projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'ipvs-for-v4.7' of https://git.kernel.org/pub/scm/linux/kernel/git/horms...
[deliverable/linux.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 hlist_head vni_list[VNI_HASH_SIZE];
91 u32 flags;
92 };
93
94 static inline __u32 geneve_net_vni_hash(u8 vni[3])
95 {
96 __u32 vnid;
97
98 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
99 return hash_32(vnid, VNI_HASH_BITS);
100 }
101
102 static __be64 vni_to_tunnel_id(const __u8 *vni)
103 {
104 #ifdef __BIG_ENDIAN
105 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
106 #else
107 return (__force __be64)(((__force u64)vni[0] << 40) |
108 ((__force u64)vni[1] << 48) |
109 ((__force u64)vni[2] << 56));
110 #endif
111 }
112
113 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
114 {
115 return gs->sock->sk->sk_family;
116 }
117
118 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
119 __be32 addr, u8 vni[])
120 {
121 struct hlist_head *vni_list_head;
122 struct geneve_dev *geneve;
123 __u32 hash;
124
125 /* Find the device for this VNI */
126 hash = geneve_net_vni_hash(vni);
127 vni_list_head = &gs->vni_list[hash];
128 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
129 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
130 addr == geneve->remote.sin.sin_addr.s_addr)
131 return geneve;
132 }
133 return NULL;
134 }
135
136 #if IS_ENABLED(CONFIG_IPV6)
137 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
138 struct in6_addr addr6, u8 vni[])
139 {
140 struct hlist_head *vni_list_head;
141 struct geneve_dev *geneve;
142 __u32 hash;
143
144 /* Find the device for this VNI */
145 hash = geneve_net_vni_hash(vni);
146 vni_list_head = &gs->vni_list[hash];
147 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
148 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
149 ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
150 return geneve;
151 }
152 return NULL;
153 }
154 #endif
155
156 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
157 {
158 return (struct genevehdr *)(udp_hdr(skb) + 1);
159 }
160
161 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
162 struct sk_buff *skb)
163 {
164 u8 *vni;
165 __be32 addr;
166 static u8 zero_vni[3];
167 #if IS_ENABLED(CONFIG_IPV6)
168 static struct in6_addr zero_addr6;
169 #endif
170
171 if (geneve_get_sk_family(gs) == AF_INET) {
172 struct iphdr *iph;
173
174 iph = ip_hdr(skb); /* outer IP header... */
175
176 if (gs->collect_md) {
177 vni = zero_vni;
178 addr = 0;
179 } else {
180 vni = geneve_hdr(skb)->vni;
181 addr = iph->saddr;
182 }
183
184 return geneve_lookup(gs, addr, vni);
185 #if IS_ENABLED(CONFIG_IPV6)
186 } else if (geneve_get_sk_family(gs) == AF_INET6) {
187 struct ipv6hdr *ip6h;
188 struct in6_addr addr6;
189
190 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
191
192 if (gs->collect_md) {
193 vni = zero_vni;
194 addr6 = zero_addr6;
195 } else {
196 vni = geneve_hdr(skb)->vni;
197 addr6 = ip6h->saddr;
198 }
199
200 return geneve6_lookup(gs, addr6, vni);
201 #endif
202 }
203 return NULL;
204 }
205
206 /* geneve receive/decap routine */
207 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
208 struct sk_buff *skb)
209 {
210 struct genevehdr *gnvh = geneve_hdr(skb);
211 struct metadata_dst *tun_dst = NULL;
212 struct pcpu_sw_netstats *stats;
213 int err = 0;
214 void *oiph;
215
216 if (ip_tunnel_collect_metadata() || gs->collect_md) {
217 __be16 flags;
218
219 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
220 (gnvh->oam ? TUNNEL_OAM : 0) |
221 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
222
223 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
224 vni_to_tunnel_id(gnvh->vni),
225 gnvh->opt_len * 4);
226 if (!tun_dst)
227 goto drop;
228 /* Update tunnel dst according to Geneve options. */
229 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
230 gnvh->options, gnvh->opt_len * 4);
231 } else {
232 /* Drop packets w/ critical options,
233 * since we don't support any...
