• 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 
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 int geneve_net_id;
47 
48 union geneve_addr {
49 	struct sockaddr_in sin;
50 	struct sockaddr_in6 sin6;
51 	struct sockaddr sa;
52 };
53 
54 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
55 
56 /* Pseudo network device */
57 struct geneve_dev {
58 	struct hlist_node  hlist;	/* vni hash table */
59 	struct net	   *net;	/* netns for packet i/o */
60 	struct net_device  *dev;	/* netdev for geneve tunnel */
61 	struct geneve_sock __rcu *sock4;	/* IPv4 socket used for geneve tunnel */
62 #if IS_ENABLED(CONFIG_IPV6)
63 	struct geneve_sock __rcu *sock6;	/* IPv6 socket used for geneve tunnel */
64 #endif
65 	u8                 vni[3];	/* virtual network ID for tunnel */
66 	u8                 ttl;		/* TTL override */
67 	u8                 tos;		/* TOS override */
68 	union geneve_addr  remote;	/* IP address for link partner */
69 	struct list_head   next;	/* geneve's per namespace list */
70 	__be32		   label;	/* IPv6 flowlabel override */
71 	__be16		   dst_port;
72 	bool		   collect_md;
73 	struct gro_cells   gro_cells;
74 	u32		   flags;
75 	struct dst_cache   dst_cache;
76 };
77 
78 /* Geneve device flags */
79 #define GENEVE_F_UDP_ZERO_CSUM_TX	BIT(0)
80 #define GENEVE_F_UDP_ZERO_CSUM6_TX	BIT(1)
81 #define GENEVE_F_UDP_ZERO_CSUM6_RX	BIT(2)
82 
83 struct geneve_sock {
84 	bool			collect_md;
85 	struct list_head	list;
86 	struct socket		*sock;
87 	struct rcu_head		rcu;
88 	int			refcnt;
89 	struct hlist_head	vni_list[VNI_HASH_SIZE];
90 	u32			flags;
91 };
92 
geneve_net_vni_hash(u8 vni[3])93 static inline __u32 geneve_net_vni_hash(u8 vni[3])
94 {
95 	__u32 vnid;
96 
97 	vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
98 	return hash_32(vnid, VNI_HASH_BITS);
99 }
100 
vni_to_tunnel_id(const __u8 * vni)101 static __be64 vni_to_tunnel_id(const __u8 *vni)
102 {
103 #ifdef __BIG_ENDIAN
104 	return (vni[0] << 16) | (vni[1] << 8) | vni[2];
105 #else
106 	return (__force __be64)(((__force u64)vni[0] << 40) |
107 				((__force u64)vni[1] << 48) |
108 				((__force u64)vni[2] << 56));
109 #endif
110 }
111 
geneve_get_sk_family(struct geneve_sock * gs)112 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
113 {
114 	return gs->sock->sk->sk_family;
115 }
116 
geneve_lookup(struct geneve_sock * gs,__be32 addr,u8 vni[])117 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
118 					__be32 addr, u8 vni[])
119 {
120 	struct hlist_head *vni_list_head;
121 	struct geneve_dev *geneve;
122 	__u32 hash;
123 
124 	/* Find the device for this VNI */
125 	hash = geneve_net_vni_hash(vni);
126 	vni_list_head = &gs->vni_list[hash];
127 	hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
128 		if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
129 		    addr == geneve->remote.sin.sin_addr.s_addr)
130 			return geneve;
131 	}
132 	return NULL;
133 }
134 
135 #if IS_ENABLED(CONFIG_IPV6)
geneve6_lookup(struct geneve_sock * gs,struct in6_addr addr6,u8 vni[])136 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
137 					 struct in6_addr addr6, u8 vni[])
138 {
139 	struct hlist_head *vni_list_head;
140 	struct geneve_dev *geneve;
141 	__u32 hash;
142 
143 	/* Find the device for this VNI */
144 	hash = geneve_net_vni_hash(vni);
145 	vni_list_head = &gs->vni_list[hash];
146 	hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
147 		if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
148 		    ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
149 			return geneve;
150 	}
151 	return NULL;
152 }
153 #endif
154 
geneve_hdr(const struct sk_buff * skb)155 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
156 {
157 	return (struct genevehdr *)(udp_hdr(skb) + 1);
158 }
159 
geneve_lookup_skb(struct geneve_sock * gs,struct sk_buff * skb)160 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
161 					    struct sk_buff *skb)
162 {
163 	u8 *vni;
164 	__be32 addr;
165 	static u8 zero_vni[3];
166 #if IS_ENABLED(CONFIG_IPV6)
167 	static struct in6_addr zero_addr6;
168 #endif
169 
170 	if (geneve_get_sk_family(gs) == AF_INET) {
171 		struct iphdr *iph;
172 
173 		iph = ip_hdr(skb); /* outer IP header... */
174 
175 		if (gs->collect_md) {
176 			vni = zero_vni;
177 			addr = 0;
178 		} else {
179 			vni = geneve_hdr(skb)->vni;
180 			addr = iph->saddr;
181 		}
182 
183 		return geneve_lookup(gs, addr, vni);
184 #if IS_ENABLED(CONFIG_IPV6)
185 	} else if (geneve_get_sk_family(gs) == AF_INET6) {
186 		struct ipv6hdr *ip6h;
187 		struct in6_addr addr6;
188 
189 		ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
190 
191 		if (gs->collect_md) {
192 			vni = zero_vni;
193 			addr6 = zero_addr6;
194 		} else {
195 			vni = geneve_hdr(skb)->vni;
196 			addr6 = ip6h->saddr;
197 		}
198 
199 		return geneve6_lookup(gs, addr6, vni);
200 #endif
201 	}
202 	return NULL;
203 }
204 
205 /* geneve receive/decap routine */
geneve_rx(struct geneve_dev * geneve,struct geneve_sock * gs,struct sk_buff * skb)206 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
207 		      struct sk_buff *skb)
208 {
209 	struct genevehdr *gnvh = geneve_hdr(skb);
210 	struct metadata_dst *tun_dst = NULL;
211 	struct pcpu_sw_netstats *stats;
212 	unsigned int len;
213 	int err = 0;
214 	void *oiph;
215 
216 	if (ip_tunnel_collect_metadata() || gs->collect_md) {
217 		__be16 flags;
218 
219 		flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
220 			(gnvh->oam ? TUNNEL_OAM : 0) |
221 			(gnvh->critical ? TUNNEL_CRIT_OPT : 0);
222 
223 		tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
224 					 vni_to_tunnel_id(gnvh->vni),
225 					 gnvh->opt_len * 4);
226 		if (!tun_dst) {
227 			geneve->dev->stats.rx_dropped++;
228 			goto drop;
229 		}
230 		/* Update tunnel dst according to Geneve options. */
231 		ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
232 					gnvh->options, gnvh->opt_len * 4);
233 	} else {
234 		/* Drop packets w/ critical options,
235 		 * since we don't support any...
