Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * GENEVE: Generic Network Virtualization Encapsulation
4 *
5 * Copyright (c) 2015 Red Hat, Inc.
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/kernel.h>
11#include <linux/module.h>
12#include <linux/etherdevice.h>
13#include <linux/hash.h>
14#include <net/ipv6_stubs.h>
15#include <net/dst_metadata.h>
16#include <net/gro_cells.h>
17#include <net/rtnetlink.h>
18#include <net/geneve.h>
19#include <net/protocol.h>
20
21#define GENEVE_NETDEV_VER "0.6"
22
23#define GENEVE_N_VID (1u << 24)
24#define GENEVE_VID_MASK (GENEVE_N_VID - 1)
25
26#define VNI_HASH_BITS 10
27#define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
28
29static bool log_ecn_error = true;
30module_param(log_ecn_error, bool, 0644);
31MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
32
33#define GENEVE_VER 0
34#define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
35#define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
36#define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
37
38/* per-network namespace private data for this module */
39struct geneve_net {
40 struct list_head geneve_list;
41 struct list_head sock_list;
42};
43
44static unsigned int geneve_net_id;
45
46struct geneve_dev_node {
47 struct hlist_node hlist;
48 struct geneve_dev *geneve;
49};
50
51/* Pseudo network device */
52struct geneve_dev {
53 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
54#if IS_ENABLED(CONFIG_IPV6)
55 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
56#endif
57 struct net *net; /* netns for packet i/o */
58 struct net_device *dev; /* netdev for geneve tunnel */
59 struct ip_tunnel_info info;
60 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
61#if IS_ENABLED(CONFIG_IPV6)
62 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
63#endif
64 struct list_head next; /* geneve's per namespace list */
65 struct gro_cells gro_cells;
66 bool collect_md;
67 bool use_udp6_rx_checksums;
68 bool ttl_inherit;
69 enum ifla_geneve_df df;
70};
71
72struct geneve_sock {
73 bool collect_md;
74 struct list_head list;
75 struct socket *sock;
76 struct rcu_head rcu;
77 int refcnt;
78 struct hlist_head vni_list[VNI_HASH_SIZE];
79};
80
81static inline __u32 geneve_net_vni_hash(u8 vni[3])
82{
83 __u32 vnid;
84
85 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
86 return hash_32(vnid, VNI_HASH_BITS);
87}
88
89static __be64 vni_to_tunnel_id(const __u8 *vni)
90{
91#ifdef __BIG_ENDIAN
92 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
93#else
94 return (__force __be64)(((__force u64)vni[0] << 40) |
95 ((__force u64)vni[1] << 48) |
96 ((__force u64)vni[2] << 56));
97#endif
98}
99
100/* Convert 64 bit tunnel ID to 24 bit VNI. */
101static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
102{
103#ifdef __BIG_ENDIAN
104 vni[0] = (__force __u8)(tun_id >> 16);
105 vni[1] = (__force __u8)(tun_id >> 8);
106 vni[2] = (__force __u8)tun_id;
107#else
108 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
109 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
110 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
111#endif
112}
113
114static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
115{
116 return !memcmp(vni, &tun_id[5], 3);
117}
118
119static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
120{
121 return gs->sock->sk->sk_family;
122}
123
124static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
125 __be32 addr, u8 vni[])
126{
127 struct hlist_head *vni_list_head;
128 struct geneve_dev_node *node;
129 __u32 hash;
130
131 /* Find the device for this VNI */
132 hash = geneve_net_vni_hash(vni);
133 vni_list_head = &gs->vni_list[hash];
134 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
135 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
136 addr == node->geneve->info.key.u.ipv4.dst)
137 return node->geneve;
138 }
139 return NULL;
140}
141
142#if IS_ENABLED(CONFIG_IPV6)
143static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
144 struct in6_addr addr6, u8 vni[])
145{
146 struct hlist_head *vni_list_head;
147 struct geneve_dev_node *node;
148 __u32 hash;
149
150 /* Find the device for this VNI */
151 hash = geneve_net_vni_hash(vni);
152 vni_list_head = &gs->vni_list[hash];
153 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
154 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
155 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
156 return node->geneve;
157 }
158 return NULL;
159}
160#endif
161
162static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
163{
164 return (struct genevehdr *)(udp_hdr(skb) + 1);
165}
166
167static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
168 struct sk_buff *skb)
169{
170 static u8 zero_vni[3];
171 u8 *vni;
172
173 if (geneve_get_sk_family(gs) == AF_INET) {
174 struct iphdr *iph;
175 __be32 addr;
176
177 iph = ip_hdr(skb); /* outer IP header... */
178
179 if (gs->collect_md) {
180 vni = zero_vni;
181 addr = 0;
182 } else {
183 vni = geneve_hdr(skb)->vni;
184 addr = iph->saddr;
185 }
186
187 return geneve_lookup(gs, addr, vni);
188#if IS_ENABLED(CONFIG_IPV6)
189 } else if (geneve_get_sk_family(gs) == AF_INET6) {
190 static struct in6_addr zero_addr6;
191 struct ipv6hdr *ip6h;
192 struct in6_addr addr6;
193
194 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
195
196 if (gs->collect_md) {
197 vni = zero_vni;
198 addr6 = zero_addr6;
199 } else {
200 vni = geneve_hdr(skb)->vni;
201 addr6 = ip6h->saddr;
202 }
203
204 return geneve6_lookup(gs, addr6, vni);
205#endif
206 }
207 return NULL;
208}
209
210/* geneve receive/decap routine */
211static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
212 struct sk_buff *skb)
213{
214 struct genevehdr *gnvh = geneve_hdr(skb);
215 struct metadata_dst *tun_dst = NULL;
216 struct pcpu_sw_netstats *stats;
217 unsigned int len;
218 int err = 0;
219 void *oiph;
220
221 if (ip_tunnel_collect_metadata() || gs->collect_md) {
222 __be16 flags;
223
224 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
225 (gnvh->oam ? TUNNEL_OAM : 0) |
226 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
227
228 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
229 vni_to_tunnel_id(gnvh->vni),
230 gnvh->opt_len * 4);
231 if (!tun_dst) {
232 geneve->dev->stats.rx_dropped++;
233 goto drop;
234 }
235 /* Update tunnel dst according to Geneve options. */
236 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
237 gnvh->options, gnvh->opt_len * 4,
238 TUNNEL_GENEVE_OPT);
239 } else {
240 /* Drop packets w/ critical options,
241 * since we don't support any...
242 */
243 if (gnvh->critical) {
244 geneve->dev->stats.rx_frame_errors++;
245 geneve->dev->stats.rx_errors++;
246 goto drop;
247 }
248 }
249
250 skb_reset_mac_header(skb);
251 skb->protocol = eth_type_trans(skb, geneve->dev);
252 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
253
254 if (tun_dst)
255 skb_dst_set(skb, &tun_dst->dst);
256
257 /* Ignore packet loops (and multicast echo) */
258 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
259 geneve->dev->stats.rx_errors++;
260 goto drop;
261 }
262
263 oiph = skb_network_header(skb);
264 skb_reset_network_header(skb);
265
266 if (geneve_get_sk_family(gs) == AF_INET)
267 err = IP_ECN_decapsulate(oiph, skb);
268#if IS_ENABLED(CONFIG_IPV6)
269 else
270 err = IP6_ECN_decapsulate(oiph, skb);
271#endif
272
273 if (unlikely(err)) {
274 if (log_ecn_error) {
275 if (geneve_get_sk_family(gs) == AF_INET)
276 net_info_ratelimited("non-ECT from %pI4 "
277 "with TOS=%#x\n",
278 &((struct iphdr *)oiph)->saddr,
279 ((struct iphdr *)oiph)->tos);
280#if IS_ENABLED(CONFIG_IPV6)
281 else
282 net_info_ratelimited("non-ECT from %pI6\n",
283 &((struct ipv6hdr *)oiph)->saddr);
284#endif
285 }
286 if (err > 1) {
287 ++geneve->dev->stats.rx_frame_errors;
288 ++geneve->dev->stats.rx_errors;
289 goto drop;
290 }
291 }
292
293 len = skb->len;
294 err = gro_cells_receive(&geneve->gro_cells, skb);
295 if (likely(err == NET_RX_SUCCESS)) {
296 stats = this_cpu_ptr(geneve->dev->tstats);
297 u64_stats_update_begin(&stats->syncp);
298 stats->rx_packets++;
299 stats->rx_bytes += len;
300 u64_stats_update_end(&stats->syncp);
301 }
302 return;
303drop:
304 /* Consume bad packet */
305 kfree_skb(skb);
306}
307
308/* Setup stats when device is created */
309static int geneve_init(struct net_device *dev)
310{
311 struct geneve_dev *geneve = netdev_priv(dev);
312 int err;
313
314 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
315 if (!dev->tstats)
316 return -ENOMEM;
317
318 err = gro_cells_init(&geneve->gro_cells, dev);
319 if (err) {
320 free_percpu(dev->tstats);
321 return err;
322 }
323
324 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
325 if (err) {
326 free_percpu(dev->tstats);
327 gro_cells_destroy(&geneve->gro_cells);
328 return err;
329 }
330 return 0;
331}
332
333static void geneve_uninit(struct net_device *dev)
334{
335 struct geneve_dev *geneve = netdev_priv(dev);
336
337 dst_cache_destroy(&geneve->info.dst_cache);
338 gro_cells_destroy(&geneve->gro_cells);
339 free_percpu(dev->tstats);
340}
341
342/* Callback from net/ipv4/udp.c to receive packets */
343static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
344{
345 struct genevehdr *geneveh;
346 struct geneve_dev *geneve;
347 struct geneve_sock *gs;
348 int opts_len;
349
350 /* Need UDP and Geneve header to be present */
351 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
352 goto drop;
353
354 /* Return packets with reserved bits set */
355 geneveh = geneve_hdr(skb);
356 if (unlikely(geneveh->ver != GENEVE_VER))
357 goto drop;
358
359 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
360 goto drop;
361
362 gs = rcu_dereference_sk_user_data(sk);
363 if (!gs)
364 goto drop;
365
366 geneve = geneve_lookup_skb(gs, skb);
367 if (!geneve)
368 goto drop;
369
370 opts_len = geneveh->opt_len * 4;
371 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
372 htons(ETH_P_TEB),
373 !net_eq(geneve->net, dev_net(geneve->dev)))) {
374 geneve->dev->stats.rx_dropped++;
375 goto drop;
376 }
377
378 geneve_rx(geneve, gs, skb);
379 return 0;
380
381drop:
382 /* Consume bad packet */
383 kfree_skb(skb);
384 return 0;
385}
386
387/* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
388static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
389{
390 struct genevehdr *geneveh;
391 struct geneve_sock *gs;
392 u8 zero_vni[3] = { 0 };
393 u8 *vni = zero_vni;
394
395 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
396 return -EINVAL;
397
398 geneveh = geneve_hdr(skb);
399 if (geneveh->ver != GENEVE_VER)
400 return -EINVAL;
401
402 if (geneveh->proto_type != htons(ETH_P_TEB))
403 return -EINVAL;
404
405 gs = rcu_dereference_sk_user_data(sk);
406 if (!gs)
407 return -ENOENT;
408
409 if (geneve_get_sk_family(gs) == AF_INET) {
410 struct iphdr *iph = ip_hdr(skb);
411 __be32 addr4 = 0;
412
413 if (!gs->collect_md) {
414 vni = geneve_hdr(skb)->vni;
415 addr4 = iph->daddr;
416 }
417
418 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
419 }
420
421#if IS_ENABLED(CONFIG_IPV6)
422 if (geneve_get_sk_family(gs) == AF_INET6) {
423 struct ipv6hdr *ip6h = ipv6_hdr(skb);
424 struct in6_addr addr6;
425
426 memset(&addr6, 0, sizeof(struct in6_addr));
427
428 if (!gs->collect_md) {
429 vni = geneve_hdr(skb)->vni;
430 addr6 = ip6h->daddr;
431 }
432
433 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
434 }
435#endif
436
437 return -EPFNOSUPPORT;
438}
439
440static struct socket *geneve_create_sock(struct net *net, bool ipv6,
441 __be16 port, bool ipv6_rx_csum)
442{
443 struct socket *sock;
444 struct udp_port_cfg udp_conf;
445 int err;
446
447 memset(&udp_conf, 0, sizeof(udp_conf));
448
449 if (ipv6) {
450 udp_conf.family = AF_INET6;
451 udp_conf.ipv6_v6only = 1;
452 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
453 } else {
454 udp_conf.family = AF_INET;
455 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
456 }
457
458 udp_conf.local_udp_port = port;
459
460 /* Open UDP socket */
461 err = udp_sock_create(net, &udp_conf, &sock);
462 if (err < 0)
463 return ERR_PTR(err);
464
465 return sock;
466}
467
468static int geneve_hlen(struct genevehdr *gh)
469{
470 return sizeof(*gh) + gh->opt_len * 4;
471}
472
473static struct sk_buff *geneve_gro_receive(struct sock *sk,
474 struct list_head *head,
475 struct sk_buff *skb)
476{
477 struct sk_buff *pp = NULL;
478 struct sk_buff *p;
479 struct genevehdr *gh, *gh2;
480 unsigned int hlen, gh_len, off_gnv;
481 const struct packet_offload *ptype;
482 __be16 type;
483 int flush = 1;
484
485 off_gnv = skb_gro_offset(skb);
486 hlen = off_gnv + sizeof(*gh);
487 gh = skb_gro_header_fast(skb, off_gnv);
488 if (skb_gro_header_hard(skb, hlen)) {
489 gh = skb_gro_header_slow(skb, hlen, off_gnv);
490 if (unlikely(!gh))
491 goto out;
492 }
493
494 if (gh->ver != GENEVE_VER || gh->oam)
495 goto out;
496 gh_len = geneve_hlen(gh);
497
498 hlen = off_gnv + gh_len;
499 if (skb_gro_header_hard(skb, hlen)) {
500 gh = skb_gro_header_slow(skb, hlen, off_gnv);
501 if (unlikely(!gh))
502 goto out;
503 }
504
505 list_for_each_entry(p, head, list) {
506 if (!NAPI_GRO_CB(p)->same_flow)
507 continue;
508
509 gh2 = (struct genevehdr *)(p->data + off_gnv);
510 if (gh->opt_len != gh2->opt_len ||
511 memcmp(gh, gh2, gh_len)) {
512 NAPI_GRO_CB(p)->same_flow = 0;
513 continue;
514 }
515 }
516
517 type = gh->proto_type;
518
519 rcu_read_lock();
520 ptype = gro_find_receive_by_type(type);
521 if (!ptype)
522 goto out_unlock;
523
524 skb_gro_pull(skb, gh_len);
525 skb_gro_postpull_rcsum(skb, gh, gh_len);
526 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
527 flush = 0;
528
529out_unlock:
530 rcu_read_unlock();
531out:
532 skb_gro_flush_final(skb, pp, flush);
533
534 return pp;
535}
536
537static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
538 int nhoff)
539{
540 struct genevehdr *gh;
541 struct packet_offload *ptype;
542 __be16 type;
543 int gh_len;
544 int err = -ENOSYS;
545
546 gh = (struct genevehdr *)(skb->data + nhoff);
547 gh_len = geneve_hlen(gh);
548 type = gh->proto_type;
549
550 rcu_read_lock();
551 ptype = gro_find_complete_by_type(type);
552 if (ptype)
553 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
554
555 rcu_read_unlock();
556
557 skb_set_inner_mac_header(skb, nhoff + gh_len);
558
559 return err;
560}
561
562/* Create new listen socket if needed */
563static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
564 bool ipv6, bool ipv6_rx_csum)
565{
566 struct geneve_net *gn = net_generic(net, geneve_net_id);
567 struct geneve_sock *gs;
568 struct socket *sock;
569 struct udp_tunnel_sock_cfg tunnel_cfg;
570 int h;
571
572 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
573 if (!gs)
574 return ERR_PTR(-ENOMEM);
575
576 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
577 if (IS_ERR(sock)) {
578 kfree(gs);
579 return ERR_CAST(sock);