234 */
235 if (gnvh->critical)
236 goto drop;
237 }
238
239 skb_reset_mac_header(skb);
240 skb->protocol = eth_type_trans(skb, geneve->dev);
241 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
242
243 if (tun_dst)
244 skb_dst_set(skb, &tun_dst->dst);
245
246 /* Ignore packet loops (and multicast echo) */
247 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
248 goto drop;
249
250 oiph = skb_network_header(skb);
251 skb_reset_network_header(skb);
252
253 if (geneve_get_sk_family(gs) == AF_INET)
254 err = IP_ECN_decapsulate(oiph, skb);
255 #if IS_ENABLED(CONFIG_IPV6)
256 else
257 err = IP6_ECN_decapsulate(oiph, skb);
258 #endif
259
260 if (unlikely(err)) {
261 if (log_ecn_error) {
262 if (geneve_get_sk_family(gs) == AF_INET)
263 net_info_ratelimited("non-ECT from %pI4 "
264 "with TOS=%#x\n",
265 &((struct iphdr *)oiph)->saddr,
266 ((struct iphdr *)oiph)->tos);
267 #if IS_ENABLED(CONFIG_IPV6)
268 else
269 net_info_ratelimited("non-ECT from %pI6\n",
270 &((struct ipv6hdr *)oiph)->saddr);
271 #endif
272 }
273 if (err > 1) {
274 ++geneve->dev->stats.rx_frame_errors;
275 ++geneve->dev->stats.rx_errors;
276 goto drop;
277 }
278 }
279
280 stats = this_cpu_ptr(geneve->dev->tstats);
281 u64_stats_update_begin(&stats->syncp);
282 stats->rx_packets++;
283 stats->rx_bytes += skb->len;
284 u64_stats_update_end(&stats->syncp);
285
286 gro_cells_receive(&geneve->gro_cells, skb);
287 return;
288 drop:
289 /* Consume bad packet */
290 kfree_skb(skb);
291 }
292
293 /* Setup stats when device is created */
294 static int geneve_init(struct net_device *dev)
295 {
296 struct geneve_dev *geneve = netdev_priv(dev);
297 int err;
298
299 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
300 if (!dev->tstats)
301 return -ENOMEM;
302
303 err = gro_cells_init(&geneve->gro_cells, dev);
304 if (err) {
305 free_percpu(dev->tstats);
306 return err;
307 }
308
309 err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
310 if (err) {
311 free_percpu(dev->tstats);
312 gro_cells_destroy(&geneve->gro_cells);
313 return err;
314 }
315
316 return 0;
317 }
318
319 static void geneve_uninit(struct net_device *dev)
320 {
321 struct geneve_dev *geneve = netdev_priv(dev);
322
323 dst_cache_destroy(&geneve->dst_cache);
324 gro_cells_destroy(&geneve->gro_cells);
325 free_percpu(dev->tstats);
326 }
327
328 /* Callback from net/ipv4/udp.c to receive packets */
329 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
330 {
331 struct genevehdr *geneveh;
332 struct geneve_dev *geneve;
333 struct geneve_sock *gs;
334 int opts_len;
335
336 /* Need Geneve and inner Ethernet header to be present */
337 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
338 goto error;
339
340 /* Return packets with reserved bits set */
341 geneveh = geneve_hdr(skb);
342 if (unlikely(geneveh->ver != GENEVE_VER))
343 goto error;
344
345 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
346 goto error;
347
348 gs = rcu_dereference_sk_user_data(sk);
349 if (!gs)
350 goto drop;
351
352 geneve = geneve_lookup_skb(gs, skb);
353 if (!geneve)
354 goto drop;
355
356 opts_len = geneveh->opt_len * 4;
357 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
358 htons(ETH_P_TEB),
359 !net_eq(geneve->net, dev_net(geneve->dev))))
360 goto drop;
361
362 geneve_rx(geneve, gs, skb);
363 return 0;
364
365 drop:
366 /* Consume bad packet */
367 kfree_skb(skb);
368 return 0;
369
370 error:
371 /* Let the UDP layer deal with the skb */
372 return 1;
373 }
374
375 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
376 __be16 port, u32 flags)
377 {
378 struct socket *sock;
379 struct udp_port_cfg udp_conf;
380 int err;
381
382 memset(&udp_conf, 0, sizeof(udp_conf));
383
384 if (ipv6) {
385 udp_conf.family = AF_INET6;
386 udp_conf.ipv6_v6only = 1;
387 udp_conf.use_udp6_rx_checksums =
388 !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
389 } else {
390 udp_conf.family = AF_INET;
391 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
392 }
393
394 udp_conf.local_udp_port = port;
395
396 /* Open UDP socket */
397 err = udp_sock_create(net, &udp_conf, &sock);
398 if (err < 0)
399 return ERR_PTR(err);
400
401 return sock;
402 }
403
404 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
405 {
406 struct net_device *dev;
407 struct sock *sk = gs->sock->sk;
408 struct net *net = sock_net(sk);
409 sa_family_t sa_family = geneve_get_sk_family(gs);
410 __be16 port = inet_sk(sk)->inet_sport;
411
412 rcu_read_lock();
413 for_each_netdev_rcu(net, dev) {
414 if (dev->netdev_ops->ndo_add_geneve_port)
415 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
416 port);
417 }
418 rcu_read_unlock();
419 }
420
421 static int geneve_hlen(struct genevehdr *gh)
422 {
423 return sizeof(*gh) + gh->opt_len * 4;
424 }
425
426 static struct sk_buff **geneve_gro_receive(struct sock *sk,
427 struct sk_buff **head,
428 struct sk_buff *skb)
429 {
430 struct sk_buff *p, **pp = NULL;
431 struct genevehdr *gh, *gh2;
432 unsigned int hlen, gh_len, off_gnv;
433 const struct packet_offload *ptype;
434 __be16 type;
435 int flush = 1;
436
437 off_gnv = skb_gro_offset(skb);
438 hlen = off_gnv + sizeof(*gh);
439 gh = skb_gro_header_fast(skb, off_gnv);
440 if (skb_gro_header_hard(skb, hlen)) {
441 gh = skb_gro_header_slow(skb, hlen, off_gnv);
442 if (unlikely(!gh))
443 goto out;
444 }
445
446 if (gh->ver != GENEVE_VER || gh->oam)
447 goto out;
448 gh_len = geneve_hlen(gh);
449
450 hlen = off_gnv + gh_len;
451 if (skb_gro_header_hard(skb, hlen)) {
452 gh = skb_gro_header_slow(skb, hlen, off_gnv);
453 if (unlikely(!gh))
454 goto out;
455 }
456
457 for (p = *head; p; p = p->next) {
