]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/linux/compat/geneve.c
projects / mirror_ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
datapath: backport: udp_offload: Set encapsulation before inner completes.
[mirror_ovs.git] / datapath / linux / compat / 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 <linux/if_link.h>
19 #include <linux/if_vlan.h>
20
21 #include <net/addrconf.h>
22 #include <net/dst_cache.h>
23 #include <net/dst_metadata.h>
24 #include <net/net_namespace.h>
25 #include <net/netns/generic.h>
26 #include <net/rtnetlink.h>
27 #include <net/geneve.h>
28 #include <net/protocol.h>
29 #include <net/udp_tunnel.h>
30 #include <net/ip6_route.h>
31 #if IS_ENABLED(CONFIG_IPV6)
32 #include <net/ipv6.h>
33 #include <net/addrconf.h>
34 #include <net/ip6_tunnel.h>
35 #include <net/ip6_checksum.h>
36 #endif
37
38
39 #include "gso.h"
40 #include "vport-netdev.h"
41 #include "compat.h"
42
43 #ifndef USE_UPSTREAM_TUNNEL
44
45 #define GENEVE_NETDEV_VER "0.6"
46
47 #define GENEVE_UDP_PORT 6081
48
49 #define GENEVE_N_VID (1u << 24)
50 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
51
52 #define VNI_HASH_BITS 10
53 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
54
55 #define GENEVE_VER 0
56 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
57
58 /* per-network namespace private data for this module */
59 struct geneve_net {
60 struct list_head geneve_list;
61 struct list_head sock_list;
62 };
63
64 static int geneve_net_id;
65
66 union geneve_addr {
67 struct sockaddr_in sin;
68 struct sockaddr_in6 sin6;
69 struct sockaddr sa;
70 };
71
72 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, };
73
74 /* Pseudo network device */
75 struct geneve_dev {
76 struct hlist_node hlist; /* vni hash table */
77 struct net *net; /* netns for packet i/o */
78 struct net_device *dev; /* netdev for geneve tunnel */
79 struct geneve_sock *sock4; /* IPv4 socket used for geneve tunnel */
80 #if IS_ENABLED(CONFIG_IPV6)
81 struct geneve_sock *sock6; /* IPv6 socket used for geneve tunnel */
82 #endif
83 u8 vni[3]; /* virtual network ID for tunnel */
84 u8 ttl; /* TTL override */
85 u8 tos; /* TOS override */
86 union geneve_addr remote; /* IP address for link partner */
87 struct list_head next; /* geneve's per namespace list */
88 __be32 label; /* IPv6 flowlabel override */
89 __be16 dst_port;
90 bool collect_md;
91 u32 flags;
92 struct dst_cache dst_cache;
93 };
94
95 /* Geneve device flags */
96 #define GENEVE_F_UDP_ZERO_CSUM_TX BIT(0)
97 #define GENEVE_F_UDP_ZERO_CSUM6_TX BIT(1)
98 #define GENEVE_F_UDP_ZERO_CSUM6_RX BIT(2)
99
100 struct geneve_sock {
101 bool collect_md;
102 struct list_head list;
103 struct socket *sock;
104 struct rcu_head rcu;
105 int refcnt;
106 struct hlist_head vni_list[VNI_HASH_SIZE];
107 u32 flags;
108 #ifdef HAVE_UDP_OFFLOAD
109 struct udp_offload udp_offloads;
110 #endif
111 };
112
113 static inline __u32 geneve_net_vni_hash(u8 vni[3])
114 {
115 __u32 vnid;
116
117 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
118 return hash_32(vnid, VNI_HASH_BITS);
119 }
120
121 static __be64 vni_to_tunnel_id(const __u8 *vni)
122 {
123 #ifdef __BIG_ENDIAN
124 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
125 #else
126 return (__force __be64)(((__force u64)vni[0] << 40) |
127 ((__force u64)vni[1] << 48) |
128 ((__force u64)vni[2] << 56));
129 #endif
130 }
131
132 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
133 {
134 return gs->sock->sk->sk_family;
135 }
136
137 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
138 __be32 addr, 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 addr == geneve->remote.sin.sin_addr.s_addr)
150 return geneve;
151 }
152 return NULL;
153 }
154
155 #if IS_ENABLED(CONFIG_IPV6)
156 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
157 struct in6_addr addr6, u8 vni[])
158 {
159 struct hlist_head *vni_list_head;
160 struct geneve_dev *geneve;
161 __u32 hash;
162
163 /* Find the device for this VNI */
164 hash = geneve_net_vni_hash(vni);
165 vni_list_head = &gs->vni_list[hash];
166 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
167 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
168 ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr))
169 return geneve;
170 }
171 return NULL;
172 }
173 #endif
174
175 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
176 {
177 return (struct genevehdr *)(udp_hdr(skb) + 1);
178 }
179
180 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
181 struct sk_buff *skb)
182 {
183 u8 *vni;
184 __be32 addr;
185 static u8 zero_vni[3];
186 #if IS_ENABLED(CONFIG_IPV6)
187 static struct in6_addr zero_addr6;
188 #endif
189
190 if (geneve_get_sk_family(gs) == AF_INET) {
191 struct iphdr *iph;
192
193 iph = ip_hdr(skb); /* outer IP header... */
194
195 if (gs->collect_md) {
196 vni = zero_vni;
197 addr = 0;
198 } else {
199 vni = geneve_hdr(skb)->vni;
200 addr = iph->saddr;
201 }
202
203 return geneve_lookup(gs, addr, vni);
204 #if IS_ENABLED(CONFIG_IPV6)
205 } else if (geneve_get_sk_family(gs) == AF_INET6) {
206 struct ipv6hdr *ip6h;
207 struct in6_addr addr6;
208
209 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
210
211 if (gs->collect_md) {
212 vni = zero_vni;
213 addr6 = zero_addr6;
214 } else {
215 vni = geneve_hdr(skb)->vni;
216 addr6 = ip6h->saddr;
217 }
218
219 return geneve6_lookup(gs, addr6, vni);
220 #endif
221 }
222 return NULL;
223 }
224
225 /* geneve receive/decap routine */
226 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
227 struct sk_buff *skb)
228 {
229 struct genevehdr *gnvh = geneve_hdr(skb);
230 struct metadata_dst *tun_dst = NULL;
231 struct pcpu_sw_netstats *stats;
232 int err = 0;
233 void *oiph;
234 union {
235 struct metadata_dst dst;
236 char buf[sizeof(struct metadata_dst) + 256];
237 } buf;
238
239 if (ip_tunnel_collect_metadata() || gs->collect_md) {
240 __be16 flags;
241
242 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
243 (gnvh->oam ? TUNNEL_OAM : 0) |
244 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
245
246 tun_dst = &buf.dst;
247 ovs_udp_tun_rx_dst(&tun_dst->u.tun_info,
248 skb, geneve_get_sk_family(gs), flags,
249 vni_to_tunnel_id(gnvh->vni),
250 gnvh->opt_len * 4);
251 if (!tun_dst)
252 goto drop;
253 /* Update tunnel dst according to Geneve options. */
254 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
255 gnvh->options, gnvh->opt_len * 4);
256 } else {
257 /* Drop packets w/ critical options,
258 * since we don't support any...