580 }
581
582 gs->sock = sock;
583 gs->refcnt = 1;
584 for (h = 0; h < VNI_HASH_SIZE; ++h)
585 INIT_HLIST_HEAD(&gs->vni_list[h]);
586
587 /* Initialize the geneve udp offloads structure */
588 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
589
590 /* Mark socket as an encapsulation socket */
591 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
592 tunnel_cfg.sk_user_data = gs;
593 tunnel_cfg.encap_type = 1;
594 tunnel_cfg.gro_receive = geneve_gro_receive;
595 tunnel_cfg.gro_complete = geneve_gro_complete;
596 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
597 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
598 tunnel_cfg.encap_destroy = NULL;
599 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
600 list_add(&gs->list, &gn->sock_list);
601 return gs;
602}
603
604static void __geneve_sock_release(struct geneve_sock *gs)
605{
606 if (!gs || --gs->refcnt)
607 return;
608
609 list_del(&gs->list);
610 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
611 udp_tunnel_sock_release(gs->sock);
612 kfree_rcu(gs, rcu);
613}
614
615static void geneve_sock_release(struct geneve_dev *geneve)
616{
617 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
618#if IS_ENABLED(CONFIG_IPV6)
619 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
620
621 rcu_assign_pointer(geneve->sock6, NULL);
622#endif
623
624 rcu_assign_pointer(geneve->sock4, NULL);
625 synchronize_net();
626
627 __geneve_sock_release(gs4);
628#if IS_ENABLED(CONFIG_IPV6)
629 __geneve_sock_release(gs6);
630#endif
631}
632
633static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
634 sa_family_t family,
635 __be16 dst_port)
636{
637 struct geneve_sock *gs;
638
639 list_for_each_entry(gs, &gn->sock_list, list) {
640 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
641 geneve_get_sk_family(gs) == family) {
642 return gs;
643 }
644 }
645 return NULL;
646}
647
648static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
649{
650 struct net *net = geneve->net;
651 struct geneve_net *gn = net_generic(net, geneve_net_id);
652 struct geneve_dev_node *node;
653 struct geneve_sock *gs;
654 __u8 vni[3];
655 __u32 hash;
656
657 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
658 if (gs) {
659 gs->refcnt++;
660 goto out;
661 }
662
663 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
664 geneve->use_udp6_rx_checksums);
665 if (IS_ERR(gs))
666 return PTR_ERR(gs);
667
668out:
669 gs->collect_md = geneve->collect_md;
670#if IS_ENABLED(CONFIG_IPV6)
671 if (ipv6) {
672 rcu_assign_pointer(geneve->sock6, gs);
673 node = &geneve->hlist6;
674 } else
675#endif
676 {
677 rcu_assign_pointer(geneve->sock4, gs);
678 node = &geneve->hlist4;
679 }
680 node->geneve = geneve;
681
682 tunnel_id_to_vni(geneve->info.key.tun_id, vni);
683 hash = geneve_net_vni_hash(vni);
684 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
685 return 0;
686}
687
688static int geneve_open(struct net_device *dev)
689{
690 struct geneve_dev *geneve = netdev_priv(dev);
691 bool metadata = geneve->collect_md;
692 bool ipv4, ipv6;
693 int ret = 0;
694
695 ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
696 ipv4 = !ipv6 || metadata;
697#if IS_ENABLED(CONFIG_IPV6)
698 if (ipv6) {
699 ret = geneve_sock_add(geneve, true);
700 if (ret < 0 && ret != -EAFNOSUPPORT)
701 ipv4 = false;
702 }
703#endif
704 if (ipv4)
705 ret = geneve_sock_add(geneve, false);
706 if (ret < 0)
707 geneve_sock_release(geneve);
708
709 return ret;
710}
711
712static int geneve_stop(struct net_device *dev)
713{
714 struct geneve_dev *geneve = netdev_priv(dev);
715
716 hlist_del_init_rcu(&geneve->hlist4.hlist);
717#if IS_ENABLED(CONFIG_IPV6)
718 hlist_del_init_rcu(&geneve->hlist6.hlist);
719#endif
720 geneve_sock_release(geneve);
721 return 0;
722}
723
724static void geneve_build_header(struct genevehdr *geneveh,
725 const struct ip_tunnel_info *info)
726{
727 geneveh->ver = GENEVE_VER;
728 geneveh->opt_len = info->options_len / 4;
729 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
730 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
731 geneveh->rsvd1 = 0;
732 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
733 geneveh->proto_type = htons(ETH_P_TEB);
734 geneveh->rsvd2 = 0;
735
736 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
737 ip_tunnel_info_opts_get(geneveh->options, info);
738}
739
740static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
741 const struct ip_tunnel_info *info,
742 bool xnet, int ip_hdr_len)
743{
744 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
745 struct genevehdr *gnvh;
746 int min_headroom;
747 int err;
748
749 skb_reset_mac_header(skb);
750 skb_scrub_packet(skb, xnet);
751
752 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
753 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
754 err = skb_cow_head(skb, min_headroom);
755 if (unlikely(err))
756 goto free_dst;
757
758 err = udp_tunnel_handle_offloads(skb, udp_sum);
759 if (err)
760 goto free_dst;
761
762 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
763 geneve_build_header(gnvh, info);
764 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
765 return 0;
766
767free_dst:
768 dst_release(dst);
769 return err;
770}
771
772static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
773 struct net_device *dev,
774 struct geneve_sock *gs4,
775 struct flowi4 *fl4,
776 const struct ip_tunnel_info *info)
777{
778 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
779 struct geneve_dev *geneve = netdev_priv(dev);
780 struct dst_cache *dst_cache;
781 struct rtable *rt = NULL;
782 __u8 tos;
783
784 if (!gs4)
785 return ERR_PTR(-EIO);
786
787 memset(fl4, 0, sizeof(*fl4));
788 fl4->flowi4_mark = skb->mark;
789 fl4->flowi4_proto = IPPROTO_UDP;
790 fl4->daddr = info->key.u.ipv4.dst;
791 fl4->saddr = info->key.u.ipv4.src;
792
793 tos = info->key.tos;
794 if ((tos == 1) && !geneve->collect_md) {
795 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
796 use_cache = false;
797 }
798 fl4->flowi4_tos = RT_TOS(tos);
799
800 dst_cache = (struct dst_cache *)&info->dst_cache;
801 if (use_cache) {
802 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
803 if (rt)
804 return rt;
805 }
806 rt = ip_route_output_key(geneve->net, fl4);
807 if (IS_ERR(rt)) {
808 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
809 return ERR_PTR(-ENETUNREACH);
810 }
811 if (rt->dst.dev == dev) { /* is this necessary? */
812 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
813 ip_rt_put(rt);
814 return ERR_PTR(-ELOOP);
815 }
816 if (use_cache)
817 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
818 return rt;
819}
820
821#if IS_ENABLED(CONFIG_IPV6)
822static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
823 struct net_device *dev,
824 struct geneve_sock *gs6,
825 struct flowi6 *fl6,
826 const struct ip_tunnel_info *info)
827{
828 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
829 struct geneve_dev *geneve = netdev_priv(dev);
830 struct dst_entry *dst = NULL;
831 struct dst_cache *dst_cache;
832 __u8 prio;
833
834 if (!gs6)
835 return ERR_PTR(-EIO);
836
837 memset(fl6, 0, sizeof(*fl6));
838 fl6->flowi6_mark = skb->mark;
839 fl6->flowi6_proto = IPPROTO_UDP;
840 fl6->daddr = info->key.u.ipv6.dst;
841 fl6->saddr = info->key.u.ipv6.src;
842 prio = info->key.tos;
843 if ((prio == 1) && !geneve->collect_md) {
844 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
845 use_cache = false;
846 }
847
848 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
849 info->key.label);
850 dst_cache = (struct dst_cache *)&info->dst_cache;
851 if (use_cache) {
852 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
853 if (dst)
854 return dst;
855 }
856 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
857 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
858 return ERR_PTR(-ENETUNREACH);
859 }
860 if (dst->dev == dev) { /* is this necessary? */
861 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
862 dst_release(dst);
863 return ERR_PTR(-ELOOP);
864 }
865
866 if (use_cache)
867 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
868 return dst;
869}
870#endif
871
872static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
873 struct geneve_dev *geneve,
874 const struct ip_tunnel_info *info)
875{
876 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
877 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
878 const struct ip_tunnel_key *key = &info->key;
879 struct rtable *rt;
880 struct flowi4 fl4;
881 __u8 tos, ttl;
882 __be16 df = 0;
883 __be16 sport;
884 int err;
885
886 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
887 if (IS_ERR(rt))
888 return PTR_ERR(rt);
889
890 skb_tunnel_check_pmtu(skb, &rt->dst,
891 GENEVE_IPV4_HLEN + info->options_len);
892
893 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
894 if (geneve->collect_md) {
895 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
896 ttl = key->ttl;
897
898 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
899 } else {
900 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
901 if (geneve->ttl_inherit)
902 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
903 else
904 ttl = key->ttl;
905 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
906
907 if (geneve->df == GENEVE_DF_SET) {
908 df = htons(IP_DF);
909 } else if (geneve->df == GENEVE_DF_INHERIT) {
910 struct ethhdr *eth = eth_hdr(skb);
911
912 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
913 df = htons(IP_DF);
914 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
915 struct iphdr *iph = ip_hdr(skb);
916
917 if (iph->frag_off & htons(IP_DF))
918 df = htons(IP_DF);
919 }
920 }
921 }
922
923 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
924 if (unlikely(err))
925 return err;
926
927 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
928 tos, ttl, df, sport, geneve->info.key.tp_dst,
929 !net_eq(geneve->net, dev_net(geneve->dev)),
930 !(info->key.tun_flags & TUNNEL_CSUM));
931 return 0;
932}
933
934#if IS_ENABLED(CONFIG_IPV6)
935static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
936 struct geneve_dev *geneve,
937 const struct ip_tunnel_info *info)
938{
939 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
940 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
941 const struct ip_tunnel_key *key = &info->key;
942 struct dst_entry *dst = NULL;
943 struct flowi6 fl6;
944 __u8 prio, ttl;
945 __be16 sport;
946 int err;
947
948 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
949 if (IS_ERR(dst))
950 return PTR_ERR(dst);
951
952 skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
953
954 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
955 if (geneve->collect_md) {
956 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
957 ttl = key->ttl;
958 } else {
959 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
960 ip_hdr(skb), skb);
961 if (geneve->ttl_inherit)
962 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
963 else
964 ttl = key->ttl;
965 ttl = ttl ? : ip6_dst_hoplimit(dst);
966 }
967 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
968 if (unlikely(err))
969 return err;
970
971 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
972 &fl6.saddr, &fl6.daddr, prio, ttl,
973 info->key.label, sport, geneve->info.key.tp_dst,
974 !(info->key.tun_flags & TUNNEL_CSUM));
975 return 0;
976}
977#endif
978
979static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
980{
981 struct geneve_dev *geneve = netdev_priv(dev);
982 struct ip_tunnel_info *info = NULL;
983 int err;
984
985 if (geneve->collect_md) {
986 info = skb_tunnel_info(skb);
987 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
988 err = -EINVAL;
989 netdev_dbg(dev, "no tunnel metadata\n");
990 goto tx_error;
991 }
992 } else {
993 info = &geneve->info;
994 }
995
996 rcu_read_lock();
997#if IS_ENABLED(CONFIG_IPV6)
998 if (info->mode & IP_TUNNEL_INFO_IPV6)
999 err = geneve6_xmit_skb(skb, dev, geneve, info);
1000 else
1001#endif
1002 err = geneve_xmit_skb(skb, dev, geneve, info);
1003 rcu_read_unlock();
1004
1005 if (likely(!err))
1006 return NETDEV_TX_OK;
1007tx_error:
1008 dev_kfree_skb(skb);
1009
1010 if (err == -ELOOP)
1011 dev->stats.collisions++;
1012 else if (err == -ENETUNREACH)
1013 dev->stats.tx_carrier_errors++;
1014
1015 dev->stats.tx_errors++;
1016 return NETDEV_TX_OK;
1017}
1018
1019static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1020{
1021 if (new_mtu > dev->max_mtu)
1022 new_mtu = dev->max_mtu;
1023 else if (new_mtu < dev->min_mtu)
1024 new_mtu = dev->min_mtu;
1025
1026 dev->mtu = new_mtu;
1027 return 0;
1028}
1029
1030static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1031{
1032 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1033 struct geneve_dev *geneve = netdev_priv(dev);
1034
1035 if (ip_tunnel_info_af(info) == AF_INET) {
1036 struct rtable *rt;
1037 struct flowi4 fl4;
1038 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1039
1040 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
1041 if (IS_ERR(rt))
1042 return PTR_ERR(rt);
1043
1044 ip_rt_put(rt);
1045 info->key.u.ipv4.src = fl4.saddr;
1046#if IS_ENABLED(CONFIG_IPV6)
1047 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1048 struct dst_entry *dst;
1049 struct flowi6 fl6;
1050 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1051
1052 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
1053 if (IS_ERR(dst))
1054 return PTR_ERR(dst);
1055
1056 dst_release(dst);
1057 info->key.u.ipv6.src = fl6.saddr;
1058#endif
1059 } else {
1060 return -EINVAL;
1061 }
1062
1063 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1064 1, USHRT_MAX, true);
1065 info->key.tp_dst = geneve->info.key.tp_dst;
1066 return 0;
1067}
1068
1069static const struct net_device_ops geneve_netdev_ops = {
1070 .ndo_init = geneve_init,
1071 .ndo_uninit = geneve_uninit,
1072 .ndo_open = geneve_open,
1073 .ndo_stop = geneve_stop,
1074 .ndo_start_xmit = geneve_xmit,
1075 .ndo_get_stats64 = ip_tunnel_get_stats64,
1076 .ndo_change_mtu = geneve_change_mtu,
1077 .ndo_validate_addr = eth_validate_addr,
1078 .ndo_set_mac_address = eth_mac_addr,
1079 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1080};
1081
1082static void geneve_get_drvinfo(struct net_device *dev,
1083 struct ethtool_drvinfo *drvinfo)
1084{
1085 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1086 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1087}
1088
1089static const struct ethtool_ops geneve_ethtool_ops = {
1090 .get_drvinfo = geneve_get_drvinfo,
1091 .get_link = ethtool_op_get_link,
1092};
1093
1094/* Info for udev, that this is a virtual tunnel endpoint */
1095static struct device_type geneve_type = {
1096 .name = "geneve",
1097};
1098
1099/* Calls the ndo_udp_tunnel_add of the caller in order to
1100 * supply the listening GENEVE udp ports. Callers are expected
1101 * to implement the ndo_udp_tunnel_add.
