]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/geneve.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target...
[mirror_ubuntu-bionic-kernel.git] / drivers / net / geneve.c
1 /*
2 * GENEVE: Generic Network Virtualization Encapsulation
3 *
4 * Copyright (c) 2015 Red Hat, Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 */
10
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/netdevice.h>
16 #include <linux/etherdevice.h>
17 #include <linux/hash.h>
18 #include <net/dst_metadata.h>
19 #include <net/gro_cells.h>
20 #include <net/rtnetlink.h>
21 #include <net/geneve.h>
22 #include <net/protocol.h>
23
24 #define GENEVE_NETDEV_VER "0.6"
25
26 #define GENEVE_UDP_PORT 6081
27
28 #define GENEVE_N_VID (1u << 24)
29 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
30
31 #define VNI_HASH_BITS 10
32 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
33
34 static bool log_ecn_error = true;
35 module_param(log_ecn_error, bool, 0644);
36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
37
38 #define GENEVE_VER 0
39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
40
41 /* per-network namespace private data for this module */
42 struct geneve_net {
43 struct list_head geneve_list;
44 struct list_head sock_list;
45 };
46
47 static int geneve_net_id;
48
49 union geneve_addr {
50 struct sockaddr_in sin;
51 struct sockaddr_in6 sin6;
52 struct sockaddr sa;
53 };
54
55 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
56
57 /* Pseudo network device */
58 struct geneve_dev {
59 struct hlist_node hlist; /* vni hash table */
60 struct net *net; /* netns for packet i/o */
61 struct net_device *dev; /* netdev for geneve tunnel */
62 struct geneve_sock *sock4; /* IPv4 socket used for geneve tunnel */
63 #if IS_ENABLED(CONFIG_IPV6)
64 struct geneve_sock *sock6; /* IPv6 socket used for geneve tunnel */
65 #endif
66 u8 vni[3]; /* virtual network ID for tunnel */
67 u8 ttl; /* TTL override */
68 u8 tos; /* TOS override */
69 union geneve_addr remote; /* IP address for link partner */
70 struct list_head next; /* geneve's per namespace list */
71 __be32 label; /* IPv6 flowlabel override */
72 __be16 dst_port;
73 bool collect_md;
74 struct gro_cells gro_cells;
75 u32 flags;
76 struct dst_cache dst_cache;
77 };
78
79 /* Geneve device flags */
80 #define GENEVE_F_UDP_ZERO_CSUM_TX BIT(0)
81 #define GENEVE_F_UDP_ZERO_CSUM6_TX BIT(1)
82 #define GENEVE_F_UDP_ZERO_CSUM6_RX BIT(2)
83
84 struct geneve_sock {
85 bool collect_md;
86 struct list_head list;
87 struct socket *sock;
88 struct rcu_head rcu;
89 int refcnt;
90 struct udp_offload udp_offloads;
91 struct hlist_head vni_list[VNI_HASH_SIZE];
92 u32 flags;
93 };
94
95 static inline __u32 geneve_net_vni_hash(u8 vni[3])
96 {
97 __u32 vnid;
98
99 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
100 return hash_32(vnid, VNI_HASH_BITS);
101 }
102
103 static __be64 vni_to_tunnel_id(const __u8 *vni)
104 {
105 #ifdef __BIG_ENDIAN
106 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
107 #else
108 return (__force __be64)(((__force u64)vni[0] << 40) |
109 ((__force u64)vni[1] << 48) |
110 ((__force u64)vni[2] << 56));
111 #endif
112 }
113
114 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
115 {
116 return gs->sock->sk->sk_family;
117 }
118
119 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
120 __be32 addr, u8 vni[])
121 {
122 struct hlist_head *vni_list_head;
123 struct geneve_dev *geneve;
124 __u32 hash;
125
126 /* Find the device for this VNI */
127 hash = geneve_net_vni_hash(vni);
128 vni_list_head = &gs->vni_list[hash];
129 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
130 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
131 addr == geneve->remote.sin.sin_addr.s_addr)
132 return geneve;
133 }
134 return NULL;
135 }
136
137 #if IS_ENABLED(CONFIG_IPV6)
138 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
139 struct in6_addr addr6, u8 vni[])
140 {
141 struct hlist_head *vni_list_head;
142 struct geneve_dev *geneve;
143 __u32 hash;
144
145 /* Find the device for this VNI */
146 hash = geneve_net_vni_hash(vni);
147 vni_list_head = &gs->vni_list[hash];
148 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
149 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
150 ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
151 return geneve;
152 }
153 return NULL;
154 }
155 #endif
156
157 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
158 {
159 return (struct genevehdr *)(udp_hdr(skb) + 1);
160 }
161
162 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
163 struct sk_buff *skb)
164 {
165 u8 *vni;
166 __be32 addr;
167 static u8 zero_vni[3];
168 #if IS_ENABLED(CONFIG_IPV6)
169 static struct in6_addr zero_addr6;
170 #endif
171
172 if (geneve_get_sk_family(gs) == AF_INET) {
173 struct iphdr *iph;
174
175 iph = ip_hdr(skb); /* outer IP header... */
176
177 if (gs->collect_md) {
178 vni = zero_vni;
179 addr = 0;
180 } else {
181 vni = geneve_hdr(skb)->vni;
182 addr = iph->saddr;
183 }
184
185 return geneve_lookup(gs, addr, vni);
186 #if IS_ENABLED(CONFIG_IPV6)
187 } else if (geneve_get_sk_family(gs) == AF_INET6) {
188 struct ipv6hdr *ip6h;
189 struct in6_addr addr6;
190
191 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
192
193 if (gs->collect_md) {
194 vni = zero_vni;
195 addr6 = zero_addr6;
196 } else {
197 vni = geneve_hdr(skb)->vni;
198 addr6 = ip6h->saddr;
199 }
200
201 return geneve6_lookup(gs, addr6, vni);
202 #endif
203 }
204 return NULL;
205 }
206
207 /* geneve receive/decap routine */
208 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
209 struct sk_buff *skb)
210 {
211 struct genevehdr *gnvh = geneve_hdr(skb);
212 struct metadata_dst *tun_dst = NULL;
213 struct pcpu_sw_netstats *stats;
214 int err = 0;
215 void *oiph;
216
217 if (ip_tunnel_collect_metadata() || gs->collect_md) {
218 __be16 flags;
219
220 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
221 (gnvh->oam ? TUNNEL_OAM : 0) |
222 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
223
224 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
225 vni_to_tunnel_id(gnvh->vni),
226 gnvh->opt_len * 4);
227 if (!tun_dst)
228 goto drop;
229 /* Update tunnel dst according to Geneve options. */
230 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
231 gnvh->options, gnvh->opt_len * 4);
232 } else {
233 /* Drop packets w/ critical options,
234 * since we don't support any...