259 */
260 if (gnvh->critical)
261 goto drop;
262 }
263
264 skb_reset_mac_header(skb);
265 skb->protocol = eth_type_trans(skb, geneve->dev);
266 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
267
268 if (tun_dst)
269 ovs_skb_dst_set(skb, &tun_dst->dst);
270
271 /* Ignore packet loops (and multicast echo) */
272 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
273 goto drop;
274
275 oiph = skb_network_header(skb);
276 skb_reset_network_header(skb);
277
278 if (geneve_get_sk_family(gs) == AF_INET)
279 err = IP_ECN_decapsulate(oiph, skb);
280 #if IS_ENABLED(CONFIG_IPV6)
281 else
282 err = IP6_ECN_decapsulate(oiph, skb);
283 #endif
284 if (unlikely(err > 1)) {
285 ++geneve->dev->stats.rx_frame_errors;
286 ++geneve->dev->stats.rx_errors;
287 goto drop;
288 }
289
290 stats = this_cpu_ptr(geneve->dev->tstats);
291 u64_stats_update_begin(&stats->syncp);
292 stats->rx_packets++;
293 stats->rx_bytes += skb->len;
294 u64_stats_update_end(&stats->syncp);
295
296 netdev_port_receive(skb, skb_tunnel_info(skb));
297 return;
298 drop:
299 /* Consume bad packet */
300 kfree_skb(skb);
301 }
302
303 /* Setup stats when device is created */
304 static int geneve_init(struct net_device *dev)
305 {
306 struct geneve_dev *geneve = netdev_priv(dev);
307 int err;
308
309 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
310 if (!dev->tstats)
311 return -ENOMEM;
312
313 err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL);
314 if (err) {
315 free_percpu(dev->tstats);
316 return err;
317 }
318
319 return 0;
320 }
321
322 static void geneve_uninit(struct net_device *dev)
323 {
324 struct geneve_dev *geneve = netdev_priv(dev);
325
326 dst_cache_destroy(&geneve->dst_cache);
327 free_percpu(dev->tstats);
328 }
329
330 /* Callback from net/ipv4/udp.c to receive packets */
331 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
332 {
333 struct genevehdr *geneveh;
334 struct geneve_dev *geneve;
335 struct geneve_sock *gs;
336 int opts_len;
337
338 /* Need Geneve and inner Ethernet header to be present */
339 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
340 goto error;
341
342 /* Return packets with reserved bits set */
343 geneveh = geneve_hdr(skb);
344 if (unlikely(geneveh->ver != GENEVE_VER))
345 goto error;
346
347 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
348 goto error;
349
350 gs = rcu_dereference_sk_user_data(sk);
351 if (!gs)
352 goto drop;
353
354 #if IS_ENABLED(CONFIG_IPV6)
355 #ifdef OVS_CHECK_UDP_TUNNEL_ZERO_CSUM
356 if (geneve_get_sk_family(gs) == AF_INET6 &&
357 !udp_hdr(skb)->check &&
358 !(gs->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)) {
359 udp6_csum_zero_error(skb);
360 goto drop;
361 }
362 #endif
363 #endif
364 geneve = geneve_lookup_skb(gs, skb);
365 if (!geneve)
366 goto drop;
367
368 opts_len = geneveh->opt_len * 4;
369 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
370 htons(ETH_P_TEB),
371 !net_eq(geneve->net, dev_net(geneve->dev))))
372 goto drop;
373
374 geneve_rx(geneve, gs, skb);
375 return 0;
376
377 drop:
378 /* Consume bad packet */
379 kfree_skb(skb);
380 return 0;
381
382 error:
383 /* Let the UDP layer deal with the skb */
384 return 1;
385 }
386
387 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
388 __be16 port, u32 flags)
389 {
390 struct socket *sock;
391 struct udp_port_cfg udp_conf;
392 int err;
393
394 memset(&udp_conf, 0, sizeof(udp_conf));
395
396 if (ipv6) {
397 udp_conf.family = AF_INET6;
398 udp_conf.ipv6_v6only = 1;
399 udp_conf.use_udp6_rx_checksums =
400 !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX);
401 } else {
402 udp_conf.family = AF_INET;
403 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
404 }
405
406 udp_conf.local_udp_port = port;
407
408 /* Open UDP socket */
409 err = udp_sock_create(net, &udp_conf, &sock);
410 if (err < 0)
411 return ERR_PTR(err);
412
413 return sock;
414 }
415
416 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
417 {
418 #ifdef HAVE_NDO_ADD_GENEVE_PORT
419 struct net_device *dev;
420 struct sock *sk = gs->sock->sk;
421 struct net *net = sock_net(sk);
422 sa_family_t sa_family = geneve_get_sk_family(gs);
423 __be16 port = inet_sk(sk)->inet_sport;
424
425 rcu_read_lock();
426 for_each_netdev_rcu(net, dev) {
427 if (dev->netdev_ops->ndo_add_geneve_port)
428 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family,
429 port);
430 }
431 rcu_read_unlock();
432 #else
433 #ifdef HAVE_UDP_OFFLOAD
434 struct sock *sk = gs->sock->sk;
435 sa_family_t sa_family = sk->sk_family;
436 int err;
437
438 if (sa_family == AF_INET) {
439 err = udp_add_offload(&gs->udp_offloads);
440 if (err)
441 pr_warn("geneve: udp_add_offload failed with status %d\n",
442 err);
443 }
444 #endif
445 #endif
446
447 }
448
449 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
450 {
451 #ifdef HAVE_NDO_ADD_GENEVE_PORT
452 struct net_device *dev;
453 struct sock *sk = gs->sock->sk;
454 struct net *net = sock_net(sk);
455 sa_family_t sa_family = geneve_get_sk_family(gs);
456 __be16 port = inet_sk(sk)->inet_sport;
457
458 rcu_read_lock();
459 for_each_netdev_rcu(net, dev) {
460 if (dev->netdev_ops->ndo_del_geneve_port)
461 dev->netdev_ops->ndo_del_geneve_port(dev, sa_family,
462 port);
463 }
464
465 rcu_read_unlock();
466 #else
467 #ifdef HAVE_UDP_OFFLOAD
468 struct sock *sk = gs->sock->sk;
469 sa_family_t sa_family = sk->sk_family;
470
471 if (sa_family == AF_INET)
472 udp_del_offload(&gs->udp_offloads);
473 #endif
474
475 #endif
476 }
477
478 #if defined(HAVE_UDP_OFFLOAD) || \
479 defined(HAVE_UDP_TUNNEL_SOCK_CFG_GRO_RECEIVE)
480
481 static int geneve_hlen(struct genevehdr *gh)
482 {
483 return sizeof(*gh) + gh->opt_len * 4;
484 }
485
486 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
487 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
488 struct sk_buff *skb)
489 #else
490 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
491 struct sk_buff *skb,
492 struct udp_offload *uoff)
493 #endif
494 {
495 struct sk_buff *p, **pp = NULL;