1102 */
1103static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1104{
1105 struct net *net = dev_net(dev);
1106 struct geneve_net *gn = net_generic(net, geneve_net_id);
1107 struct geneve_sock *gs;
1108
1109 rcu_read_lock();
1110 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1111 if (push) {
1112 udp_tunnel_push_rx_port(dev, gs->sock,
1113 UDP_TUNNEL_TYPE_GENEVE);
1114 } else {
1115 udp_tunnel_drop_rx_port(dev, gs->sock,
1116 UDP_TUNNEL_TYPE_GENEVE);
1117 }
1118 }
1119 rcu_read_unlock();
1120}
1121
1122/* Initialize the device structure. */
1123static void geneve_setup(struct net_device *dev)
1124{
1125 ether_setup(dev);
1126
1127 dev->netdev_ops = &geneve_netdev_ops;
1128 dev->ethtool_ops = &geneve_ethtool_ops;
1129 dev->needs_free_netdev = true;
1130
1131 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1132
1133 dev->features |= NETIF_F_LLTX;
1134 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1135 dev->features |= NETIF_F_RXCSUM;
1136 dev->features |= NETIF_F_GSO_SOFTWARE;
1137
1138 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1139 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1140
1141 /* MTU range: 68 - (something less than 65535) */
1142 dev->min_mtu = ETH_MIN_MTU;
1143 /* The max_mtu calculation does not take account of GENEVE
1144 * options, to avoid excluding potentially valid
1145 * configurations. This will be further reduced by IPvX hdr size.
1146 */
1147 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1148
1149 netif_keep_dst(dev);
1150 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1151 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1152 eth_hw_addr_random(dev);
1153}
1154
1155static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1156 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1157 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1158 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1159 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1160 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1161 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1162 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1163 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1164 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1165 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1166 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1167 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1168 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1169};
1170
1171static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1172 struct netlink_ext_ack *extack)
1173{
1174 if (tb[IFLA_ADDRESS]) {
1175 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1176 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1177 "Provided link layer address is not Ethernet");
1178 return -EINVAL;
1179 }
1180
1181 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1182 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1183 "Provided Ethernet address is not unicast");
1184 return -EADDRNOTAVAIL;
1185 }
1186 }
1187
1188 if (!data) {
1189 NL_SET_ERR_MSG(extack,
1190 "Not enough attributes provided to perform the operation");
1191 return -EINVAL;
1192 }
1193
1194 if (data[IFLA_GENEVE_ID]) {
1195 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1196
1197 if (vni >= GENEVE_N_VID) {
1198 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1199 "Geneve ID must be lower than 16777216");
1200 return -ERANGE;
1201 }
1202 }
1203
1204 if (data[IFLA_GENEVE_DF]) {
1205 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1206
1207 if (df < 0 || df > GENEVE_DF_MAX) {
1208 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_GENEVE_DF],
1209 "Invalid DF attribute");
1210 return -EINVAL;
1211 }
1212 }
1213
1214 return 0;
1215}
1216
1217static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1218 const struct ip_tunnel_info *info,
1219 bool *tun_on_same_port,
1220 bool *tun_collect_md)
1221{
1222 struct geneve_dev *geneve, *t = NULL;
1223
1224 *tun_on_same_port = false;
1225 *tun_collect_md = false;
1226 list_for_each_entry(geneve, &gn->geneve_list, next) {
1227 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1228 *tun_collect_md = geneve->collect_md;
1229 *tun_on_same_port = true;
1230 }
1231 if (info->key.tun_id == geneve->info.key.tun_id &&
1232 info->key.tp_dst == geneve->info.key.tp_dst &&
1233 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1234 t = geneve;
1235 }
1236 return t;
1237}
1238
1239static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1240{
1241 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1242 info->key.ttl || info->key.label || info->key.tp_src ||
1243 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1244}
1245
1246static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1247 struct ip_tunnel_info *b)
1248{
1249 if (ip_tunnel_info_af(a) == AF_INET)
1250 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1251 else
1252 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1253}
1254
1255static int geneve_configure(struct net *net, struct net_device *dev,
1256 struct netlink_ext_ack *extack,
1257 const struct ip_tunnel_info *info,
1258 bool metadata, bool ipv6_rx_csum,
1259 bool ttl_inherit, enum ifla_geneve_df df)
1260{
1261 struct geneve_net *gn = net_generic(net, geneve_net_id);
1262 struct geneve_dev *t, *geneve = netdev_priv(dev);
1263 bool tun_collect_md, tun_on_same_port;
1264 int err, encap_len;
1265
1266 if (metadata && !is_tnl_info_zero(info)) {
1267 NL_SET_ERR_MSG(extack,
1268 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1269 return -EINVAL;
1270 }
1271
1272 geneve->net = net;
1273 geneve->dev = dev;
1274
1275 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1276 if (t)
1277 return -EBUSY;
1278
1279 /* make enough headroom for basic scenario */
1280 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1281 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1282 encap_len += sizeof(struct iphdr);
1283 dev->max_mtu -= sizeof(struct iphdr);
1284 } else {
1285 encap_len += sizeof(struct ipv6hdr);
1286 dev->max_mtu -= sizeof(struct ipv6hdr);
1287 }
1288 dev->needed_headroom = encap_len + ETH_HLEN;
1289
1290 if (metadata) {
1291 if (tun_on_same_port) {
1292 NL_SET_ERR_MSG(extack,
1293 "There can be only one externally controlled device on a destination port");
1294 return -EPERM;
1295 }
1296 } else {
1297 if (tun_collect_md) {
1298 NL_SET_ERR_MSG(extack,
1299 "There already exists an externally controlled device on this destination port");
1300 return -EPERM;
1301 }
1302 }
1303
1304 dst_cache_reset(&geneve->info.dst_cache);
1305 geneve->info = *info;
1306 geneve->collect_md = metadata;
1307 geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1308 geneve->ttl_inherit = ttl_inherit;
1309 geneve->df = df;
1310
1311 err = register_netdevice(dev);
1312 if (err)
1313 return err;
1314
1315 list_add(&geneve->next, &gn->geneve_list);
1316 return 0;
1317}
1318
1319static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1320{
1321 memset(info, 0, sizeof(*info));
1322 info->key.tp_dst = htons(dst_port);
1323}
1324
1325static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1326 struct netlink_ext_ack *extack,
1327 struct ip_tunnel_info *info, bool *metadata,
1328 bool *use_udp6_rx_checksums, bool *ttl_inherit,
1329 enum ifla_geneve_df *df, bool changelink)
1330{
1331 int attrtype;
1332
1333 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1334 NL_SET_ERR_MSG(extack,
1335 "Cannot specify both IPv4 and IPv6 Remote addresses");
1336 return -EINVAL;
1337 }
1338
1339 if (data[IFLA_GENEVE_REMOTE]) {
1340 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1341 attrtype = IFLA_GENEVE_REMOTE;
1342 goto change_notsup;
1343 }
1344
1345 info->key.u.ipv4.dst =
1346 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1347
1348 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1349 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1350 "Remote IPv4 address cannot be Multicast");
1351 return -EINVAL;
1352 }
1353 }
1354
1355 if (data[IFLA_GENEVE_REMOTE6]) {
1356#if IS_ENABLED(CONFIG_IPV6)
1357 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1358 attrtype = IFLA_GENEVE_REMOTE6;
1359 goto change_notsup;
1360 }
1361
1362 info->mode = IP_TUNNEL_INFO_IPV6;
1363 info->key.u.ipv6.dst =
1364 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1365
1366 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1367 IPV6_ADDR_LINKLOCAL) {
1368 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1369 "Remote IPv6 address cannot be link-local");
1370 return -EINVAL;
1371 }
1372 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1373 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1374 "Remote IPv6 address cannot be Multicast");
1375 return -EINVAL;
1376 }
1377 info->key.tun_flags |= TUNNEL_CSUM;
1378 *use_udp6_rx_checksums = true;
1379#else
1380 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1381 "IPv6 support not enabled in the kernel");
1382 return -EPFNOSUPPORT;
1383#endif
1384 }
1385
1386 if (data[IFLA_GENEVE_ID]) {
1387 __u32 vni;
1388 __u8 tvni[3];
1389 __be64 tunid;
1390
1391 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1392 tvni[0] = (vni & 0x00ff0000) >> 16;
1393 tvni[1] = (vni & 0x0000ff00) >> 8;
1394 tvni[2] = vni & 0x000000ff;
1395
1396 tunid = vni_to_tunnel_id(tvni);
1397 if (changelink && (tunid != info->key.tun_id)) {
1398 attrtype = IFLA_GENEVE_ID;
1399 goto change_notsup;
1400 }
1401 info->key.tun_id = tunid;
1402 }
1403
1404 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1405 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1406 *ttl_inherit = true;
1407 else
1408 *ttl_inherit = false;
1409 } else if (data[IFLA_GENEVE_TTL]) {
1410 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1411 *ttl_inherit = false;
1412 }
1413
1414 if (data[IFLA_GENEVE_TOS])
1415 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1416
1417 if (data[IFLA_GENEVE_DF])
1418 *df = nla_get_u8(data[IFLA_GENEVE_DF]);
1419
1420 if (data[IFLA_GENEVE_LABEL]) {
1421 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1422 IPV6_FLOWLABEL_MASK;
1423 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1424 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1425 "Label attribute only applies for IPv6 Geneve devices");
1426 return -EINVAL;
1427 }
1428 }
1429
1430 if (data[IFLA_GENEVE_PORT]) {
1431 if (changelink) {
1432 attrtype = IFLA_GENEVE_PORT;
1433 goto change_notsup;
1434 }
1435 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1436 }
1437
1438 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1439 if (changelink) {
1440 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1441 goto change_notsup;
1442 }
1443 *metadata = true;
1444 }
1445
1446 if (data[IFLA_GENEVE_UDP_CSUM]) {
1447 if (changelink) {
1448 attrtype = IFLA_GENEVE_UDP_CSUM;
1449 goto change_notsup;
1450 }
1451 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1452 info->key.tun_flags |= TUNNEL_CSUM;
1453 }
1454
1455 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1456#if IS_ENABLED(CONFIG_IPV6)
1457 if (changelink) {
1458 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1459 goto change_notsup;
1460 }
1461 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1462 info->key.tun_flags &= ~TUNNEL_CSUM;
1463#else
1464 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1465 "IPv6 support not enabled in the kernel");
1466 return -EPFNOSUPPORT;
1467#endif
1468 }
1469
1470 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1471#if IS_ENABLED(CONFIG_IPV6)
1472 if (changelink) {
1473 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1474 goto change_notsup;
1475 }
1476 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1477 *use_udp6_rx_checksums = false;
1478#else
1479 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1480 "IPv6 support not enabled in the kernel");
1481 return -EPFNOSUPPORT;
1482#endif
1483 }
1484
1485 return 0;
1486change_notsup:
1487 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1488 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1489 return -EOPNOTSUPP;
1490}
1491
1492static void geneve_link_config(struct net_device *dev,
1493 struct ip_tunnel_info *info, struct nlattr *tb[])
1494{
1495 struct geneve_dev *geneve = netdev_priv(dev);
1496 int ldev_mtu = 0;
1497
1498 if (tb[IFLA_MTU]) {
1499 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1500 return;
1501 }
1502
1503 switch (ip_tunnel_info_af(info)) {
1504 case AF_INET: {
1505 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1506 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1507
1508 if (!IS_ERR(rt) && rt->dst.dev) {
1509 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1510 ip_rt_put(rt);
1511 }
1512 break;
1513 }
1514#if IS_ENABLED(CONFIG_IPV6)
1515 case AF_INET6: {
1516 struct rt6_info *rt;
1517
1518 if (!__in6_dev_get(dev))
1519 break;
1520
1521 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1522 NULL, 0);
1523
1524 if (rt && rt->dst.dev)
1525 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1526 ip6_rt_put(rt);
1527 break;
1528 }
1529#endif
1530 }
1531
1532 if (ldev_mtu <= 0)
1533 return;
1534
1535 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1536}
1537
1538static int geneve_newlink(struct net *net, struct net_device *dev,
1539 struct nlattr *tb[], struct nlattr *data[],
1540 struct netlink_ext_ack *extack)
1541{
1542 enum ifla_geneve_df df = GENEVE_DF_UNSET;
1543 bool use_udp6_rx_checksums = false;
1544 struct ip_tunnel_info info;
1545 bool ttl_inherit = false;
1546 bool metadata = false;
1547 int err;
1548
1549 init_tnl_info(&info, GENEVE_UDP_PORT);
1550 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1551 &use_udp6_rx_checksums, &ttl_inherit, &df, false);
1552 if (err)
1553 return err;
1554
1555 err = geneve_configure(net, dev, extack, &info, metadata,
1556 use_udp6_rx_checksums, ttl_inherit, df);
1557 if (err)
1558 return err;
1559
1560 geneve_link_config(dev, &info, tb);
1561
1562 return 0;
1563}
1564
1565/* Quiesces the geneve device data path for both TX and RX.
