(skb);
253 	skb->protocol = eth_type_trans(skb, geneve->dev);
254 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
255 
256 	if (tun_dst)
257 		skb_dst_set(skb, &tun_dst->dst);
258 
259 	/* Ignore packet loops (and multicast echo) */
260 	if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
261 		geneve->dev->stats.rx_errors++;
262 		goto drop;
263 	}
264 
265 	oiph = skb_network_header(skb);
266 	skb_reset_network_header(skb);
267 
268 	if (geneve_get_sk_family(gs) == AF_INET)
269 		err = IP_ECN_decapsulate(oiph, skb);
270 #if IS_ENABLED(CONFIG_IPV6)
271 	else
272 		err = IP6_ECN_decapsulate(oiph, skb);
273 #endif
274 
275 	if (unlikely(err)) {
276 		if (log_ecn_error) {
277 			if (geneve_get_sk_family(gs) == AF_INET)
278 				net_info_ratelimited("non-ECT from %pI4 "
279 						     "with TOS=%#x\n",
280 						     &((struct iphdr *)oiph)->saddr,
281 						     ((struct iphdr *)oiph)->tos);
282 #if IS_ENABLED(CONFIG_IPV6)
283 			else
284 				net_info_ratelimited("non-ECT from %pI6\n",
285 						     &((struct ipv6hdr *)oiph)->saddr);
286 #endif
287 		}
288 		if (err > 1) {
289 			++geneve->dev->stats.rx_frame_errors;
290 			++geneve->dev->stats.rx_errors;
291 			goto drop;
292 		}
293 	}
294 
295 	len = skb->len;
296 	err = gro_cells_receive(&geneve->gro_cells, skb);
297 	if (likely(err == NET_RX_SUCCESS)) {
298 		stats = this_cpu_ptr(geneve->dev->tstats);
299 		u64_stats_update_begin(&stats->syncp);
300 		stats->rx_packets++;
301 		stats->rx_bytes += len;
302 		u64_stats_update_end(&stats->syncp);
303 	}
304 	return;
305 drop:
306 	/* Consume bad packet */
307 	kfree_skb(skb);
308 }
309 
310 /* Setup stats when device is created */
geneve_init(struct net_device * dev)311 static int geneve_init(struct net_device *dev)
312 {
313 	struct geneve_dev *geneve = netdev_priv(dev);
314 	int err;
315 
316 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
317 	if (!dev->tstats)
318 		return -ENOMEM;
319 
320 	err = gro_cells_init(&geneve->gro_cells, dev);
321 	if (err) {
322 		free_percpu(dev->tstats);
323 		return err;
324 	}
325 
326 	err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
327 	if (err) {
328 		free_percpu(dev->tstats);
329 		gro_cells_destroy(&geneve->gro_cells);
330 		return err;
331 	}
332 	return 0;
333 }
334 
geneve_uninit(struct net_device * dev)335 static void geneve_uninit(struct net_device *dev)
336 {
337 	struct geneve_dev *geneve = netdev_priv(dev);
338 
339 	dst_cache_destroy(&geneve->info.dst_cache);
340 	gro_cells_destroy(&geneve->gro_cells);
341 	free_percpu(dev->tstats);
342 }
343 
344 /* Callback from net/ipv4/udp.c to receive packets */
geneve_udp_encap_recv(struct sock * sk,struct sk_buff * skb)345 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
346 {
347 	struct genevehdr *geneveh;
348 	struct geneve_dev *geneve;
349 	struct geneve_sock *gs;
350 	int opts_len;
351 
352 	/* Need UDP and Geneve header to be present */
353 	if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
354 		goto drop;
355 
356 	/* Return packets with reserved bits set */
357 	geneveh = geneve_hdr(skb);
358 	if (unlikely(geneveh->ver != GENEVE_VER))
359 		goto drop;
360 
361 	if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
362 		goto drop;
363 
364 	gs = rcu_dereference_sk_user_data(sk);
365 	if (!gs)
366 		goto drop;
367 
368 	geneve = geneve_lookup_skb(gs, skb);
369 	if (!geneve)
370 		goto drop;
371 
372 	opts_len = geneveh->opt_len * 4;
373 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
374 				 htons(ETH_P_TEB),
375 				 !net_eq(geneve->net, dev_net(geneve->dev)))) {
376 		geneve->dev->stats.rx_dropped++;
377 		goto drop;
378 	}
379 
380 	geneve_rx(geneve, gs, skb);
381 	return 0;
382 
383 drop:
384 	/* Consume bad packet */
385 	kfree_skb(skb);
386 	return 0;
387 }
388 
geneve_create_sock(struct net * net,bool ipv6,__be16 port,bool ipv6_rx_csum)389 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
390 					 __be16 port, bool ipv6_rx_csum)
391 {
392 	struct socket *sock;
393 	struct udp_port_cfg udp_conf;
394 	int err;
395 
396 	memset(&udp_conf, 0, sizeof(udp_conf));
397 
398 	if (ipv6) {
399 		udp_conf.family = AF_INET6;
400 		udp_conf.ipv6_v6only = 1;
401 		udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
402 	} else {
403 		udp_conf.family = AF_INET;
404 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
405 	}
406 
407 	udp_conf.local_udp_port = port;
408 
409 	/* Open UDP socket */
410 	err = udp_sock_create(net, &udp_conf, &sock);
411 	if (err < 0)
412 		return ERR_PTR(err);
413 
414 	return sock;
415 }
416 
geneve_hlen(struct genevehdr * gh)417 static int geneve_hlen(struct genevehdr *gh)
418 {
419 	return sizeof(*gh) + gh->opt_len * 4;
420 }
421 
geneve_gro_receive(struct sock * sk,struct list_head * head,struct sk_buff * skb)422 static struct sk_buff *geneve_gro_receive(struct sock *sk,
423 					  struct list_head *head,
424 					  struct sk_buff *skb)
425 {
426 	struct sk_buff *pp = NULL;
427 	struct sk_buff *p;
428 	struct genevehdr *gh, *gh2;
429 	unsigned int hlen, gh_len, off_gnv;
430 	const struct packet_offload *ptype;
431 	__be16 type;
432 	int flush = 1;
433 
434 	off_gnv = skb_gro_offset(skb);
435 	hlen = off_gnv + sizeof(*gh);
436 	gh = skb_gro_header_fast(skb, off_gnv);
437 	if (skb_gro_header_hard(skb, hlen)) {
438 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
439 		if (unlikely(!gh))
440 			goto out;
441 	}
442 
443 	if (gh->ver != GENEVE_VER || gh->oam)
444 		goto out;
445 	gh_len = geneve_hlen(gh);
446 
447 	hlen = off_gnv + gh_len;
448 	if (skb_gro_header_hard(skb, hlen)) {
449 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
450 		if (unlikely(!gh))
451 			goto out;
452 	}
453 
454 	list_for_each_entry(p, head, list) {
455 		if (!NAPI_GRO_CB(p)->same_flow)
456 			continue;
457 
458 		gh2 = (struct genevehdr *)(p->data + off_gnv);
459 		if (gh->opt_len != gh2->opt_len ||
460 		    memcmp(gh, gh2, gh_len)) {
461 			NAPI_GRO_CB(p)->same_flow = 0;
462 			continue;
463 		}
464 	}
465 
466 	type = gh->proto_type;
467 
468 	rcu_read_lock();
469 	ptype = gro_find_receive_by_type(type);
470 	if (!ptype)
471 		goto out_unlock;