235 */
236 if (gnvh->critical)
237 goto drop;
238 }
239
240 skb_reset_mac_header(skb);
241 skb->protocol = eth_type_trans(skb, geneve->dev);
242 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
243
244 if (tun_dst)
245 skb_dst_set(skb, &tun_dst->dst);
246
247 /* Ignore packet loops (and multicast echo) */
248 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
249 goto drop;
250
251 oiph = skb_network_header(skb);
252 skb_reset_network_header(skb);
253
254 if (geneve_get_sk_family(gs) == AF_INET)
255 err = IP_ECN_decapsulate(oiph, skb);
256 #if IS_ENABLED(CONFIG_IPV6)
257 else
258 err = IP6_ECN_decapsulate(oiph, skb);
259 #endif
260
261 if (unlikely(err)) {
262 if (log_ecn_error) {
263 if (geneve_get_sk_family(gs) == AF_INET)
264 net_info_ratelimited("non-ECT from %pI4 "
265 "with TOS=%#x\n",
266 &((struct iphdr *)oiph)->saddr,
267 ((struct iphdr *)oiph)->tos);
268 #if IS_ENABLED(CONFIG_IPV6)
269 else
270 net_info_ratelimited("non-ECT from %pI6\n",
271 &((struct ipv6hdr *)oiph)->saddr);
272 #endif
273 }
274 if (err > 1) {
275 ++geneve->dev->stats.rx_frame_errors;
276 ++geneve->dev->stats.rx_errors;
277 goto drop;
278 }
279 }
280
281 stats = this_cpu_ptr(geneve->dev->tstats);
282 u64_stats_update_begin(&stats->syncp);
283 stats->rx_packets++;
284 stats->rx_bytes += skb->len;
285 u64_stats_update_end(&stats->syncp);
286
287 gro_cells_receive(&geneve->gro_cells, skb);
288 return;
289 drop:
290 /* Consume bad packet */
291 kfree_skb(skb);
292 }
293
294 /* Setup stats when device is created */
295 static int geneve_init(struct net_device *dev)
296 {
297 struct geneve_dev *geneve = netdev_priv(dev);
298 int err;
299
300 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
301 if (!dev->tstats)
302 return -ENOMEM;
303
304 err = gro_cells_init(&geneve->gro_cells, dev);
305 if (err) {
306 free_percpu(dev->tstats);
307 return err;
308 }
309
310 err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
311 if (err) {
312 free_percpu(dev->tstats);
313 gro_cells_destroy(&geneve->gro_cells);
314 return err;
315 }
316
317 return 0;
318 }
319
320 static void geneve_uninit(struct net_device *dev)
321 {
322 struct geneve_dev *geneve = netdev_priv(dev);
323
324 dst_cache_destroy(&geneve->dst_cache);
325 gro_cells_destroy(&geneve->gro_cells);
326 free_percpu(dev->tstats);
327 }
328
329 /* Callback from net/ipv4/udp.c to receive packets */
330 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
331 {
332 struct genevehdr *geneveh;
333 struct geneve_dev *geneve;
334 struct geneve_sock *gs;
335 int opts_len;
336
337 /* Need Geneve and inner Ethernet header to be present */
338 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
339 goto error;
340
341 /* Return packets with reserved bits set */
342 geneveh = geneve_hdr(skb);
343 if (unlikely(geneveh->ver != GENEVE_VER))
344 goto error;
345
346 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
347 goto error;
348
349 gs = rcu_dereference_sk_user_data(sk);
350 if (!gs)
351 goto drop;
352
353 geneve = geneve_lookup_skb(gs, skb);
354 if (!geneve)
355 goto drop;
356
357 opts_len = geneveh->opt_len * 4;
358 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
359 htons(ETH_P_TEB),
360 !net_eq(geneve->net, dev_net(geneve->dev))))
361 goto drop;
362
363 geneve_rx(geneve, gs, skb);
364 return 0;
365
366 drop:
367 /* Consume bad packet */
368 kfree_skb(skb);
369 return 0;
370
371 error:
372 /* Let the UDP layer deal with the skb */
373 return 1;
374 }
375
376 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
377 __be16 port, u32 flags)
378 {
379 struct socket *sock;
380 struct udp_port_cfg udp_conf;
381 int err;
382
383 memset(&udp_conf, 0, sizeof(udp_conf));
384
385 if (ipv6) {
386 udp_conf.family = AF_INET6;
387 udp_conf.ipv6_v6only = 1;
388 udp_conf.use_udp6_rx_checksums =
389 !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
390 } else {
391 udp_conf.family = AF_INET;
392 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
393 }
394
395 udp_conf.local_udp_port = port;
396
397 /* Open UDP socket */
398 err = udp_sock_create(net, &udp_conf, &sock);
399 if (err < 0)
400 return ERR_PTR(err);
401
402 return sock;
403 }
404
405 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
406 {
407 struct net_device *dev;
408 struct sock *sk = gs->sock->sk;
409 struct net *net = sock_net(sk);
410 sa_family_t sa_family = geneve_get_sk_family(gs);
411 __be16 port = inet_sk(sk)->inet_sport;
412 int err;
413
414 if (sa_family == AF_INET) {
415 err = udp_add_offload(sock_net(sk), &gs->udp_offloads);
416 if (err)
417 pr_warn("geneve: udp_add_offload failed with status %d\n",
418 err);
419 }
420
421 rcu_read_lock();
422 for_each_netdev_rcu(net, dev) {
423 if (dev->netdev_ops->ndo_add_geneve_port)
424 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
425 port);
426 }
427 rcu_read_unlock();
428 }
429
430 static int geneve_hlen(struct genevehdr *gh)
431 {
432 return sizeof(*gh) + gh->opt_len * 4;
433 }
434
435 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
436 struct sk_buff *skb,
437 struct udp_offload *uoff)
438 {
439 struct sk_buff *p, **pp = NULL;