1566 *
1567 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1568 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1569 * to complete for the existing set of in-flight packets to be transmitted,
1570 * then we would have quiesced the transmit data path. All the future packets
1571 * will get dropped until we unquiesce the data path.
1572 *
1573 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1574 * if we set that to NULL under RCU and wait for synchronize_net() to
1575 * complete, then we would have quiesced the receive data path.
1576 */
1577static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1578 struct geneve_sock **gs6)
1579{
1580 *gs4 = rtnl_dereference(geneve->sock4);
1581 rcu_assign_pointer(geneve->sock4, NULL);
1582 if (*gs4)
1583 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1584#if IS_ENABLED(CONFIG_IPV6)
1585 *gs6 = rtnl_dereference(geneve->sock6);
1586 rcu_assign_pointer(geneve->sock6, NULL);
1587 if (*gs6)
1588 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1589#else
1590 *gs6 = NULL;
1591#endif
1592 synchronize_net();
1593}
1594
1595/* Resumes the geneve device data path for both TX and RX. */
1596static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1597 struct geneve_sock __maybe_unused *gs6)
1598{
1599 rcu_assign_pointer(geneve->sock4, gs4);
1600 if (gs4)
1601 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1602#if IS_ENABLED(CONFIG_IPV6)
1603 rcu_assign_pointer(geneve->sock6, gs6);
1604 if (gs6)
1605 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1606#endif
1607 synchronize_net();
1608}
1609
1610static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1611 struct nlattr *data[],
1612 struct netlink_ext_ack *extack)
1613{
1614 struct geneve_dev *geneve = netdev_priv(dev);
1615 struct geneve_sock *gs4, *gs6;
1616 struct ip_tunnel_info info;
1617 bool metadata;
1618 bool use_udp6_rx_checksums;
1619 enum ifla_geneve_df df;
1620 bool ttl_inherit;
1621 int err;
1622
1623 /* If the geneve device is configured for metadata (or externally
1624 * controlled, for example, OVS), then nothing can be changed.
1625 */
1626 if (geneve->collect_md)
1627 return -EOPNOTSUPP;
1628
1629 /* Start with the existing info. */
1630 memcpy(&info, &geneve->info, sizeof(info));
1631 metadata = geneve->collect_md;
1632 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1633 ttl_inherit = geneve->ttl_inherit;
1634 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1635 &use_udp6_rx_checksums, &ttl_inherit, &df, true);
1636 if (err)
1637 return err;
1638
1639 if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1640 dst_cache_reset(&info.dst_cache);
1641 geneve_link_config(dev, &info, tb);
1642 }
1643
1644 geneve_quiesce(geneve, &gs4, &gs6);
1645 geneve->info = info;
1646 geneve->collect_md = metadata;
1647 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1648 geneve->ttl_inherit = ttl_inherit;
1649 geneve_unquiesce(geneve, gs4, gs6);
1650
1651 return 0;
1652}
1653
1654static void geneve_dellink(struct net_device *dev, struct list_head *head)
1655{
1656 struct geneve_dev *geneve = netdev_priv(dev);
1657
1658 list_del(&geneve->next);
1659 unregister_netdevice_queue(dev, head);
1660}
1661
1662static size_t geneve_get_size(const struct net_device *dev)
1663{
1664 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1665 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1666 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1667 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1668 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1669 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1670 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1671 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1672 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1673 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1674 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1675 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1676 0;
1677}
1678
1679static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1680{
1681 struct geneve_dev *geneve = netdev_priv(dev);
1682 struct ip_tunnel_info *info = &geneve->info;
1683 bool ttl_inherit = geneve->ttl_inherit;
1684 bool metadata = geneve->collect_md;
1685 __u8 tmp_vni[3];
1686 __u32 vni;
1687
1688 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1689 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1690 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1691 goto nla_put_failure;
1692
1693 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1694 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1695 info->key.u.ipv4.dst))
1696 goto nla_put_failure;
1697 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1698 !!(info->key.tun_flags & TUNNEL_CSUM)))
1699 goto nla_put_failure;
1700
1701#if IS_ENABLED(CONFIG_IPV6)
1702 } else if (!metadata) {
1703 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1704 &info->key.u.ipv6.dst))
1705 goto nla_put_failure;
1706 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1707 !(info->key.tun_flags & TUNNEL_CSUM)))
1708 goto nla_put_failure;
1709#endif
1710 }
1711
1712 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1713 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1714 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1715 goto nla_put_failure;
1716
1717 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->df))
1718 goto nla_put_failure;
1719
1720 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1721 goto nla_put_failure;
1722
1723 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1724 goto nla_put_failure;
1725
1726#if IS_ENABLED(CONFIG_IPV6)
1727 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1728 !geneve->use_udp6_rx_checksums))
1729 goto nla_put_failure;
1730#endif
1731
1732 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1733 goto nla_put_failure;
1734
1735 return 0;
1736
1737nla_put_failure:
1738 return -EMSGSIZE;
1739}
1740
1741static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1742 .kind = "geneve",
1743 .maxtype = IFLA_GENEVE_MAX,
1744 .policy = geneve_policy,
1745 .priv_size = sizeof(struct geneve_dev),
1746 .setup = geneve_setup,
1747 .validate = geneve_validate,
1748 .newlink = geneve_newlink,
1749 .changelink = geneve_changelink,
1750 .dellink = geneve_dellink,
1751 .get_size = geneve_get_size,
1752 .fill_info = geneve_fill_info,
1753};
1754
1755struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1756 u8 name_assign_type, u16 dst_port)
1757{
1758 struct nlattr *tb[IFLA_MAX + 1];
1759 struct ip_tunnel_info info;
1760 struct net_device *dev;
1761 LIST_HEAD(list_kill);
1762 int err;
1763
1764 memset(tb, 0, sizeof(tb));
1765 dev = rtnl_create_link(net, name, name_assign_type,
1766 &geneve_link_ops, tb, NULL);
1767 if (IS_ERR(dev))
1768 return dev;
1769
1770 init_tnl_info(&info, dst_port);
1771 err = geneve_configure(net, dev, NULL, &info,
1772 true, true, false, GENEVE_DF_UNSET);
1773 if (err) {
1774 free_netdev(dev);
1775 return ERR_PTR(err);
1776 }
1777
1778 /* openvswitch users expect packet sizes to be unrestricted,
1779 * so set the largest MTU we can.
1780 */
1781 err = geneve_change_mtu(dev, IP_MAX_MTU);
1782 if (err)
1783 goto err;
1784
1785 err = rtnl_configure_link(dev, NULL);
1786 if (err < 0)
1787 goto err;
1788
1789 return dev;
1790err:
1791 geneve_dellink(dev, &list_kill);
1792 unregister_netdevice_many(&list_kill);
1793 return ERR_PTR(err);
1794}
1795EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1796
1797static int geneve_netdevice_event(struct notifier_block *unused,
1798 unsigned long event, void *ptr)
1799{
1800 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1801
1802 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1803 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1804 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1805 } else if (event == NETDEV_UNREGISTER) {
1806 geneve_offload_rx_ports(dev, false);
1807 } else if (event == NETDEV_REGISTER) {
1808 geneve_offload_rx_ports(dev, true);
1809 }
1810
1811 return NOTIFY_DONE;
1812}
1813
1814static struct notifier_block geneve_notifier_block __read_mostly = {
1815 .notifier_call = geneve_netdevice_event,
1816};
1817
1818static __net_init int geneve_init_net(struct net *net)
1819{
1820 struct geneve_net *gn = net_generic(net, geneve_net_id);
1821
1822 INIT_LIST_HEAD(&gn->geneve_list);
1823 INIT_LIST_HEAD(&gn->sock_list);
1824 return 0;
1825}
1826
1827static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1828{
1829 struct geneve_net *gn = net_generic(net, geneve_net_id);
1830 struct geneve_dev *geneve, *next;
1831 struct net_device *dev, *aux;
1832
1833 /* gather any geneve devices that were moved into this ns */
1834 for_each_netdev_safe(net, dev, aux)
1835 if (dev->rtnl_link_ops == &geneve_link_ops)
1836 unregister_netdevice_queue(dev, head);
1837
1838 /* now gather any other geneve devices that were created in this ns */
1839 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1840 /* If geneve->dev is in the same netns, it was already added
1841 * to the list by the previous loop.
1842 */
1843 if (!net_eq(dev_net(geneve->dev), net))
1844 unregister_netdevice_queue(geneve->dev, head);
1845 }
1846
1847 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1848}
1849
1850static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1851{
1852 struct net *net;
1853 LIST_HEAD(list);
1854
1855 rtnl_lock();
1856 list_for_each_entry(net, net_list, exit_list)
1857 geneve_destroy_tunnels(net, &list);
1858
1859 /* unregister the devices gathered above */
1860 unregister_netdevice_many(&list);
1861 rtnl_unlock();
1862}
1863
1864static struct pernet_operations geneve_net_ops = {
1865 .init = geneve_init_net,
1866 .exit_batch = geneve_exit_batch_net,
1867 .id = &geneve_net_id,
1868 .size = sizeof(struct geneve_net),
1869};
1870
1871static int __init geneve_init_module(void)
1872{
1873 int rc;
1874
1875 rc = register_pernet_subsys(&geneve_net_ops);
1876 if (rc)
1877 goto out1;
1878
1879 rc = register_netdevice_notifier(&geneve_notifier_block);
1880 if (rc)
1881 goto out2;
1882
1883 rc = rtnl_link_register(&geneve_link_ops);
1884 if (rc)
1885 goto out3;
1886
1887 return 0;
1888out3:
1889 unregister_netdevice_notifier(&geneve_notifier_block);
1890out2:
1891 unregister_pernet_subsys(&geneve_net_ops);
1892out1:
1893 return rc;
1894}
1895late_initcall(geneve_init_module);
1896
1897static void __exit geneve_cleanup_module(void)
1898{
1899 rtnl_link_unregister(&geneve_link_ops);
1900 unregister_netdevice_notifier(&geneve_notifier_block);
1901 unregister_pernet_subsys(&geneve_net_ops);
1902}
1903module_exit(geneve_cleanup_module);
1904
1905MODULE_LICENSE("GPL");
1906MODULE_VERSION(GENEVE_NETDEV_VER);
1907MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1908MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1909MODULE_ALIAS_RTNL_LINK("geneve");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * GENEVE: Generic Network Virtualization Encapsulation
4 *
5 * Copyright (c) 2015 Red Hat, Inc.