236 		 */
237 		if (gnvh->critical) {
238 			geneve->dev->stats.rx_frame_errors++;
239 			geneve->dev->stats.rx_errors++;
240 			goto drop;
241 		}
242 	}
243 
244 	skb_reset_mac_header(skb);
245 	skb->protocol = eth_type_trans(skb, geneve->dev);
246 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
247 
248 	if (tun_dst)
249 		skb_dst_set(skb, &tun_dst->dst);
250 
251 	/* Ignore packet loops (and multicast echo) */
252 	if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
253 		geneve->dev->stats.rx_errors++;
254 		goto drop;
255 	}
256 
257 	oiph = skb_network_header(skb);
258 	skb_reset_network_header(skb);
259 
260 	if (geneve_get_sk_family(gs) == AF_INET)
261 		err = IP_ECN_decapsulate(oiph, skb);
262 #if IS_ENABLED(CONFIG_IPV6)
263 	else
264 		err = IP6_ECN_decapsulate(oiph, skb);
265 #endif
266 
267 	if (unlikely(err)) {
268 		if (log_ecn_error) {
269 			if (geneve_get_sk_family(gs) == AF_INET)
270 				net_info_ratelimited("non-ECT from %pI4 "
271 						     "with TOS=%#x\n",
272 						     &((struct iphdr *)oiph)->saddr,
273 						     ((struct iphdr *)oiph)->tos);
274 #if IS_ENABLED(CONFIG_IPV6)
275 			else
276 				net_info_ratelimited("non-ECT from %pI6\n",
277 						     &((struct ipv6hdr *)oiph)->saddr);
278 #endif
279 		}
280 		if (err > 1) {
281 			++geneve->dev->stats.rx_frame_errors;
282 			++geneve->dev->stats.rx_errors;
283 			goto drop;
284 		}
285 	}
286 
287 	len = skb->len;
288 	err = gro_cells_receive(&geneve->gro_cells, skb);
289 	if (likely(err == NET_RX_SUCCESS)) {
290 		stats = this_cpu_ptr(geneve->dev->tstats);
291 		u64_stats_update_begin(&stats->syncp);
292 		stats->rx_packets++;
293 		stats->rx_bytes += len;
294 		u64_stats_update_end(&stats->syncp);
295 	}
296 	return;
297 drop:
298 	/* Consume bad packet */
299 	kfree_skb(skb);
300 }
301 
302 /* Setup stats when device is created */
geneve_init(struct net_device * dev)303 static int geneve_init(struct net_device *dev)
304 {
305 	struct geneve_dev *geneve = netdev_priv(dev);
306 	int err;
307 
308 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
309 	if (!dev->tstats)
310 		return -ENOMEM;
311 
312 	err = gro_cells_init(&geneve->gro_cells, dev);
313 	if (err) {
314 		free_percpu(dev->tstats);
315 		return err;
316 	}
317 
318 	err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
319 	if (err) {
320 		free_percpu(dev->tstats);
321 		gro_cells_destroy(&geneve->gro_cells);
322 		return err;
323 	}
324 
325 	return 0;
326 }
327 
geneve_uninit(struct net_device * dev)328 static void geneve_uninit(struct net_device *dev)
329 {
330 	struct geneve_dev *geneve = netdev_priv(dev);
331 
332 	dst_cache_destroy(&geneve->dst_cache);
333 	gro_cells_destroy(&geneve->gro_cells);
334 	free_percpu(dev->tstats);
335 }
336 
337 /* Callback from net/ipv4/udp.c to receive packets */
geneve_udp_encap_recv(struct sock * sk,struct sk_buff * skb)338 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
339 {
340 	struct genevehdr *geneveh;
341 	struct geneve_dev *geneve;
342 	struct geneve_sock *gs;
343 	int opts_len;
344 
345 	/* Need UDP and Geneve header to be present */
346 	if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
347 		goto drop;
348 
349 	/* Return packets with reserved bits set */
350 	geneveh = geneve_hdr(skb);
351 	if (unlikely(geneveh->ver != GENEVE_VER))
352 		goto drop;
353 
354 	if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
355 		goto drop;
356 
357 	gs = rcu_dereference_sk_user_data(sk);
358 	if (!gs)
359 		goto drop;
360 
361 	geneve = geneve_lookup_skb(gs, skb);
362 	if (!geneve)
363 		goto drop;
364 
365 	opts_len = geneveh->opt_len * 4;
366 	if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
367 				 htons(ETH_P_TEB),
368 				 !net_eq(geneve->net, dev_net(geneve->dev)))) {
369 		geneve->dev->stats.rx_dropped++;
370 		goto drop;
371 	}
372 
373 	geneve_rx(geneve, gs, skb);
374 	return 0;
375 
376 drop:
377 	/* Consume bad packet */
378 	kfree_skb(skb);
379 	return 0;
380 }
381 
geneve_create_sock(struct net * net,bool ipv6,__be16 port,u32 flags)382 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
383 					 __be16 port, u32 flags)
384 {
385 	struct socket *sock;
386 	struct udp_port_cfg udp_conf;
387 	int err;
388 
389 	memset(&udp_conf, 0, sizeof(udp_conf));
390 
391 	if (ipv6) {
392 		udp_conf.family = AF_INET6;
393 		udp_conf.ipv6_v6only = 1;
394 		udp_conf.use_udp6_rx_checksums =
395 		    !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
396 	} else {
397 		udp_conf.family = AF_INET;
398 		udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
399 	}
400 
401 	udp_conf.local_udp_port = port;
402 
403 	/* Open UDP socket */
404 	err = udp_sock_create(net, &udp_conf, &sock);
405 	if (err < 0)
406 		return ERR_PTR(err);
407 
408 	return sock;
409 }
410 
geneve_hlen(struct genevehdr * gh)411 static int geneve_hlen(struct genevehdr *gh)
412 {
413 	return sizeof(*gh) + gh->opt_len * 4;
414 }
415 
geneve_gro_receive(struct sock * sk,struct sk_buff ** head,struct sk_buff * skb)416 static struct sk_buff **geneve_gro_receive(struct sock *sk,
417 					   struct sk_buff **head,
418 					   struct sk_buff *skb)
419 {
420 	struct sk_buff *p, **pp = NULL;
421 	struct genevehdr *gh, *gh2;
422 	unsigned int hlen, gh_len, off_gnv;
423 	const struct packet_offload *ptype;
424 	__be16 type;
425 	int flush = 1;
426 
427 	off_gnv = skb_gro_offset(skb);
428 	hlen = off_gnv + sizeof(*gh);
429 	gh = skb_gro_header_fast(skb, off_gnv);
430 	if (skb_gro_header_hard(skb, hlen)) {
431 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
432 		if (unlikely(!gh))
433 			goto out;
434 	}
435 
436 	if (gh->ver != GENEVE_VER || gh->oam)
437 		goto out;
438 	gh_len = geneve_hlen(gh);
439 
440 	hlen = off_gnv + gh_len;
441 	if (skb_gro_header_hard(skb, hlen)) {
442 		gh = skb_gro_header_slow(skb, hlen, off_gnv);
443 		if (unlikely(!gh))
444 			goto out;
445 	}
446 
447 	for (p = *head; p; p = p->next) {
448 		if (!NAPI_GRO_CB(p)->same_flow)
449 			continue;
450 
451 		gh2 = (struct genevehdr *)(p->data + off_gnv);
452 		if (gh->opt_len != gh2->opt_len ||
453 		    memcmp(gh, gh2, gh_len)) {
454 			NAPI_GRO_CB(p)->same_flow = 0;
455 			continue;
456 		}
457 	}
458 
459 	type = gh->proto_type;
460 
461 	rcu_read_lock();
462 	ptype = gro_find_receive_by_type(type);