472 
473 	skb_gro_pull(skb, gh_len);
474 	skb_gro_postpull_rcsum(skb, gh, gh_len);
475 	pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
476 	flush = 0;
477 
478 out_unlock:
479 	rcu_read_unlock();
480 out:
481 	skb_gro_flush_final(skb, pp, flush);
482 
483 	return pp;
484 }
485 
geneve_gro_complete(struct sock * sk,struct sk_buff * skb,int nhoff)486 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
487 			       int nhoff)
488 {
489 	struct genevehdr *gh;
490 	struct packet_offload *ptype;
491 	__be16 type;
492 	int gh_len;
493 	int err = -ENOSYS;
494 
495 	gh = (struct genevehdr *)(skb->data + nhoff);
496 	gh_len = geneve_hlen(gh);
497 	type = gh->proto_type;
498 
499 	rcu_read_lock();
500 	ptype = gro_find_complete_by_type(type);
501 	if (ptype)
502 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
503 
504 	rcu_read_unlock();
505 
506 	skb_set_inner_mac_header(skb, nhoff + gh_len);
507 
508 	return err;
509 }
510 
511 /* Create new listen socket if needed */
geneve_socket_create(struct net * net,__be16 port,bool ipv6,bool ipv6_rx_csum)512 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
513 						bool ipv6, bool ipv6_rx_csum)
514 {
515 	struct geneve_net *gn = net_generic(net, geneve_net_id);
516 	struct geneve_sock *gs;
517 	struct socket *sock;
518 	struct udp_tunnel_sock_cfg tunnel_cfg;
519 	int h;
520 
521 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
522 	if (!gs)
523 		return ERR_PTR(-ENOMEM);
524 
525 	sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
526 	if (IS_ERR(sock)) {
527 		kfree(gs);
528 		return ERR_CAST(sock);
529 	}
530 
531 	gs->sock = sock;
532 	gs->refcnt = 1;
533 	for (h = 0; h < VNI_HASH_SIZE; ++h)
534 		INIT_HLIST_HEAD(&gs->vni_list[h]);
535 
536 	/* Initialize the geneve udp offloads structure */
537 	udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
538 
539 	/* Mark socket as an encapsulation socket */
540 	memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
541 	tunnel_cfg.sk_user_data = gs;
542 	tunnel_cfg.encap_type = 1;
543 	tunnel_cfg.gro_receive = geneve_gro_receive;
544 	tunnel_cfg.gro_complete = geneve_gro_complete;
545 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
546 	tunnel_cfg.encap_destroy = NULL;
547 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
548 	list_add(&gs->list, &gn->sock_list);
549 	return gs;
550 }
551 
__geneve_sock_release(struct geneve_sock * gs)552 static void __geneve_sock_release(struct geneve_sock *gs)
553 {
554 	if (!gs || --gs->refcnt)
555 		return;
556 
557 	list_del(&gs->list);
558 	udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
559 	udp_tunnel_sock_release(gs->sock);
560 	kfree_rcu(gs, rcu);
561 }
562 
geneve_sock_release(struct geneve_dev * geneve)563 static void geneve_sock_release(struct geneve_dev *geneve)
564 {
565 	struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
566 #if IS_ENABLED(CONFIG_IPV6)
567 	struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
568 
569 	rcu_assign_pointer(geneve->sock6, NULL);
570 #endif
571 
572 	rcu_assign_pointer(geneve->sock4, NULL);
573 	synchronize_net();
574 
575 	__geneve_sock_release(gs4);
576 #if IS_ENABLED(CONFIG_IPV6)
577 	__geneve_sock_release(gs6);
578 #endif
579 }
580 
geneve_find_sock(struct geneve_net * gn,sa_family_t family,__be16 dst_port)581 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
582 					    sa_family_t family,
583 					    __be16 dst_port)
584 {
585 	struct geneve_sock *gs;
586 
587 	list_for_each_entry(gs, &gn->sock_list, list) {
588 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
589 		    geneve_get_sk_family(gs) == family) {
590 			return gs;
591 		}
592 	}
593 	return NULL;
594 }
595 
geneve_sock_add(struct geneve_dev * geneve,bool ipv6)596 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
597 {
598 	struct net *net = geneve->net;
599 	struct geneve_net *gn = net_generic(net, geneve_net_id);
600 	struct geneve_dev_node *node;
601 	struct geneve_sock *gs;
602 	__u8 vni[3];
603 	__u32 hash;
604 
605 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
606 	if (gs) {
607 		gs->refcnt++;
608 		goto out;
609 	}
610 
611 	gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
612 				  geneve->use_udp6_rx_checksums);
613 	if (IS_ERR(gs))
614 		return PTR_ERR(gs);
615 
616 out:
617 	gs->collect_md = geneve->collect_md;
618 #if IS_ENABLED(CONFIG_IPV6)
619 	if (ipv6) {
620 		rcu_assign_pointer(geneve->sock6, gs);
621 		node = &geneve->hlist6;
622 	} else
623 #endif
624 	{
625 		rcu_assign_pointer(geneve->sock4, gs);
626 		node = &geneve->hlist4;
627 	}
628 	node->geneve = geneve;
629 
630 	tunnel_id_to_vni(geneve->info.key.tun_id, vni);
631 	hash = geneve_net_vni_hash(vni);
632 	hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
633 	return 0;
634 }
635 
geneve_open(struct net_device * dev)636 static int geneve_open(struct net_device *dev)
637 {
638 	struct geneve_dev *geneve = netdev_priv(dev);
639 	bool metadata = geneve->collect_md;
640 	bool ipv4, ipv6;
641 	int ret = 0;
642 
643 	ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
644 	ipv4 = !ipv6 || metadata;
645 #if IS_ENABLED(CONFIG_IPV6)
646 	if (ipv6) {
647 		ret = geneve_sock_add(geneve, true);
648 		if (ret < 0 && ret != -EAFNOSUPPORT)
649 			ipv4 = false;
650 	}
651 #endif
652 	if (ipv4)
653 		ret = geneve_sock_add(geneve, false);
654 	if (ret < 0)
655 		geneve_sock_release(geneve);
656 
657 	return ret;
658 }
659 
geneve_stop(struct net_device * dev)660 static int geneve_stop(struct net_device *dev)
661 {
662 	struct geneve_dev *geneve = netdev_priv(dev);
663 
664 	hlist_del_init_rcu(&geneve->hlist4.hlist);
665 #if IS_ENABLED(CONFIG_IPV6)
666 	hlist_del_init_rcu(&geneve->hlist6.hlist);
667 #endif
668 	geneve_sock_release(geneve);
669 	return 0;
670 }
671 
geneve_build_header(struct genevehdr * geneveh,const struct ip_tunnel_info * info)672 static void geneve_build_header(struct genevehdr *geneveh,
673 				const struct ip_tunnel_info *info)
674 {
675 	geneveh->ver = GENEVE_VER;
676 	geneveh->opt_len = info->options_len / 4;
677 	geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