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/kernel.h>
11#include <linux/module.h>
12#include <linux/etherdevice.h>
13#include <linux/hash.h>
14#include <net/ipv6_stubs.h>
15#include <net/dst_metadata.h>
16#include <net/gro_cells.h>
17#include <net/rtnetlink.h>
18#include <net/geneve.h>
19#include <net/protocol.h>
20
21#define GENEVE_NETDEV_VER "0.6"
22
23#define GENEVE_N_VID (1u << 24)
24#define GENEVE_VID_MASK (GENEVE_N_VID - 1)
25
26#define VNI_HASH_BITS 10
27#define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
28
29static bool log_ecn_error = true;
30module_param(log_ecn_error, bool, 0644);
31MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
32
33#define GENEVE_VER 0
34#define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
35#define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
36#define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
37
38/* per-network namespace private data for this module */
39struct geneve_net {
40 struct list_head geneve_list;
41 struct list_head sock_list;
42};
43
44static unsigned int geneve_net_id;
45
46struct geneve_dev_node {
47 struct hlist_node hlist;
48 struct geneve_dev *geneve;
49};
50
51struct geneve_config {
52 struct ip_tunnel_info info;
53 bool collect_md;
54 bool use_udp6_rx_checksums;
55 bool ttl_inherit;
56 enum ifla_geneve_df df;
57};
58
59/* Pseudo network device */
60struct geneve_dev {
61 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
62#if IS_ENABLED(CONFIG_IPV6)
63 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
64#endif
65 struct net *net; /* netns for packet i/o */
66 struct net_device *dev; /* netdev for geneve tunnel */
67 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
68#if IS_ENABLED(CONFIG_IPV6)
69 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
70#endif
71 struct list_head next; /* geneve's per namespace list */
72 struct gro_cells gro_cells;
73 struct geneve_config cfg;
74};
75
76struct geneve_sock {
77 bool collect_md;
78 struct list_head list;
79 struct socket *sock;
80 struct rcu_head rcu;
81 int refcnt;
82 struct hlist_head vni_list[VNI_HASH_SIZE];
83};
84
85static inline __u32 geneve_net_vni_hash(u8 vni[3])
86{
87 __u32 vnid;
88
89 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
90 return hash_32(vnid, VNI_HASH_BITS);
91}
92
93static __be64 vni_to_tunnel_id(const __u8 *vni)
94{
95#ifdef __BIG_ENDIAN
96 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
97#else
98 return (__force __be64)(((__force u64)vni[0] << 40) |
99 ((__force u64)vni[1] << 48) |
100 ((__force u64)vni[2] << 56));
101#endif
102}
103
104/* Convert 64 bit tunnel ID to 24 bit VNI. */
105static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
106{
107#ifdef __BIG_ENDIAN
108 vni[0] = (__force __u8)(tun_id >> 16);
109 vni[1] = (__force __u8)(tun_id >> 8);
110 vni[2] = (__force __u8)tun_id;
111#else
112 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
113 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
114 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
115#endif
116}
117
118static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
119{
120 return !memcmp(vni, &tun_id[5], 3);
121}
122
123static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
124{
125 return gs->sock->sk->sk_family;
126}
127
128static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
129 __be32 addr, u8 vni[])
130{
131 struct hlist_head *vni_list_head;
132 struct geneve_dev_node *node;
133 __u32 hash;
134
135 /* Find the device for this VNI */
136 hash = geneve_net_vni_hash(vni);
137 vni_list_head = &gs->vni_list[hash];
138 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
139 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
140 addr == node->geneve->cfg.info.key.u.ipv4.dst)
141 return node->geneve;
142 }
143 return NULL;
144}
145
146#if IS_ENABLED(CONFIG_IPV6)
147static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
148 struct in6_addr addr6, u8 vni[])
149{
150 struct hlist_head *vni_list_head;
151 struct geneve_dev_node *node;
152 __u32 hash;
153
154 /* Find the device for this VNI */
155 hash = geneve_net_vni_hash(vni);
156 vni_list_head = &gs->vni_list[hash];
157 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
158 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
159 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
160 return node->geneve;
161 }
162 return NULL;
163}
164#endif
165
166static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
167{
168 return (struct genevehdr *)(udp_hdr(skb) + 1);
169}
170
171static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
172 struct sk_buff *skb)
173{
174 static u8 zero_vni[3];
175 u8 *vni;
176
177 if (geneve_get_sk_family(gs) == AF_INET) {
178 struct iphdr *iph;
179 __be32 addr;
180
181 iph = ip_hdr(skb); /* outer IP header... */
182
183 if (gs->collect_md) {
184 vni = zero_vni;
185 addr = 0;
186 } else {
187 vni = geneve_hdr(skb)->vni;
188 addr = iph->saddr;
189 }
190
191 return geneve_lookup(gs, addr, vni);
192#if IS_ENABLED(CONFIG_IPV6)
193 } else if (geneve_get_sk_family(gs) == AF_INET6) {
194 static struct in6_addr zero_addr6;
195 struct ipv6hdr *ip6h;
196 struct in6_addr addr6;
197
198 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
199
200 if (gs->collect_md) {
201 vni = zero_vni;
202 addr6 = zero_addr6;
203 } else {
204 vni = geneve_hdr(skb)->vni;
205 addr6 = ip6h->saddr;
206 }
207
208 return geneve6_lookup(gs, addr6, vni);
209#endif
210 }
211 return NULL;
212}
213
214/* geneve receive/decap routine */
215static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
216 struct sk_buff *skb)
217{
218 struct genevehdr *gnvh = geneve_hdr(skb);
219 struct metadata_dst *tun_dst = NULL;
220 struct pcpu_sw_netstats *stats;
221 unsigned int len;
222 int err = 0;
223 void *oiph;
224
225 if (ip_tunnel_collect_metadata() || gs->collect_md) {
226 __be16 flags;
227
228 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
229 (gnvh->oam ? TUNNEL_OAM : 0) |
230 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
231
232 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
233 vni_to_tunnel_id(gnvh->vni),
234 gnvh->opt_len * 4);
235 if (!tun_dst) {
236 geneve->dev->stats.rx_dropped++;
237 goto drop;
238 }
239 /* Update tunnel dst according to Geneve options. */
240 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
241 gnvh->options, gnvh->opt_len * 4,
242 TUNNEL_GENEVE_OPT);
243 } else {
244 /* Drop packets w/ critical options,
245 * since we don't support any...
246 */
247 if (gnvh->critical) {
248 geneve->dev->stats.rx_frame_errors++;
249 geneve->dev->stats.rx_errors++;
250 goto drop;
251 }
252 }
253
254 skb_reset_mac_header(skb);
255 skb->protocol = eth_type_trans(skb, geneve->dev);
256 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
257
258 if (tun_dst)
259 skb_dst_set(skb, &tun_dst->dst);
260
261 /* Ignore packet loops (and multicast echo) */
262 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
263 geneve->dev->stats.rx_errors++;
264 goto drop;
265 }
266
267 oiph = skb_network_header(skb);
268 skb_reset_network_header(skb);
269
270 if (geneve_get_sk_family(gs) == AF_INET)
271 err = IP_ECN_decapsulate(oiph, skb);
272#if IS_ENABLED(CONFIG_IPV6)
273 else
274 err = IP6_ECN_decapsulate(oiph, skb);
275#endif
276
277 if (unlikely(err)) {
278 if (log_ecn_error) {
279 if (geneve_get_sk_family(gs) == AF_INET)
280 net_info_ratelimited("non-ECT from %pI4 "
281 "with TOS=%#x\n",
282 &((struct iphdr *)oiph)->saddr,
283 ((struct iphdr *)oiph)->tos);
284#if IS_ENABLED(CONFIG_IPV6)
285 else
286 net_info_ratelimited("non-ECT from %pI6\n",
287 &((struct ipv6hdr *)oiph)->saddr);
288#endif
289 }
290 if (err > 1) {
291 ++geneve->dev->stats.rx_frame_errors;
292 ++geneve->dev->stats.rx_errors;
293 goto drop;
294 }
295 }
296
297 len = skb->len;
298 err = gro_cells_receive(&geneve->gro_cells, skb);
299 if (likely(err == NET_RX_SUCCESS)) {
300 stats = this_cpu_ptr(geneve->dev->tstats);
301 u64_stats_update_begin(&stats->syncp);
302 stats->rx_packets++;
303 stats->rx_bytes += len;
304 u64_stats_update_end(&stats->syncp);
305 }
306 return;
307drop:
308 /* Consume bad packet */
309 kfree_skb(skb);
310}
311
312/* Setup stats when device is created */
313static int geneve_init(struct net_device *dev)
314{
315 struct geneve_dev *geneve = netdev_priv(dev);
316 int err;
317
318 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
319 if (!dev->tstats)
320 return -ENOMEM;
321
322 err = gro_cells_init(&geneve->gro_cells, dev);
323 if (err) {
324 free_percpu(dev->tstats);
325 return err;
326 }
327
328 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
329 if (err) {
330 free_percpu(dev->tstats);
331 gro_cells_destroy(&geneve->gro_cells);
332 return err;
333 }
334 return 0;
335}
336
337static void geneve_uninit(struct net_device *dev)
338{
339 struct geneve_dev *geneve = netdev_priv(dev);
340
341 dst_cache_destroy(&geneve->cfg.info.dst_cache);
342 gro_cells_destroy(&geneve->gro_cells);
343 free_percpu(dev->tstats);
344}
345
346/* Callback from net/ipv4/udp.c to receive packets */
347static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
348{
349 struct genevehdr *geneveh;
350 struct geneve_dev *geneve;
351 struct geneve_sock *gs;
352 int opts_len;
353
354 /* Need UDP and Geneve header to be present */
355 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
356 goto drop;
357
358 /* Return packets with reserved bits set */
359 geneveh = geneve_hdr(skb);
360 if (unlikely(geneveh->ver != GENEVE_VER))
361 goto drop;
362
363 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
364 goto drop;
365
366 gs = rcu_dereference_sk_user_data(sk);
367 if (!gs)
368 goto drop;
369
370 geneve = geneve_lookup_skb(gs, skb);
371 if (!geneve)
372 goto drop;
373
374 opts_len = geneveh->opt_len * 4;
375 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
376 htons(ETH_P_TEB),
377 !net_eq(geneve->net, dev_net(geneve->dev)))) {
378 geneve->dev->stats.rx_dropped++;
379 goto drop;
380 }
381
382 geneve_rx(geneve, gs, skb);
383 return 0;
384
385drop:
386 /* Consume bad packet */
387 kfree_skb(skb);
388 return 0;
389}
390
391/* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
392static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
393{
394 struct genevehdr *geneveh;
395 struct geneve_sock *gs;
396 u8 zero_vni[3] = { 0 };
397 u8 *vni = zero_vni;
398
399 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
400 return -EINVAL;
401
402 geneveh = geneve_hdr(skb);
403 if (geneveh->ver != GENEVE_VER)
404 return -EINVAL;
405
406 if (geneveh->proto_type != htons(ETH_P_TEB))
407 return -EINVAL;
408
409 gs = rcu_dereference_sk_user_data(sk);
410 if (!gs)
411 return -ENOENT;
412
413 if (geneve_get_sk_family(gs) == AF_INET) {
414 struct iphdr *iph = ip_hdr(skb);
415 __be32 addr4 = 0;
416
417 if (!gs->collect_md) {
418 vni = geneve_hdr(skb)->vni;
419 addr4 = iph->daddr;
420 }
421
422 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
423 }
424
425#if IS_ENABLED(CONFIG_IPV6)
426 if (geneve_get_sk_family(gs) == AF_INET6) {
427 struct ipv6hdr *ip6h = ipv6_hdr(skb);
428 struct in6_addr addr6;
429
430 memset(&addr6, 0, sizeof(struct in6_addr));
431
432 if (!gs->collect_md) {
433 vni = geneve_hdr(skb)->vni;
434 addr6 = ip6h->daddr;
435 }
436
437 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
438 }
439#endif
440
441 return -EPFNOSUPPORT;
442}
443
444static struct socket *geneve_create_sock(struct net *net, bool ipv6,
445 __be16 port, bool ipv6_rx_csum)
446{
447 struct socket *sock;
448 struct udp_port_cfg udp_conf;
449 int err;
450
451 memset(&udp_conf, 0, sizeof(udp_conf));
452
453 if (ipv6) {
454 udp_conf.family = AF_INET6;
455 udp_conf.ipv6_v6only = 1;
456 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
457 } else {
458 udp_conf.family = AF_INET;
459 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
460 }
461
462 udp_conf.local_udp_port = port;
463
464 /* Open UDP socket */
465 err = udp_sock_create(net, &udp_conf, &sock);
466 if (err < 0)
467 return ERR_PTR(err);
468
469 return sock;
470}
471
472static int geneve_hlen(struct genevehdr *gh)
473{
474 return sizeof(*gh) + gh->opt_len * 4;
475}
476
477static struct sk_buff *geneve_gro_receive(struct sock *sk,
478 struct list_head *head,
479 struct sk_buff *skb)
480{
481 struct sk_buff *pp = NULL;
482 struct sk_buff *p;
483 struct genevehdr *gh, *gh2;
484 unsigned int hlen, gh_len, off_gnv;
485 const struct packet_offload *ptype;
486 __be16 type;
487 int flush = 1;
488
489 off_gnv = skb_gro_offset(skb);
490 hlen = off_gnv + sizeof(*gh);
491 gh = skb_gro_header_fast(skb, off_gnv);
492 if (skb_gro_header_hard(skb, hlen)) {
493 gh = skb_gro_header_slow(skb, hlen, off_gnv);
494 if (unlikely(!gh))
495 goto out;
496 }
497
498 if (gh->ver != GENEVE_VER || gh->oam)
499 goto out;
500 gh_len = geneve_hlen(gh);
501
502 hlen = off_gnv + gh_len;
503 if (skb_gro_header_hard(skb, hlen)) {
504 gh = skb_gro_header_slow(skb, hlen, off_gnv);
505 if (unlikely(!gh))
506 goto out;
507 }
508
509 list_for_each_entry(p, head, list) {
510 if (!NAPI_GRO_CB(p)->same_flow)
511 continue;
512
513 gh2 = (struct genevehdr *)(p->data + off_gnv);
514 if (gh->opt_len != gh2->opt_len ||
515 memcmp(gh, gh2, gh_len)) {
516 NAPI_GRO_CB(p)->same_flow = 0;
517 continue;
518 }
519 }
520
521 type = gh->proto_type;
522
523 rcu_read_lock();
524 ptype = gro_find_receive_by_type(type);
525 if (!ptype)
526 goto out_unlock;
527
528 skb_gro_pull(skb, gh_len);
529 skb_gro_postpull_rcsum(skb, gh, gh_len);
530 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
531 flush = 0;
532
533out_unlock:
534 rcu_read_unlock();
535out:
536 skb_gro_flush_final(skb, pp, flush);
537
538 return pp;
539}
540
541static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
542 int nhoff)
543{
544 struct genevehdr *gh;
545 struct packet_offload *ptype;
546 __be16 type;
547 int gh_len;
548 int err = -ENOSYS;
549
550 gh = (struct genevehdr *)(skb->data + nhoff);
551 gh_len = geneve_hlen(gh);
552 type = gh->proto_type;
553
554 rcu_read_lock();
555 ptype = gro_find_complete_by_type(type);
556 if (ptype)
557 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
558
559 rcu_read_unlock();
560
561 skb_set_inner_mac_header(skb, nhoff + gh_len);
562
563 return err;
564}
565
566/* Create new listen socket if needed */
567static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
568 bool ipv6, bool ipv6_rx_csum)
569{
570 struct geneve_net *gn = net_generic(net, geneve_net_id);
571 struct geneve_sock *gs;
572 struct socket *sock;
573 struct udp_tunnel_sock_cfg tunnel_cfg;
574 int h;
575
576 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
577 if (!gs)
578 return ERR_PTR(-ENOMEM);
579
580 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
581 if (IS_ERR(sock)) {
582 kfree(gs);
583 return ERR_CAST(sock);
584 }
585
586 gs->sock = sock;
587 gs->refcnt = 1;
588 for (h = 0; h < VNI_HASH_SIZE; ++h)
589 INIT_HLIST_HEAD(&gs->vni_list[h]);
590
591 /* Initialize the geneve udp offloads structure */
592 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
593