463 	if (!ptype)
464 		goto out_unlock;
465 
466 	skb_gro_pull(skb, gh_len);
467 	skb_gro_postpull_rcsum(skb, gh, gh_len);
468 	pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
469 	flush = 0;
470 
471 out_unlock:
472 	rcu_read_unlock();
473 out:
474 	NAPI_GRO_CB(skb)->flush |= flush;
475 
476 	return pp;
477 }
478 
geneve_gro_complete(struct sock * sk,struct sk_buff * skb,int nhoff)479 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
480 			       int nhoff)
481 {
482 	struct genevehdr *gh;
483 	struct packet_offload *ptype;
484 	__be16 type;
485 	int gh_len;
486 	int err = -ENOSYS;
487 
488 	gh = (struct genevehdr *)(skb->data + nhoff);
489 	gh_len = geneve_hlen(gh);
490 	type = gh->proto_type;
491 
492 	rcu_read_lock();
493 	ptype = gro_find_complete_by_type(type);
494 	if (ptype)
495 		err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
496 
497 	rcu_read_unlock();
498 
499 	skb_set_inner_mac_header(skb, nhoff + gh_len);
500 
501 	return err;
502 }
503 
504 /* Create new listen socket if needed */
geneve_socket_create(struct net * net,__be16 port,bool ipv6,u32 flags)505 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
506 						bool ipv6, u32 flags)
507 {
508 	struct geneve_net *gn = net_generic(net, geneve_net_id);
509 	struct geneve_sock *gs;
510 	struct socket *sock;
511 	struct udp_tunnel_sock_cfg tunnel_cfg;
512 	int h;
513 
514 	gs = kzalloc(sizeof(*gs), GFP_KERNEL);
515 	if (!gs)
516 		return ERR_PTR(-ENOMEM);
517 
518 	sock = geneve_create_sock(net, ipv6, port, flags);
519 	if (IS_ERR(sock)) {
520 		kfree(gs);
521 		return ERR_CAST(sock);
522 	}
523 
524 	gs->sock = sock;
525 	gs->refcnt = 1;
526 	for (h = 0; h < VNI_HASH_SIZE; ++h)
527 		INIT_HLIST_HEAD(&gs->vni_list[h]);
528 
529 	/* Initialize the geneve udp offloads structure */
530 	udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
531 
532 	/* Mark socket as an encapsulation socket */
533 	memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
534 	tunnel_cfg.sk_user_data = gs;
535 	tunnel_cfg.encap_type = 1;
536 	tunnel_cfg.gro_receive = geneve_gro_receive;
537 	tunnel_cfg.gro_complete = geneve_gro_complete;
538 	tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
539 	tunnel_cfg.encap_destroy = NULL;
540 	setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
541 	list_add(&gs->list, &gn->sock_list);
542 	return gs;
543 }
544 
__geneve_sock_release(struct geneve_sock * gs)545 static void __geneve_sock_release(struct geneve_sock *gs)
546 {
547 	if (!gs || --gs->refcnt)
548 		return;
549 
550 	list_del(&gs->list);
551 	udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
552 	udp_tunnel_sock_release(gs->sock);
553 	kfree_rcu(gs, rcu);
554 }
555 
geneve_sock_release(struct geneve_dev * geneve)556 static void geneve_sock_release(struct geneve_dev *geneve)
557 {
558 	struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
559 #if IS_ENABLED(CONFIG_IPV6)
560 	struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
561 
562 	rcu_assign_pointer(geneve->sock6, NULL);
563 #endif
564 
565 	rcu_assign_pointer(geneve->sock4, NULL);
566 	synchronize_net();
567 
568 	__geneve_sock_release(gs4);
569 #if IS_ENABLED(CONFIG_IPV6)
570 	__geneve_sock_release(gs6);
571 #endif
572 }
573 
geneve_find_sock(struct geneve_net * gn,sa_family_t family,__be16 dst_port)574 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
575 					    sa_family_t family,
576 					    __be16 dst_port)
577 {
578 	struct geneve_sock *gs;
579 
580 	list_for_each_entry(gs, &gn->sock_list, list) {
581 		if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
582 		    geneve_get_sk_family(gs) == family) {
583 			return gs;
584 		}
585 	}
586 	return NULL;
587 }
588 
geneve_sock_add(struct geneve_dev * geneve,bool ipv6)589 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
590 {
591 	struct net *net = geneve->net;
592 	struct geneve_net *gn = net_generic(net, geneve_net_id);
593 	struct geneve_sock *gs;
594 	__u32 hash;
595 
596 	gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
597 	if (gs) {
598 		gs->refcnt++;
599 		goto out;
600 	}
601 
602 	gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
603 	if (IS_ERR(gs))
604 		return PTR_ERR(gs);
605 
606 out:
607 	gs->collect_md = geneve->collect_md;
608 	gs->flags = geneve->flags;
609 #if IS_ENABLED(CONFIG_IPV6)
610 	if (ipv6)
611 		rcu_assign_pointer(geneve->sock6, gs);
612 	else
613 #endif
614 		rcu_assign_pointer(geneve->sock4, gs);
615 
616 	hash = geneve_net_vni_hash(geneve->vni);
617 	hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
618 	return 0;
619 }
620 
geneve_open(struct net_device * dev)621 static int geneve_open(struct net_device *dev)
622 {
623 	struct geneve_dev *geneve = netdev_priv(dev);
624 	bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
625 	bool metadata = geneve->collect_md;
626 	int ret = 0;
627 
628 #if IS_ENABLED(CONFIG_IPV6)
629 	if (ipv6 || metadata)
630 		ret = geneve_sock_add(geneve, true);
631 #endif
632 	if (!ret && (!ipv6 || metadata))
633 		ret = geneve_sock_add(geneve, false);
634 	if (ret < 0)
635 		geneve_sock_release(geneve);
636 
637 	return ret;
638 }
639 
geneve_stop(struct net_device * dev)640 static int geneve_stop(struct net_device *dev)
641 {
642 	struct geneve_dev *geneve = netdev_priv(dev);
643 
644 	if (!hlist_unhashed(&geneve->hlist))
645 		hlist_del_rcu(&geneve->hlist);
646 	geneve_sock_release(geneve);
647 	return 0;
648 }
649 
geneve_build_header(struct genevehdr * geneveh,__be16 tun_flags,u8 vni[3],u8 options_len,u8 * options)650 static void geneve_build_header(struct genevehdr *geneveh,
651 				__be16 tun_flags, u8 vni[3],
652 				u8 options_len, u8 *options)
653 {
654 	geneveh->ver = GENEVE_VER;
655 	geneveh->opt_len = options_len / 4;
656 	geneveh->oam = !!(tun_flags & TUNNEL_OAM);
657 	geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
658 	geneveh->rsvd1 = 0;
659 	memcpy(geneveh->vni, vni, 3);
660 	geneveh->proto_type = htons(ETH_P_TEB);
661 	geneveh->rsvd2 = 0;
662 
663 	memcpy(geneveh->options, options, options_len);
664 }
665 
geneve_build_skb(struct rtable * rt,struct sk_buff * skb,__be16 tun_flags,u8 vni[3],u8 opt_len,u8 * opt,u32 flags,bool xnet)666 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
667 			    __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