678 	geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
679 	geneveh->rsvd1 = 0;
680 	tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
681 	geneveh->proto_type = htons(ETH_P_TEB);
682 	geneveh->rsvd2 = 0;
683 
684 	if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
685 		ip_tunnel_info_opts_get(geneveh->options, info);
686 }
687 
geneve_build_skb(struct dst_entry * dst,struct sk_buff * skb,const struct ip_tunnel_info * info,bool xnet,int ip_hdr_len)688 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
689 			    const struct ip_tunnel_info *info,
690 			    bool xnet, int ip_hdr_len)
691 {
692 	bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
693 	struct genevehdr *gnvh;
694 	int min_headroom;
695 	int err;
696 
697 	skb_reset_mac_header(skb);
698 	skb_scrub_packet(skb, xnet);
699 
700 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
701 		       GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
702 	err = skb_cow_head(skb, min_headroom);
703 	if (unlikely(err))
704 		goto free_dst;
705 
706 	err = udp_tunnel_handle_offloads(skb, udp_sum);
707 	if (err)
708 		goto free_dst;
709 
710 	gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
711 	geneve_build_header(gnvh, info);
712 	skb_set_inner_protocol(skb, htons(ETH_P_TEB));
713 	return 0;
714 
715 free_dst:
716 	dst_release(dst);
717 	return err;
718 }
719 
geneve_get_v4_rt(struct sk_buff * skb,struct net_device * dev,struct geneve_sock * gs4,struct flowi4 * fl4,const struct ip_tunnel_info * info,__be16 dport,__be16 sport)720 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
721 				       struct net_device *dev,
722 				       struct geneve_sock *gs4,
723 				       struct flowi4 *fl4,
724 				       const struct ip_tunnel_info *info,
725 				       __be16 dport, __be16 sport)
726 {
727 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
728 	struct geneve_dev *geneve = netdev_priv(dev);
729 	struct dst_cache *dst_cache;
730 	struct rtable *rt = NULL;
731 	__u8 tos;
732 
733 	if (!gs4)
734 		return ERR_PTR(-EIO);
735 
736 	memset(fl4, 0, sizeof(*fl4));
737 	fl4->flowi4_mark = skb->mark;
738 	fl4->flowi4_proto = IPPROTO_UDP;
739 	fl4->daddr = info->key.u.ipv4.dst;
740 	fl4->saddr = info->key.u.ipv4.src;
741 	fl4->fl4_dport = dport;
742 	fl4->fl4_sport = sport;
743 
744 	tos = info->key.tos;
745 	if ((tos == 1) && !geneve->collect_md) {
746 		tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
747 		use_cache = false;
748 	}
749 	fl4->flowi4_tos = RT_TOS(tos);
750 
751 	dst_cache = (struct dst_cache *)&info->dst_cache;
752 	if (use_cache) {
753 		rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
754 		if (rt)
755 			return rt;
756 	}
757 	rt = ip_route_output_key(geneve->net, fl4);
758 	if (IS_ERR(rt)) {
759 		netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
760 		return ERR_PTR(-ENETUNREACH);
761 	}
762 	if (rt->dst.dev == dev) { /* is this necessary? */
763 		netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
764 		ip_rt_put(rt);
765 		return ERR_PTR(-ELOOP);
766 	}
767 	if (use_cache)
768 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
769 	return rt;
770 }
771 
772 #if IS_ENABLED(CONFIG_IPV6)
geneve_get_v6_dst(struct sk_buff * skb,struct net_device * dev,struct geneve_sock * gs6,struct flowi6 * fl6,const struct ip_tunnel_info * info,__be16 dport,__be16 sport)773 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
774 					   struct net_device *dev,
775 					   struct geneve_sock *gs6,
776 					   struct flowi6 *fl6,
777 					   const struct ip_tunnel_info *info,
778 					   __be16 dport, __be16 sport)
779 {
780 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
781 	struct geneve_dev *geneve = netdev_priv(dev);
782 	struct dst_entry *dst = NULL;
783 	struct dst_cache *dst_cache;
784 	__u8 prio;
785 
786 	if (!gs6)
787 		return ERR_PTR(-EIO);
788 
789 	memset(fl6, 0, sizeof(*fl6));
790 	fl6->flowi6_mark = skb->mark;
791 	fl6->flowi6_proto = IPPROTO_UDP;
792 	fl6->daddr = info->key.u.ipv6.dst;
793 	fl6->saddr = info->key.u.ipv6.src;
794 	fl6->fl6_dport = dport;
795 	fl6->fl6_sport = sport;
796 
797 	prio = info->key.tos;
798 	if ((prio == 1) && !geneve->collect_md) {
799 		prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
800 		use_cache = false;
801 	}
802 
803 	fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
804 					   info->key.label);
805 	dst_cache = (struct dst_cache *)&info->dst_cache;
806 	if (use_cache) {
807 		dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
808 		if (dst)
809 			return dst;
810 	}
811 	dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
812 					      NULL);
813 	if (IS_ERR(dst)) {
814 		netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
815 		return ERR_PTR(-ENETUNREACH);
816 	}
817 	if (dst->dev == dev) { /* is this necessary? */
818 		netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
819 		dst_release(dst);
820 		return ERR_PTR(-ELOOP);
821 	}
822 
823 	if (use_cache)
824 		dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
825 	return dst;
826 }
827 #endif
828 
geneve_xmit_skb(struct sk_buff * skb,struct net_device * dev,struct geneve_dev * geneve,const struct ip_tunnel_info * info)829 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
830 			   struct geneve_dev *geneve,
831 			   const struct ip_tunnel_info *info)
832 {
833 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
834 	struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
835 	const struct ip_tunnel_key *key = &info->key;
836 	struct rtable *rt;
837 	struct flowi4 fl4;
838 	__u8 tos, ttl;
839 	__be16 sport;
840 	__be16 df;
841 	int err;
842 
843 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
844 	rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
845 			      geneve->info.key.tp_dst, sport);
846 	if (IS_ERR(rt))
847 		return PTR_ERR(rt);
848 
849 	skb_tunnel_check_pmtu(skb, &rt->dst,
850 			      GENEVE_IPV4_HLEN + info->options_len);
851 
852 	if (geneve->collect_md) {
853 		tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
854 		ttl = key->ttl;
855 	} else {
856 		tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