458 if (!NAPI_GRO_CB(p)->same_flow)
459 continue;
460
461 gh2 = (struct genevehdr *)(p->data + off_gnv);
462 if (gh->opt_len != gh2->opt_len ||
463 memcmp(gh, gh2, gh_len)) {
464 NAPI_GRO_CB(p)->same_flow = 0;
465 continue;
466 }
467 }
468
469 type = gh->proto_type;
470
471 rcu_read_lock();
472 ptype = gro_find_receive_by_type(type);
473 if (!ptype)
474 goto out_unlock;
475
476 skb_gro_pull(skb, gh_len);
477 skb_gro_postpull_rcsum(skb, gh, gh_len);
478 pp = ptype->callbacks.gro_receive(head, skb);
479 flush = 0;
480
481 out_unlock:
482 rcu_read_unlock();
483 out:
484 NAPI_GRO_CB(skb)->flush |= flush;
485
486 return pp;
487 }
488
489 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
490 int nhoff)
491 {
492 struct genevehdr *gh;
493 struct packet_offload *ptype;
494 __be16 type;
495 int gh_len;
496 int err = -ENOSYS;
497
498 udp_tunnel_gro_complete(skb, nhoff);
499
500 gh = (struct genevehdr *)(skb->data + nhoff);
501 gh_len = geneve_hlen(gh);
502 type = gh->proto_type;
503
504 rcu_read_lock();
505 ptype = gro_find_complete_by_type(type);
506 if (ptype)
507 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
508
509 rcu_read_unlock();
510 return err;
511 }
512
513 /* Create new listen socket if needed */
514 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
515 bool ipv6, u32 flags)
516 {
517 struct geneve_net *gn = net_generic(net, geneve_net_id);
518 struct geneve_sock *gs;
519 struct socket *sock;
520 struct udp_tunnel_sock_cfg tunnel_cfg;
521 int h;
522
523 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
524 if (!gs)
525 return ERR_PTR(-ENOMEM);
526
527 sock = geneve_create_sock(net, ipv6, port, flags);
528 if (IS_ERR(sock)) {
529 kfree(gs);
530 return ERR_CAST(sock);
531 }
532
533 gs->sock = sock;
534 gs->refcnt = 1;
535 for (h = 0; h < VNI_HASH_SIZE; ++h)
536 INIT_HLIST_HEAD(&gs->vni_list[h]);
537
538 /* Initialize the geneve udp offloads structure */
539 geneve_notify_add_rx_port(gs);
540
541 /* Mark socket as an encapsulation socket */
542 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
543 tunnel_cfg.sk_user_data = gs;
544 tunnel_cfg.encap_type = 1;
545 tunnel_cfg.gro_receive = geneve_gro_receive;
546 tunnel_cfg.gro_complete = geneve_gro_complete;
547 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
548 tunnel_cfg.encap_destroy = NULL;
549 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
550 list_add(&gs->list, &gn->sock_list);
551 return gs;
552 }
553
554 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
555 {
556 struct net_device *dev;
557 struct sock *sk = gs->sock->sk;
558 struct net *net = sock_net(sk);
559 sa_family_t sa_family = geneve_get_sk_family(gs);
560 __be16 port = inet_sk(sk)->inet_sport;
561
562 rcu_read_lock();
563 for_each_netdev_rcu(net, dev) {
564 if (dev->netdev_ops->ndo_del_geneve_port)
565 dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
566 port);
567 }
568
569 rcu_read_unlock();
570 }
571
572 static void __geneve_sock_release(struct geneve_sock *gs)
573 {
574 if (!gs || --gs->refcnt)
575 return;
576
577 list_del(&gs->list);
578 geneve_notify_del_rx_port(gs);
579 udp_tunnel_sock_release(gs->sock);
580 kfree_rcu(gs, rcu);
581 }
582
583 static void geneve_sock_release(struct geneve_dev *geneve)
584 {
585 __geneve_sock_release(geneve->sock4);
586 #if IS_ENABLED(CONFIG_IPV6)
587 __geneve_sock_release(geneve->sock6);
588 #endif
589 }
590
591 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
592 sa_family_t family,
593 __be16 dst_port)
594 {
595 struct geneve_sock *gs;
596
597 list_for_each_entry(gs, &gn->sock_list, list) {
598 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
599 geneve_get_sk_family(gs) == family) {
600 return gs;
601 }
602 }
603 return NULL;
604 }
605
606 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
607 {
608 struct net *net = geneve->net;
609 struct geneve_net *gn = net_generic(net, geneve_net_id);
610 struct geneve_sock *gs;
611 __u32 hash;
612
613 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
614 if (gs) {
615 gs->refcnt++;
616 goto out;
617 }
618
619 gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
620 if (IS_ERR(gs))
621 return PTR_ERR(gs);
622
623 out:
624 gs->collect_md = geneve->collect_md;
625 gs->flags = geneve->flags;
626 #if IS_ENABLED(CONFIG_IPV6)
627 if (ipv6)
628 geneve->sock6 = gs;
629 else
630 #endif
631 geneve->sock4 = gs;
632
633 hash = geneve_net_vni_hash(geneve->vni);
634 hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
635 return 0;
636 }
637
638 static int geneve_open(struct net_device *dev)
639 {
640 struct geneve_dev *geneve = netdev_priv(dev);
641 bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
642 bool metadata = geneve->collect_md;
643 int ret = 0;
644
645 geneve->sock4 = NULL;
646 #if IS_ENABLED(CONFIG_IPV6)
647 geneve->sock6 = NULL;
648 if (ipv6 || metadata)
649 ret = geneve_sock_add(geneve, true);
650 #endif
651 if (!ret && (!ipv6 || metadata))
652 ret = geneve_sock_add(geneve, false);
653 if (ret < 0)
654 geneve_sock_release(geneve);
655
656 return ret;
657 }
658
659 static int geneve_stop(struct net_device *dev)
660 {
661 struct geneve_dev *geneve = netdev_priv(dev);
662
663 if (!hlist_unhashed(&geneve->hlist))
664 hlist_del_rcu(&geneve->hlist);
665 geneve_sock_release(geneve);
666 return 0;
667 }
668
669 static void geneve_build_header(struct genevehdr *geneveh,
670 __be16 tun_flags, u8 vni[3],
671 u8 options_len, u8 *options)
672 {
673 geneveh->ver = GENEVE_VER;
674 geneveh->opt_len = options_len / 4;
675 geneveh->oam = !!(tun_flags & TUNNEL_OAM);
676 geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
677 geneveh->rsvd1 = 0;