496 struct genevehdr *gh, *gh2;
497 unsigned int hlen, gh_len, off_gnv;
498 const struct packet_offload *ptype;
499 __be16 type;
500 int flush = 1;
501
502 off_gnv = skb_gro_offset(skb);
503 hlen = off_gnv + sizeof(*gh);
504 gh = skb_gro_header_fast(skb, off_gnv);
505 if (skb_gro_header_hard(skb, hlen)) {
506 gh = skb_gro_header_slow(skb, hlen, off_gnv);
507 if (unlikely(!gh))
508 goto out;
509 }
510
511 if (gh->ver != GENEVE_VER || gh->oam)
512 goto out;
513 gh_len = geneve_hlen(gh);
514
515 hlen = off_gnv + gh_len;
516 if (skb_gro_header_hard(skb, hlen)) {
517 gh = skb_gro_header_slow(skb, hlen, off_gnv);
518 if (unlikely(!gh))
519 goto out;
520 }
521
522 for (p = *head; p; p = p->next) {
523 if (!NAPI_GRO_CB(p)->same_flow)
524 continue;
525
526 gh2 = (struct genevehdr *)(p->data + off_gnv);
527 if (gh->opt_len != gh2->opt_len ||
528 memcmp(gh, gh2, gh_len)) {
529 NAPI_GRO_CB(p)->same_flow = 0;
530 continue;
531 }
532 }
533
534 type = gh->proto_type;
535
536 rcu_read_lock();
537 ptype = gro_find_receive_by_type(type);
538 if (!ptype)
539 goto out_unlock;
540
541 skb_gro_pull(skb, gh_len);
542 skb_gro_postpull_rcsum(skb, gh, gh_len);
543 pp = ptype->callbacks.gro_receive(head, skb);
544 flush = 0;
545
546 out_unlock:
547 rcu_read_unlock();
548 out:
549 NAPI_GRO_CB(skb)->flush |= flush;
550
551 return pp;
552 }
553
554 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
555 static int geneve_gro_complete(struct sk_buff *skb, int nhoff)
556 #else
557 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
558 struct udp_offload *uoff)
559 #endif
560 {
561 struct genevehdr *gh;
562 struct packet_offload *ptype;
563 __be16 type;
564 int gh_len;
565 int err = -ENOSYS;
566
567 udp_tunnel_gro_complete(skb, nhoff);
568
569 gh = (struct genevehdr *)(skb->data + nhoff);
570 gh_len = geneve_hlen(gh);
571 type = gh->proto_type;
572
573 rcu_read_lock();
574 ptype = gro_find_complete_by_type(type);
575 if (ptype)
576 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
577
578 rcu_read_unlock();
579
580 skb_set_inner_mac_header(skb, nhoff + gh_len);
581 return err;
582 }
583 #endif
584
585 /* Create new listen socket if needed */
586 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
587 bool ipv6, u32 flags)
588 {
589 struct geneve_net *gn = net_generic(net, geneve_net_id);
590 struct geneve_sock *gs;
591 struct socket *sock;
592 struct udp_tunnel_sock_cfg tunnel_cfg;
593 int h;
594
595 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
596 if (!gs)
597 return ERR_PTR(-ENOMEM);
598
599 sock = geneve_create_sock(net, ipv6, port, flags);
600 if (IS_ERR(sock)) {
601 kfree(gs);
602 return ERR_CAST(sock);
603 }
604
605 gs->sock = sock;
606 gs->refcnt = 1;
607 for (h = 0; h < VNI_HASH_SIZE; ++h)
608 INIT_HLIST_HEAD(&gs->vni_list[h]);
609
610 /* Initialize the geneve udp offloads structure */
611 #ifdef HAVE_UDP_OFFLOAD
612 gs->udp_offloads.port = port;
613 gs->udp_offloads.callbacks.gro_receive = geneve_gro_receive;
614 gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
615 #endif
616
617 geneve_notify_add_rx_port(gs);
618 /* Mark socket as an encapsulation socket */
619 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
620 tunnel_cfg.sk_user_data = gs;
621 tunnel_cfg.encap_type = 1;
622 #ifdef HAVE_UDP_TUNNEL_SOCK_CFG_GRO_RECEIVE
623 tunnel_cfg.gro_receive = geneve_gro_receive;
624 tunnel_cfg.gro_complete = geneve_gro_complete;
625 #endif
626 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
627 tunnel_cfg.encap_destroy = NULL;
628 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
629 list_add(&gs->list, &gn->sock_list);
630 return gs;
631 }
632
633 static void __geneve_sock_release(struct geneve_sock *gs)
634 {
635 if (!gs || --gs->refcnt)
636 return;
637
638 list_del(&gs->list);
639 geneve_notify_del_rx_port(gs);
640 udp_tunnel_sock_release(gs->sock);
641 kfree_rcu(gs, rcu);
642 }
643
644 static void geneve_sock_release(struct geneve_dev *geneve)
645 {
646 __geneve_sock_release(geneve->sock4);
647 #if IS_ENABLED(CONFIG_IPV6)
648 __geneve_sock_release(geneve->sock6);
649 #endif
650 }
651
652 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
653 sa_family_t family,
654 __be16 dst_port)
655 {
656 struct geneve_sock *gs;
657
658 list_for_each_entry(gs, &gn->sock_list, list) {
659 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
660 geneve_get_sk_family(gs) == family) {
661 return gs;
662 }
663 }
664 return NULL;
665 }
666
667 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
668 {
669 struct net *net = geneve->net;
670 struct geneve_net *gn = net_generic(net, geneve_net_id);
671 struct geneve_sock *gs;
672 __u32 hash;
673
674 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port);
675 if (gs) {
676 gs->refcnt++;
677 goto out;
678 }
679
680 gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags);
681 if (IS_ERR(gs))
682 return PTR_ERR(gs);
683
684 out:
685 gs->collect_md = geneve->collect_md;
686 gs->flags = geneve->flags;
687 #if IS_ENABLED(CONFIG_IPV6)
688 if (ipv6)
689 geneve->sock6 = gs;
690 else
691 #endif
692 geneve->sock4 = gs;
693
694 hash = geneve_net_vni_hash(geneve->vni);
695 hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
696 return 0;
697 }
698
699 static int geneve_open(struct net_device *dev)
700 {
701 struct geneve_dev *geneve = netdev_priv(dev);
702 bool ipv6 = geneve->remote.sa.sa_family == AF_INET6;
703 bool metadata = geneve->collect_md;
704 int ret = 0;
705
706 geneve->sock4 = NULL;
707 #if IS_ENABLED(CONFIG_IPV6)
708 geneve->sock6 = NULL;
709 if (ipv6 || metadata)
710 ret = geneve_sock_add(geneve, true);
711 #endif
712
713 if (!ret && (!ipv6 || metadata))
714 ret = geneve_sock_add(geneve, false);
715 if (ret < 0)
716 geneve_sock_release(geneve);
717
718 return ret;
719 }
720
721 static int geneve_stop(struct net_device *dev)
722 {
723 struct geneve_dev *geneve = netdev_priv(dev);
724
725 if (!hlist_unhashed(&geneve->hlist))
726 hlist_del_rcu(&geneve->hlist);
727 geneve_sock_release(geneve);