594 /* Mark socket as an encapsulation socket */
595 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
596 tunnel_cfg.sk_user_data = gs;
597 tunnel_cfg.encap_type = 1;
598 tunnel_cfg.gro_receive = geneve_gro_receive;
599 tunnel_cfg.gro_complete = geneve_gro_complete;
600 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
601 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
602 tunnel_cfg.encap_destroy = NULL;
603 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
604 list_add(&gs->list, &gn->sock_list);
605 return gs;
606}
607
608static void __geneve_sock_release(struct geneve_sock *gs)
609{
610 if (!gs || --gs->refcnt)
611 return;
612
613 list_del(&gs->list);
614 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
615 udp_tunnel_sock_release(gs->sock);
616 kfree_rcu(gs, rcu);
617}
618
619static void geneve_sock_release(struct geneve_dev *geneve)
620{
621 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
622#if IS_ENABLED(CONFIG_IPV6)
623 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
624
625 rcu_assign_pointer(geneve->sock6, NULL);
626#endif
627
628 rcu_assign_pointer(geneve->sock4, NULL);
629 synchronize_net();
630
631 __geneve_sock_release(gs4);
632#if IS_ENABLED(CONFIG_IPV6)
633 __geneve_sock_release(gs6);
634#endif
635}
636
637static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
638 sa_family_t family,
639 __be16 dst_port)
640{
641 struct geneve_sock *gs;
642
643 list_for_each_entry(gs, &gn->sock_list, list) {
644 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
645 geneve_get_sk_family(gs) == family) {
646 return gs;
647 }
648 }
649 return NULL;
650}
651
652static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
653{
654 struct net *net = geneve->net;
655 struct geneve_net *gn = net_generic(net, geneve_net_id);
656 struct geneve_dev_node *node;
657 struct geneve_sock *gs;
658 __u8 vni[3];
659 __u32 hash;
660
661 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
662 if (gs) {
663 gs->refcnt++;
664 goto out;
665 }
666
667 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
668 geneve->cfg.use_udp6_rx_checksums);
669 if (IS_ERR(gs))
670 return PTR_ERR(gs);
671
672out:
673 gs->collect_md = geneve->cfg.collect_md;
674#if IS_ENABLED(CONFIG_IPV6)
675 if (ipv6) {
676 rcu_assign_pointer(geneve->sock6, gs);
677 node = &geneve->hlist6;
678 } else
679#endif
680 {
681 rcu_assign_pointer(geneve->sock4, gs);
682 node = &geneve->hlist4;
683 }
684 node->geneve = geneve;
685
686 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
687 hash = geneve_net_vni_hash(vni);
688 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
689 return 0;
690}
691
692static int geneve_open(struct net_device *dev)
693{
694 struct geneve_dev *geneve = netdev_priv(dev);
695 bool metadata = geneve->cfg.collect_md;
696 bool ipv4, ipv6;
697 int ret = 0;
698
699 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
700 ipv4 = !ipv6 || metadata;
701#if IS_ENABLED(CONFIG_IPV6)
702 if (ipv6) {
703 ret = geneve_sock_add(geneve, true);
704 if (ret < 0 && ret != -EAFNOSUPPORT)
705 ipv4 = false;
706 }
707#endif
708 if (ipv4)
709 ret = geneve_sock_add(geneve, false);
710 if (ret < 0)
711 geneve_sock_release(geneve);
712
713 return ret;
714}
715
716static int geneve_stop(struct net_device *dev)
717{
718 struct geneve_dev *geneve = netdev_priv(dev);
719
720 hlist_del_init_rcu(&geneve->hlist4.hlist);
721#if IS_ENABLED(CONFIG_IPV6)
722 hlist_del_init_rcu(&geneve->hlist6.hlist);
723#endif
724 geneve_sock_release(geneve);
725 return 0;
726}
727
728static void geneve_build_header(struct genevehdr *geneveh,
729 const struct ip_tunnel_info *info)
730{
731 geneveh->ver = GENEVE_VER;
732 geneveh->opt_len = info->options_len / 4;
733 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
734 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
735 geneveh->rsvd1 = 0;
736 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
737 geneveh->proto_type = htons(ETH_P_TEB);
738 geneveh->rsvd2 = 0;
739
740 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
741 ip_tunnel_info_opts_get(geneveh->options, info);
742}
743
744static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
745 const struct ip_tunnel_info *info,
746 bool xnet, int ip_hdr_len)
747{
748 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
749 struct genevehdr *gnvh;
750 int min_headroom;
751 int err;
752
753 skb_reset_mac_header(skb);
754 skb_scrub_packet(skb, xnet);
755
756 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
757 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
758 err = skb_cow_head(skb, min_headroom);
759 if (unlikely(err))
760 goto free_dst;
761
762 err = udp_tunnel_handle_offloads(skb, udp_sum);
763 if (err)
764 goto free_dst;
765
766 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
767 geneve_build_header(gnvh, info);
768 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
769 return 0;
770
771free_dst:
772 dst_release(dst);
773 return err;
774}
775
776static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
777 struct net_device *dev,
778 struct geneve_sock *gs4,
779 struct flowi4 *fl4,
780 const struct ip_tunnel_info *info,
781 __be16 dport, __be16 sport)
782{
783 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
784 struct geneve_dev *geneve = netdev_priv(dev);
785 struct dst_cache *dst_cache;
786 struct rtable *rt = NULL;
787 __u8 tos;
788
789 if (!gs4)
790 return ERR_PTR(-EIO);
791
792 memset(fl4, 0, sizeof(*fl4));
793 fl4->flowi4_mark = skb->mark;
794 fl4->flowi4_proto = IPPROTO_UDP;
795 fl4->daddr = info->key.u.ipv4.dst;
796 fl4->saddr = info->key.u.ipv4.src;
797 fl4->fl4_dport = dport;
798 fl4->fl4_sport = sport;
799
800 tos = info->key.tos;
801 if ((tos == 1) && !geneve->cfg.collect_md) {
802 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
803 use_cache = false;
804 }
805 fl4->flowi4_tos = RT_TOS(tos);
806
807 dst_cache = (struct dst_cache *)&info->dst_cache;
808 if (use_cache) {
809 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
810 if (rt)
811 return rt;
812 }
813 rt = ip_route_output_key(geneve->net, fl4);
814 if (IS_ERR(rt)) {
815 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
816 return ERR_PTR(-ENETUNREACH);
817 }
818 if (rt->dst.dev == dev) { /* is this necessary? */
819 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
820 ip_rt_put(rt);
821 return ERR_PTR(-ELOOP);
822 }
823 if (use_cache)
824 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
825 return rt;
826}
827
828#if IS_ENABLED(CONFIG_IPV6)
829static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
830 struct net_device *dev,
831 struct geneve_sock *gs6,
832 struct flowi6 *fl6,
833 const struct ip_tunnel_info *info,
834 __be16 dport, __be16 sport)
835{
836 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
837 struct geneve_dev *geneve = netdev_priv(dev);
838 struct dst_entry *dst = NULL;
839 struct dst_cache *dst_cache;
840 __u8 prio;
841
842 if (!gs6)
843 return ERR_PTR(-EIO);
844
845 memset(fl6, 0, sizeof(*fl6));
846 fl6->flowi6_mark = skb->mark;
847 fl6->flowi6_proto = IPPROTO_UDP;
848 fl6->daddr = info->key.u.ipv6.dst;
849 fl6->saddr = info->key.u.ipv6.src;
850 fl6->fl6_dport = dport;
851 fl6->fl6_sport = sport;
852
853 prio = info->key.tos;
854 if ((prio == 1) && !geneve->cfg.collect_md) {
855 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
856 use_cache = false;
857 }
858
859 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
860 info->key.label);
861 dst_cache = (struct dst_cache *)&info->dst_cache;
862 if (use_cache) {
863 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
864 if (dst)
865 return dst;
866 }
867 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
868 NULL);
869 if (IS_ERR(dst)) {
870 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
871 return ERR_PTR(-ENETUNREACH);
872 }
873 if (dst->dev == dev) { /* is this necessary? */
874 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
875 dst_release(dst);
876 return ERR_PTR(-ELOOP);
877 }
878
879 if (use_cache)
880 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
881 return dst;
882}
883#endif
884
885static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
886 struct geneve_dev *geneve,
887 const struct ip_tunnel_info *info)
888{
889 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
890 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
891 const struct ip_tunnel_key *key = &info->key;
892 struct rtable *rt;
893 struct flowi4 fl4;
894 __u8 tos, ttl;
895 __be16 df = 0;
896 __be16 sport;
897 int err;
898
899 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
900 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
901 geneve->cfg.info.key.tp_dst, sport);
902 if (IS_ERR(rt))
903 return PTR_ERR(rt);
904
905 err = skb_tunnel_check_pmtu(skb, &rt->dst,
906 GENEVE_IPV4_HLEN + info->options_len,
907 netif_is_any_bridge_port(dev));
908 if (err < 0) {
909 dst_release(&rt->dst);
910 return err;
911 } else if (err) {
912 struct ip_tunnel_info *info;
913
914 info = skb_tunnel_info(skb);
915 if (info) {
916 info->key.u.ipv4.dst = fl4.saddr;
917 info->key.u.ipv4.src = fl4.daddr;
918 }
919
920 if (!pskb_may_pull(skb, ETH_HLEN)) {
921 dst_release(&rt->dst);
922 return -EINVAL;
923 }
924
925 skb->protocol = eth_type_trans(skb, geneve->dev);
926 netif_rx(skb);
927 dst_release(&rt->dst);
928 return -EMSGSIZE;
929 }
930
931 if (geneve->cfg.collect_md) {
932 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
933 ttl = key->ttl;
934
935 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
936 } else {
937 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
938 if (geneve->cfg.ttl_inherit)
939 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
940 else
941 ttl = key->ttl;
942 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
943
944 if (geneve->cfg.df == GENEVE_DF_SET) {
945 df = htons(IP_DF);
946 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
947 struct ethhdr *eth = eth_hdr(skb);
948
949 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
950 df = htons(IP_DF);
951 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
952 struct iphdr *iph = ip_hdr(skb);
953
954 if (iph->frag_off & htons(IP_DF))
955 df = htons(IP_DF);
956 }
957 }
958 }
959
960 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
961 if (unlikely(err))
962 return err;
963
964 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
965 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
966 !net_eq(geneve->net, dev_net(geneve->dev)),
967 !(info->key.tun_flags & TUNNEL_CSUM));
968 return 0;
969}
970
971#if IS_ENABLED(CONFIG_IPV6)
972static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
973 struct geneve_dev *geneve,
974 const struct ip_tunnel_info *info)
975{
976 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
977 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
978 const struct ip_tunnel_key *key = &info->key;
979 struct dst_entry *dst = NULL;
980 struct flowi6 fl6;
981 __u8 prio, ttl;
982 __be16 sport;
983 int err;
984
985 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
986 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
987 geneve->cfg.info.key.tp_dst, sport);
988 if (IS_ERR(dst))
989 return PTR_ERR(dst);
990
991 err = skb_tunnel_check_pmtu(skb, dst,
992 GENEVE_IPV6_HLEN + info->options_len,
993 netif_is_any_bridge_port(dev));
994 if (err < 0) {
995 dst_release(dst);
996 return err;
997 } else if (err) {
998 struct ip_tunnel_info *info = skb_tunnel_info(skb);
999
1000 if (info) {
1001 info->key.u.ipv6.dst = fl6.saddr;
1002 info->key.u.ipv6.src = fl6.daddr;
1003 }
1004
1005 if (!pskb_may_pull(skb, ETH_HLEN)) {
1006 dst_release(dst);
1007 return -EINVAL;
1008 }
1009
1010 skb->protocol = eth_type_trans(skb, geneve->dev);
1011 netif_rx(skb);
1012 dst_release(dst);
1013 return -EMSGSIZE;
1014 }
1015
1016 if (geneve->cfg.collect_md) {
1017 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1018 ttl = key->ttl;
1019 } else {
1020 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1021 ip_hdr(skb), skb);
1022 if (geneve->cfg.ttl_inherit)
1023 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1024 else
1025 ttl = key->ttl;
1026 ttl = ttl ? : ip6_dst_hoplimit(dst);
1027 }
1028 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1029 if (unlikely(err))
1030 return err;
1031
1032 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1033 &fl6.saddr, &fl6.daddr, prio, ttl,
1034 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1035 !(info->key.tun_flags & TUNNEL_CSUM));
1036 return 0;
1037}
1038#endif
1039
1040static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1041{
1042 struct geneve_dev *geneve = netdev_priv(dev);
1043 struct ip_tunnel_info *info = NULL;
1044 int err;
1045
1046 if (geneve->cfg.collect_md) {
1047 info = skb_tunnel_info(skb);
1048 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1049 netdev_dbg(dev, "no tunnel metadata\n");
1050 dev_kfree_skb(skb);
1051 dev->stats.tx_dropped++;
1052 return NETDEV_TX_OK;
1053 }
1054 } else {
1055 info = &geneve->cfg.info;
1056 }
1057
1058 rcu_read_lock();
1059#if IS_ENABLED(CONFIG_IPV6)
1060 if (info->mode & IP_TUNNEL_INFO_IPV6)
1061 err = geneve6_xmit_skb(skb, dev, geneve, info);
1062 else
1063#endif
1064 err = geneve_xmit_skb(skb, dev, geneve, info);
1065 rcu_read_unlock();
1066
1067 if (likely(!err))
1068 return NETDEV_TX_OK;
1069
1070 if (err != -EMSGSIZE)
1071 dev_kfree_skb(skb);
1072
1073 if (err == -ELOOP)
1074 dev->stats.collisions++;
1075 else if (err == -ENETUNREACH)
1076 dev->stats.tx_carrier_errors++;
1077
1078 dev->stats.tx_errors++;
1079 return NETDEV_TX_OK;
1080}
1081
1082static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1083{
1084 if (new_mtu > dev->max_mtu)
1085 new_mtu = dev->max_mtu;
1086 else if (new_mtu < dev->min_mtu)
1087 new_mtu = dev->min_mtu;
1088
1089 dev->mtu = new_mtu;
1090 return 0;
1091}
1092
1093static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1094{
1095 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1096 struct geneve_dev *geneve = netdev_priv(dev);
1097 __be16 sport;
1098
1099 if (ip_tunnel_info_af(info) == AF_INET) {
1100 struct rtable *rt;
1101 struct flowi4 fl4;
1102
1103 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1104 sport = udp_flow_src_port(geneve->net, skb,
1105 1, USHRT_MAX, true);
1106
1107 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1108 geneve->cfg.info.key.tp_dst, sport);
1109 if (IS_ERR(rt))
1110 return PTR_ERR(rt);
1111
1112 ip_rt_put(rt);
1113 info->key.u.ipv4.src = fl4.saddr;
1114#if IS_ENABLED(CONFIG_IPV6)
1115 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1116 struct dst_entry *dst;
1117 struct flowi6 fl6;
1118
1119 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1120 sport = udp_flow_src_port(geneve->net, skb,
1121 1, USHRT_MAX, true);
1122
1123 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1124 geneve->cfg.info.key.tp_dst, sport);
1125 if (IS_ERR(dst))
1126 return PTR_ERR(dst);
1127
1128 dst_release(dst);
1129 info->key.u.ipv6.src = fl6.saddr;
1130#endif
1131 } else {
1132 return -EINVAL;
1133 }
1134
1135 info->key.tp_src = sport;