668 			    u32 flags, bool xnet)
669 {
670 	struct genevehdr *gnvh;
671 	int min_headroom;
672 	int err;
673 	bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
674 
675 	skb_scrub_packet(skb, xnet);
676 
677 	min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
678 			+ GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
679 	err = skb_cow_head(skb, min_headroom);
680 	if (unlikely(err))
681 		goto free_rt;
682 
683 	err = udp_tunnel_handle_offloads(skb, udp_sum);
684 	if (err)
685 		goto free_rt;
686 
687 	gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
688 	geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
689 
690 	skb_set_inner_protocol(skb, htons(ETH_P_TEB));
691 	return 0;
692 
693 free_rt:
694 	ip_rt_put(rt);
695 	return err;
696 }
697 
698 #if IS_ENABLED(CONFIG_IPV6)
geneve6_build_skb(struct dst_entry * dst,struct sk_buff * skb,__be16 tun_flags,u8 vni[3],u8 opt_len,u8 * opt,u32 flags,bool xnet)699 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
700 			     __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
701 			     u32 flags, bool xnet)
702 {
703 	struct genevehdr *gnvh;
704 	int min_headroom;
705 	int err;
706 	bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
707 
708 	skb_scrub_packet(skb, xnet);
709 
710 	min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
711 			+ GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
712 	err = skb_cow_head(skb, min_headroom);
713 	if (unlikely(err))
714 		goto free_dst;
715 
716 	err = udp_tunnel_handle_offloads(skb, udp_sum);
717 	if (err)
718 		goto free_dst;
719 
720 	gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
721 	geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
722 
723 	skb_set_inner_protocol(skb, htons(ETH_P_TEB));
724 	return 0;
725 
726 free_dst:
727 	dst_release(dst);
728 	return err;
729 }
730 #endif
731 
geneve_get_v4_rt(struct sk_buff * skb,struct net_device * dev,struct flowi4 * fl4,struct ip_tunnel_info * info)732 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
733 				       struct net_device *dev,
734 				       struct flowi4 *fl4,
735 				       struct ip_tunnel_info *info)
736 {
737 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
738 	struct geneve_dev *geneve = netdev_priv(dev);
739 	struct dst_cache *dst_cache;
740 	struct rtable *rt = NULL;
741 	__u8 tos;
742 
743 	if (!rcu_dereference(geneve->sock4))
744 		return ERR_PTR(-EIO);
745 
746 	memset(fl4, 0, sizeof(*fl4));
747 	fl4->flowi4_mark = skb->mark;
748 	fl4->flowi4_proto = IPPROTO_UDP;
749 
750 	if (info) {
751 		fl4->daddr = info->key.u.ipv4.dst;
752 		fl4->saddr = info->key.u.ipv4.src;
753 		fl4->flowi4_tos = RT_TOS(info->key.tos);
754 		dst_cache = &info->dst_cache;
755 	} else {
756 		tos = geneve->tos;
757 		if (tos == 1) {
758 			const struct iphdr *iip = ip_hdr(skb);
759 
760 			tos = ip_tunnel_get_dsfield(iip, skb);
761 			use_cache = false;
762 		}
763 
764 		fl4->flowi4_tos = RT_TOS(tos);
765 		fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
766 		dst_cache = &geneve->dst_cache;
767 	}
768 
769 	if (use_cache) {
770 		rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
771 		if (rt)
772 			return rt;
773 	}
774 
775 	rt = ip_route_output_key(geneve->net, fl4);
776 	if (IS_ERR(rt)) {
777 		netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
778 		return ERR_PTR(-ENETUNREACH);
779 	}
780 	if (rt->dst.dev == dev) { /* is this necessary? */
781 		netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
782 		ip_rt_put(rt);
783 		return ERR_PTR(-ELOOP);
784 	}
785 	if (use_cache)
786 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
787 	return rt;
788 }
789 
790 #if IS_ENABLED(CONFIG_IPV6)
geneve_get_v6_dst(struct sk_buff * skb,struct net_device * dev,struct flowi6 * fl6,struct ip_tunnel_info * info)791 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
792 					   struct net_device *dev,
793 					   struct flowi6 *fl6,
794 					   struct ip_tunnel_info *info)
795 {
796 	bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
797 	struct geneve_dev *geneve = netdev_priv(dev);
798 	struct dst_entry *dst = NULL;
799 	struct dst_cache *dst_cache;
800 	struct geneve_sock *gs6;
801 	__u8 prio;
802 
803 	gs6 = rcu_dereference(geneve->sock6);
804 	if (!gs6)
805 		return ERR_PTR(-EIO);
806 
807 	memset(fl6, 0, sizeof(*fl6));
808 	fl6->flowi6_mark = skb->mark;
809 	fl6->flowi6_proto = IPPROTO_UDP;
810 
811 	if (info) {
812 		fl6->daddr = info->key.u.ipv6.dst;
813 		fl6->saddr = info->key.u.ipv6.src;
814 		fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
815 						   info->key.label);
816 		dst_cache = &info->dst_cache;
817 	} else {
818 		prio = geneve->tos;
819 		if (prio == 1) {
820 			const struct iphdr *iip = ip_hdr(skb);
821 
822 			prio = ip_tunnel_get_dsfield(iip, skb);
823 			use_cache = false;
824 		}
825 
826 		fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
827 						   geneve->label);
828 		fl6->daddr = geneve->remote.sin6.sin6_addr;
829 		dst_cache = &geneve->dst_cache;
830 	}
831 
832 	if (use_cache) {
833 		dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
834 		if (dst)
835 			return dst;
836 	}
837 
838 	if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
839 		netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
840 		return ERR_PTR(-ENETUNREACH);
841 	}
842 	if (dst->dev == dev) { /* is this necessary? */
843 		netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
844 		dst_release(dst);
845 		return ERR_PTR(-ELOOP);
846 	}
847 
848 	if (use_cache)
849 		dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
850 	return dst;
851 }
852 #endif
853 
854 /* Convert 64 bit tunnel ID to 24 bit VNI. */
tunnel_id_to_vni(__be64 tun_id,__u8 * vni)855 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
856 {
857 #ifdef __BIG_ENDIAN
858 	vni[0] = (__force __u8)(tun_id >> 16);
859 	vni[1] = (__force __u8)(tun_id >> 8);
860 	vni[2] = (__force __u8)tun_id;
861 #else
862 	vni[0] = (__force __u8)((__force u64)tun_id >> 40);
863 	vni[1] = (__force __u8)((__force u64)tun_id >> 48);
864 	vni[2] = (__force __u8)((__force u64)tun_id >> 56);
865 #endif
866 }
867 
geneve_xmit_skb(struct sk_buff * skb,struct net_device * dev,struct ip_tunnel_info * info)868 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