857 		ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
858 	}
859 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
860 
861 	err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
862 	if (unlikely(err))
863 		return err;
864 
865 	udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
866 			    tos, ttl, df, sport, geneve->info.key.tp_dst,
867 			    !net_eq(geneve->net, dev_net(geneve->dev)),
868 			    !(info->key.tun_flags & TUNNEL_CSUM));
869 	return 0;
870 }
871 
872 #if IS_ENABLED(CONFIG_IPV6)
geneve6_xmit_skb(struct sk_buff * skb,struct net_device * dev,struct geneve_dev * geneve,const struct ip_tunnel_info * info)873 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
874 			    struct geneve_dev *geneve,
875 			    const struct ip_tunnel_info *info)
876 {
877 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
878 	struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
879 	const struct ip_tunnel_key *key = &info->key;
880 	struct dst_entry *dst = NULL;
881 	struct flowi6 fl6;
882 	__u8 prio, ttl;
883 	__be16 sport;
884 	int err;
885 
886 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
887 	dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
888 				geneve->info.key.tp_dst, sport);
889 	if (IS_ERR(dst))
890 		return PTR_ERR(dst);
891 
892 	skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
893 
894 	if (geneve->collect_md) {
895 		prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
896 		ttl = key->ttl;
897 	} else {
898 		prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
899 					   ip_hdr(skb), skb);
900 		ttl = key->ttl ? : ip6_dst_hoplimit(dst);
901 	}
902 	err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
903 	if (unlikely(err))
904 		return err;
905 
906 	udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
907 			     &fl6.saddr, &fl6.daddr, prio, ttl,
908 			     info->key.label, sport, geneve->info.key.tp_dst,
909 			     !(info->key.tun_flags & TUNNEL_CSUM));
910 	return 0;
911 }
912 #endif
913 
geneve_xmit(struct sk_buff * skb,struct net_device * dev)914 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
915 {
916 	struct geneve_dev *geneve = netdev_priv(dev);
917 	struct ip_tunnel_info *info = NULL;
918 	int err;
919 
920 	if (geneve->collect_md) {
921 		info = skb_tunnel_info(skb);
922 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
923 			netdev_dbg(dev, "no tunnel metadata\n");
924 			dev_kfree_skb(skb);
925 			dev->stats.tx_dropped++;
926 			return NETDEV_TX_OK;
927 		}
928 	} else {
929 		info = &geneve->info;
930 	}
931 
932 	rcu_read_lock();
933 #if IS_ENABLED(CONFIG_IPV6)
934 	if (info->mode & IP_TUNNEL_INFO_IPV6)
935 		err = geneve6_xmit_skb(skb, dev, geneve, info);
936 	else
937 #endif
938 		err = geneve_xmit_skb(skb, dev, geneve, info);
939 	rcu_read_unlock();
940 
941 	if (likely(!err))
942 		return NETDEV_TX_OK;
943 
944 	dev_kfree_skb(skb);
945 
946 	if (err == -ELOOP)
947 		dev->stats.collisions++;
948 	else if (err == -ENETUNREACH)
949 		dev->stats.tx_carrier_errors++;
950 
951 	dev->stats.tx_errors++;
952 	return NETDEV_TX_OK;
953 }
954 
geneve_change_mtu(struct net_device * dev,int new_mtu)955 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
956 {
957 	if (new_mtu > dev->max_mtu)
958 		new_mtu = dev->max_mtu;
959 	else if (new_mtu < dev->min_mtu)
960 		new_mtu = dev->min_mtu;
961 
962 	dev->mtu = new_mtu;
963 	return 0;
964 }
965 
geneve_fill_metadata_dst(struct net_device * dev,struct sk_buff * skb)966 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
967 {
968 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
969 	struct geneve_dev *geneve = netdev_priv(dev);
970 	__be16 sport;
971 
972 	if (ip_tunnel_info_af(info) == AF_INET) {
973 		struct rtable *rt;
974 		struct flowi4 fl4;
975 
976 		struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
977 		sport = udp_flow_src_port(geneve->net, skb,
978 					  1, USHRT_MAX, true);
979 
980 		rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
981 				      geneve->info.key.tp_dst, sport);
982 		if (IS_ERR(rt))
983 			return PTR_ERR(rt);
984 
985 		ip_rt_put(rt);
986 		info->key.u.ipv4.src = fl4.saddr;
987 #if IS_ENABLED(CONFIG_IPV6)
988 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
989 		struct dst_entry *dst;
990 		struct flowi6 fl6;
991 
992 		struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
993 		sport = udp_flow_src_port(geneve->net, skb,
994 					  1, USHRT_MAX, true);
995 
996 		dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
997 					geneve->info.key.tp_dst, sport);
998 		if (IS_ERR(dst))
999 			return PTR_ERR(dst);
1000 
1001 		dst_release(dst);
1002 		info->key.u.ipv6.src = fl6.saddr;
1003 #endif
1004 	} else {
1005 		return -EINVAL;
1006 	}
1007 
1008 	info->key.tp_src = sport;
1009 	info->key.tp_dst = geneve->info.key.tp_dst;
1010 	return 0;
1011 }
1012 
1013 static const struct net_device_ops geneve_netdev_ops = {
1014 	.ndo_init		= geneve_init,
1015 	.ndo_uninit		= geneve_uninit,
1016 	.ndo_open		= geneve_open,
1017 	.ndo_stop		= geneve_stop,
1018 	.ndo_start_xmit		= geneve_xmit,
1019 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1020 	.ndo_change_mtu		= geneve_change_mtu,
1021 	.ndo_validate_addr	= eth_validate_addr,
1022 	.ndo_set_mac_address	= eth_mac_addr,
1023 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1024 };
1025 
geneve_get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * drvinfo)1026 static void geneve_get_drvinfo(struct net_device *dev,
1027 			       struct ethtool_drvinfo *drvinfo)
1028 {
1029 	strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1030 	strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1031 }
1032 
1033 static const struct ethtool_ops geneve_ethtool_ops = {
1034 	.get_drvinfo	= geneve_get_drvinfo,
1035 	.get_link	= ethtool_op_get_link,
1036 };
1037 
1038 /* Info for udev, that this is a virtual tunnel endpoint */
1039 static struct device_type geneve_type = {
1040 	.name = "geneve",
1041 };
1042 
1043 /* Calls the ndo_udp_tunnel_add of the caller in order to
1044  * supply the listening GENEVE udp ports. Callers are expected
1045  * to implement the ndo_udp_tunnel_add.