728 return 0;
729 }
730
731 static void geneve_build_header(struct genevehdr *geneveh,
732 __be16 tun_flags, u8 vni[3],
733 u8 options_len, u8 *options)
734 {
735 geneveh->ver = GENEVE_VER;
736 geneveh->opt_len = options_len / 4;
737 geneveh->oam = !!(tun_flags & TUNNEL_OAM);
738 geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
739 geneveh->rsvd1 = 0;
740 memcpy(geneveh->vni, vni, 3);
741 geneveh->proto_type = htons(ETH_P_TEB);
742 geneveh->rsvd2 = 0;
743
744 memcpy(geneveh->options, options, options_len);
745 }
746
747 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
748 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
749 u32 flags, bool xnet)
750 {
751 struct genevehdr *gnvh;
752 int min_headroom;
753 int err;
754 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX);
755
756 skb_scrub_packet(skb, xnet);
757
758 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
759 + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
760 err = skb_cow_head(skb, min_headroom);
761 if (unlikely(err))
762 goto free_rt;
763
764 err = udp_tunnel_handle_offloads(skb, udp_sum, false);
765 if (err)
766 goto free_rt;
767
768 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
769 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
770
771 ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
772 return 0;
773
774 free_rt:
775 ip_rt_put(rt);
776 return err;
777 }
778
779 #if IS_ENABLED(CONFIG_IPV6)
780 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb,
781 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
782 u32 flags, bool xnet)
783 {
784 struct genevehdr *gnvh;
785 int min_headroom;
786 int err;
787 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX);
788
789 skb_scrub_packet(skb, xnet);
790
791 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
792 + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr);
793 err = skb_cow_head(skb, min_headroom);
794 if (unlikely(err))
795 goto free_dst;
796
797 err = udp_tunnel_handle_offloads(skb, udp_sum, false);
798 if (err)
799 goto free_dst;
800
801 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
802 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt);
803
804 ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
805 return 0;
806
807 free_dst:
808 dst_release(dst);
809 return err;
810 }
811 #endif
812
813 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
814 struct net_device *dev,
815 struct flowi4 *fl4,
816 struct ip_tunnel_info *info)
817 {
818 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
819 struct geneve_dev *geneve = netdev_priv(dev);
820 struct dst_cache *dst_cache;
821 struct rtable *rt = NULL;
822 __u8 tos;
823
824 memset(fl4, 0, sizeof(*fl4));
825 fl4->flowi4_mark = skb->mark;
826 fl4->flowi4_proto = IPPROTO_UDP;
827
828 if (info) {
829 fl4->daddr = info->key.u.ipv4.dst;
830 fl4->saddr = info->key.u.ipv4.src;
831 fl4->flowi4_tos = RT_TOS(info->key.tos);
832 dst_cache = &info->dst_cache;
833 } else {
834 tos = geneve->tos;
835 if (tos == 1) {
836 const struct iphdr *iip = ip_hdr(skb);
837
838 tos = ip_tunnel_get_dsfield(iip, skb);
839 use_cache = false;
840 }
841
842 fl4->flowi4_tos = RT_TOS(tos);
843 fl4->daddr = geneve->remote.sin.sin_addr.s_addr;
844 dst_cache = &geneve->dst_cache;
845 }
846
847 if (use_cache) {
848 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
849 if (rt)
850 return rt;
851 }
852
853 rt = ip_route_output_key(geneve->net, fl4);
854 if (IS_ERR(rt)) {
855 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
856 return ERR_PTR(-ENETUNREACH);
857 }
858 if (rt->dst.dev == dev) { /* is this necessary? */
859 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
860 ip_rt_put(rt);
861 return ERR_PTR(-ELOOP);
862 }
863 if (use_cache)
864 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
865 return rt;
866 }
867
868 #if IS_ENABLED(CONFIG_IPV6)
869 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
870 struct net_device *dev,
871 struct flowi6 *fl6,
872 struct ip_tunnel_info *info)
873 {
874 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
875 struct geneve_dev *geneve = netdev_priv(dev);
876 struct geneve_sock *gs6 = geneve->sock6;
877 struct dst_entry *dst = NULL;
878 struct dst_cache *dst_cache;
879 __u8 prio;
880
881 memset(fl6, 0, sizeof(*fl6));
882 fl6->flowi6_mark = skb->mark;
883 fl6->flowi6_proto = IPPROTO_UDP;
884
885 if (info) {
886 fl6->daddr = info->key.u.ipv6.dst;
887 fl6->saddr = info->key.u.ipv6.src;
888 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos),
889 info->key.label);
890 dst_cache = &info->dst_cache;
891 } else {
892 prio = geneve->tos;
893 if (prio == 1) {
894 const struct iphdr *iip = ip_hdr(skb);
895
896 prio = ip_tunnel_get_dsfield(iip, skb);
897 use_cache = false;
898 }
899
900 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
901 geneve->label);
902 fl6->daddr = geneve->remote.sin6.sin6_addr;
903 dst_cache = &geneve->dst_cache;
904 }
905
906 if (use_cache) {
907 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
908 if (dst)
909 return dst;
910 }
911
912 #ifdef HAVE_IPV6_DST_LOOKUP_NET
913 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
914 #else
915 #ifdef HAVE_IPV6_STUB
916 if (ipv6_stub->ipv6_dst_lookup(gs6->sock->sk, &dst, fl6)) {
917 #else
918 if (ip6_dst_lookup(gs6->sock->sk, &dst, fl6)) {
919 #endif
920 #endif
921 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
922 return ERR_PTR(-ENETUNREACH);
923 }
924 if (dst->dev == dev) { /* is this necessary? */
925 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
926 dst_release(dst);
927 return ERR_PTR(-ELOOP);
928 }
929
930 if (use_cache)
931 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
932 return dst;
933 }
934 #endif
935
936 /* Convert 64 bit tunnel ID to 24 bit VNI. */
937 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
938 {
939 #ifdef __BIG_ENDIAN
940 vni[0] = (__force __u8)(tun_id >> 16);
941 vni[1] = (__force __u8)(tun_id >> 8);
942 vni[2] = (__force __u8)tun_id;
943 #else