440 struct genevehdr *gh, *gh2;
441 unsigned int hlen, gh_len, off_gnv;
442 const struct packet_offload *ptype;
443 __be16 type;
444 int flush = 1;
445
446 off_gnv = skb_gro_offset(skb);
447 hlen = off_gnv + sizeof(*gh);
448 gh = skb_gro_header_fast(skb, off_gnv);
449 if (skb_gro_header_hard(skb, hlen)) {
450 gh = skb_gro_header_slow(skb, hlen, off_gnv);
451 if (unlikely(!gh))
452 goto out;
453 }
454
455 if (gh->ver != GENEVE_VER || gh->oam)
456 goto out;
457 gh_len = geneve_hlen(gh);
458
459 hlen = off_gnv + gh_len;
460 if (skb_gro_header_hard(skb, hlen)) {
461 gh = skb_gro_header_slow(skb, hlen, off_gnv);
462 if (unlikely(!gh))
463 goto out;
464 }
465
466 for (p = *head; p; p = p->next) {
467 if (!NAPI_GRO_CB(p)->same_flow)
468 continue;
469
470 gh2 = (struct genevehdr *)(p->data + off_gnv);
471 if (gh->opt_len != gh2->opt_len ||
472 memcmp(gh, gh2, gh_len)) {
473 NAPI_GRO_CB(p)->same_flow = 0;
474 continue;
475 }
476 }
477
478 type = gh->proto_type;
479
480 rcu_read_lock();
481 ptype = gro_find_receive_by_type(type);
482 if (!ptype)
483 goto out_unlock;
484
485 skb_gro_pull(skb, gh_len);
486 skb_gro_postpull_rcsum(skb, gh, gh_len);
487 pp = ptype->callbacks.gro_receive(head, skb);
488 flush = 0;
489
490 out_unlock:
491 rcu_read_unlock();
492 out:
493 NAPI_GRO_CB(skb)->flush |= flush;
494
495 return pp;
496 }
497
498 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
499 struct udp_offload *uoff)
500 {
501 struct genevehdr *gh;
502 struct packet_offload *ptype;
503 __be16 type;
504 int gh_len;
505 int err = -ENOSYS;
506
507 udp_tunnel_gro_complete(skb, nhoff);
508
509 gh = (struct genevehdr *)(skb->data + nhoff);
510 gh_len = geneve_hlen(gh);
511 type = gh->proto_type;
512
513 rcu_read_lock();
514 ptype = gro_find_complete_by_type(type);
515 if (ptype)
516 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
517
518 rcu_read_unlock();
519 return err;
520 }
521
522 /* Create new listen socket if needed */
523 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
524 bool ipv6, u32 flags)
525 {
526 struct geneve_net *gn = net_generic(net, geneve_net_id);
527 struct geneve_sock *gs;
528 struct socket *sock;
529 struct udp_tunnel_sock_cfg tunnel_cfg;
530 int h;
531
532 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
533 if (!gs)
534 return ERR_PTR(-ENOMEM);
535
536 sock = geneve_create_sock(net, ipv6, port, flags);
537 if (IS_ERR(sock)) {
538 kfree(gs);
539 return ERR_CAST(sock);
540 }
541
542 gs->sock = sock;
543 gs->refcnt = 1;
544 for (h = 0; h < VNI_HASH_SIZE; ++h)
545 INIT_HLIST_HEAD(&gs->vni_list[h]);
546
547 /* Initialize the geneve udp offloads structure */
548 gs->udp_offloads.port = port;
549 gs->udp_offloads.callbacks.gro_receive = geneve_gro_receive;
550 gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
551 geneve_notify_add_rx_port(gs);
552
553 /* Mark socket as an encapsulation socket */
554 tunnel_cfg.sk_user_data = gs;
555 tunnel_cfg.encap_type = 1;
556 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
557 tunnel_cfg.encap_destroy = NULL;
558 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
559 list_add(&gs->list, &gn->sock_list);
560 return gs;
561 }
562
563 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
564 {
565 struct net_device *dev;
566 struct sock *sk = gs->sock->sk;
567 struct net *net = sock_net(sk);
568 sa_family_t sa_family = geneve_get_sk_family(gs);
569 __be16 port = inet_sk(sk)->inet_sport;
570
571 rcu_read_lock();
572 for_each_netdev_rcu(net, dev) {
573 if (dev->netdev_ops->ndo_del_geneve_port)
574 dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
575 port);
576 }
577
578 rcu_read_unlock();
579
580 if (sa_family == AF_INET)
581 udp_del_offload(&gs->udp_offloads);
582 }
583
584 static void __geneve_sock_release(struct geneve_sock *gs)
585 {
586 if (!gs || --gs->refcnt)
587 return;
588
589 list_del(&gs->list);
590 geneve_notify_del_rx_port(gs);
591 udp_tunnel_sock_release(gs->sock);
592 kfree_rcu(gs, rcu);
593 }
594
595 static void geneve_sock_release(struct geneve_dev *geneve)
596 {
597 __geneve_sock_release(geneve->sock4);
598 #if IS_ENABLED(CONFIG_IPV6)
599 __geneve_sock_release(geneve->sock6);
600 #endif
601 }
602
603 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
604 sa_family_t family,
605 __be16 dst_port)
606 {
607 struct geneve_sock *gs;
608
609 list_for_each_entry(gs, &gn->sock_list, list) {
610 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
611 geneve_get_sk_family(gs) == family) {
612 return gs;
613 }
614 }
615 return NULL;
616 }
617
618 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
619 {
620 struct net *net = geneve->net;
621 struct geneve_net *gn = net_generic(net, geneve_net_id);
622 struct geneve_sock *gs;
623 __u32 hash;
624
625 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
626 if (gs) {
627 gs->refcnt++;
628 goto out;
629 }
630
631 gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
632 if (IS_ERR(gs))
633 return PTR_ERR(gs);
634
635 out:
636 gs->collect_md = geneve->collect_md;
637 gs->flags = geneve->flags;
638 #if IS_ENABLED(CONFIG_IPV6)
639 if (ipv6)
640 geneve->sock6 = gs;