678 memcpy(geneveh->vni, vni, 3);
679 geneveh->proto_type = htons(ETH_P_TEB);
680 geneveh->rsvd2 = 0;
681
682 memcpy(geneveh->options, options, options_len);
683 }
684
685 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
686 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
687 u32 flags, bool xnet)
688 {
689 struct genevehdr *gnvh;
690 int min_headroom;
691 int err;
692 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
693
694 skb_scrub_packet(skb, xnet);
695
696 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
697 + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
698 err = skb_cow_head(skb, min_headroom);
699 if (unlikely(err))
700 goto free_rt;
701
702 err = udp_tunnel_handle_offloads(skb, udp_sum);
703 if (err)
704 goto free_rt;
705
706 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
707 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
708
709 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
710 return 0;
711
712 free_rt:
713 ip_rt_put(rt);
714 return err;
715 }
716
717 #if IS_ENABLED(CONFIG_IPV6)
718 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
719 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
720 u32 flags, bool xnet)
721 {
722 struct genevehdr *gnvh;
723 int min_headroom;
724 int err;
725 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
726
727 skb_scrub_packet(skb, xnet);
728
729 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
730 + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
731 err = skb_cow_head(skb, min_headroom);
732 if (unlikely(err))
733 goto free_dst;
734
735 err = udp_tunnel_handle_offloads(skb, udp_sum);
736 if (err)
737 goto free_dst;
738
739 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
740 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
741
742 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
743 return 0;
744
745 free_dst:
746 dst_release(dst);
747 return err;
748 }
749 #endif
750
751 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
752 struct net_device *dev,
753 struct flowi4 *fl4,
754 struct ip_tunnel_info *info)
755 {
756 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
757 struct geneve_dev *geneve = netdev_priv(dev);
758 struct dst_cache *dst_cache;
759 struct rtable *rt = NULL;
760 __u8 tos;
761
762 memset(fl4, 0, sizeof(*fl4));
763 fl4->flowi4_mark = skb->mark;
764 fl4->flowi4_proto = IPPROTO_UDP;
765
766 if (info) {
767 fl4->daddr = info->key.u.ipv4.dst;
768 fl4->saddr = info->key.u.ipv4.src;
769 fl4->flowi4_tos = RT_TOS(info->key.tos);
770 dst_cache = &info->dst_cache;
771 } else {
772 tos = geneve->tos;
773 if (tos == 1) {
774 const struct iphdr *iip = ip_hdr(skb);
775
776 tos = ip_tunnel_get_dsfield(iip, skb);
777 use_cache = false;
778 }
779
780 fl4->flowi4_tos = RT_TOS(tos);
781 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
782 dst_cache = &geneve->dst_cache;
783 }
784
785 if (use_cache) {
786 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
787 if (rt)
788 return rt;
789 }
790
791 rt = ip_route_output_key(geneve->net, fl4);
792 if (IS_ERR(rt)) {
793 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
794 return ERR_PTR(-ENETUNREACH);
795 }
796 if (rt->dst.dev == dev) { /* is this necessary? */
797 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
798 ip_rt_put(rt);
799 return ERR_PTR(-ELOOP);
800 }
801 if (use_cache)
802 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
803 return rt;
804 }
805
806 #if IS_ENABLED(CONFIG_IPV6)
807 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
808 struct net_device *dev,
809 struct flowi6 *fl6,
810 struct ip_tunnel_info *info)
811 {
812 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
813 struct geneve_dev *geneve = netdev_priv(dev);
814 struct geneve_sock *gs6 = geneve->sock6;
815 struct dst_entry *dst = NULL;
816 struct dst_cache *dst_cache;
817 __u8 prio;
818
819 memset(fl6, 0, sizeof(*fl6));
820 fl6->flowi6_mark = skb->mark;
821 fl6->flowi6_proto = IPPROTO_UDP;
822
823 if (info) {
824 fl6->daddr = info->key.u.ipv6.dst;
825 fl6->saddr = info->key.u.ipv6.src;
826 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
827 info->key.label);
828 dst_cache = &info->dst_cache;
829 } else {
830 prio = geneve->tos;
831 if (prio == 1) {
832 const struct iphdr *iip = ip_hdr(skb);
833
834 prio = ip_tunnel_get_dsfield(iip, skb);
835 use_cache = false;
836 }
837
838 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
839 geneve->label);
840 fl6->daddr = geneve->remote.sin6.sin6_addr;
841 dst_cache = &geneve->dst_cache;
842 }
843
844 if (use_cache) {
845 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
846 if (dst)
847 return dst;
848 }
849
850 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
851 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
852 return ERR_PTR(-ENETUNREACH);
853 }
854 if (dst->dev == dev) { /* is this necessary? */
855 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
856 dst_release(dst);
857 return ERR_PTR(-ELOOP);
858 }
859
860 if (use_cache)
861 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
862 return dst;
863 }
864 #endif
865
866 /* Convert 64 bit tunnel ID to 24 bit VNI. */
867 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
868 {
869 #ifdef __BIG_ENDIAN
870 vni[0] = (__force __u8)(tun_id >> 16);
871 vni[1] = (__force __u8)(tun_id >> 8);
872 vni[2] = (__force __u8)tun_id;
873 #else
874 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
875 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
876 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
877 #endif
878 }
879
880 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