1046  */
geneve_offload_rx_ports(struct net_device * dev,bool push)1047 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1048 {
1049 	struct net *net = dev_net(dev);
1050 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1051 	struct geneve_sock *gs;
1052 
1053 	rcu_read_lock();
1054 	list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1055 		if (push) {
1056 			udp_tunnel_push_rx_port(dev, gs->sock,
1057 						UDP_TUNNEL_TYPE_GENEVE);
1058 		} else {
1059 			udp_tunnel_drop_rx_port(dev, gs->sock,
1060 						UDP_TUNNEL_TYPE_GENEVE);
1061 		}
1062 	}
1063 	rcu_read_unlock();
1064 }
1065 
1066 /* Initialize the device structure. */
geneve_setup(struct net_device * dev)1067 static void geneve_setup(struct net_device *dev)
1068 {
1069 	ether_setup(dev);
1070 
1071 	dev->netdev_ops = &geneve_netdev_ops;
1072 	dev->ethtool_ops = &geneve_ethtool_ops;
1073 	dev->needs_free_netdev = true;
1074 
1075 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1076 
1077 	dev->features    |= NETIF_F_LLTX;
1078 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1079 	dev->features    |= NETIF_F_RXCSUM;
1080 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1081 
1082 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1083 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1084 
1085 	/* MTU range: 68 - (something less than 65535) */
1086 	dev->min_mtu = ETH_MIN_MTU;
1087 	/* The max_mtu calculation does not take account of GENEVE
1088 	 * options, to avoid excluding potentially valid
1089 	 * configurations. This will be further reduced by IPvX hdr size.
1090 	 */
1091 	dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1092 
1093 	netif_keep_dst(dev);
1094 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1095 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1096 	eth_hw_addr_random(dev);
1097 }
1098 
1099 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1100 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1101 	[IFLA_GENEVE_REMOTE]		= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1102 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1103 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1104 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1105 	[IFLA_GENEVE_LABEL]		= { .type = NLA_U32 },
1106 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1107 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1108 	[IFLA_GENEVE_UDP_CSUM]		= { .type = NLA_U8 },
1109 	[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]	= { .type = NLA_U8 },
1110 	[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]	= { .type = NLA_U8 },
1111 };
1112 
geneve_validate(struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack)1113 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1114 			   struct netlink_ext_ack *extack)
1115 {
1116 	if (tb[IFLA_ADDRESS]) {
1117 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1118 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1119 					    "Provided link layer address is not Ethernet");
1120 			return -EINVAL;
1121 		}
1122 
1123 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1124 			NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1125 					    "Provided Ethernet address is not unicast");
1126 			return -EADDRNOTAVAIL;
1127 		}
1128 	}
1129 
1130 	if (!data) {
1131 		NL_SET_ERR_MSG(extack,
1132 			       "Not enough attributes provided to perform the operation");
1133 		return -EINVAL;
1134 	}
1135 
1136 	if (data[IFLA_GENEVE_ID]) {
1137 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1138 
1139 		if (vni >= GENEVE_N_VID) {
1140 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1141 					    "Geneve ID must be lower than 16777216");
1142 			return -ERANGE;
1143 		}
1144 	}
1145 
1146 	return 0;
1147 }
1148 
geneve_find_dev(struct geneve_net * gn,const struct ip_tunnel_info * info,bool * tun_on_same_port,bool * tun_collect_md)1149 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1150 					  const struct ip_tunnel_info *info,
1151 					  bool *tun_on_same_port,
1152 					  bool *tun_collect_md)
1153 {
1154 	struct geneve_dev *geneve, *t = NULL;
1155 
1156 	*tun_on_same_port = false;
1157 	*tun_collect_md = false;
1158 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1159 		if (info->key.tp_dst == geneve->info.key.tp_dst) {
1160 			*tun_collect_md = geneve->collect_md;
1161 			*tun_on_same_port = true;
1162 		}
1163 		if (info->key.tun_id == geneve->info.key.tun_id &&
1164 		    info->key.tp_dst == geneve->info.key.tp_dst &&
1165 		    !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1166 			t = geneve;
1167 	}
1168 	return t;
1169 }
1170 
is_tnl_info_zero(const struct ip_tunnel_info * info)1171 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1172 {
1173 	return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1174 		 info->key.ttl || info->key.label || info->key.tp_src ||
1175 		 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1176 }
1177 
geneve_dst_addr_equal(struct ip_tunnel_info * a,struct ip_tunnel_info * b)1178 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1179 				  struct ip_tunnel_info *b)
1180 {
1181 	if (ip_tunnel_info_af(a) == AF_INET)
1182 		return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1183 	else
1184 		return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1185 }
1186 
geneve_configure(struct net * net,struct net_device * dev,struct netlink_ext_ack * extack,const struct ip_tunnel_info * info,bool metadata,bool ipv6_rx_csum)1187 static int geneve_configure(struct net *net, struct net_device *dev,
1188 			    struct netlink_ext_ack *extack,
1189 			    const struct ip_tunnel_info *info,
1190 			    bool metadata, bool ipv6_rx_csum)
1191 {
1192 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1193 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1194 	bool tun_collect_md, tun_on_same_port;
1195 	int err, encap_len;
1196 
1197 	if (metadata && !is_tnl_info_zero(info)) {
1198 		NL_SET_ERR_MSG(extack,
1199 			       "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1200 		return -EINVAL;
1201 	}
1202 
1203 	geneve->net = net;
1204 	geneve->dev = dev;
1205 
1206 	t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1207 	if (t)
1208 		return -EBUSY;
1209 
1210 	/* make enough headroom for basic scenario */
1211 	encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1212 	if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1213 		encap_len += sizeof(struct iphdr);
1214 		dev->max_mtu -= sizeof(struct iphdr);
1215 	} else {
1216 		encap_len += sizeof(struct ipv6hdr);
1217 		dev->max_mtu -= sizeof(struct ipv6hdr);
1218 	}
1219 	dev->needed_headroom = encap_len + ETH_HLEN;
1220 
1221 	if (metadata) {
1222 		if (tun_on_same_port) {
1223 			NL_SET_ERR_MSG(extack,
1224 				       "There can be only one externally controlled device on a destination port");
1225 			return -EPERM;
1226 		}
1227 	} else {
1228 		if (tun_collect_md) {
1229 			NL_SET_ERR_MSG(extack,
1230 				       "There already exists an externally controlled device on this destination port");
1231 			return -EPERM;
1232 		}
1233 	}
1234 
1235 	dst_cache_reset(&geneve->info.dst_cache);
1236 	geneve->info = *info;
1237 	geneve->collect_md = metadata;
1238 	geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1239 
1240 	err = register_netdevice(dev);
1241 	if (err)
1242 		return err;
1243 
1244 	list_add(&geneve->next, &gn->geneve_list);
1245 	return 0;
1246 }
1247 