641 else
642 #endif
643 geneve->sock4 = gs;
644
645 hash = geneve_net_vni_hash(geneve->vni);
646 hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
647 return 0;
648 }
649
650 static int geneve_open(struct net_device *dev)
651 {
652 struct geneve_dev *geneve = netdev_priv(dev);
653 bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
654 bool metadata = geneve->collect_md;
655 int ret = 0;
656
657 geneve->sock4 = NULL;
658 #if IS_ENABLED(CONFIG_IPV6)
659 geneve->sock6 = NULL;
660 if (ipv6 || metadata)
661 ret = geneve_sock_add(geneve, true);
662 #endif
663 if (!ret && (!ipv6 || metadata))
664 ret = geneve_sock_add(geneve, false);
665 if (ret < 0)
666 geneve_sock_release(geneve);
667
668 return ret;
669 }
670
671 static int geneve_stop(struct net_device *dev)
672 {
673 struct geneve_dev *geneve = netdev_priv(dev);
674
675 if (!hlist_unhashed(&geneve->hlist))
676 hlist_del_rcu(&geneve->hlist);
677 geneve_sock_release(geneve);
678 return 0;
679 }
680
681 static void geneve_build_header(struct genevehdr *geneveh,
682 __be16 tun_flags, u8 vni[3],
683 u8 options_len, u8 *options)
684 {
685 geneveh->ver = GENEVE_VER;
686 geneveh->opt_len = options_len / 4;
687 geneveh->oam = !!(tun_flags & TUNNEL_OAM);
688 geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
689 geneveh->rsvd1 = 0;
690 memcpy(geneveh->vni, vni, 3);
691 geneveh->proto_type = htons(ETH_P_TEB);
692 geneveh->rsvd2 = 0;
693
694 memcpy(geneveh->options, options, options_len);
695 }
696
697 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
698 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
699 u32 flags, bool xnet)
700 {
701 struct genevehdr *gnvh;
702 int min_headroom;
703 int err;
704 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
705
706 skb_scrub_packet(skb, xnet);
707
708 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
709 + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
710 err = skb_cow_head(skb, min_headroom);
711 if (unlikely(err)) {
712 kfree_skb(skb);
713 goto free_rt;
714 }
715
716 skb = udp_tunnel_handle_offloads(skb, udp_sum);
717 if (IS_ERR(skb)) {
718 err = PTR_ERR(skb);
719 goto free_rt;
720 }
721
722 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
723 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
724
725 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
726 return 0;
727
728 free_rt:
729 ip_rt_put(rt);
730 return err;
731 }
732
733 #if IS_ENABLED(CONFIG_IPV6)
734 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
735 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
736 u32 flags, bool xnet)
737 {
738 struct genevehdr *gnvh;
739 int min_headroom;
740 int err;
741 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
742
743 skb_scrub_packet(skb, xnet);
744
745 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
746 + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
747 err = skb_cow_head(skb, min_headroom);
748 if (unlikely(err)) {
749 kfree_skb(skb);
750 goto free_dst;
751 }
752
753 skb = udp_tunnel_handle_offloads(skb, udp_sum);
754 if (IS_ERR(skb)) {
755 err = PTR_ERR(skb);
756 goto free_dst;
757 }
758
759 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
760 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
761
762 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
763 return 0;
764
765 free_dst:
766 dst_release(dst);
767 return err;
768 }
769 #endif
770
771 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
772 struct net_device *dev,
773 struct flowi4 *fl4,
774 struct ip_tunnel_info *info)
775 {
776 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
777 struct geneve_dev *geneve = netdev_priv(dev);
778 struct dst_cache *dst_cache;
779 struct rtable *rt = NULL;
780 __u8 tos;
781
782 memset(fl4, 0, sizeof(*fl4));
783 fl4->flowi4_mark = skb->mark;
784 fl4->flowi4_proto = IPPROTO_UDP;
785
786 if (info) {
787 fl4->daddr = info->key.u.ipv4.dst;
788 fl4->saddr = info->key.u.ipv4.src;
789 fl4->flowi4_tos = RT_TOS(info->key.tos);
790 dst_cache = &info->dst_cache;
791 } else {
792 tos = geneve->tos;
793 if (tos == 1) {
794 const struct iphdr *iip = ip_hdr(skb);
795
796 tos = ip_tunnel_get_dsfield(iip, skb);
797 use_cache = false;
798 }
799
800 fl4->flowi4_tos = RT_TOS(tos);
801 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
802 dst_cache = &geneve->dst_cache;
803 }
804
805 if (use_cache) {
806 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
807 if (rt)
808 return rt;
809 }
810
811 rt = ip_route_output_key(geneve->net, fl4);
812 if (IS_ERR(rt)) {
813 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
814 return ERR_PTR(-ENETUNREACH);
815 }
816 if (rt->dst.dev == dev) { /* is this necessary? */
817 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
818 ip_rt_put(rt);
819 return ERR_PTR(-ELOOP);
820 }
821 if (use_cache)
822 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
823 return rt;
824 }
825
826 #if IS_ENABLED(CONFIG_IPV6)
827 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
828 struct net_device *dev,
829 struct flowi6 *fl6,
830 struct ip_tunnel_info *info)
831 {
832 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