881 struct ip_tunnel_info *info)
882 {
883 struct geneve_dev *geneve = netdev_priv(dev);
884 struct geneve_sock *gs4 = geneve->sock4;
885 struct rtable *rt = NULL;
886 const struct iphdr *iip; /* interior IP header */
887 int err = -EINVAL;
888 struct flowi4 fl4;
889 __u8 tos, ttl;
890 __be16 sport;
891 __be16 df;
892 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
893 u32 flags = geneve->flags;
894
895 if (geneve->collect_md) {
896 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
897 netdev_dbg(dev, "no tunnel metadata\n");
898 goto tx_error;
899 }
900 if (info && ip_tunnel_info_af(info) != AF_INET)
901 goto tx_error;
902 }
903
904 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
905 if (IS_ERR(rt)) {
906 err = PTR_ERR(rt);
907 goto tx_error;
908 }
909
910 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
911 skb_reset_mac_header(skb);
912
913 iip = ip_hdr(skb);
914
915 if (info) {
916 const struct ip_tunnel_key *key = &info->key;
917 u8 *opts = NULL;
918 u8 vni[3];
919
920 tunnel_id_to_vni(key->tun_id, vni);
921 if (info->options_len)
922 opts = ip_tunnel_info_opts(info);
923
924 if (key->tun_flags & TUNNEL_CSUM)
925 flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
926 else
927 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
928
929 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
930 info->options_len, opts, flags, xnet);
931 if (unlikely(err))
932 goto tx_error;
933
934 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
935 ttl = key->ttl;
936 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
937 } else {
938 err = geneve_build_skb(rt, skb, 0, geneve->vni,
939 0, NULL, flags, xnet);
940 if (unlikely(err))
941 goto tx_error;
942
943 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
944 ttl = geneve->ttl;
945 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
946 ttl = 1;
947 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
948 df = 0;
949 }
950 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
951 tos, ttl, df, sport, geneve->dst_port,
952 !net_eq(geneve->net, dev_net(geneve->dev)),
953 !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
954
955 return NETDEV_TX_OK;
956
957 tx_error:
958 dev_kfree_skb(skb);
959
960 if (err == -ELOOP)
961 dev->stats.collisions++;
962 else if (err == -ENETUNREACH)
963 dev->stats.tx_carrier_errors++;
964 else
965 dev->stats.tx_errors++;
966 return NETDEV_TX_OK;
967 }
968
969 #if IS_ENABLED(CONFIG_IPV6)
970 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
971 struct ip_tunnel_info *info)
972 {
973 struct geneve_dev *geneve = netdev_priv(dev);
974 struct geneve_sock *gs6 = geneve->sock6;
975 struct dst_entry *dst = NULL;
976 const struct iphdr *iip; /* interior IP header */
977 int err = -EINVAL;
978 struct flowi6 fl6;
979 __u8 prio, ttl;
980 __be16 sport;
981 __be32 label;
982 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
983 u32 flags = geneve->flags;
984
985 if (geneve->collect_md) {
986 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
987 netdev_dbg(dev, "no tunnel metadata\n");
988 goto tx_error;
989 }
990 }
991
992 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
993 if (IS_ERR(dst)) {
994 err = PTR_ERR(dst);
995 goto tx_error;
996 }
997
998 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
999 skb_reset_mac_header(skb);
1000
1001 iip = ip_hdr(skb);
1002
1003 if (info) {
1004 const struct ip_tunnel_key *key = &info->key;
1005 u8 *opts = NULL;
1006 u8 vni[3];
1007
1008 tunnel_id_to_vni(key->tun_id, vni);
1009 if (info->options_len)
1010 opts = ip_tunnel_info_opts(info);
1011
1012 if (key->tun_flags & TUNNEL_CSUM)
1013 flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1014 else
1015 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1016
1017 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1018 info->options_len, opts,
1019 flags, xnet);
1020 if (unlikely(err))
1021 goto tx_error;
1022
1023 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1024 ttl = key->ttl;
1025 label = info->key.label;
1026 } else {
1027 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1028 0, NULL, flags, xnet);
1029 if (unlikely(err))
1030 goto tx_error;
1031
1032 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1033 iip, skb);
1034 ttl = geneve->ttl;
1035 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1036 ttl = 1;
1037 ttl = ttl ? : ip6_dst_hoplimit(dst);
1038 label = geneve->label;
1039 }
1040
1041 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1042 &fl6.saddr, &fl6.daddr, prio, ttl, label,
1043 sport, geneve->dst_port,
1044 !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1045 return NETDEV_TX_OK;
1046
1047 tx_error:
1048 dev_kfree_skb(skb);
1049
1050 if (err == -ELOOP)
1051 dev->stats.collisions++;
1052 else if (err == -ENETUNREACH)
1053 dev->stats.tx_carrier_errors++;
1054 else
1055 dev->stats.tx_errors++;
1056 return NETDEV_TX_OK;
1057 }
1058 #endif
1059
1060 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1061 {
1062 struct geneve_dev *geneve = netdev_priv(dev);
1063 struct ip_tunnel_info *info = NULL;
1064
1065 if (geneve->collect_md)
1066 info = skb_tunnel_info(skb);
1067
1068 #if IS_ENABLED(CONFIG_IPV6)
1069 if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1070 (!info && geneve->remote.sa.sa_family == AF_INET6))
1071 return geneve6_xmit_skb(skb, dev, info);
1072 #endif
1073 return geneve_xmit_skb(skb, dev, info);
1074 }
1075
1076 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1077 {
1078 /* The max_mtu calculation does not take account of GENEVE
1079 * options, to avoid excluding potentially valid
1080 * configurations.