1136 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1137 return 0;
1138}
1139
1140static const struct net_device_ops geneve_netdev_ops = {
1141 .ndo_init = geneve_init,
1142 .ndo_uninit = geneve_uninit,
1143 .ndo_open = geneve_open,
1144 .ndo_stop = geneve_stop,
1145 .ndo_start_xmit = geneve_xmit,
1146 .ndo_get_stats64 = ip_tunnel_get_stats64,
1147 .ndo_change_mtu = geneve_change_mtu,
1148 .ndo_validate_addr = eth_validate_addr,
1149 .ndo_set_mac_address = eth_mac_addr,
1150 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1151};
1152
1153static void geneve_get_drvinfo(struct net_device *dev,
1154 struct ethtool_drvinfo *drvinfo)
1155{
1156 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1157 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1158}
1159
1160static const struct ethtool_ops geneve_ethtool_ops = {
1161 .get_drvinfo = geneve_get_drvinfo,
1162 .get_link = ethtool_op_get_link,
1163};
1164
1165/* Info for udev, that this is a virtual tunnel endpoint */
1166static struct device_type geneve_type = {
1167 .name = "geneve",
1168};
1169
1170/* Calls the ndo_udp_tunnel_add of the caller in order to
1171 * supply the listening GENEVE udp ports. Callers are expected
1172 * to implement the ndo_udp_tunnel_add.
1173 */
1174static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1175{
1176 struct net *net = dev_net(dev);
1177 struct geneve_net *gn = net_generic(net, geneve_net_id);
1178 struct geneve_sock *gs;
1179
1180 rcu_read_lock();
1181 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1182 if (push) {
1183 udp_tunnel_push_rx_port(dev, gs->sock,
1184 UDP_TUNNEL_TYPE_GENEVE);
1185 } else {
1186 udp_tunnel_drop_rx_port(dev, gs->sock,
1187 UDP_TUNNEL_TYPE_GENEVE);
1188 }
1189 }
1190 rcu_read_unlock();
1191}
1192
1193/* Initialize the device structure. */
1194static void geneve_setup(struct net_device *dev)
1195{
1196 ether_setup(dev);
1197
1198 dev->netdev_ops = &geneve_netdev_ops;
1199 dev->ethtool_ops = &geneve_ethtool_ops;
1200 dev->needs_free_netdev = true;
1201
1202 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1203
1204 dev->features |= NETIF_F_LLTX;
1205 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1206 dev->features |= NETIF_F_RXCSUM;
1207 dev->features |= NETIF_F_GSO_SOFTWARE;
1208
1209 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1210 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1211
1212 /* MTU range: 68 - (something less than 65535) */
1213 dev->min_mtu = ETH_MIN_MTU;
1214 /* The max_mtu calculation does not take account of GENEVE
1215 * options, to avoid excluding potentially valid
1216 * configurations. This will be further reduced by IPvX hdr size.
1217 */
1218 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1219
1220 netif_keep_dst(dev);
1221 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1222 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1223 eth_hw_addr_random(dev);
1224}
1225
1226static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1227 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1228 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1229 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1230 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1231 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1232 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1233 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1234 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1235 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1236 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1237 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1238 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1239 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1240};
1241
1242static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1243 struct netlink_ext_ack *extack)
1244{
1245 if (tb[IFLA_ADDRESS]) {
1246 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1247 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1248 "Provided link layer address is not Ethernet");
1249 return -EINVAL;
1250 }
1251
1252 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1253 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1254 "Provided Ethernet address is not unicast");
1255 return -EADDRNOTAVAIL;
1256 }
1257 }
1258
1259 if (!data) {
1260 NL_SET_ERR_MSG(extack,
1261 "Not enough attributes provided to perform the operation");
1262 return -EINVAL;
1263 }
1264
1265 if (data[IFLA_GENEVE_ID]) {
1266 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1267
1268 if (vni >= GENEVE_N_VID) {
1269 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1270 "Geneve ID must be lower than 16777216");
1271 return -ERANGE;
1272 }
1273 }
1274
1275 if (data[IFLA_GENEVE_DF]) {
1276 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1277
1278 if (df < 0 || df > GENEVE_DF_MAX) {
1279 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1280 "Invalid DF attribute");
1281 return -EINVAL;
1282 }
1283 }
1284
1285 return 0;
1286}
1287
1288static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1289 const struct ip_tunnel_info *info,
1290 bool *tun_on_same_port,
1291 bool *tun_collect_md)
1292{
1293 struct geneve_dev *geneve, *t = NULL;
1294
1295 *tun_on_same_port = false;
1296 *tun_collect_md = false;
1297 list_for_each_entry(geneve, &gn->geneve_list, next) {
1298 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1299 *tun_collect_md = geneve->cfg.collect_md;
1300 *tun_on_same_port = true;
1301 }
1302 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1303 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1304 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1305 t = geneve;
1306 }
1307 return t;
1308}
1309
1310static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1311{
1312 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1313 info->key.ttl || info->key.label || info->key.tp_src ||
1314 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1315}
1316
1317static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1318 struct ip_tunnel_info *b)
1319{
1320 if (ip_tunnel_info_af(a) == AF_INET)
1321 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1322 else
1323 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1324}
1325
1326static int geneve_configure(struct net *net, struct net_device *dev,
1327 struct netlink_ext_ack *extack,
1328 const struct geneve_config *cfg)
1329{
1330 struct geneve_net *gn = net_generic(net, geneve_net_id);
1331 struct geneve_dev *t, *geneve = netdev_priv(dev);
1332 const struct ip_tunnel_info *info = &cfg->info;
1333 bool tun_collect_md, tun_on_same_port;
1334 int err, encap_len;
1335
1336 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1337 NL_SET_ERR_MSG(extack,
1338 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1339 return -EINVAL;
1340 }
1341
1342 geneve->net = net;
1343 geneve->dev = dev;
1344
1345 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1346 if (t)
1347 return -EBUSY;
1348
1349 /* make enough headroom for basic scenario */
1350 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1351 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1352 encap_len += sizeof(struct iphdr);
1353 dev->max_mtu -= sizeof(struct iphdr);
1354 } else {
1355 encap_len += sizeof(struct ipv6hdr);
1356 dev->max_mtu -= sizeof(struct ipv6hdr);
1357 }
1358 dev->needed_headroom = encap_len + ETH_HLEN;
1359
1360 if (cfg->collect_md) {
1361 if (tun_on_same_port) {
1362 NL_SET_ERR_MSG(extack,
1363 "There can be only one externally controlled device on a destination port");
1364 return -EPERM;
1365 }
1366 } else {
1367 if (tun_collect_md) {
1368 NL_SET_ERR_MSG(extack,
1369 "There already exists an externally controlled device on this destination port");
1370 return -EPERM;
1371 }
1372 }
1373
1374 dst_cache_reset(&geneve->cfg.info.dst_cache);
1375 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1376
1377 err = register_netdevice(dev);
1378 if (err)
1379 return err;
1380
1381 list_add(&geneve->next, &gn->geneve_list);
1382 return 0;
1383}
1384
1385static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1386{
1387 memset(info, 0, sizeof(*info));
1388 info->key.tp_dst = htons(dst_port);
1389}
1390
1391static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1392 struct netlink_ext_ack *extack,
1393 struct geneve_config *cfg, bool changelink)
1394{
1395 struct ip_tunnel_info *info = &cfg->info;
1396 int attrtype;
1397
1398 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1399 NL_SET_ERR_MSG(extack,
1400 "Cannot specify both IPv4 and IPv6 Remote addresses");
1401 return -EINVAL;
1402 }
1403
1404 if (data[IFLA_GENEVE_REMOTE]) {
1405 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1406 attrtype = IFLA_GENEVE_REMOTE;
1407 goto change_notsup;
1408 }
1409
1410 info->key.u.ipv4.dst =
1411 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1412
1413 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1414 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1415 "Remote IPv4 address cannot be Multicast");
1416 return -EINVAL;
1417 }
1418 }
1419
1420 if (data[IFLA_GENEVE_REMOTE6]) {
1421#if IS_ENABLED(CONFIG_IPV6)
1422 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1423 attrtype = IFLA_GENEVE_REMOTE6;
1424 goto change_notsup;
1425 }
1426
1427 info->mode = IP_TUNNEL_INFO_IPV6;
1428 info->key.u.ipv6.dst =
1429 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1430
1431 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1432 IPV6_ADDR_LINKLOCAL) {
1433 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1434 "Remote IPv6 address cannot be link-local");
1435 return -EINVAL;
1436 }
1437 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1438 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1439 "Remote IPv6 address cannot be Multicast");
1440 return -EINVAL;
1441 }
1442 info->key.tun_flags |= TUNNEL_CSUM;
1443 cfg->use_udp6_rx_checksums = true;
1444#else
1445 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1446 "IPv6 support not enabled in the kernel");
1447 return -EPFNOSUPPORT;
1448#endif
1449 }
1450
1451 if (data[IFLA_GENEVE_ID]) {
1452 __u32 vni;
1453 __u8 tvni[3];
1454 __be64 tunid;
1455
1456 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1457 tvni[0] = (vni & 0x00ff0000) >> 16;
1458 tvni[1] = (vni & 0x0000ff00) >> 8;
1459 tvni[2] = vni & 0x000000ff;
1460
1461 tunid = vni_to_tunnel_id(tvni);
1462 if (changelink && (tunid != info->key.tun_id)) {
1463 attrtype = IFLA_GENEVE_ID;
1464 goto change_notsup;
1465 }
1466 info->key.tun_id = tunid;
1467 }
1468
1469 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1470 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1471 cfg->ttl_inherit = true;
1472 else
1473 cfg->ttl_inherit = false;
1474 } else if (data[IFLA_GENEVE_TTL]) {
1475 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1476 cfg->ttl_inherit = false;
1477 }
1478
1479 if (data[IFLA_GENEVE_TOS])
1480 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1481
1482 if (data[IFLA_GENEVE_DF])
1483 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1484
1485 if (data[IFLA_GENEVE_LABEL]) {
1486 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1487 IPV6_FLOWLABEL_MASK;
1488 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1489 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1490 "Label attribute only applies for IPv6 Geneve devices");
1491 return -EINVAL;
1492 }
1493 }
1494
1495 if (data[IFLA_GENEVE_PORT]) {
1496 if (changelink) {
1497 attrtype = IFLA_GENEVE_PORT;
1498 goto change_notsup;
1499 }
1500 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1501 }
1502
1503 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1504 if (changelink) {
1505 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1506 goto change_notsup;
1507 }
1508 cfg->collect_md = true;
1509 }
1510
1511 if (data[IFLA_GENEVE_UDP_CSUM]) {
1512 if (changelink) {
1513 attrtype = IFLA_GENEVE_UDP_CSUM;
1514 goto change_notsup;
1515 }
1516 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1517 info->key.tun_flags |= TUNNEL_CSUM;
1518 }
1519
1520 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1521#if IS_ENABLED(CONFIG_IPV6)
1522 if (changelink) {
1523 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1524 goto change_notsup;
1525 }
1526 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1527 info->key.tun_flags &= ~TUNNEL_CSUM;
1528#else
1529 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1530 "IPv6 support not enabled in the kernel");
1531 return -EPFNOSUPPORT;
1532#endif
1533 }
1534
1535 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1536#if IS_ENABLED(CONFIG_IPV6)
1537 if (changelink) {
1538 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1539 goto change_notsup;
1540 }
1541 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1542 cfg->use_udp6_rx_checksums = false;
1543#else
1544 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1545 "IPv6 support not enabled in the kernel");
1546 return -EPFNOSUPPORT;
1547#endif
1548 }
1549
1550 return 0;
1551change_notsup:
1552 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1553 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1554 return -EOPNOTSUPP;
1555}
1556
1557static void geneve_link_config(struct net_device *dev,
1558 struct ip_tunnel_info *info, struct nlattr *tb[])
1559{
1560 struct geneve_dev *geneve = netdev_priv(dev);
1561 int ldev_mtu = 0;
1562
1563 if (tb[IFLA_MTU]) {
1564 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1565 return;
1566 }
1567
1568 switch (ip_tunnel_info_af(info)) {
1569 case AF_INET: {
1570 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1571 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1572
1573 if (!IS_ERR(rt) && rt->dst.dev) {
1574 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1575 ip_rt_put(rt);
1576 }
1577 break;
1578 }
1579#if IS_ENABLED(CONFIG_IPV6)
1580 case AF_INET6: {
1581 struct rt6_info *rt;
1582
1583 if (!__in6_dev_get(dev))
1584 break;
1585
1586 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1587 NULL, 0);
1588
1589 if (rt && rt->dst.dev)
1590 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1591 ip6_rt_put(rt);
1592 break;
1593 }
1594#endif
1595 }
1596
1597 if (ldev_mtu <= 0)
1598 return;
1599
1600 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1601}
1602
1603static int geneve_newlink(struct net *net, struct net_device *dev,
1604 struct nlattr *tb[], struct nlattr *data[],
1605 struct netlink_ext_ack *extack)
1606{
1607 struct geneve_config cfg = {
1608 .df = GENEVE_DF_UNSET,
1609 .use_udp6_rx_checksums = false,
1610 .ttl_inherit = false,
1611 .collect_md = false,
1612 };
1613 int err;
1614
1615 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1616 err = geneve_nl2info(tb, data, extack, &cfg, false);
1617 if (err)
1618 return err;
1619
1620 err = geneve_configure(net, dev, extack, &cfg);
1621 if (err)
1622 return err;
1623
1624 geneve_link_config(dev, &cfg.info, tb);
1625
1626 return 0;
1627}
1628
1629/* Quiesces the geneve device data path for both TX and RX.