833 struct geneve_dev *geneve = netdev_priv(dev);
834 struct geneve_sock *gs6 = geneve->sock6;
835 struct dst_entry *dst = NULL;
836 struct dst_cache *dst_cache;
837 __u8 prio;
838
839 memset(fl6, 0, sizeof(*fl6));
840 fl6->flowi6_mark = skb->mark;
841 fl6->flowi6_proto = IPPROTO_UDP;
842
843 if (info) {
844 fl6->daddr = info->key.u.ipv6.dst;
845 fl6->saddr = info->key.u.ipv6.src;
846 fl6->flowi6_tos = RT_TOS(info->key.tos);
847 fl6->flowlabel = info->key.label;
848 dst_cache = &info->dst_cache;
849 } else {
850 prio = geneve->tos;
851 if (prio == 1) {
852 const struct iphdr *iip = ip_hdr(skb);
853
854 prio = ip_tunnel_get_dsfield(iip, skb);
855 use_cache = false;
856 }
857
858 fl6->flowi6_tos = RT_TOS(prio);
859 fl6->flowlabel = geneve->label;
860 fl6->daddr = geneve->remote.sin6.sin6_addr;
861 dst_cache = &geneve->dst_cache;
862 }
863
864 if (use_cache) {
865 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
866 if (dst)
867 return dst;
868 }
869
870 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
871 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
872 return ERR_PTR(-ENETUNREACH);
873 }
874 if (dst->dev == dev) { /* is this necessary? */
875 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
876 dst_release(dst);
877 return ERR_PTR(-ELOOP);
878 }
879
880 if (use_cache)
881 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
882 return dst;
883 }
884 #endif
885
886 /* Convert 64 bit tunnel ID to 24 bit VNI. */
887 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
888 {
889 #ifdef __BIG_ENDIAN
890 vni[0] = (__force __u8)(tun_id >> 16);
891 vni[1] = (__force __u8)(tun_id >> 8);
892 vni[2] = (__force __u8)tun_id;
893 #else
894 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
895 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
896 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
897 #endif
898 }
899
900 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
901 struct ip_tunnel_info *info)
902 {
903 struct geneve_dev *geneve = netdev_priv(dev);
904 struct geneve_sock *gs4 = geneve->sock4;
905 struct rtable *rt = NULL;
906 const struct iphdr *iip; /* interior IP header */
907 int err = -EINVAL;
908 struct flowi4 fl4;
909 __u8 tos, ttl;
910 __be16 sport;
911 __be16 df;
912 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
913 u32 flags = geneve->flags;
914
915 if (geneve->collect_md) {
916 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
917 netdev_dbg(dev, "no tunnel metadata\n");
918 goto tx_error;
919 }
920 if (info && ip_tunnel_info_af(info) != AF_INET)
921 goto tx_error;
922 }
923
924 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
925 if (IS_ERR(rt)) {
926 err = PTR_ERR(rt);
927 goto tx_error;
928 }
929
930 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
931 skb_reset_mac_header(skb);
932
933 iip = ip_hdr(skb);
934
935 if (info) {
936 const struct ip_tunnel_key *key = &info->key;
937 u8 *opts = NULL;
938 u8 vni[3];
939
940 tunnel_id_to_vni(key->tun_id, vni);
941 if (info->options_len)
942 opts = ip_tunnel_info_opts(info);
943
944 if (key->tun_flags & TUNNEL_CSUM)
945 flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
946 else
947 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
948
949 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
950 info->options_len, opts, flags, xnet);
951 if (unlikely(err))
952 goto err;
953
954 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
955 ttl = key->ttl;
956 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
957 } else {
958 err = geneve_build_skb(rt, skb, 0, geneve->vni,
959 0, NULL, flags, xnet);
960 if (unlikely(err))
961 goto err;
962
963 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
964 ttl = geneve->ttl;
965 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
966 ttl = 1;
967 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
968 df = 0;
969 }
970 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
971 tos, ttl, df, sport, geneve->dst_port,
972 !net_eq(geneve->net, dev_net(geneve->dev)),
973 !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
974
975 return NETDEV_TX_OK;
976
977 tx_error:
978 dev_kfree_skb(skb);
979 err:
980 if (err == -ELOOP)
981 dev->stats.collisions++;
982 else if (err == -ENETUNREACH)
983 dev->stats.tx_carrier_errors++;
984 else
985 dev->stats.tx_errors++;
986 return NETDEV_TX_OK;
987 }
988
989 #if IS_ENABLED(CONFIG_IPV6)
990 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
991 struct ip_tunnel_info *info)
992 {
993 struct geneve_dev *geneve = netdev_priv(dev);
994 struct geneve_sock *gs6 = geneve->sock6;
995 struct dst_entry *dst = NULL;
996 const struct iphdr *iip; /* interior IP header */
997 int err = -EINVAL;
998 struct flowi6 fl6;
999 __u8 prio, ttl;
1000 __be16 sport;
1001 __be32 label;
1002 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1003 u32 flags = geneve->flags;
1004
1005 if (geneve->collect_md) {
1006 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1007 netdev_dbg(dev, "no tunnel metadata\n");
1008 goto tx_error;
1009 }
1010 }
1011
1012 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1013 if (IS_ERR(dst)) {