869 				   struct ip_tunnel_info *info)
870 {
871 	struct geneve_dev *geneve = netdev_priv(dev);
872 	struct geneve_sock *gs4;
873 	struct rtable *rt = NULL;
874 	int err = -EINVAL;
875 	struct flowi4 fl4;
876 	__u8 tos, ttl;
877 	__be16 sport;
878 	__be16 df;
879 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
880 	u32 flags = geneve->flags;
881 
882 	gs4 = rcu_dereference(geneve->sock4);
883 	if (!gs4)
884 		goto tx_error;
885 
886 	if (geneve->collect_md) {
887 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
888 			netdev_dbg(dev, "no tunnel metadata\n");
889 			goto tx_error;
890 		}
891 		if (info && ip_tunnel_info_af(info) != AF_INET)
892 			goto tx_error;
893 	}
894 
895 	rt = geneve_get_v4_rt(skb, dev, &fl4, info);
896 	if (IS_ERR(rt)) {
897 		err = PTR_ERR(rt);
898 		goto tx_error;
899 	}
900 
901 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
902 	skb_reset_mac_header(skb);
903 
904 	if (info) {
905 		const struct ip_tunnel_key *key = &info->key;
906 		u8 *opts = NULL;
907 		u8 vni[3];
908 
909 		tunnel_id_to_vni(key->tun_id, vni);
910 		if (info->options_len)
911 			opts = ip_tunnel_info_opts(info);
912 
913 		if (key->tun_flags & TUNNEL_CSUM)
914 			flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
915 		else
916 			flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
917 
918 		err = geneve_build_skb(rt, skb, key->tun_flags, vni,
919 				       info->options_len, opts, flags, xnet);
920 		if (unlikely(err))
921 			goto tx_error;
922 
923 		tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
924 		ttl = key->ttl;
925 		df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
926 	} else {
927 		err = geneve_build_skb(rt, skb, 0, geneve->vni,
928 				       0, NULL, flags, xnet);
929 		if (unlikely(err))
930 			goto tx_error;
931 
932 		tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
933 		ttl = geneve->ttl;
934 		if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
935 			ttl = 1;
936 		ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
937 		df = 0;
938 	}
939 	udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
940 			    tos, ttl, df, sport, geneve->dst_port,
941 			    !net_eq(geneve->net, dev_net(geneve->dev)),
942 			    !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
943 
944 	return NETDEV_TX_OK;
945 
946 tx_error:
947 	dev_kfree_skb(skb);
948 
949 	if (err == -ELOOP)
950 		dev->stats.collisions++;
951 	else if (err == -ENETUNREACH)
952 		dev->stats.tx_carrier_errors++;
953 
954 	dev->stats.tx_errors++;
955 	return NETDEV_TX_OK;
956 }
957 
958 #if IS_ENABLED(CONFIG_IPV6)
geneve6_xmit_skb(struct sk_buff * skb,struct net_device * dev,struct ip_tunnel_info * info)959 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
960 				    struct ip_tunnel_info *info)
961 {
962 	struct geneve_dev *geneve = netdev_priv(dev);
963 	struct dst_entry *dst = NULL;
964 	struct geneve_sock *gs6;
965 	int err = -EINVAL;
966 	struct flowi6 fl6;
967 	__u8 prio, ttl;
968 	__be16 sport;
969 	__be32 label;
970 	bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
971 	u32 flags = geneve->flags;
972 
973 	gs6 = rcu_dereference(geneve->sock6);
974 	if (!gs6)
975 		goto tx_error;
976 
977 	if (geneve->collect_md) {
978 		if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
979 			netdev_dbg(dev, "no tunnel metadata\n");
980 			goto tx_error;
981 		}
982 	}
983 
984 	dst = geneve_get_v6_dst(skb, dev, &fl6, info);
985 	if (IS_ERR(dst)) {
986 		err = PTR_ERR(dst);
987 		goto tx_error;
988 	}
989 
990 	sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
991 	skb_reset_mac_header(skb);
992 
993 	if (info) {
994 		const struct ip_tunnel_key *key = &info->key;
995 		u8 *opts = NULL;
996 		u8 vni[3];
997 
998 		tunnel_id_to_vni(key->tun_id, vni);
999 		if (info->options_len)
1000 			opts = ip_tunnel_info_opts(info);
1001 
1002 		if (key->tun_flags & TUNNEL_CSUM)
1003 			flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1004 		else
1005 			flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1006 
1007 		err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1008 					info->options_len, opts,
1009 					flags, xnet);
1010 		if (unlikely(err))
1011 			goto tx_error;
1012 
1013 		prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1014 		ttl = key->ttl;
1015 		label = info->key.label;
1016 	} else {
1017 		err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1018 					0, NULL, flags, xnet);
1019 		if (unlikely(err))
1020 			goto tx_error;
1021 
1022 		prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1023 					   ip_hdr(skb), skb);
1024 		ttl = geneve->ttl;
1025 		if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1026 			ttl = 1;
1027 		ttl = ttl ? : ip6_dst_hoplimit(dst);
1028 		label = geneve->label;
1029 	}
1030 
1031 	udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1032 			     &fl6.saddr, &fl6.daddr, prio, ttl, label,
1033 			     sport, geneve->dst_port,
1034 			     !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1035 	return NETDEV_TX_OK;
1036 
1037 tx_error:
1038 	dev_kfree_skb(skb);
1039 
1040 	if (err == -ELOOP)
1041 		dev->stats.collisions++;
1042 	else if (err == -ENETUNREACH)
1043 		dev->stats.tx_carrier_errors++;
1044 
1045 	dev->stats.tx_errors++;
1046 	return NETDEV_TX_OK;
1047 }
1048 #endif
1049 
geneve_xmit(struct sk_buff * skb,struct net_device * dev)1050 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1051 {
1052 	struct geneve_dev *geneve = netdev_priv(dev);
1053 	struct ip_tunnel_info *info = NULL;
1054 	int err;
1055 
1056 	if (geneve->collect_md)
1057 		info = skb_tunnel_info(skb);
1058 
1059 	rcu_read_lock();
1060 #if IS_ENABLED(CONFIG_IPV6)
1061 	if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1062 	    (!info && geneve->remote.sa.sa_family == AF_INET6))
1063 		err = geneve6_xmit_skb(skb, dev, info);
1064 	else
1065 #endif
1066 		err = geneve_xmit_skb(skb, dev, info);
1067 	rcu_read_unlock();
1068 
1069 	return err;
1070 }
1071 
__geneve_change_mtu(struct net_device * dev,int new_mtu,bool strict)1072 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1073 {
1074 	struct geneve_dev *geneve = netdev_priv(dev);
1075 	/* The max_mtu calculation does not take account of GENEVE
1076 	 * options, to avoid excluding potentially valid
1077 	 * configurations.
