Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * GENEVE: Generic Network Virtualization Encapsulation
4 *
5 * Copyright (c) 2015 Red Hat, Inc.
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/ethtool.h>
11#include <linux/kernel.h>
12#include <linux/module.h>
13#include <linux/etherdevice.h>
14#include <linux/hash.h>
15#include <net/ipv6_stubs.h>
16#include <net/dst_metadata.h>
17#include <net/gro_cells.h>
18#include <net/rtnetlink.h>
19#include <net/geneve.h>
20#include <net/gro.h>
21#include <net/protocol.h>
22
23#define GENEVE_NETDEV_VER "0.6"
24
25#define GENEVE_N_VID (1u << 24)
26#define GENEVE_VID_MASK (GENEVE_N_VID - 1)
27
28#define VNI_HASH_BITS 10
29#define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
30
31static bool log_ecn_error = true;
32module_param(log_ecn_error, bool, 0644);
33MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
34
35#define GENEVE_VER 0
36#define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
37#define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
38#define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
39
40/* per-network namespace private data for this module */
41struct geneve_net {
42 struct list_head geneve_list;
43 struct list_head sock_list;
44};
45
46static unsigned int geneve_net_id;
47
48struct geneve_dev_node {
49 struct hlist_node hlist;
50 struct geneve_dev *geneve;
51};
52
53struct geneve_config {
54 struct ip_tunnel_info info;
55 bool collect_md;
56 bool use_udp6_rx_checksums;
57 bool ttl_inherit;
58 enum ifla_geneve_df df;
59 bool inner_proto_inherit;
60};
61
62/* Pseudo network device */
63struct geneve_dev {
64 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
65#if IS_ENABLED(CONFIG_IPV6)
66 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
67#endif
68 struct net *net; /* netns for packet i/o */
69 struct net_device *dev; /* netdev for geneve tunnel */
70 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
71#if IS_ENABLED(CONFIG_IPV6)
72 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
73#endif
74 struct list_head next; /* geneve's per namespace list */
75 struct gro_cells gro_cells;
76 struct geneve_config cfg;
77};
78
79struct geneve_sock {
80 bool collect_md;
81 struct list_head list;
82 struct socket *sock;
83 struct rcu_head rcu;
84 int refcnt;
85 struct hlist_head vni_list[VNI_HASH_SIZE];
86};
87
88static inline __u32 geneve_net_vni_hash(u8 vni[3])
89{
90 __u32 vnid;
91
92 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
93 return hash_32(vnid, VNI_HASH_BITS);
94}
95
96static __be64 vni_to_tunnel_id(const __u8 *vni)
97{
98#ifdef __BIG_ENDIAN
99 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
100#else
101 return (__force __be64)(((__force u64)vni[0] << 40) |
102 ((__force u64)vni[1] << 48) |
103 ((__force u64)vni[2] << 56));
104#endif
105}
106
107/* Convert 64 bit tunnel ID to 24 bit VNI. */
108static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
109{
110#ifdef __BIG_ENDIAN
111 vni[0] = (__force __u8)(tun_id >> 16);
112 vni[1] = (__force __u8)(tun_id >> 8);
113 vni[2] = (__force __u8)tun_id;
114#else
115 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
116 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
117 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
118#endif
119}
120
121static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
122{
123 return !memcmp(vni, &tun_id[5], 3);
124}
125
126static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
127{
128 return gs->sock->sk->sk_family;
129}
130
131static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
132 __be32 addr, u8 vni[])
133{
134 struct hlist_head *vni_list_head;
135 struct geneve_dev_node *node;
136 __u32 hash;
137
138 /* Find the device for this VNI */
139 hash = geneve_net_vni_hash(vni);
140 vni_list_head = &gs->vni_list[hash];
141 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
142 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
143 addr == node->geneve->cfg.info.key.u.ipv4.dst)
144 return node->geneve;
145 }
146 return NULL;
147}
148
149#if IS_ENABLED(CONFIG_IPV6)
150static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
151 struct in6_addr addr6, u8 vni[])
152{
153 struct hlist_head *vni_list_head;
154 struct geneve_dev_node *node;
155 __u32 hash;
156
157 /* Find the device for this VNI */
158 hash = geneve_net_vni_hash(vni);
159 vni_list_head = &gs->vni_list[hash];
160 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
161 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
162 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
163 return node->geneve;
164 }
165 return NULL;
166}
167#endif
168
169static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
170{
171 return (struct genevehdr *)(udp_hdr(skb) + 1);
172}
173
174static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
175 struct sk_buff *skb)
176{
177 static u8 zero_vni[3];
178 u8 *vni;
179
180 if (geneve_get_sk_family(gs) == AF_INET) {
181 struct iphdr *iph;
182 __be32 addr;
183
184 iph = ip_hdr(skb); /* outer IP header... */
185
186 if (gs->collect_md) {
187 vni = zero_vni;
188 addr = 0;
189 } else {
190 vni = geneve_hdr(skb)->vni;
191 addr = iph->saddr;
192 }
193
194 return geneve_lookup(gs, addr, vni);
195#if IS_ENABLED(CONFIG_IPV6)
196 } else if (geneve_get_sk_family(gs) == AF_INET6) {
197 static struct in6_addr zero_addr6;
198 struct ipv6hdr *ip6h;
199 struct in6_addr addr6;
200
201 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
202
203 if (gs->collect_md) {
204 vni = zero_vni;
205 addr6 = zero_addr6;
206 } else {
207 vni = geneve_hdr(skb)->vni;
208 addr6 = ip6h->saddr;
209 }
210
211 return geneve6_lookup(gs, addr6, vni);
212#endif
213 }
214 return NULL;
215}
216
217/* geneve receive/decap routine */
218static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
219 struct sk_buff *skb)
220{
221 struct genevehdr *gnvh = geneve_hdr(skb);
222 struct metadata_dst *tun_dst = NULL;
223 unsigned int len;
224 int nh, err = 0;
225 void *oiph;
226
227 if (ip_tunnel_collect_metadata() || gs->collect_md) {
228 __be16 flags;
229
230 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
231 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
232
233 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
234 vni_to_tunnel_id(gnvh->vni),
235 gnvh->opt_len * 4);
236 if (!tun_dst) {
237 DEV_STATS_INC(geneve->dev, rx_dropped);
238 goto drop;
239 }
240 /* Update tunnel dst according to Geneve options. */
241 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
242 gnvh->options, gnvh->opt_len * 4,
243 TUNNEL_GENEVE_OPT);
244 } else {
245 /* Drop packets w/ critical options,
246 * since we don't support any...