init_tnl_info(struct ip_tunnel_info * info,__u16 dst_port)1248 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1249 {
1250 	memset(info, 0, sizeof(*info));
1251 	info->key.tp_dst = htons(dst_port);
1252 }
1253 
geneve_nl2info(struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack,struct ip_tunnel_info * info,bool * metadata,bool * use_udp6_rx_checksums,bool changelink)1254 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1255 			  struct netlink_ext_ack *extack,
1256 			  struct ip_tunnel_info *info, bool *metadata,
1257 			  bool *use_udp6_rx_checksums, bool changelink)
1258 {
1259 	int attrtype;
1260 
1261 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1262 		NL_SET_ERR_MSG(extack,
1263 			       "Cannot specify both IPv4 and IPv6 Remote addresses");
1264 		return -EINVAL;
1265 	}
1266 
1267 	if (data[IFLA_GENEVE_REMOTE]) {
1268 		if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1269 			attrtype = IFLA_GENEVE_REMOTE;
1270 			goto change_notsup;
1271 		}
1272 
1273 		info->key.u.ipv4.dst =
1274 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1275 
1276 		if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1277 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1278 					    "Remote IPv4 address cannot be Multicast");
1279 			return -EINVAL;
1280 		}
1281 	}
1282 
1283 	if (data[IFLA_GENEVE_REMOTE6]) {
1284 #if IS_ENABLED(CONFIG_IPV6)
1285 		if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1286 			attrtype = IFLA_GENEVE_REMOTE6;
1287 			goto change_notsup;
1288 		}
1289 
1290 		info->mode = IP_TUNNEL_INFO_IPV6;
1291 		info->key.u.ipv6.dst =
1292 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1293 
1294 		if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1295 		    IPV6_ADDR_LINKLOCAL) {
1296 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1297 					    "Remote IPv6 address cannot be link-local");
1298 			return -EINVAL;
1299 		}
1300 		if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1301 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1302 					    "Remote IPv6 address cannot be Multicast");
1303 			return -EINVAL;
1304 		}
1305 		info->key.tun_flags |= TUNNEL_CSUM;
1306 		*use_udp6_rx_checksums = true;
1307 #else
1308 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1309 				    "IPv6 support not enabled in the kernel");
1310 		return -EPFNOSUPPORT;
1311 #endif
1312 	}
1313 
1314 	if (data[IFLA_GENEVE_ID]) {
1315 		__u32 vni;
1316 		__u8 tvni[3];
1317 		__be64 tunid;
1318 
1319 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1320 		tvni[0] = (vni & 0x00ff0000) >> 16;
1321 		tvni[1] = (vni & 0x0000ff00) >> 8;
1322 		tvni[2] =  vni & 0x000000ff;
1323 
1324 		tunid = vni_to_tunnel_id(tvni);
1325 		if (changelink && (tunid != info->key.tun_id)) {
1326 			attrtype = IFLA_GENEVE_ID;
1327 			goto change_notsup;
1328 		}
1329 		info->key.tun_id = tunid;
1330 	}
1331 
1332 	if (data[IFLA_GENEVE_TTL])
1333 		info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1334 
1335 	if (data[IFLA_GENEVE_TOS])
1336 		info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1337 
1338 	if (data[IFLA_GENEVE_LABEL]) {
1339 		info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1340 				  IPV6_FLOWLABEL_MASK;
1341 		if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1342 			NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1343 					    "Label attribute only applies for IPv6 Geneve devices");
1344 			return -EINVAL;
1345 		}
1346 	}
1347 
1348 	if (data[IFLA_GENEVE_PORT]) {
1349 		if (changelink) {
1350 			attrtype = IFLA_GENEVE_PORT;
1351 			goto change_notsup;
1352 		}
1353 		info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1354 	}
1355 
1356 	if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1357 		if (changelink) {
1358 			attrtype = IFLA_GENEVE_COLLECT_METADATA;
1359 			goto change_notsup;
1360 		}
1361 		*metadata = true;
1362 	}
1363 
1364 	if (data[IFLA_GENEVE_UDP_CSUM]) {
1365 		if (changelink) {
1366 			attrtype = IFLA_GENEVE_UDP_CSUM;
1367 			goto change_notsup;
1368 		}
1369 		if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1370 			info->key.tun_flags |= TUNNEL_CSUM;
1371 	}
1372 
1373 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1374 #if IS_ENABLED(CONFIG_IPV6)
1375 		if (changelink) {
1376 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1377 			goto change_notsup;
1378 		}
1379 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1380 			info->key.tun_flags &= ~TUNNEL_CSUM;
1381 #else
1382 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1383 				    "IPv6 support not enabled in the kernel");
1384 		return -EPFNOSUPPORT;
1385 #endif
1386 	}
1387 
1388 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1389 #if IS_ENABLED(CONFIG_IPV6)
1390 		if (changelink) {
1391 			attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1392 			goto change_notsup;
1393 		}
1394 		if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1395 			*use_udp6_rx_checksums = false;
1396 #else
1397 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1398 				    "IPv6 support not enabled in the kernel");
1399 		return -EPFNOSUPPORT;
1400 #endif
1401 	}
1402 
1403 	return 0;
1404 change_notsup:
1405 	NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1406 			    "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1407 	return -EOPNOTSUPP;
1408 }
1409 
geneve_link_config(struct net_device * dev,struct ip_tunnel_info * info,struct nlattr * tb[])1410 static void geneve_link_config(struct net_device *dev,
1411 			       struct ip_tunnel_info *info, struct nlattr *tb[])
1412 {
1413 	struct geneve_dev *geneve = netdev_priv(dev);
1414 	int ldev_mtu = 0;
1415 
1416 	if (tb[IFLA_MTU]) {
1417 		geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1418 		return;
1419 	}
1420 
1421 	switch (ip_tunnel_info_af(info)) {
1422 	case AF_INET: {
1423 		struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1424 		struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1425 
1426 		if (!IS_ERR(rt) && rt->dst.dev) {
1427 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1428 			ip_rt_put(rt);
1429 		}
1430 		break;
1431 	}
1432 #if IS_ENABLED(CONFIG_IPV6)
1433 	case AF_INET6: {
1434 		struct rt6_info *rt;
1435 
1436 		if (!__in6_dev_get(dev))
1437 			break;
1438 
1439 		rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1440 				NULL, 0);
1441 
1442 		if (rt && rt->dst.dev)
1443 			ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1444 		ip6_rt_put(rt);
1445 		break;
1446 	}
1447 #endif
1448 	}
1449 
1450 	if (ldev_mtu <= 0)
1451 		return;
1452 
1453 	geneve_change_mtu(dev, ldev_mtu - info->options_len);
1454 }
1455 
geneve_newlink(struct net * net,struct net_device * dev,struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack)1456 static int geneve_newlink(struct net *net, struct net_device *dev,
1457 			  struct nlattr *tb[], struct nlattr *data[],
1458 			  struct netlink_ext_ack *extack)
1459 {
1460 	bool use_udp6_rx_checksums = false;
1461 	struct ip_tunnel_info info;
1462 	bool metadata = false;
1463 	int err;
1464 
1465 	init_tnl_info(&info, GENEVE_UDP_PORT);
1466 	err = geneve_nl2info(tb, data, extack, &info, &metadata,
1467 			     &use_udp6_rx_checksums, false);
1468 	if (err)
1469 		return err;
1470 
1471 	err = geneve_configure(net, dev, extack, &info, metadata,
1472 			       use_udp6_rx_checksums);
1473 	if (err)
1474 		return err;
1475 
1476 	geneve_link_config(dev, &info, tb);
1477 
1478 	return 0;
1479 }
1480 
1481 /* Quiesces the geneve device data path for both TX and RX.