1630 *
1631 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1632 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1633 * to complete for the existing set of in-flight packets to be transmitted,
1634 * then we would have quiesced the transmit data path. All the future packets
1635 * will get dropped until we unquiesce the data path.
1636 *
1637 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1638 * if we set that to NULL under RCU and wait for synchronize_net() to
1639 * complete, then we would have quiesced the receive data path.
1640 */
1641static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1642 struct geneve_sock **gs6)
1643{
1644 *gs4 = rtnl_dereference(geneve->sock4);
1645 rcu_assign_pointer(geneve->sock4, NULL);
1646 if (*gs4)
1647 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1648#if IS_ENABLED(CONFIG_IPV6)
1649 *gs6 = rtnl_dereference(geneve->sock6);
1650 rcu_assign_pointer(geneve->sock6, NULL);
1651 if (*gs6)
1652 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1653#else
1654 *gs6 = NULL;
1655#endif
1656 synchronize_net();
1657}
1658
1659/* Resumes the geneve device data path for both TX and RX. */
1660static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1661 struct geneve_sock __maybe_unused *gs6)
1662{
1663 rcu_assign_pointer(geneve->sock4, gs4);
1664 if (gs4)
1665 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1666#if IS_ENABLED(CONFIG_IPV6)
1667 rcu_assign_pointer(geneve->sock6, gs6);
1668 if (gs6)
1669 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1670#endif
1671 synchronize_net();
1672}
1673
1674static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1675 struct nlattr *data[],
1676 struct netlink_ext_ack *extack)
1677{
1678 struct geneve_dev *geneve = netdev_priv(dev);
1679 struct geneve_sock *gs4, *gs6;
1680 struct geneve_config cfg;
1681 int err;
1682
1683 /* If the geneve device is configured for metadata (or externally
1684 * controlled, for example, OVS), then nothing can be changed.
1685 */
1686 if (geneve->cfg.collect_md)
1687 return -EOPNOTSUPP;
1688
1689 /* Start with the existing info. */
1690 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1691 err = geneve_nl2info(tb, data, extack, &cfg, true);
1692 if (err)
1693 return err;
1694
1695 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1696 dst_cache_reset(&cfg.info.dst_cache);
1697 geneve_link_config(dev, &cfg.info, tb);
1698 }
1699
1700 geneve_quiesce(geneve, &gs4, &gs6);
1701 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1702 geneve_unquiesce(geneve, gs4, gs6);
1703
1704 return 0;
1705}
1706
1707static void geneve_dellink(struct net_device *dev, struct list_head *head)
1708{
1709 struct geneve_dev *geneve = netdev_priv(dev);
1710
1711 list_del(&geneve->next);
1712 unregister_netdevice_queue(dev, head);
1713}
1714
1715static size_t geneve_get_size(const struct net_device *dev)
1716{
1717 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1718 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1719 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1720 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1721 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1722 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1723 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1724 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1725 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1726 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1727 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1728 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1729 0;
1730}
1731
1732static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1733{
1734 struct geneve_dev *geneve = netdev_priv(dev);
1735 struct ip_tunnel_info *info = &geneve->cfg.info;
1736 bool ttl_inherit = geneve->cfg.ttl_inherit;
1737 bool metadata = geneve->cfg.collect_md;
1738 __u8 tmp_vni[3];
1739 __u32 vni;
1740
1741 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1742 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1743 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1744 goto nla_put_failure;
1745
1746 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1747 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1748 info->key.u.ipv4.dst))
1749 goto nla_put_failure;
1750 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1751 !!(info->key.tun_flags & TUNNEL_CSUM)))
1752 goto nla_put_failure;
1753
1754#if IS_ENABLED(CONFIG_IPV6)
1755 } else if (!metadata) {
1756 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1757 &info->key.u.ipv6.dst))
1758 goto nla_put_failure;
1759 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1760 !(info->key.tun_flags & TUNNEL_CSUM)))
1761 goto nla_put_failure;
1762#endif
1763 }
1764
1765 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1766 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1767 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1768 goto nla_put_failure;
1769
1770 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1771 goto nla_put_failure;
1772
1773 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1774 goto nla_put_failure;
1775
1776 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1777 goto nla_put_failure;
1778
1779#if IS_ENABLED(CONFIG_IPV6)
1780 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1781 !geneve->cfg.use_udp6_rx_checksums))
1782 goto nla_put_failure;
1783#endif
1784
1785 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1786 goto nla_put_failure;
1787
1788 return 0;
1789
1790nla_put_failure:
1791 return -EMSGSIZE;
1792}
1793
1794static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1795 .kind = "geneve",
1796 .maxtype = IFLA_GENEVE_MAX,
1797 .policy = geneve_policy,
1798 .priv_size = sizeof(struct geneve_dev),
1799 .setup = geneve_setup,
1800 .validate = geneve_validate,
1801 .newlink = geneve_newlink,
1802 .changelink = geneve_changelink,
1803 .dellink = geneve_dellink,
1804 .get_size = geneve_get_size,
1805 .fill_info = geneve_fill_info,
1806};
1807
1808struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1809 u8 name_assign_type, u16 dst_port)
1810{
1811 struct nlattr *tb[IFLA_MAX + 1];
1812 struct net_device *dev;
1813 LIST_HEAD(list_kill);
1814 int err;
1815 struct geneve_config cfg = {
1816 .df = GENEVE_DF_UNSET,
1817 .use_udp6_rx_checksums = true,
1818 .ttl_inherit = false,
1819 .collect_md = true,
1820 };
1821
1822 memset(tb, 0, sizeof(tb));
1823 dev = rtnl_create_link(net, name, name_assign_type,
1824 &geneve_link_ops, tb, NULL);
1825 if (IS_ERR(dev))
1826 return dev;
1827
1828 init_tnl_info(&cfg.info, dst_port);
1829 err = geneve_configure(net, dev, NULL, &cfg);
1830 if (err) {
1831 free_netdev(dev);
1832 return ERR_PTR(err);
1833 }
1834
1835 /* openvswitch users expect packet sizes to be unrestricted,
1836 * so set the largest MTU we can.
1837 */
1838 err = geneve_change_mtu(dev, IP_MAX_MTU);
1839 if (err)
1840 goto err;
1841
1842 err = rtnl_configure_link(dev, NULL);
1843 if (err < 0)
1844 goto err;
1845
1846 return dev;
1847err:
1848 geneve_dellink(dev, &list_kill);
1849 unregister_netdevice_many(&list_kill);
1850 return ERR_PTR(err);
1851}
1852EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1853
1854static int geneve_netdevice_event(struct notifier_block *unused,
1855 unsigned long event, void *ptr)
1856{
1857 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1858
1859 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1860 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1861 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1862 } else if (event == NETDEV_UNREGISTER) {
1863 if (!dev->udp_tunnel_nic_info)
1864 geneve_offload_rx_ports(dev, false);
1865 } else if (event == NETDEV_REGISTER) {
1866 if (!dev->udp_tunnel_nic_info)
1867 geneve_offload_rx_ports(dev, true);
1868 }
1869
1870 return NOTIFY_DONE;
1871}
1872
1873static struct notifier_block geneve_notifier_block __read_mostly = {
1874 .notifier_call = geneve_netdevice_event,
1875};
1876
1877static __net_init int geneve_init_net(struct net *net)
1878{
1879 struct geneve_net *gn = net_generic(net, geneve_net_id);
1880
1881 INIT_LIST_HEAD(&gn->geneve_list);
1882 INIT_LIST_HEAD(&gn->sock_list);
1883 return 0;
1884}
1885
1886static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1887{
1888 struct geneve_net *gn = net_generic(net, geneve_net_id);
1889 struct geneve_dev *geneve, *next;
1890 struct net_device *dev, *aux;
1891
1892 /* gather any geneve devices that were moved into this ns */
1893 for_each_netdev_safe(net, dev, aux)
1894 if (dev->rtnl_link_ops == &geneve_link_ops)
1895 unregister_netdevice_queue(dev, head);
1896
1897 /* now gather any other geneve devices that were created in this ns */
1898 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1899 /* If geneve->dev is in the same netns, it was already added
1900 * to the list by the previous loop.
1901 */
1902 if (!net_eq(dev_net(geneve->dev), net))
1903 unregister_netdevice_queue(geneve->dev, head);
1904 }
1905}
1906
1907static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1908{
1909 struct net *net;
1910 LIST_HEAD(list);
1911
1912 rtnl_lock();
1913 list_for_each_entry(net, net_list, exit_list)
1914 geneve_destroy_tunnels(net, &list);
1915
1916 /* unregister the devices gathered above */
1917 unregister_netdevice_many(&list);
1918 rtnl_unlock();
1919
1920 list_for_each_entry(net, net_list, exit_list) {
1921 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1922
1923 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1924 }
1925}
1926
1927static struct pernet_operations geneve_net_ops = {
1928 .init = geneve_init_net,
1929 .exit_batch = geneve_exit_batch_net,
1930 .id = &geneve_net_id,
1931 .size = sizeof(struct geneve_net),
1932};
1933
1934static int __init geneve_init_module(void)
1935{
1936 int rc;
1937
1938 rc = register_pernet_subsys(&geneve_net_ops);
1939 if (rc)
1940 goto out1;
1941
1942 rc = register_netdevice_notifier(&geneve_notifier_block);
1943 if (rc)
1944 goto out2;
1945
1946 rc = rtnl_link_register(&geneve_link_ops);
1947 if (rc)
1948 goto out3;
1949
1950 return 0;
1951out3:
1952 unregister_netdevice_notifier(&geneve_notifier_block);
1953out2:
1954 unregister_pernet_subsys(&geneve_net_ops);
1955out1:
1956 return rc;
1957}
1958late_initcall(geneve_init_module);
1959
1960static void __exit geneve_cleanup_module(void)
1961{
1962 rtnl_link_unregister(&geneve_link_ops);
1963 unregister_netdevice_notifier(&geneve_notifier_block);
1964 unregister_pernet_subsys(&geneve_net_ops);
1965}
1966module_exit(geneve_cleanup_module);
1967
1968MODULE_LICENSE("GPL");
1969MODULE_VERSION(GENEVE_NETDEV_VER);
1970MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1971MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1972MODULE_ALIAS_RTNL_LINK("geneve");