247 */
248 if (gnvh->critical) {
249 DEV_STATS_INC(geneve->dev, rx_frame_errors);
250 DEV_STATS_INC(geneve->dev, rx_errors);
251 goto drop;
252 }
253 }
254
255 if (tun_dst)
256 skb_dst_set(skb, &tun_dst->dst);
257
258 if (gnvh->proto_type == htons(ETH_P_TEB)) {
259 skb_reset_mac_header(skb);
260 skb->protocol = eth_type_trans(skb, geneve->dev);
261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
262
263 /* Ignore packet loops (and multicast echo) */
264 if (ether_addr_equal(eth_hdr(skb)->h_source,
265 geneve->dev->dev_addr)) {
266 DEV_STATS_INC(geneve->dev, rx_errors);
267 goto drop;
268 }
269 } else {
270 skb_reset_mac_header(skb);
271 skb->dev = geneve->dev;
272 skb->pkt_type = PACKET_HOST;
273 }
274
275 /* Save offset of outer header relative to skb->head,
276 * because we are going to reset the network header to the inner header
277 * and might change skb->head.
278 */
279 nh = skb_network_header(skb) - skb->head;
280
281 skb_reset_network_header(skb);
282
283 if (!pskb_inet_may_pull(skb)) {
284 DEV_STATS_INC(geneve->dev, rx_length_errors);
285 DEV_STATS_INC(geneve->dev, rx_errors);
286 goto drop;
287 }
288
289 /* Get the outer header. */
290 oiph = skb->head + nh;
291
292 if (geneve_get_sk_family(gs) == AF_INET)
293 err = IP_ECN_decapsulate(oiph, skb);
294#if IS_ENABLED(CONFIG_IPV6)
295 else
296 err = IP6_ECN_decapsulate(oiph, skb);
297#endif
298
299 if (unlikely(err)) {
300 if (log_ecn_error) {
301 if (geneve_get_sk_family(gs) == AF_INET)
302 net_info_ratelimited("non-ECT from %pI4 "
303 "with TOS=%#x\n",
304 &((struct iphdr *)oiph)->saddr,
305 ((struct iphdr *)oiph)->tos);
306#if IS_ENABLED(CONFIG_IPV6)
307 else
308 net_info_ratelimited("non-ECT from %pI6\n",
309 &((struct ipv6hdr *)oiph)->saddr);
310#endif
311 }
312 if (err > 1) {
313 DEV_STATS_INC(geneve->dev, rx_frame_errors);
314 DEV_STATS_INC(geneve->dev, rx_errors);
315 goto drop;
316 }
317 }
318
319 len = skb->len;
320 err = gro_cells_receive(&geneve->gro_cells, skb);
321 if (likely(err == NET_RX_SUCCESS))
322 dev_sw_netstats_rx_add(geneve->dev, len);
323
324 return;
325drop:
326 /* Consume bad packet */
327 kfree_skb(skb);
328}
329
330/* Setup stats when device is created */
331static int geneve_init(struct net_device *dev)
332{
333 struct geneve_dev *geneve = netdev_priv(dev);
334 int err;
335
336 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
337 if (!dev->tstats)
338 return -ENOMEM;
339
340 err = gro_cells_init(&geneve->gro_cells, dev);
341 if (err) {
342 free_percpu(dev->tstats);
343 return err;
344 }
345
346 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
347 if (err) {
348 free_percpu(dev->tstats);
349 gro_cells_destroy(&geneve->gro_cells);
350 return err;
351 }
352 return 0;
353}
354
355static void geneve_uninit(struct net_device *dev)
356{
357 struct geneve_dev *geneve = netdev_priv(dev);
358
359 dst_cache_destroy(&geneve->cfg.info.dst_cache);
360 gro_cells_destroy(&geneve->gro_cells);
361 free_percpu(dev->tstats);
362}
363
364/* Callback from net/ipv4/udp.c to receive packets */
365static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
366{
367 struct genevehdr *geneveh;
368 struct geneve_dev *geneve;
369 struct geneve_sock *gs;
370 __be16 inner_proto;
371 int opts_len;
372
373 /* Need UDP and Geneve header to be present */
374 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
375 goto drop;
376
377 /* Return packets with reserved bits set */
378 geneveh = geneve_hdr(skb);
379 if (unlikely(geneveh->ver != GENEVE_VER))