944 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
945 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
946 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
947 #endif
948 }
949
950 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
951 struct ip_tunnel_info *info)
952 {
953 struct geneve_dev *geneve = netdev_priv(dev);
954 struct geneve_sock *gs4 = geneve->sock4;
955 struct rtable *rt = NULL;
956 const struct iphdr *iip; /* interior IP header */
957 int err = -EINVAL;
958 struct flowi4 fl4;
959 __u8 tos, ttl;
960 __be16 sport;
961 __be16 df;
962 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
963 u32 flags = geneve->flags;
964
965 if (geneve->collect_md) {
966 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
967 netdev_dbg(dev, "no tunnel metadata\n");
968 goto tx_error;
969 }
970 if (info && ip_tunnel_info_af(info) != AF_INET)
971 goto tx_error;
972 }
973
974 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
975 if (IS_ERR(rt)) {
976 err = PTR_ERR(rt);
977 goto tx_error;
978 }
979
980 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
981 skb_reset_mac_header(skb);
982
983 iip = ip_hdr(skb);
984
985 if (info) {
986 const struct ip_tunnel_key *key = &info->key;
987 u8 *opts = NULL;
988 u8 vni[3];
989
990 tunnel_id_to_vni(key->tun_id, vni);
991 if (info->options_len)
992 opts = ip_tunnel_info_opts(info);
993
994 if (key->tun_flags & TUNNEL_CSUM)
995 flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX;
996 else
997 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
998
999 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
1000 info->options_len, opts, flags, xnet);
1001 if (unlikely(err))
1002 goto tx_error;
1003
1004 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
1005 ttl = key->ttl;
1006 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
1007 } else {
1008 err = geneve_build_skb(rt, skb, 0, geneve->vni,
1009 0, NULL, flags, xnet);
1010 if (unlikely(err))
1011 goto tx_error;
1012
1013 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
1014 ttl = geneve->ttl;
1015 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
1016 ttl = 1;
1017 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1018 df = 0;
1019 }
1020 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
1021 tos, ttl, df, sport, geneve->dst_port,
1022 !net_eq(geneve->net, dev_net(geneve->dev)),
1023 !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX));
1024
1025 return NETDEV_TX_OK;
1026
1027 tx_error:
1028 dev_kfree_skb(skb);
1029
1030 if (err == -ELOOP)
1031 dev->stats.collisions++;
1032 else if (err == -ENETUNREACH)
1033 dev->stats.tx_carrier_errors++;
1034 else
1035 dev->stats.tx_errors++;
1036 return NETDEV_TX_OK;
1037 }
1038
1039 #if IS_ENABLED(CONFIG_IPV6)
1040 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
1041 struct ip_tunnel_info *info)
1042 {
1043 struct geneve_dev *geneve = netdev_priv(dev);
1044 struct geneve_sock *gs6 = geneve->sock6;
1045 struct dst_entry *dst = NULL;
1046 const struct iphdr *iip; /* interior IP header */
1047 int err = -EINVAL;
1048 struct flowi6 fl6;
1049 __u8 prio, ttl;
1050 __be16 sport;
1051 __be32 label;
1052 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1053 u32 flags = geneve->flags;
1054
1055 if (geneve->collect_md) {
1056 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1057 netdev_dbg(dev, "no tunnel metadata\n");
1058 goto tx_error;
1059 }
1060 }
1061
1062 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1063 if (IS_ERR(dst)) {
1064 err = PTR_ERR(dst);
1065 goto tx_error;
1066 }
1067
1068 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1069 skb_reset_mac_header(skb);
1070
1071 iip = ip_hdr(skb);
1072
1073 if (info) {
1074 const struct ip_tunnel_key *key = &info->key;
1075 u8 *opts = NULL;
1076 u8 vni[3];
1077
1078 tunnel_id_to_vni(key->tun_id, vni);
1079 if (info->options_len)
1080 opts = ip_tunnel_info_opts(info);
1081
1082 if (key->tun_flags & TUNNEL_CSUM)
1083 flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX;
1084 else
1085 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1086
1087 err = geneve6_build_skb(dst, skb, key->tun_flags, vni,
1088 info->options_len, opts,
1089 flags, xnet);
1090 if (unlikely(err))
1091 goto tx_error;
1092
1093 prio = ip_tunnel_ecn_encap(key->tos, iip, skb);
1094 ttl = key->ttl;
1095 label = info->key.label;
1096 } else {
1097 err = geneve6_build_skb(dst, skb, 0, geneve->vni,
1098 0, NULL, flags, xnet);
1099 if (unlikely(err))
1100 goto tx_error;
1101
1102 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1103 iip, skb);
1104 ttl = geneve->ttl;
1105 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr))
1106 ttl = 1;
1107 ttl = ttl ? : ip6_dst_hoplimit(dst);
1108 label = geneve->label;
1109 }
1110 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1111 &fl6.saddr, &fl6.daddr, prio, ttl, label,
1112 sport, geneve->dst_port,
1113 !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX));
1114 return NETDEV_TX_OK;
1115
1116 tx_error:
1117 dev_kfree_skb(skb);
1118
1119 if (err == -ELOOP)
1120 dev->stats.collisions++;
1121 else if (err == -ENETUNREACH)
1122 dev->stats.tx_carrier_errors++;
1123 else
1124 dev->stats.tx_errors++;
1125 return NETDEV_TX_OK;
1126 }
1127 #endif
1128
1129 netdev_tx_t rpl_geneve_xmit(struct sk_buff *skb)
1130 {
1131 struct net_device *dev = skb->dev;
1132 struct geneve_dev *geneve = netdev_priv(dev);
1133 struct ip_tunnel_info *info = NULL;
1134
1135 if (geneve->collect_md)
1136 info = skb_tunnel_info(skb);
1137
1138 #if IS_ENABLED(CONFIG_IPV6)
1139 if ((info && ip_tunnel_info_af(info) == AF_INET6) ||
1140 (!info && geneve->remote.sa.sa_family == AF_INET6))
1141 return geneve6_xmit_skb(skb, dev, info);
1142 #endif
1143 return geneve_xmit_skb(skb, dev, info);
1144 }
1145 EXPORT_SYMBOL_GPL(rpl_geneve_xmit);
1146
1147 static netdev_tx_t geneve_dev_xmit(struct sk_buff *skb, struct net_device *dev)
1148 {
1149 /* Drop All packets coming from networking stack. OVS-CB is
1150 * not initialized for these packets.