1014 err = PTR_ERR(dst);
1015 goto tx_error;
1016 }
1017
1018 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1019 skb_reset_mac_header(skb);
1020
1021 iip = ip_hdr(skb);
1022
1023 if (info) {
1024 const struct ip_tunnel_key *key = &info->key;
1025 u8 *opts = NULL;
1026 u8 vni[3];
1027
1028 tunnel_id_to_vni(key->tun_id, vni);
1029 if (info->options_len)
1030 opts = ip_tunnel_info_opts(info);
1031
1032 if (key->tun_flags & TUNNEL_CSUM)
1033 flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1034 else
1035 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1036
1037 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1038 info->options_len, opts,
1039 flags, xnet);
1040 if (unlikely(err))
1041 goto err;
1042
1043 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1044 ttl = key->ttl;
1045 label = info->key.label;
1046 } else {
1047 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1048 0, NULL, flags, xnet);
1049 if (unlikely(err))
1050 goto err;
1051
1052 prio = ip_tunnel_ecn_encap(fl6.flowi6_tos, iip, skb);
1053 ttl = geneve->ttl;
1054 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1055 ttl = 1;
1056 ttl = ttl ? : ip6_dst_hoplimit(dst);
1057 label = geneve->label;
1058 }
1059
1060 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1061 &fl6.saddr, &fl6.daddr, prio, ttl, label,
1062 sport, geneve->dst_port,
1063 !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1064 return NETDEV_TX_OK;
1065
1066 tx_error:
1067 dev_kfree_skb(skb);
1068 err:
1069 if (err == -ELOOP)
1070 dev->stats.collisions++;
1071 else if (err == -ENETUNREACH)
1072 dev->stats.tx_carrier_errors++;
1073 else
1074 dev->stats.tx_errors++;
1075 return NETDEV_TX_OK;
1076 }
1077 #endif
1078
1079 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1080 {
1081 struct geneve_dev *geneve = netdev_priv(dev);
1082 struct ip_tunnel_info *info = NULL;
1083
1084 if (geneve->collect_md)
1085 info = skb_tunnel_info(skb);
1086
1087 #if IS_ENABLED(CONFIG_IPV6)
1088 if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1089 (!info && geneve->remote.sa.sa_family == AF_INET6))
1090 return geneve6_xmit_skb(skb, dev, info);
1091 #endif
1092 return geneve_xmit_skb(skb, dev, info);
1093 }
1094
1095 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1096 {
1097 /* The max_mtu calculation does not take account of GENEVE
1098 * options, to avoid excluding potentially valid
1099 * configurations.
1100 */
1101 int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1102 - dev->hard_header_len;
1103
1104 if (new_mtu < 68)
1105 return -EINVAL;
1106
1107 if (new_mtu > max_mtu) {
1108 if (strict)
1109 return -EINVAL;
1110
1111 new_mtu = max_mtu;
1112 }
1113
1114 dev->mtu = new_mtu;
1115 return 0;
1116 }
1117
1118 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1119 {
1120 return __geneve_change_mtu(dev, new_mtu, true);
1121 }
1122
1123 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1124 {
1125 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1126 struct geneve_dev *geneve = netdev_priv(dev);
1127 struct rtable *rt;
1128 struct flowi4 fl4;
1129 #if IS_ENABLED(CONFIG_IPV6)
1130 struct dst_entry *dst;
1131 struct flowi6 fl6;
1132 #endif
1133
1134 if (ip_tunnel_info_af(info) == AF_INET) {
1135 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1136 if (IS_ERR(rt))
1137 return PTR_ERR(rt);
1138
1139 ip_rt_put(rt);
1140 info->key.u.ipv4.src = fl4.saddr;
1141 #if IS_ENABLED(CONFIG_IPV6)
1142 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1143 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1144 if (IS_ERR(dst))
1145 return PTR_ERR(dst);
1146
1147 dst_release(dst);
1148 info->key.u.ipv6.src = fl6.saddr;
1149 #endif
1150 } else {
1151 return -EINVAL;
1152 }
1153
1154 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1155 1, USHRT_MAX, true);
1156 info->key.tp_dst = geneve->dst_port;
1157 return 0;
1158 }
1159
1160 static const struct net_device_ops geneve_netdev_ops = {
1161 .ndo_init = geneve_init,
1162 .ndo_uninit = geneve_uninit,
1163 .ndo_open = geneve_open,
1164 .ndo_stop = geneve_stop,
1165 .ndo_start_xmit = geneve_xmit,
1166 .ndo_get_stats64 = ip_tunnel_get_stats64,
1167 .ndo_change_mtu = geneve_change_mtu,
1168 .ndo_validate_addr = eth_validate_addr,
1169 .ndo_set_mac_address = eth_mac_addr,
1170 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1171 };
1172
1173 static void geneve_get_drvinfo(struct net_device *dev,
1174 struct ethtool_drvinfo *drvinfo)
1175 {
1176 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1177 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1178 }
1179
1180 static const struct ethtool_ops geneve_ethtool_ops = {
1181 .get_drvinfo = geneve_get_drvinfo,
1182 .get_link = ethtool_op_get_link,
1183 };
1184
1185 /* Info for udev, that this is a virtual tunnel endpoint */
1186 static struct device_type geneve_type = {
1187 .name = "geneve",
1188 };
1189
1190 /* Calls the ndo_add_geneve_port of the caller in order to
1191 * supply the listening GENEVE udp ports. Callers are expected
1192 * to implement the ndo_add_geneve_port.