1081 */
1082 int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1083 - dev->hard_header_len;
1084
1085 if (new_mtu < 68)
1086 return -EINVAL;
1087
1088 if (new_mtu > max_mtu) {
1089 if (strict)
1090 return -EINVAL;
1091
1092 new_mtu = max_mtu;
1093 }
1094
1095 dev->mtu = new_mtu;
1096 return 0;
1097 }
1098
1099 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1100 {
1101 return __geneve_change_mtu(dev, new_mtu, true);
1102 }
1103
1104 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1105 {
1106 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1107 struct geneve_dev *geneve = netdev_priv(dev);
1108 struct rtable *rt;
1109 struct flowi4 fl4;
1110 #if IS_ENABLED(CONFIG_IPV6)
1111 struct dst_entry *dst;
1112 struct flowi6 fl6;
1113 #endif
1114
1115 if (ip_tunnel_info_af(info) == AF_INET) {
1116 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1117 if (IS_ERR(rt))
1118 return PTR_ERR(rt);
1119
1120 ip_rt_put(rt);
1121 info->key.u.ipv4.src = fl4.saddr;
1122 #if IS_ENABLED(CONFIG_IPV6)
1123 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1124 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
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 = udp_flow_src_port(geneve->net, skb,
1136 1, USHRT_MAX, true);
1137 info->key.tp_dst = geneve->dst_port;
1138 return 0;
1139 }
1140
1141 static const struct net_device_ops geneve_netdev_ops = {
1142 .ndo_init = geneve_init,
1143 .ndo_uninit = geneve_uninit,
1144 .ndo_open = geneve_open,
1145 .ndo_stop = geneve_stop,
1146 .ndo_start_xmit = geneve_xmit,
1147 .ndo_get_stats64 = ip_tunnel_get_stats64,
1148 .ndo_change_mtu = geneve_change_mtu,
1149 .ndo_validate_addr = eth_validate_addr,
1150 .ndo_set_mac_address = eth_mac_addr,
1151 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1152 };
1153
1154 static void geneve_get_drvinfo(struct net_device *dev,
1155 struct ethtool_drvinfo *drvinfo)
1156 {
1157 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1158 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1159 }
1160
1161 static const struct ethtool_ops geneve_ethtool_ops = {
1162 .get_drvinfo = geneve_get_drvinfo,
1163 .get_link = ethtool_op_get_link,
1164 };
1165
1166 /* Info for udev, that this is a virtual tunnel endpoint */
1167 static struct device_type geneve_type = {
1168 .name = "geneve",
1169 };
1170
1171 /* Calls the ndo_add_geneve_port of the caller in order to
1172 * supply the listening GENEVE udp ports. Callers are expected
1173 * to implement the ndo_add_geneve_port.
1174 */
1175 static void geneve_push_rx_ports(struct net_device *dev)
1176 {
1177 struct net *net = dev_net(dev);
1178 struct geneve_net *gn = net_generic(net, geneve_net_id);
1179 struct geneve_sock *gs;
1180 sa_family_t sa_family;
1181 struct sock *sk;
1182 __be16 port;
1183
1184 if (!dev->netdev_ops->ndo_add_geneve_port)
1185 return;
1186
1187 rcu_read_lock();
1188 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1189 sk = gs->sock->sk;
1190 sa_family = sk->sk_family;
1191 port = inet_sk(sk)->inet_sport;
1192 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1193 }
1194 rcu_read_unlock();
1195 }
1196
1197 /* Initialize the device structure. */
1198 static void geneve_setup(struct net_device *dev)
1199 {
1200 ether_setup(dev);
1201
1202 dev->netdev_ops = &geneve_netdev_ops;
1203 dev->ethtool_ops = &geneve_ethtool_ops;
1204 dev->destructor = free_netdev;
1205
1206 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1207
1208 dev->features |= NETIF_F_LLTX;
1209 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1210 dev->features |= NETIF_F_RXCSUM;
1211 dev->features |= NETIF_F_GSO_SOFTWARE;
1212
1213 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1214 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1215
1216 netif_keep_dst(dev);
1217 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1218 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1219 eth_hw_addr_random(dev);
1220 }
1221
1222 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1223 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1224 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1225 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1226 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1227 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1228 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1229 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1230 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1231 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1232 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1233 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1234 };
1235
1236 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1237 {
1238 if (tb[IFLA_ADDRESS]) {
1239 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1240 return -EINVAL;
1241
1242 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1243 return -EADDRNOTAVAIL;
1244 }
1245
1246 if (!data)
1247 return -EINVAL;
1248
1249 if (data[IFLA_GENEVE_ID]) {
1250 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1251
1252 if (vni >= GENEVE_VID_MASK)
1253 return -ERANGE;
1254 }
1255
1256 return 0;
1257 }
1258
1259 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1260 __be16 dst_port,
1261 union geneve_addr *remote,
1262 u8 vni[],
1263 bool *tun_on_same_port,
1264 bool *tun_collect_md)
1265 {
1266 struct geneve_dev *geneve, *t;
1267
1268 *tun_on_same_port = false;
1269 *tun_collect_md = false;
1270 t = NULL;
1271 list_for_each_entry(geneve, &gn->geneve_list, next) {
1272 if (geneve->dst_port == dst_port) {
1273 *tun_collect_md = geneve->collect_md;