1151 */
1152
1153 dev_kfree_skb(skb);
1154 dev->stats.tx_dropped++;
1155 return NETDEV_TX_OK;
1156 }
1157
1158 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1159 {
1160 /* The max_mtu calculation does not take account of GENEVE
1161 * options, to avoid excluding potentially valid
1162 * configurations.
1163 */
1164 int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
1165 - dev->hard_header_len;
1166
1167 if (new_mtu < 68)
1168 return -EINVAL;
1169
1170 if (new_mtu > max_mtu) {
1171 if (strict)
1172 return -EINVAL;
1173
1174 new_mtu = max_mtu;
1175 }
1176
1177 dev->mtu = new_mtu;
1178 return 0;
1179 }
1180
1181 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1182 {
1183 return __geneve_change_mtu(dev, new_mtu, true);
1184 }
1185
1186 int ovs_geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1187 {
1188 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1189 struct geneve_dev *geneve = netdev_priv(dev);
1190 struct rtable *rt;
1191 struct flowi4 fl4;
1192 #if IS_ENABLED(CONFIG_IPV6)
1193 struct dst_entry *dst;
1194 struct flowi6 fl6;
1195 #endif
1196
1197 if (ip_tunnel_info_af(info) == AF_INET) {
1198 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
1199 if (IS_ERR(rt))
1200 return PTR_ERR(rt);
1201
1202 ip_rt_put(rt);
1203 info->key.u.ipv4.src = fl4.saddr;
1204 #if IS_ENABLED(CONFIG_IPV6)
1205 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1206 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
1207 if (IS_ERR(dst))
1208 return PTR_ERR(dst);
1209
1210 dst_release(dst);
1211 info->key.u.ipv6.src = fl6.saddr;
1212 #endif
1213 } else {
1214 return -EINVAL;
1215 }
1216
1217 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1218 1, USHRT_MAX, true);
1219 info->key.tp_dst = geneve->dst_port;
1220 return 0;
1221 }
1222 EXPORT_SYMBOL_GPL(ovs_geneve_fill_metadata_dst);
1223
1224 static const struct net_device_ops geneve_netdev_ops = {
1225 .ndo_init = geneve_init,
1226 .ndo_uninit = geneve_uninit,
1227 .ndo_open = geneve_open,
1228 .ndo_stop = geneve_stop,
1229 .ndo_start_xmit = geneve_dev_xmit,
1230 .ndo_get_stats64 = ip_tunnel_get_stats64,
1231 .ndo_change_mtu = geneve_change_mtu,
1232 .ndo_validate_addr = eth_validate_addr,
1233 .ndo_set_mac_address = eth_mac_addr,
1234 #ifdef HAVE_NDO_FILL_METADATA_DST
1235 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1236 #endif
1237 };
1238
1239 static void geneve_get_drvinfo(struct net_device *dev,
1240 struct ethtool_drvinfo *drvinfo)
1241 {
1242 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1243 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1244 }
1245
1246 static const struct ethtool_ops geneve_ethtool_ops = {
1247 .get_drvinfo = geneve_get_drvinfo,
1248 .get_link = ethtool_op_get_link,
1249 };
1250
1251 /* Info for udev, that this is a virtual tunnel endpoint */
1252 static struct device_type geneve_type = {
1253 .name = "geneve",
1254 };
1255
1256 /* Calls the ndo_add_geneve_port of the caller in order to
1257 * supply the listening GENEVE udp ports. Callers are expected
1258 * to implement the ndo_add_geneve_port.