1193 */
1194 void geneve_get_rx_port(struct net_device *dev)
1195 {
1196 struct net *net = dev_net(dev);
1197 struct geneve_net *gn = net_generic(net, geneve_net_id);
1198 struct geneve_sock *gs;
1199 sa_family_t sa_family;
1200 struct sock *sk;
1201 __be16 port;
1202
1203 rcu_read_lock();
1204 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1205 sk = gs->sock->sk;
1206 sa_family = sk->sk_family;
1207 port = inet_sk(sk)->inet_sport;
1208 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1209 }
1210 rcu_read_unlock();
1211 }
1212 EXPORT_SYMBOL_GPL(geneve_get_rx_port);
1213
1214 /* Initialize the device structure. */
1215 static void geneve_setup(struct net_device *dev)
1216 {
1217 ether_setup(dev);
1218
1219 dev->netdev_ops = &geneve_netdev_ops;
1220 dev->ethtool_ops = &geneve_ethtool_ops;
1221 dev->destructor = free_netdev;
1222
1223 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1224
1225 dev->features |= NETIF_F_LLTX;
1226 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1227 dev->features |= NETIF_F_RXCSUM;
1228 dev->features |= NETIF_F_GSO_SOFTWARE;
1229
1230 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1231 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1232
1233 netif_keep_dst(dev);
1234 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1235 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1236 eth_hw_addr_random(dev);
1237 }
1238
1239 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1240 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1241 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1242 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1243 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1244 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1245 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1246 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1247 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1248 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1249 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1250 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1251 };
1252
1253 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1254 {
1255 if (tb[IFLA_ADDRESS]) {
1256 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1257 return -EINVAL;
1258
1259 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1260 return -EADDRNOTAVAIL;
1261 }
1262
1263 if (!data)
1264 return -EINVAL;
1265
1266 if (data[IFLA_GENEVE_ID]) {
1267 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1268
1269 if (vni >= GENEVE_VID_MASK)
1270 return -ERANGE;
1271 }
1272
1273 return 0;
1274 }
1275
1276 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1277 __be16 dst_port,
1278 union geneve_addr *remote,
1279 u8 vni[],
1280 bool *tun_on_same_port,
1281 bool *tun_collect_md)
1282 {
1283 struct geneve_dev *geneve, *t;
1284
1285 *tun_on_same_port = false;
1286 *tun_collect_md = false;
1287 t = NULL;
1288 list_for_each_entry(geneve, &gn->geneve_list, next) {
1289 if (geneve->dst_port == dst_port) {
1290 *tun_collect_md = geneve->collect_md;
1291 *tun_on_same_port = true;
1292 }
1293 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1294 !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1295 dst_port == geneve->dst_port)
1296 t = geneve;
1297 }
1298 return t;
1299 }
1300
1301 static int geneve_configure(struct net *net, struct net_device *dev,
1302 union geneve_addr *remote,
1303 __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1304 __be16 dst_port, bool metadata, u32 flags)
1305 {
1306 struct geneve_net *gn = net_generic(net, geneve_net_id);
1307 struct geneve_dev *t, *geneve = netdev_priv(dev);
1308 bool tun_collect_md, tun_on_same_port;
1309 int err, encap_len;
1310
1311 if (!remote)
1312 return -EINVAL;
1313 if (metadata &&
1314 (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1315 return -EINVAL;
1316
1317 geneve->net = net;
1318 geneve->dev = dev;
1319
1320 geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1321 geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1322 geneve->vni[2] = vni & 0x000000ff;
1323
1324 if ((remote->sa.sa_family == AF_INET &&
1325 IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1326 (remote->sa.sa_family == AF_INET6 &&
1327 ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1328 return -EINVAL;
1329 if (label && remote->sa.sa_family != AF_INET6)
1330 return -EINVAL;
1331
1332 geneve->remote = *remote;
1333
1334 geneve->ttl = ttl;
1335 geneve->tos = tos;
1336 geneve->label = label;
1337 geneve->dst_port = dst_port;
1338 geneve->collect_md = metadata;
1339 geneve->flags = flags;
1340
1341 t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1342 &tun_on_same_port, &tun_collect_md);
1343 if (t)
1344 return -EBUSY;
1345
1346 /* make enough headroom for basic scenario */
1347 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1348 if (remote->sa.sa_family == AF_INET)
1349 encap_len += sizeof(struct iphdr);
1350 else
1351 encap_len += sizeof(struct ipv6hdr);
1352 dev->needed_headroom = encap_len + ETH_HLEN;
1353
1354 if (metadata) {
1355 if (tun_on_same_port)
1356 return -EPERM;
1357 } else {
1358 if (tun_collect_md)
1359 return -EPERM;
1360 }
1361
1362 dst_cache_reset(&geneve->dst_cache);
1363
1364 err = register_netdevice(dev);
1365 if (err)
1366 return err;
1367
1368 list_add(&geneve->next, &gn->geneve_list);
1369 return 0;
1370 }
1371
1372 static int geneve_newlink(struct net *net, struct net_device *dev,
1373 struct nlattr *tb[], struct nlattr *data[])
1374 {
1375 __be16 dst_port = htons(GENEVE_UDP_PORT);
1376 __u8 ttl = 0, tos = 0;
1377 bool metadata = false;
1378 union geneve_addr remote = geneve_remote_unspec;
1379 __be32 label = 0;
1380 __u32 vni = 0;
1381 u32 flags = 0;
1382
1383 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1384 return -EINVAL;
1385
1386 if (data[IFLA_GENEVE_REMOTE]) {
1387 remote.sa.sa_family = AF_INET;
1388 remote.sin.sin_addr.s_addr =
1389 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1390 }
1391
1392 if (data[IFLA_GENEVE_REMOTE6]) {
1393 if (!IS_ENABLED(CONFIG_IPV6))
1394 return -EPFNOSUPPORT;
1395
1396 remote.sa.sa_family = AF_INET6;
1397 remote.sin6.sin6_addr =
1398 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1399
1400 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1401 IPV6_ADDR_LINKLOCAL) {
1402 netdev_dbg(dev, "link-local remote is unsupported\n");
1403 return -EINVAL;
1404 }
1405 }
1406
1407 if (data[IFLA_GENEVE_ID])
1408 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1409
1410 if (data[IFLA_GENEVE_TTL])
1411 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1412
1413 if (data[IFLA_GENEVE_TOS])
1414 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1415
1416 if (data[IFLA_GENEVE_LABEL])
1417 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1418 IPV6_FLOWLABEL_MASK;