1078 	 */
1079 	int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1080 
1081 	if (geneve->remote.sa.sa_family == AF_INET6)
1082 		max_mtu -= sizeof(struct ipv6hdr);
1083 	else
1084 		max_mtu -= sizeof(struct iphdr);
1085 
1086 	if (new_mtu < 68)
1087 		return -EINVAL;
1088 
1089 	if (new_mtu > max_mtu) {
1090 		if (strict)
1091 			return -EINVAL;
1092 
1093 		new_mtu = max_mtu;
1094 	}
1095 
1096 	dev->mtu = new_mtu;
1097 	return 0;
1098 }
1099 
geneve_change_mtu(struct net_device * dev,int new_mtu)1100 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1101 {
1102 	return __geneve_change_mtu(dev, new_mtu, true);
1103 }
1104 
geneve_fill_metadata_dst(struct net_device * dev,struct sk_buff * skb)1105 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1106 {
1107 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
1108 	struct geneve_dev *geneve = netdev_priv(dev);
1109 	struct rtable *rt;
1110 	struct flowi4 fl4;
1111 #if IS_ENABLED(CONFIG_IPV6)
1112 	struct dst_entry *dst;
1113 	struct flowi6 fl6;
1114 #endif
1115 
1116 	if (ip_tunnel_info_af(info) == AF_INET) {
1117 		rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1118 		if (IS_ERR(rt))
1119 			return PTR_ERR(rt);
1120 
1121 		ip_rt_put(rt);
1122 		info->key.u.ipv4.src = fl4.saddr;
1123 #if IS_ENABLED(CONFIG_IPV6)
1124 	} else if (ip_tunnel_info_af(info) == AF_INET6) {
1125 		dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1126 		if (IS_ERR(dst))
1127 			return PTR_ERR(dst);
1128 
1129 		dst_release(dst);
1130 		info->key.u.ipv6.src = fl6.saddr;
1131 #endif
1132 	} else {
1133 		return -EINVAL;
1134 	}
1135 
1136 	info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1137 					     1, USHRT_MAX, true);
1138 	info->key.tp_dst = geneve->dst_port;
1139 	return 0;
1140 }
1141 
1142 static const struct net_device_ops geneve_netdev_ops = {
1143 	.ndo_init		= geneve_init,
1144 	.ndo_uninit		= geneve_uninit,
1145 	.ndo_open		= geneve_open,
1146 	.ndo_stop		= geneve_stop,
1147 	.ndo_start_xmit		= geneve_xmit,
1148 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1149 	.ndo_change_mtu		= geneve_change_mtu,
1150 	.ndo_validate_addr	= eth_validate_addr,
1151 	.ndo_set_mac_address	= eth_mac_addr,
1152 	.ndo_fill_metadata_dst	= geneve_fill_metadata_dst,
1153 };
1154 
geneve_get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * drvinfo)1155 static void geneve_get_drvinfo(struct net_device *dev,
1156 			       struct ethtool_drvinfo *drvinfo)
1157 {
1158 	strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1159 	strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1160 }
1161 
1162 static const struct ethtool_ops geneve_ethtool_ops = {
1163 	.get_drvinfo	= geneve_get_drvinfo,
1164 	.get_link	= ethtool_op_get_link,
1165 };
1166 
1167 /* Info for udev, that this is a virtual tunnel endpoint */
1168 static struct device_type geneve_type = {
1169 	.name = "geneve",
1170 };
1171 
1172 /* Calls the ndo_udp_tunnel_add of the caller in order to
1173  * supply the listening GENEVE udp ports. Callers are expected
1174  * to implement the ndo_udp_tunnel_add.
1175  */
geneve_push_rx_ports(struct net_device * dev)1176 static void geneve_push_rx_ports(struct net_device *dev)
1177 {
1178 	struct net *net = dev_net(dev);
1179 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1180 	struct geneve_sock *gs;
1181 
1182 	rcu_read_lock();
1183 	list_for_each_entry_rcu(gs, &gn->sock_list, list)
1184 		udp_tunnel_push_rx_port(dev, gs->sock,
1185 					UDP_TUNNEL_TYPE_GENEVE);
1186 	rcu_read_unlock();
1187 }
1188 
1189 /* Initialize the device structure. */
geneve_setup(struct net_device * dev)1190 static void geneve_setup(struct net_device *dev)
1191 {
1192 	ether_setup(dev);
1193 
1194 	dev->netdev_ops = &geneve_netdev_ops;
1195 	dev->ethtool_ops = &geneve_ethtool_ops;
1196 	dev->destructor = free_netdev;
1197 
1198 	SET_NETDEV_DEVTYPE(dev, &geneve_type);
1199 
1200 	dev->features    |= NETIF_F_LLTX;
1201 	dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1202 	dev->features    |= NETIF_F_RXCSUM;
1203 	dev->features    |= NETIF_F_GSO_SOFTWARE;
1204 
1205 	dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1206 	dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1207 
1208 	netif_keep_dst(dev);
1209 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1210 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1211 	eth_hw_addr_random(dev);
1212 }
1213 
1214 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1215 	[IFLA_GENEVE_ID]		= { .type = NLA_U32 },
1216 	[IFLA_GENEVE_REMOTE]		= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1217 	[IFLA_GENEVE_REMOTE6]		= { .len = sizeof(struct in6_addr) },
1218 	[IFLA_GENEVE_TTL]		= { .type = NLA_U8 },
1219 	[IFLA_GENEVE_TOS]		= { .type = NLA_U8 },
1220 	[IFLA_GENEVE_LABEL]		= { .type = NLA_U32 },
1221 	[IFLA_GENEVE_PORT]		= { .type = NLA_U16 },
1222 	[IFLA_GENEVE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1223 	[IFLA_GENEVE_UDP_CSUM]		= { .type = NLA_U8 },
1224 	[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]	= { .type = NLA_U8 },
1225 	[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]	= { .type = NLA_U8 },
1226 };
1227 
geneve_validate(struct nlattr * tb[],struct nlattr * data[])1228 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1229 {
1230 	if (tb[IFLA_ADDRESS]) {
1231 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1232 			return -EINVAL;
1233 
1234 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1235 			return -EADDRNOTAVAIL;
1236 	}
1237 
1238 	if (!data)
1239 		return -EINVAL;
1240 
1241 	if (data[IFLA_GENEVE_ID]) {
1242 		__u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1243 
1244 		if (vni >= GENEVE_VID_MASK)
1245 			return -ERANGE;
1246 	}
1247 
1248 	return 0;
1249 }
1250 
geneve_find_dev(struct geneve_net * gn,__be16 dst_port,union geneve_addr * remote,u8 vni[],bool * tun_on_same_port,bool * tun_collect_md)1251 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1252 					  __be16 dst_port,
1253 					  union geneve_addr *remote,
1254 					  u8 vni[],
1255 					  bool *tun_on_same_port,
1256 					  bool *tun_collect_md)
1257 {
1258 	struct geneve_dev *geneve, *t;
1259 
1260 	*tun_on_same_port = false;
1261 	*tun_collect_md = false;
1262 	t = NULL;
1263 	list_for_each_entry(geneve, &gn->geneve_list, next) {
1264 		if (geneve->dst_port == dst_port) {