1259 */
1260 static void geneve_push_rx_ports(struct net_device *dev)
1261 {
1262 #ifdef HAVE_NDO_ADD_GENEVE_PORT
1263 struct net *net = dev_net(dev);
1264 struct geneve_net *gn = net_generic(net, geneve_net_id);
1265 struct geneve_sock *gs;
1266 sa_family_t sa_family;
1267 struct sock *sk;
1268 __be16 port;
1269
1270 if (!dev->netdev_ops->ndo_add_geneve_port)
1271 return;
1272
1273 rcu_read_lock();
1274 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1275 sk = gs->sock->sk;
1276 sa_family = sk->sk_family;
1277 port = inet_sk(sk)->inet_sport;
1278 dev->netdev_ops->ndo_add_geneve_port(dev, sa_family, port);
1279 }
1280 rcu_read_unlock();
1281 #endif
1282 }
1283
1284 /* Initialize the device structure. */
1285 static void geneve_setup(struct net_device *dev)
1286 {
1287 ether_setup(dev);
1288
1289 dev->netdev_ops = &geneve_netdev_ops;
1290 dev->ethtool_ops = &geneve_ethtool_ops;
1291 dev->destructor = free_netdev;
1292
1293 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1294
1295 dev->features |= NETIF_F_LLTX;
1296 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1297 dev->features |= NETIF_F_RXCSUM;
1298 dev->features |= NETIF_F_GSO_SOFTWARE;
1299
1300 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1301 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1302
1303 #if 0
1304 netif_keep_dst(dev);
1305 #endif
1306 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1307 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1308 eth_hw_addr_random(dev);
1309 }
1310
1311 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1312 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1313 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1314 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1315 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1316 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1317 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1318 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1319 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1320 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1321 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1322 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1323 };
1324
1325 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1326 {
1327 if (tb[IFLA_ADDRESS]) {
1328 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1329 return -EINVAL;
1330
1331 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1332 return -EADDRNOTAVAIL;
1333 }
1334
1335 if (!data)
1336 return -EINVAL;
1337
1338 if (data[IFLA_GENEVE_ID]) {
1339 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1340
1341 if (vni >= GENEVE_VID_MASK)
1342 return -ERANGE;
1343 }
1344
1345 return 0;
1346 }
1347
1348 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1349 __be16 dst_port,
1350 union geneve_addr *remote,
1351 u8 vni[],
1352 bool *tun_on_same_port,
1353 bool *tun_collect_md)
1354 {
1355 struct geneve_dev *geneve, *t;
1356
1357 *tun_on_same_port = false;
1358 *tun_collect_md = false;
1359 t = NULL;
1360 list_for_each_entry(geneve, &gn->geneve_list, next) {
1361 if (geneve->dst_port == dst_port) {
1362 *tun_collect_md = geneve->collect_md;
1363 *tun_on_same_port = true;
1364 }
1365 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
1366 !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) &&
1367 dst_port == geneve->dst_port)
1368 t = geneve;
1369 }
1370 return t;
1371 }
1372
1373 static int geneve_configure(struct net *net, struct net_device *dev,
1374 union geneve_addr *remote,
1375 __u32 vni, __u8 ttl, __u8 tos, __be32 label,
1376 __be16 dst_port, bool metadata, u32 flags)
1377 {
1378 struct geneve_net *gn = net_generic(net, geneve_net_id);
1379 struct geneve_dev *t, *geneve = netdev_priv(dev);
1380 bool tun_collect_md, tun_on_same_port;
1381 int err, encap_len;
1382
1383 if (!remote)
1384 return -EINVAL;
1385 if (metadata &&
1386 (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label))
1387 return -EINVAL;
1388
1389 geneve->net = net;
1390 geneve->dev = dev;
1391
1392 geneve->vni[0] = (vni & 0x00ff0000) >> 16;
1393 geneve->vni[1] = (vni & 0x0000ff00) >> 8;
1394 geneve->vni[2] = vni & 0x000000ff;
1395
1396 if ((remote->sa.sa_family == AF_INET &&
1397 IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) ||
1398 (remote->sa.sa_family == AF_INET6 &&
1399 ipv6_addr_is_multicast(&remote->sin6.sin6_addr)))
1400 return -EINVAL;
1401 if (label && remote->sa.sa_family != AF_INET6)
1402 return -EINVAL;
1403
1404 geneve->remote = *remote;
1405
1406 geneve->ttl = ttl;
1407 geneve->tos = tos;
1408 geneve->label = label;
1409 geneve->dst_port = dst_port;
1410 geneve->collect_md = metadata;
1411 geneve->flags = flags;
1412
1413 t = geneve_find_dev(gn, dst_port, remote, geneve->vni,
1414 &tun_on_same_port, &tun_collect_md);
1415 if (t)
1416 return -EBUSY;
1417
1418 /* make enough headroom for basic scenario */
1419 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1420 if (remote->sa.sa_family == AF_INET)
1421 encap_len += sizeof(struct iphdr);
1422 else
1423 encap_len += sizeof(struct ipv6hdr);
1424 dev->needed_headroom = encap_len + ETH_HLEN;
1425
1426 if (metadata) {
1427 if (tun_on_same_port)
1428 return -EPERM;
1429 } else {
1430 if (tun_collect_md)
1431 return -EPERM;
1432 }
1433
1434 dst_cache_reset(&geneve->dst_cache);
1435
1436 err = register_netdevice(dev);
1437 if (err)
1438 return err;
1439
1440 list_add(&geneve->next, &gn->geneve_list);
1441 return 0;
1442 }
1443
1444 static int geneve_newlink(struct net *net, struct net_device *dev,
1445 struct nlattr *tb[], struct nlattr *data[])
1446 {
1447 __be16 dst_port = htons(GENEVE_UDP_PORT);
1448 __u8 ttl = 0, tos = 0;
1449 bool metadata = false;
1450 union geneve_addr remote = geneve_remote_unspec;
1451 __be32 label = 0;
1452 __u32 vni = 0;
1453 u32 flags = 0;
1454
1455 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1456 return -EINVAL;
1457
1458 if (data[IFLA_GENEVE_REMOTE]) {
1459 remote.sa.sa_family = AF_INET;
1460 remote.sin.sin_addr.s_addr =
1461 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1462 }
1463
1464 if (data[IFLA_GENEVE_REMOTE6]) {
1465 if (!IS_ENABLED(CONFIG_IPV6))
1466 return -EPFNOSUPPORT;
1467
1468 remote.sa.sa_family = AF_INET6;
1469 remote.sin6.sin6_addr =
1470 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1471
1472 if (ipv6_addr_type(&remote.sin6.sin6_addr) &
1473 IPV6_ADDR_LINKLOCAL) {
1474 netdev_dbg(dev, "link-local remote is unsupported\n");
1475 return -EINVAL;
1476 }
1477 }
1478
1479 if (data[IFLA_GENEVE_ID])
1480 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1481
1482 if (data[IFLA_GENEVE_TTL])
1483 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1484
1485 if (data[IFLA_GENEVE_TOS])
1486 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1487
1488 if (data[IFLA_GENEVE_LABEL])
1489 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1490 IPV6_FLOWLABEL_MASK;
1491
1492 if (data[IFLA_GENEVE_PORT])
1493 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
1494
1495 if (data[IFLA_GENEVE_COLLECT_METADATA])