1419
1420 if (data[IFLA_GENEVE_PORT])
1421 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1422
1423 if (data[IFLA_GENEVE_COLLECT_METADATA])
1424 metadata = true;
1425
1426 if (data[IFLA_GENEVE_UDP_CSUM] &&
1427 !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1428 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1429
1430 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1431 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1432 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1433
1434 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1435 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1436 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1437
1438 return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1439 dst_port, metadata, flags);
1440 }
1441
1442 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1443 {
1444 struct geneve_dev *geneve = netdev_priv(dev);
1445
1446 list_del(&geneve->next);
1447 unregister_netdevice_queue(dev, head);
1448 }
1449
1450 static size_t geneve_get_size(const struct net_device *dev)
1451 {
1452 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1453 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1454 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1455 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1456 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1457 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1458 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1459 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1460 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1461 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1462 0;
1463 }
1464
1465 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1466 {
1467 struct geneve_dev *geneve = netdev_priv(dev);
1468 __u32 vni;
1469
1470 vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1471 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1472 goto nla_put_failure;
1473
1474 if (geneve->remote.sa.sa_family == AF_INET) {
1475 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1476 geneve->remote.sin.sin_addr.s_addr))
1477 goto nla_put_failure;
1478 #if IS_ENABLED(CONFIG_IPV6)
1479 } else {
1480 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1481 &geneve->remote.sin6.sin6_addr))
1482 goto nla_put_failure;
1483 #endif
1484 }
1485
1486 if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1487 nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1488 nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1489 goto nla_put_failure;
1490
1491 if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1492 goto nla_put_failure;
1493
1494 if (geneve->collect_md) {
1495 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1496 goto nla_put_failure;
1497 }
1498
1499 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1500 !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1501 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1502 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1503 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1504 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1505 goto nla_put_failure;
1506
1507 return 0;
1508
1509 nla_put_failure:
1510 return -EMSGSIZE;
1511 }
1512
1513 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1514 .kind = "geneve",
1515 .maxtype = IFLA_GENEVE_MAX,
1516 .policy = geneve_policy,
1517 .priv_size = sizeof(struct geneve_dev),
1518 .setup = geneve_setup,
1519 .validate = geneve_validate,
1520 .newlink = geneve_newlink,
1521 .dellink = geneve_dellink,
1522 .get_size = geneve_get_size,
1523 .fill_info = geneve_fill_info,
1524 };
1525
1526 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1527 u8 name_assign_type, u16 dst_port)
1528 {
1529 struct nlattr *tb[IFLA_MAX + 1];
1530 struct net_device *dev;
1531 int err;
1532
1533 memset(tb, 0, sizeof(tb));
1534 dev = rtnl_create_link(net, name, name_assign_type,
1535 &geneve_link_ops, tb);
1536 if (IS_ERR(dev))
1537 return dev;
1538
1539 err = geneve_configure(net, dev, &geneve_remote_unspec,
1540 0, 0, 0, 0, htons(dst_port), true,
1541 GENEVE_F_UDP_ZERO_CSUM6_RX);
1542 if (err)
1543 goto err;
1544
1545 /* openvswitch users expect packet sizes to be unrestricted,
1546 * so set the largest MTU we can.
1547 */
1548 err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1549 if (err)
1550 goto err;
1551
1552 return dev;
1553
1554 err:
1555 free_netdev(dev);
1556 return ERR_PTR(err);
1557 }
1558 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1559
1560 static __net_init int geneve_init_net(struct net *net)
1561 {
1562 struct geneve_net *gn = net_generic(net, geneve_net_id);
1563
1564 INIT_LIST_HEAD(&gn->geneve_list);
1565 INIT_LIST_HEAD(&gn->sock_list);
1566 return 0;
1567 }
1568
1569 static void __net_exit geneve_exit_net(struct net *net)
1570 {
1571 struct geneve_net *gn = net_generic(net, geneve_net_id);
1572 struct geneve_dev *geneve, *next;
1573 struct net_device *dev, *aux;
1574 LIST_HEAD(list);
1575
1576 rtnl_lock();
1577
1578 /* gather any geneve devices that were moved into this ns */
1579 for_each_netdev_safe(net, dev, aux)
1580 if (dev->rtnl_link_ops == &geneve_link_ops)
1581 unregister_netdevice_queue(dev, &list);
1582
1583 /* now gather any other geneve devices that were created in this ns */
1584 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1585 /* If geneve->dev is in the same netns, it was already added
1586 * to the list by the previous loop.
1587 */
1588 if (!net_eq(dev_net(geneve->dev), net))
1589 unregister_netdevice_queue(geneve->dev, &list);
1590 }
1591
1592 /* unregister the devices gathered above */
1593 unregister_netdevice_many(&list);
1594 rtnl_unlock();
1595 }
1596
1597 static struct pernet_operations geneve_net_ops = {
1598 .init = geneve_init_net,
1599 .exit = geneve_exit_net,
1600 .id = &geneve_net_id,
1601 .size = sizeof(struct geneve_net),
1602 };
1603
1604 static int __init geneve_init_module(void)
1605 {
1606 int rc;
1607
1608 rc = register_pernet_subsys(&geneve_net_ops);
1609 if (rc)
1610 goto out1;
1611
1612 rc = rtnl_link_register(&geneve_link_ops);
1613 if (rc)
1614 goto out2;
1615
1616 return 0;
1617 out2:
1618 unregister_pernet_subsys(&geneve_net_ops);
1619 out1:
1620 return rc;
1621 }
1622 late_initcall(geneve_init_module);
1623
1624 static void __exit geneve_cleanup_module(void)
1625 {
1626 rtnl_link_unregister(&geneve_link_ops);
1627 unregister_pernet_subsys(&geneve_net_ops);
1628 }
1629 module_exit(geneve_cleanup_module);
1630
1631 MODULE_LICENSE("GPL");
1632 MODULE_VERSION(GENEVE_NETDEV_VER);
1633 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1634 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1635 MODULE_ALIAS_RTNL_LINK("geneve");
ubuntu-bionic-kernel mirror