1482  *
1483  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1484  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1485  * to complete for the existing set of in-flight packets to be transmitted,
1486  * then we would have quiesced the transmit data path. All the future packets
1487  * will get dropped until we unquiesce the data path.
1488  *
1489  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1490  * if we set that to NULL under RCU and wait for synchronize_net() to
1491  * complete, then we would have quiesced the receive data path.
1492  */
geneve_quiesce(struct geneve_dev * geneve,struct geneve_sock ** gs4,struct geneve_sock ** gs6)1493 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1494 			   struct geneve_sock **gs6)
1495 {
1496 	*gs4 = rtnl_dereference(geneve->sock4);
1497 	rcu_assign_pointer(geneve->sock4, NULL);
1498 	if (*gs4)
1499 		rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1500 #if IS_ENABLED(CONFIG_IPV6)
1501 	*gs6 = rtnl_dereference(geneve->sock6);
1502 	rcu_assign_pointer(geneve->sock6, NULL);
1503 	if (*gs6)
1504 		rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1505 #else
1506 	*gs6 = NULL;
1507 #endif
1508 	synchronize_net();
1509 }
1510 
1511 /* Resumes the geneve device data path for both TX and RX. */
geneve_unquiesce(struct geneve_dev * geneve,struct geneve_sock * gs4,struct geneve_sock __maybe_unused * gs6)1512 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1513 			     struct geneve_sock __maybe_unused *gs6)
1514 {
1515 	rcu_assign_pointer(geneve->sock4, gs4);
1516 	if (gs4)
1517 		rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1518 #if IS_ENABLED(CONFIG_IPV6)
1519 	rcu_assign_pointer(geneve->sock6, gs6);
1520 	if (gs6)
1521 		rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1522 #endif
1523 	synchronize_net();
1524 }
1525 
geneve_changelink(struct net_device * dev,struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack)1526 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1527 			     struct nlattr *data[],
1528 			     struct netlink_ext_ack *extack)
1529 {
1530 	struct geneve_dev *geneve = netdev_priv(dev);
1531 	struct geneve_sock *gs4, *gs6;
1532 	struct ip_tunnel_info info;
1533 	bool metadata;
1534 	bool use_udp6_rx_checksums;
1535 	int err;
1536 
1537 	/* If the geneve device is configured for metadata (or externally
1538 	 * controlled, for example, OVS), then nothing can be changed.
1539 	 */
1540 	if (geneve->collect_md)
1541 		return -EOPNOTSUPP;
1542 
1543 	/* Start with the existing info. */
1544 	memcpy(&info, &geneve->info, sizeof(info));
1545 	metadata = geneve->collect_md;
1546 	use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1547 	err = geneve_nl2info(tb, data, extack, &info, &metadata,
1548 			     &use_udp6_rx_checksums, true);
1549 	if (err)
1550 		return err;
1551 
1552 	if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1553 		dst_cache_reset(&info.dst_cache);
1554 		geneve_link_config(dev, &info, tb);
1555 	}
1556 
1557 	geneve_quiesce(geneve, &gs4, &gs6);
1558 	geneve->info = info;
1559 	geneve->collect_md = metadata;
1560 	geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1561 	geneve_unquiesce(geneve, gs4, gs6);
1562 
1563 	return 0;
1564 }
1565 
geneve_dellink(struct net_device * dev,struct list_head * head)1566 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1567 {
1568 	struct geneve_dev *geneve = netdev_priv(dev);
1569 
1570 	list_del(&geneve->next);
1571 	unregister_netdevice_queue(dev, head);
1572 }
1573 
geneve_get_size(const struct net_device * dev)1574 static size_t geneve_get_size(const struct net_device *dev)
1575 {
1576 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1577 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1578 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1579 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1580 		nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1581 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1582 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1583 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1584 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1585 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1586 		0;
1587 }
1588 
geneve_fill_info(struct sk_buff * skb,const struct net_device * dev)1589 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1590 {
1591 	struct geneve_dev *geneve = netdev_priv(dev);
1592 	struct ip_tunnel_info *info = &geneve->info;
1593 	bool metadata = geneve->collect_md;
1594 	__u8 tmp_vni[3];
1595 	__u32 vni;
1596 
1597 	tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1598 	vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1599 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1600 		goto nla_put_failure;
1601 
1602 	if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1603 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1604 				    info->key.u.ipv4.dst))
1605 			goto nla_put_failure;
1606 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1607 			       !!(info->key.tun_flags & TUNNEL_CSUM)))
1608 			goto nla_put_failure;
1609 
1610 #if IS_ENABLED(CONFIG_IPV6)
1611 	} else if (!metadata) {
1612 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1613 				     &info->key.u.ipv6.dst))
1614 			goto nla_put_failure;
1615 		if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1616 			       !(info->key.tun_flags & TUNNEL_CSUM)))
1617 			goto nla_put_failure;
1618 #endif
1619 	}
1620 
1621 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1622 	    nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1623 	    nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1624 		goto nla_put_failure;
1625 
1626 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1627 		goto nla_put_failure;
1628 
1629 	if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1630 		goto nla_put_failure;
1631 
1632 #if IS_ENABLED(CONFIG_IPV6)
1633 	if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1634 		       !geneve->use_udp6_rx_checksums))
1635 		goto nla_put_failure;
1636 #endif
1637 
1638 	return 0;
1639 
1640 nla_put_failure:
1641 	return -EMSGSIZE;
1642 }
1643 
1644 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1645 	.kind		= "geneve",
1646 	.maxtype	= IFLA_GENEVE_MAX,
1647 	.policy		= geneve_policy,
1648 	.priv_size	= sizeof(struct geneve_dev),
1649 	.setup		= geneve_setup,
1650 	.validate	= geneve_validate,
1651 	.newlink	= geneve_newlink,
1652 	.changelink	= geneve_changelink,
1653 	.dellink	= geneve_dellink,
1654 	.get_size	= geneve_get_size,
1655 	.fill_info	= geneve_fill_info,
1656 };
1657 
geneve_dev_create_fb(struct net * net,const char * name,u8 name_assign_type,u16 dst_port)1658 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1659 					u8 name_assign_type, u16 dst_port)
1660 {
1661 	struct nlattr *tb[IFLA_MAX + 1];
1662 	struct ip_tunnel_info info;
1663 	struct net_device *dev;
1664 	LIST_HEAD(list_kill);
1665 	int err;
1666 
1667 	memset(tb, 0, sizeof(tb));
1668 	dev = rtnl_create_link(net, name, name_assign_type,
1669 			       &geneve_link_ops, tb);
1670 	if (IS_ERR(dev))
1671 		return dev;
1672 
1673 	init_tnl_info(&info, dst_port);
1674 	err = geneve_configure(net, dev, NULL, &info, true, true);
1675 	if (err) {
1676 		free_netdev(dev);
1677 		return ERR_PTR(err);
1678 	}
1679 
1680 	/* openvswitch users expect packet sizes to be unrestricted,
1681 	 * so set the largest MTU we can.