1274 *tun_on_same_port = true;
1275 }
1276 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1277 !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1278 dst_port == geneve->dst_port)
1279 t = geneve;
1280 }
1281 return t;
1282 }
1283
1284 static int geneve_configure(struct net *net, struct net_device *dev,
1285 union geneve_addr *remote,
1286 __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1287 __be16 dst_port, bool metadata, u32 flags)
1288 {
1289 struct geneve_net *gn = net_generic(net, geneve_net_id);
1290 struct geneve_dev *t, *geneve = netdev_priv(dev);
1291 bool tun_collect_md, tun_on_same_port;
1292 int err, encap_len;
1293
1294 if (!remote)
1295 return -EINVAL;
1296 if (metadata &&
1297 (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1298 return -EINVAL;
1299
1300 geneve->net = net;
1301 geneve->dev = dev;
1302
1303 geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1304 geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1305 geneve->vni[2] = vni & 0x000000ff;
1306
1307 if ((remote->sa.sa_family == AF_INET &&
1308 IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1309 (remote->sa.sa_family == AF_INET6 &&
1310 ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1311 return -EINVAL;
1312 if (label && remote->sa.sa_family != AF_INET6)
1313 return -EINVAL;
1314
1315 geneve->remote = *remote;
1316
1317 geneve->ttl = ttl;
1318 geneve->tos = tos;
1319 geneve->label = label;
1320 geneve->dst_port = dst_port;
1321 geneve->collect_md = metadata;
1322 geneve->flags = flags;
1323
1324 t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1325 &tun_on_same_port, &tun_collect_md);
1326 if (t)
1327 return -EBUSY;
1328
1329 /* make enough headroom for basic scenario */
1330 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1331 if (remote->sa.sa_family == AF_INET)
1332 encap_len += sizeof(struct iphdr);
1333 else
1334 encap_len += sizeof(struct ipv6hdr);
1335 dev->needed_headroom = encap_len + ETH_HLEN;
1336
1337 if (metadata) {
1338 if (tun_on_same_port)
1339 return -EPERM;
1340 } else {
1341 if (tun_collect_md)
1342 return -EPERM;
1343 }
1344
1345 dst_cache_reset(&geneve->dst_cache);
1346
1347 err = register_netdevice(dev);
1348 if (err)
1349 return err;
1350
1351 list_add(&geneve->next, &gn->geneve_list);
1352 return 0;
1353 }
1354
1355 static int geneve_newlink(struct net *net, struct net_device *dev,
1356 struct nlattr *tb[], struct nlattr *data[])
1357 {
1358 __be16 dst_port = htons(GENEVE_UDP_PORT);
1359 __u8 ttl = 0, tos = 0;
1360 bool metadata = false;
1361 union geneve_addr remote = geneve_remote_unspec;
1362 __be32 label = 0;
1363 __u32 vni = 0;
1364 u32 flags = 0;
1365
1366 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1367 return -EINVAL;
1368
1369 if (data[IFLA_GENEVE_REMOTE]) {
1370 remote.sa.sa_family = AF_INET;
1371 remote.sin.sin_addr.s_addr =
1372 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1373 }
1374
1375 if (data[IFLA_GENEVE_REMOTE6]) {
1376 if (!IS_ENABLED(CONFIG_IPV6))
1377 return -EPFNOSUPPORT;
1378
1379 remote.sa.sa_family = AF_INET6;
1380 remote.sin6.sin6_addr =
1381 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1382
1383 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1384 IPV6_ADDR_LINKLOCAL) {
1385 netdev_dbg(dev, "link-local remote is unsupported\n");
1386 return -EINVAL;
1387 }
1388 }
1389
1390 if (data[IFLA_GENEVE_ID])
1391 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1392
1393 if (data[IFLA_GENEVE_TTL])
1394 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1395
1396 if (data[IFLA_GENEVE_TOS])
1397 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1398
1399 if (data[IFLA_GENEVE_LABEL])
1400 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1401 IPV6_FLOWLABEL_MASK;
1402
1403 if (data[IFLA_GENEVE_PORT])
1404 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1405
1406 if (data[IFLA_GENEVE_COLLECT_METADATA])
1407 metadata = true;
1408
1409 if (data[IFLA_GENEVE_UDP_CSUM] &&
1410 !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1411 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1412
1413 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1414 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1415 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1416
1417 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1418 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1419 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1420
1421 return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1422 dst_port, metadata, flags);
1423 }
1424
1425 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1426 {
1427 struct geneve_dev *geneve = netdev_priv(dev);
1428
1429 list_del(&geneve->next);
1430 unregister_netdevice_queue(dev, head);
1431 }
1432
1433 static size_t geneve_get_size(const struct net_device *dev)
1434 {
1435 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1436 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1437 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1438 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1439 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1440 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1441 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1442 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1443 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1444 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1445 0;
1446 }
1447
1448 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1449 {
1450 struct geneve_dev *geneve = netdev_priv(dev);
1451 __u32 vni;
1452