1265 			*tun_collect_md = geneve->collect_md;
1266 			*tun_on_same_port = true;
1267 		}
1268 		if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1269 		    !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1270 		    dst_port == geneve->dst_port)
1271 			t = geneve;
1272 	}
1273 	return t;
1274 }
1275 
geneve_configure(struct net * net,struct net_device * dev,union geneve_addr * remote,__u32 vni,__u8 ttl,__u8 tos,__be32 label,__be16 dst_port,bool metadata,u32 flags)1276 static int geneve_configure(struct net *net, struct net_device *dev,
1277 			    union geneve_addr *remote,
1278 			    __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1279 			    __be16 dst_port, bool metadata, u32 flags)
1280 {
1281 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1282 	struct geneve_dev *t, *geneve = netdev_priv(dev);
1283 	bool tun_collect_md, tun_on_same_port;
1284 	int err, encap_len;
1285 
1286 	if (!remote)
1287 		return -EINVAL;
1288 	if (metadata &&
1289 	    (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1290 		return -EINVAL;
1291 
1292 	geneve->net = net;
1293 	geneve->dev = dev;
1294 
1295 	geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1296 	geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1297 	geneve->vni[2] =  vni & 0x000000ff;
1298 
1299 	if ((remote->sa.sa_family == AF_INET &&
1300 	     IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1301 	    (remote->sa.sa_family == AF_INET6 &&
1302 	     ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1303 		return -EINVAL;
1304 	if (label && remote->sa.sa_family != AF_INET6)
1305 		return -EINVAL;
1306 
1307 	geneve->remote = *remote;
1308 
1309 	geneve->ttl = ttl;
1310 	geneve->tos = tos;
1311 	geneve->label = label;
1312 	geneve->dst_port = dst_port;
1313 	geneve->collect_md = metadata;
1314 	geneve->flags = flags;
1315 
1316 	t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1317 			    &tun_on_same_port, &tun_collect_md);
1318 	if (t)
1319 		return -EBUSY;
1320 
1321 	/* make enough headroom for basic scenario */
1322 	encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1323 	if (remote->sa.sa_family == AF_INET)
1324 		encap_len += sizeof(struct iphdr);
1325 	else
1326 		encap_len += sizeof(struct ipv6hdr);
1327 	dev->needed_headroom = encap_len + ETH_HLEN;
1328 
1329 	if (metadata) {
1330 		if (tun_on_same_port)
1331 			return -EPERM;
1332 	} else {
1333 		if (tun_collect_md)
1334 			return -EPERM;
1335 	}
1336 
1337 	dst_cache_reset(&geneve->dst_cache);
1338 
1339 	err = register_netdevice(dev);
1340 	if (err)
1341 		return err;
1342 
1343 	list_add(&geneve->next, &gn->geneve_list);
1344 	return 0;
1345 }
1346 
geneve_newlink(struct net * net,struct net_device * dev,struct nlattr * tb[],struct nlattr * data[])1347 static int geneve_newlink(struct net *net, struct net_device *dev,
1348 			  struct nlattr *tb[], struct nlattr *data[])
1349 {
1350 	__be16 dst_port = htons(GENEVE_UDP_PORT);
1351 	__u8 ttl = 0, tos = 0;
1352 	bool metadata = false;
1353 	union geneve_addr remote = geneve_remote_unspec;
1354 	__be32 label = 0;
1355 	__u32 vni = 0;
1356 	u32 flags = 0;
1357 
1358 	if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1359 		return -EINVAL;
1360 
1361 	if (data[IFLA_GENEVE_REMOTE]) {
1362 		remote.sa.sa_family = AF_INET;
1363 		remote.sin.sin_addr.s_addr =
1364 			nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1365 	}
1366 
1367 	if (data[IFLA_GENEVE_REMOTE6]) {
1368 		if (!IS_ENABLED(CONFIG_IPV6))
1369 			return -EPFNOSUPPORT;
1370 
1371 		remote.sa.sa_family = AF_INET6;
1372 		remote.sin6.sin6_addr =
1373 			nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1374 
1375 		if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1376 		    IPV6_ADDR_LINKLOCAL) {
1377 			netdev_dbg(dev, "link-local remote is unsupported\n");
1378 			return -EINVAL;
1379 		}
1380 	}
1381 
1382 	if (data[IFLA_GENEVE_ID])
1383 		vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1384 
1385 	if (data[IFLA_GENEVE_TTL])
1386 		ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1387 
1388 	if (data[IFLA_GENEVE_TOS])
1389 		tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1390 
1391 	if (data[IFLA_GENEVE_LABEL])
1392 		label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1393 			IPV6_FLOWLABEL_MASK;
1394 
1395 	if (data[IFLA_GENEVE_PORT])
1396 		dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1397 
1398 	if (data[IFLA_GENEVE_COLLECT_METADATA])
1399 		metadata = true;
1400 
1401 	if (data[IFLA_GENEVE_UDP_CSUM] &&
1402 	    !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1403 		flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1404 
1405 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1406 	    nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1407 		flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1408 
1409 	if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1410 	    nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1411 		flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1412 
1413 	return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1414 				dst_port, metadata, flags);
1415 }
1416 
geneve_dellink(struct net_device * dev,struct list_head * head)1417 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1418 {
1419 	struct geneve_dev *geneve = netdev_priv(dev);
1420 
1421 	list_del(&geneve->next);
1422 	unregister_netdevice_queue(dev, head);
1423 }
1424 
geneve_get_size(const struct net_device * dev)1425 static size_t geneve_get_size(const struct net_device *dev)
1426 {
1427 	return nla_total_size(sizeof(__u32)) +	/* IFLA_GENEVE_ID */
1428 		nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1429 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1430 		nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1431 		nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1432 		nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1433 		nla_total_size(0) +	 /* IFLA_GENEVE_COLLECT_METADATA */
1434 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1435 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1436 		nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1437 		0;
1438 }
1439 
geneve_fill_info(struct sk_buff * skb,const struct net_device * dev)1440 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1441 {
1442 	struct geneve_dev *geneve = netdev_priv(dev);
1443 	__u32 vni;
1444 