1682 	 */
1683 	err = geneve_change_mtu(dev, IP_MAX_MTU);
1684 	if (err)
1685 		goto err;
1686 
1687 	err = rtnl_configure_link(dev, NULL);
1688 	if (err < 0)
1689 		goto err;
1690 
1691 	return dev;
1692 err:
1693 	geneve_dellink(dev, &list_kill);
1694 	unregister_netdevice_many(&list_kill);
1695 	return ERR_PTR(err);
1696 }
1697 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1698 
geneve_netdevice_event(struct notifier_block * unused,unsigned long event,void * ptr)1699 static int geneve_netdevice_event(struct notifier_block *unused,
1700 				  unsigned long event, void *ptr)
1701 {
1702 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1703 
1704 	if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1705 	    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1706 		geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1707 	} else if (event == NETDEV_UNREGISTER) {
1708 		geneve_offload_rx_ports(dev, false);
1709 	} else if (event == NETDEV_REGISTER) {
1710 		geneve_offload_rx_ports(dev, true);
1711 	}
1712 
1713 	return NOTIFY_DONE;
1714 }
1715 
1716 static struct notifier_block geneve_notifier_block __read_mostly = {
1717 	.notifier_call = geneve_netdevice_event,
1718 };
1719 
geneve_init_net(struct net * net)1720 static __net_init int geneve_init_net(struct net *net)
1721 {
1722 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1723 
1724 	INIT_LIST_HEAD(&gn->geneve_list);
1725 	INIT_LIST_HEAD(&gn->sock_list);
1726 	return 0;
1727 }
1728 
geneve_destroy_tunnels(struct net * net,struct list_head * head)1729 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1730 {
1731 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1732 	struct geneve_dev *geneve, *next;
1733 	struct net_device *dev, *aux;
1734 
1735 	/* gather any geneve devices that were moved into this ns */
1736 	for_each_netdev_safe(net, dev, aux)
1737 		if (dev->rtnl_link_ops == &geneve_link_ops)
1738 			unregister_netdevice_queue(dev, head);
1739 
1740 	/* now gather any other geneve devices that were created in this ns */
1741 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1742 		/* If geneve->dev is in the same netns, it was already added
1743 		 * to the list by the previous loop.
1744 		 */
1745 		if (!net_eq(dev_net(geneve->dev), net))
1746 			unregister_netdevice_queue(geneve->dev, head);
1747 	}
1748 }
1749 
geneve_exit_batch_net(struct list_head * net_list)1750 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1751 {
1752 	struct net *net;
1753 	LIST_HEAD(list);
1754 
1755 	rtnl_lock();
1756 	list_for_each_entry(net, net_list, exit_list)
1757 		geneve_destroy_tunnels(net, &list);
1758 
1759 	/* unregister the devices gathered above */
1760 	unregister_netdevice_many(&list);
1761 	rtnl_unlock();
1762 
1763 	list_for_each_entry(net, net_list, exit_list) {
1764 		const struct geneve_net *gn = net_generic(net, geneve_net_id);
1765 
1766 		WARN_ON_ONCE(!list_empty(&gn->sock_list));
1767 	}
1768 }
1769 
1770 static struct pernet_operations geneve_net_ops = {
1771 	.init = geneve_init_net,
1772 	.exit_batch = geneve_exit_batch_net,
1773 	.id   = &geneve_net_id,
1774 	.size = sizeof(struct geneve_net),
1775 };
1776 
geneve_init_module(void)1777 static int __init geneve_init_module(void)
1778 {
1779 	int rc;
1780 
1781 	rc = register_pernet_subsys(&geneve_net_ops);
1782 	if (rc)
1783 		goto out1;
1784 
1785 	rc = register_netdevice_notifier(&geneve_notifier_block);
1786 	if (rc)
1787 		goto out2;
1788 
1789 	rc = rtnl_link_register(&geneve_link_ops);
1790 	if (rc)
1791 		goto out3;
1792 
1793 	return 0;
1794 out3:
1795 	unregister_netdevice_notifier(&geneve_notifier_block);
1796 out2:
1797 	unregister_pernet_subsys(&geneve_net_ops);
1798 out1:
1799 	return rc;
1800 }
1801 late_initcall(geneve_init_module);
1802 
geneve_cleanup_module(void)1803 static void __exit geneve_cleanup_module(void)
1804 {
1805 	rtnl_link_unregister(&geneve_link_ops);
1806 	unregister_netdevice_notifier(&geneve_notifier_block);
1807 	unregister_pernet_subsys(&geneve_net_ops);
1808 }
1809 module_exit(geneve_cleanup_module);
1810 
1811 MODULE_LICENSE("GPL");
1812 MODULE_VERSION(GENEVE_NETDEV_VER);
1813 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1814 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1815 MODULE_ALIAS_RTNL_LINK("geneve");
1816