1496 metadata = true;
1497
1498 if (data[IFLA_GENEVE_UDP_CSUM] &&
1499 !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1500 flags |= GENEVE_F_UDP_ZERO_CSUM_TX;
1501
1502 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1503 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1504 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX;
1505
1506 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1507 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1508 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX;
1509
1510 return geneve_configure(net, dev, &remote, vni, ttl, tos, label,
1511 dst_port, metadata, flags);
1512 }
1513
1514 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1515 {
1516 struct geneve_dev *geneve = netdev_priv(dev);
1517
1518 list_del(&geneve->next);
1519 unregister_netdevice_queue(dev, head);
1520 }
1521
1522 static size_t geneve_get_size(const struct net_device *dev)
1523 {
1524 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1525 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1526 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1527 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1528 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1529 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1530 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1531 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1532 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1533 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1534 0;
1535 }
1536
1537 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1538 {
1539 struct geneve_dev *geneve = netdev_priv(dev);
1540 __u32 vni;
1541
1542 vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1543 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1544 goto nla_put_failure;
1545
1546 if (geneve->remote.sa.sa_family == AF_INET) {
1547 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1548 geneve->remote.sin.sin_addr.s_addr))
1549 goto nla_put_failure;
1550 #if IS_ENABLED(CONFIG_IPV6)
1551 } else {
1552 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1553 &geneve->remote.sin6.sin6_addr))
1554 goto nla_put_failure;
1555 #endif
1556 }
1557
1558 if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1559 nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) ||
1560 nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label))
1561 goto nla_put_failure;
1562
1563 if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1564 goto nla_put_failure;
1565
1566 if (geneve->collect_md) {
1567 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1568 goto nla_put_failure;
1569 }
1570
1571 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1572 !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) ||
1573 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1574 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) ||
1575 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1576 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX)))
1577 goto nla_put_failure;
1578
1579 return 0;
1580
1581 nla_put_failure:
1582 return -EMSGSIZE;
1583 }
1584
1585 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1586 .kind = "geneve",
1587 .maxtype = IFLA_GENEVE_MAX,
1588 .policy = geneve_policy,
1589 .priv_size = sizeof(struct geneve_dev),
1590 .setup = geneve_setup,
1591 .validate = geneve_validate,
1592 .newlink = geneve_newlink,
1593 .dellink = geneve_dellink,
1594 .get_size = geneve_get_size,
1595 .fill_info = geneve_fill_info,
1596 };
1597
1598 struct net_device *rpl_geneve_dev_create_fb(struct net *net, const char *name,
1599 u8 name_assign_type, u16 dst_port)
1600 {
1601 struct nlattr *tb[IFLA_MAX + 1];
1602 struct net_device *dev;
1603 int err;
1604
1605 memset(tb, 0, sizeof(tb));
1606 dev = rtnl_create_link(net, name, name_assign_type,
1607 &geneve_link_ops, tb);
1608 if (IS_ERR(dev))
1609 return dev;
1610
1611 err = geneve_configure(net, dev, &geneve_remote_unspec,
1612 0, 0, 0, 0, htons(dst_port), true,
1613 GENEVE_F_UDP_ZERO_CSUM6_RX);
1614 if (err)
1615 goto err;
1616
1617 /* openvswitch users expect packet sizes to be unrestricted,
1618 * so set the largest MTU we can.
1619 */
1620 err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1621 if (err)
1622 goto err;
1623
1624 return dev;
1625
1626 err:
1627 free_netdev(dev);
1628 return ERR_PTR(err);
1629 }
1630 EXPORT_SYMBOL_GPL(rpl_geneve_dev_create_fb);
1631
1632 static int geneve_netdevice_event(struct notifier_block *unused,
1633 unsigned long event, void *ptr)
1634 {
1635 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1636
1637 if (event == NETDEV_OFFLOAD_PUSH_GENEVE)
1638 geneve_push_rx_ports(dev);
1639
1640 return NOTIFY_DONE;
1641 }
1642
1643 static struct notifier_block geneve_notifier_block __read_mostly = {
1644 .notifier_call = geneve_netdevice_event,
1645 };
1646
1647 static __net_init int geneve_init_net(struct net *net)
1648 {
1649 struct geneve_net *gn = net_generic(net, geneve_net_id);
1650
1651 INIT_LIST_HEAD(&gn->geneve_list);
1652 INIT_LIST_HEAD(&gn->sock_list);
1653 return 0;
1654 }
1655
1656 static void __net_exit geneve_exit_net(struct net *net)
1657 {
1658 struct geneve_net *gn = net_generic(net, geneve_net_id);
1659 struct geneve_dev *geneve, *next;
1660 struct net_device *dev, *aux;
1661 LIST_HEAD(list);
1662
1663 rtnl_lock();
1664
1665 /* gather any geneve devices that were moved into this ns */
1666 for_each_netdev_safe(net, dev, aux)
1667 if (dev->rtnl_link_ops == &geneve_link_ops)
1668 unregister_netdevice_queue(dev, &list);
1669
1670 /* now gather any other geneve devices that were created in this ns */
1671 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1672 /* If geneve->dev is in the same netns, it was already added
1673 * to the list by the previous loop.
1674 */
1675 if (!net_eq(dev_net(geneve->dev), net))
1676 unregister_netdevice_queue(geneve->dev, &list);
1677 }
1678
1679 /* unregister the devices gathered above */
1680 unregister_netdevice_many(&list);
1681 rtnl_unlock();
1682 }
1683
1684 static struct pernet_operations geneve_net_ops = {
1685 .init = geneve_init_net,
1686 .exit = geneve_exit_net,
1687 .id = &geneve_net_id,
1688 .size = sizeof(struct geneve_net),
1689 };
1690
1691 int rpl_geneve_init_module(void)
1692 {
1693 int rc;
1694
1695 rc = register_pernet_subsys(&geneve_net_ops);
1696 if (rc)
1697 goto out1;
1698
1699 rc = register_netdevice_notifier(&geneve_notifier_block);
1700 if (rc)
1701 goto out2;
1702
1703 rc = rtnl_link_register(&geneve_link_ops);
1704 if (rc)
1705 goto out3;
1706
1707 pr_info("Geneve tunneling driver\n");
1708 return 0;
1709
1710 out3:
1711 unregister_netdevice_notifier(&geneve_notifier_block);
1712 out2:
1713 unregister_pernet_subsys(&geneve_net_ops);
1714 out1:
1715 return rc;
1716 }
1717
1718 void rpl_geneve_cleanup_module(void)
1719 {
1720 rtnl_link_unregister(&geneve_link_ops);
1721 unregister_netdevice_notifier(&geneve_notifier_block);
1722 unregister_pernet_subsys(&geneve_net_ops);
1723 }
1724
1725 #endif
openvswitch mirror