1445 	vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1446 	if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1447 		goto nla_put_failure;
1448 
1449 	if (geneve->remote.sa.sa_family == AF_INET) {
1450 		if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1451 				    geneve->remote.sin.sin_addr.s_addr))
1452 			goto nla_put_failure;
1453 #if IS_ENABLED(CONFIG_IPV6)
1454 	} else {
1455 		if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1456 				     &geneve->remote.sin6.sin6_addr))
1457 			goto nla_put_failure;
1458 #endif
1459 	}
1460 
1461 	if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1462 	    nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1463 	    nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1464 		goto nla_put_failure;
1465 
1466 	if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1467 		goto nla_put_failure;
1468 
1469 	if (geneve->collect_md) {
1470 		if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1471 			goto nla_put_failure;
1472 	}
1473 
1474 	if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1475 		       !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1476 	    nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1477 		       !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1478 	    nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1479 		       !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1480 		goto nla_put_failure;
1481 
1482 	return 0;
1483 
1484 nla_put_failure:
1485 	return -EMSGSIZE;
1486 }
1487 
1488 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1489 	.kind		= "geneve",
1490 	.maxtype	= IFLA_GENEVE_MAX,
1491 	.policy		= geneve_policy,
1492 	.priv_size	= sizeof(struct geneve_dev),
1493 	.setup		= geneve_setup,
1494 	.validate	= geneve_validate,
1495 	.newlink	= geneve_newlink,
1496 	.dellink	= geneve_dellink,
1497 	.get_size	= geneve_get_size,
1498 	.fill_info	= geneve_fill_info,
1499 };
1500 
geneve_dev_create_fb(struct net * net,const char * name,u8 name_assign_type,u16 dst_port)1501 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1502 					u8 name_assign_type, u16 dst_port)
1503 {
1504 	struct nlattr *tb[IFLA_MAX + 1];
1505 	struct net_device *dev;
1506 	LIST_HEAD(list_kill);
1507 	int err;
1508 
1509 	memset(tb, 0, sizeof(tb));
1510 	dev = rtnl_create_link(net, name, name_assign_type,
1511 			       &geneve_link_ops, tb);
1512 	if (IS_ERR(dev))
1513 		return dev;
1514 
1515 	err = geneve_configure(net, dev, &geneve_remote_unspec,
1516 			       0, 0, 0, 0, htons(dst_port), true,
1517 			       GENEVE_F_UDP_ZERO_CSUM6_RX);
1518 	if (err) {
1519 		free_netdev(dev);
1520 		return ERR_PTR(err);
1521 	}
1522 
1523 	/* openvswitch users expect packet sizes to be unrestricted,
1524 	 * so set the largest MTU we can.
1525 	 */
1526 	err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1527 	if (err)
1528 		goto err;
1529 
1530 	err = rtnl_configure_link(dev, NULL);
1531 	if (err < 0)
1532 		goto err;
1533 
1534 	return dev;
1535 
1536  err:
1537 	geneve_dellink(dev, &list_kill);
1538 	unregister_netdevice_many(&list_kill);
1539 	return ERR_PTR(err);
1540 }
1541 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1542 
geneve_netdevice_event(struct notifier_block * unused,unsigned long event,void * ptr)1543 static int geneve_netdevice_event(struct notifier_block *unused,
1544 				  unsigned long event, void *ptr)
1545 {
1546 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1547 
1548 	if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1549 		geneve_push_rx_ports(dev);
1550 
1551 	return NOTIFY_DONE;
1552 }
1553 
1554 static struct notifier_block geneve_notifier_block __read_mostly = {
1555 	.notifier_call = geneve_netdevice_event,
1556 };
1557 
geneve_init_net(struct net * net)1558 static __net_init int geneve_init_net(struct net *net)
1559 {
1560 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1561 
1562 	INIT_LIST_HEAD(&gn->geneve_list);
1563 	INIT_LIST_HEAD(&gn->sock_list);
1564 	return 0;
1565 }
1566 
geneve_exit_net(struct net * net)1567 static void __net_exit geneve_exit_net(struct net *net)
1568 {
1569 	struct geneve_net *gn = net_generic(net, geneve_net_id);
1570 	struct geneve_dev *geneve, *next;
1571 	struct net_device *dev, *aux;
1572 	LIST_HEAD(list);
1573 
1574 	rtnl_lock();
1575 
1576 	/* gather any geneve devices that were moved into this ns */
1577 	for_each_netdev_safe(net, dev, aux)
1578 		if (dev->rtnl_link_ops == &geneve_link_ops)
1579 			unregister_netdevice_queue(dev, &list);
1580 
1581 	/* now gather any other geneve devices that were created in this ns */
1582 	list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1583 		/* If geneve->dev is in the same netns, it was already added
1584 		 * to the list by the previous loop.
1585 		 */
1586 		if (!net_eq(dev_net(geneve->dev), net))
1587 			unregister_netdevice_queue(geneve->dev, &list);
1588 	}
1589 
1590 	/* unregister the devices gathered above */
1591 	unregister_netdevice_many(&list);
1592 	rtnl_unlock();
1593 }
1594 
1595 static struct pernet_operations geneve_net_ops = {
1596 	.init = geneve_init_net,
1597 	.exit = geneve_exit_net,
1598 	.id   = &geneve_net_id,
1599 	.size = sizeof(struct geneve_net),
1600 };
1601 
geneve_init_module(void)1602 static int __init geneve_init_module(void)
1603 {
1604 	int rc;
1605 
1606 	rc = register_pernet_subsys(&geneve_net_ops);
1607 	if (rc)
1608 		goto out1;
1609 
1610 	rc = register_netdevice_notifier(&geneve_notifier_block);
1611 	if (rc)
1612 		goto out2;
1613 
1614 	rc = rtnl_link_register(&geneve_link_ops);
1615 	if (rc)
1616 		goto out3;
1617 
1618 	return 0;
1619 
1620 out3:
1621 	unregister_netdevice_notifier(&geneve_notifier_block);
1622 out2:
1623 	unregister_pernet_subsys(&geneve_net_ops);
1624 out1:
1625 	return rc;
1626 }
1627 late_initcall(geneve_init_module);
1628 
geneve_cleanup_module(void)1629 static void __exit geneve_cleanup_module(void)
1630 {
1631 	rtnl_link_unregister(&geneve_link_ops);
1632 	unregister_netdevice_notifier(&geneve_notifier_block);
1633 	unregister_pernet_subsys(&geneve_net_ops);
1634 }
1635 module_exit(geneve_cleanup_module);
1636 
1637 MODULE_LICENSE("GPL");
1638 MODULE_VERSION(GENEVE_NETDEV_VER);
1639 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1640 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1641 MODULE_ALIAS_RTNL_LINK("geneve");
1642