1453 vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1454 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1455 goto nla_put_failure;
1456
1457 if (geneve->remote.sa.sa_family == AF_INET) {
1458 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1459 geneve->remote.sin.sin_addr.s_addr))
1460 goto nla_put_failure;
1461 #if IS_ENABLED(CONFIG_IPV6)
1462 } else {
1463 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1464 &geneve->remote.sin6.sin6_addr))
1465 goto nla_put_failure;
1466 #endif
1467 }
1468
1469 if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1470 nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1471 nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1472 goto nla_put_failure;
1473
1474 if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1475 goto nla_put_failure;
1476
1477 if (geneve->collect_md) {
1478 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1479 goto nla_put_failure;
1480 }
1481
1482 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1483 !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1484 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1485 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1486 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1487 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1488 goto nla_put_failure;
1489
1490 return 0;
1491
1492 nla_put_failure:
1493 return -EMSGSIZE;
1494 }
1495
1496 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1497 .kind = "geneve",
1498 .maxtype = IFLA_GENEVE_MAX,
1499 .policy = geneve_policy,
1500 .priv_size = sizeof(struct geneve_dev),
1501 .setup = geneve_setup,
1502 .validate = geneve_validate,
1503 .newlink = geneve_newlink,
1504 .dellink = geneve_dellink,
1505 .get_size = geneve_get_size,
1506 .fill_info = geneve_fill_info,
1507 };
1508
1509 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1510 u8 name_assign_type, u16 dst_port)
1511 {
1512 struct nlattr *tb[IFLA_MAX + 1];
1513 struct net_device *dev;
1514 int err;
1515
1516 memset(tb, 0, sizeof(tb));
1517 dev = rtnl_create_link(net, name, name_assign_type,
1518 &geneve_link_ops, tb);
1519 if (IS_ERR(dev))
1520 return dev;
1521
1522 err = geneve_configure(net, dev, &geneve_remote_unspec,
1523 0, 0, 0, 0, htons(dst_port), true,
1524 GENEVE_F_UDP_ZERO_CSUM6_RX);
1525 if (err)
1526 goto err;
1527
1528 /* openvswitch users expect packet sizes to be unrestricted,
1529 * so set the largest MTU we can.
1530 */
1531 err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1532 if (err)
1533 goto err;
1534
1535 return dev;
1536
1537 err:
1538 free_netdev(dev);
1539 return ERR_PTR(err);
1540 }
1541 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1542
1543 static int geneve_netdevice_event(struct notifier_block *unused,
1544 unsigned long event, void *ptr)
1545 {
1546 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1547
1548 if (event == NETDEV_OFFLOAD_PUSH_GENEVE)
1549 geneve_push_rx_ports(dev);
1550
1551 return NOTIFY_DONE;
1552 }
1553
1554 static struct notifier_block geneve_notifier_block __read_mostly = {
1555 .notifier_call = geneve_netdevice_event,
1556 };
1557
1558 static __net_init int geneve_init_net(struct net *net)
1559 {
1560 struct geneve_net *gn = net_generic(net, geneve_net_id);
1561
1562 INIT_LIST_HEAD(&gn->geneve_list);
1563 INIT_LIST_HEAD(&gn->sock_list);
1564 return 0;
1565 }
1566
1567 static void __net_exit geneve_exit_net(struct net *net)
1568 {
1569 struct geneve_net *gn = net_generic(net, geneve_net_id);
1570 struct geneve_dev *geneve, *next;
1571 struct net_device *dev, *aux;
1572 LIST_HEAD(list);
1573
1574 rtnl_lock();
1575
1576 /* gather any geneve devices that were moved into this ns */
1577 for_each_netdev_safe(net, dev, aux)
1578 if (dev->rtnl_link_ops == &geneve_link_ops)
1579 unregister_netdevice_queue(dev, &list);
1580
1581 /* now gather any other geneve devices that were created in this ns */
1582 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1583 /* If geneve->dev is in the same netns, it was already added
1584 * to the list by the previous loop.
1585 */
1586 if (!net_eq(dev_net(geneve->dev), net))
1587 unregister_netdevice_queue(geneve->dev, &list);
1588 }
1589
1590 /* unregister the devices gathered above */
1591 unregister_netdevice_many(&list);
1592 rtnl_unlock();
1593 }
1594
1595 static struct pernet_operations geneve_net_ops = {
1596 .init = geneve_init_net,
1597 .exit = geneve_exit_net,
1598 .id = &geneve_net_id,
1599 .size = sizeof(struct geneve_net),
1600 };
1601
1602 static int __init geneve_init_module(void)
1603 {
1604 int rc;
1605
1606 rc = register_pernet_subsys(&geneve_net_ops);
1607 if (rc)
1608 goto out1;
1609
1610 rc = register_netdevice_notifier(&geneve_notifier_block);
1611 if (rc)
1612 goto out2;
1613
1614 rc = rtnl_link_register(&geneve_link_ops);
1615 if (rc)
1616 goto out3;
1617
1618 return 0;
1619
1620 out3:
1621 unregister_netdevice_notifier(&geneve_notifier_block);
1622 out2:
1623 unregister_pernet_subsys(&geneve_net_ops);
1624 out1:
1625 return rc;
1626 }
1627 late_initcall(geneve_init_module);
1628
1629 static void __exit geneve_cleanup_module(void)
1630 {
1631 rtnl_link_unregister(&geneve_link_ops);
1632 unregister_netdevice_notifier(&geneve_notifier_block);
1633 unregister_pernet_subsys(&geneve_net_ops);
1634 }
1635 module_exit(geneve_cleanup_module);
1636
1637 MODULE_LICENSE("GPL");
1638 MODULE_VERSION(GENEVE_NETDEV_VER);
1639 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1640 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1641 MODULE_ALIAS_RTNL_LINK("geneve");
Linux kernel - Deliverables
This page took 0.084929 seconds and 6 git commands to generate.