380 goto drop;
381
382 gs = rcu_dereference_sk_user_data(sk);
383 if (!gs)
384 goto drop;
385
386 geneve = geneve_lookup_skb(gs, skb);
387 if (!geneve)
388 goto drop;
389
390 inner_proto = geneveh->proto_type;
391
392 if (unlikely((!geneve->cfg.inner_proto_inherit &&
393 inner_proto != htons(ETH_P_TEB)))) {
394 DEV_STATS_INC(geneve->dev, rx_dropped);
395 goto drop;
396 }
397
398 opts_len = geneveh->opt_len * 4;
399 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
400 !net_eq(geneve->net, dev_net(geneve->dev)))) {
401 DEV_STATS_INC(geneve->dev, rx_dropped);
402 goto drop;
403 }
404
405 geneve_rx(geneve, gs, skb);
406 return 0;
407
408drop:
409 /* Consume bad packet */
410 kfree_skb(skb);
411 return 0;
412}
413
414/* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
415static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
416{
417 struct genevehdr *geneveh;
418 struct geneve_sock *gs;
419 u8 zero_vni[3] = { 0 };
420 u8 *vni = zero_vni;
421
422 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
423 return -EINVAL;
424
425 geneveh = geneve_hdr(skb);
426 if (geneveh->ver != GENEVE_VER)
427 return -EINVAL;
428
429 if (geneveh->proto_type != htons(ETH_P_TEB))
430 return -EINVAL;
431
432 gs = rcu_dereference_sk_user_data(sk);
433 if (!gs)
434 return -ENOENT;
435
436 if (geneve_get_sk_family(gs) == AF_INET) {
437 struct iphdr *iph = ip_hdr(skb);
438 __be32 addr4 = 0;
439
440 if (!gs->collect_md) {
441 vni = geneve_hdr(skb)->vni;
442 addr4 = iph->daddr;
443 }
444
445 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
446 }
447
448#if IS_ENABLED(CONFIG_IPV6)
449 if (geneve_get_sk_family(gs) == AF_INET6) {
450 struct ipv6hdr *ip6h = ipv6_hdr(skb);
451 struct in6_addr addr6;
452
453 memset(&addr6, 0, sizeof(struct in6_addr));
454
455 if (!gs->collect_md) {
456 vni = geneve_hdr(skb)->vni;
457 addr6 = ip6h->daddr;
458 }
459
460 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
461 }
462#endif
463
464 return -EPFNOSUPPORT;
465}
466
467static struct socket *geneve_create_sock(struct net *net, bool ipv6,
468 __be16 port, bool ipv6_rx_csum)
469{
470 struct socket *sock;
471 struct udp_port_cfg udp_conf;
472 int err;
473
474 memset(&udp_conf, 0, sizeof(udp_conf));
475
476 if (ipv6) {
477 udp_conf.family = AF_INET6;
478 udp_conf.ipv6_v6only = 1;
479 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
480 } else {
481 udp_conf.family = AF_INET;
482 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
483 }
484
485 udp_conf.local_udp_port = port;
486
487 /* Open UDP socket */
488 err = udp_sock_create(net, &udp_conf, &sock);
489 if (err < 0)
490 return ERR_PTR(err);
491
492 udp_allow_gso(sock->sk);
493 return sock;
494}
495
496static int geneve_hlen(struct genevehdr *gh)
497{
498 return sizeof(*gh) + gh->opt_len * 4;
499}
500
501static struct sk_buff *geneve_gro_receive(struct sock *sk,
502 struct list_head *head,
503 struct sk_buff *skb)
504{
505 struct sk_buff *pp = NULL;
506 struct sk_buff *p;
507 struct genevehdr *gh, *gh2;
508 unsigned int hlen, gh_len, off_gnv;
509 const struct packet_offload *ptype;
510 __be16 type;
511 int flush = 1;
512
513 off_gnv = skb_gro_offset(skb);
514 hlen = off_gnv + sizeof(*gh);
515 gh = skb_gro_header(skb, hlen, off_gnv);
516 if (unlikely(!gh))
517 goto out;
518
519 if (gh->ver != GENEVE_VER || gh->oam)
520 goto out;
521 gh_len = geneve_hlen(gh);
522
523 hlen = off_gnv + gh_len;
524 if (skb_gro_header_hard(skb, hlen)) {
525 gh = skb_gro_header_slow(skb, hlen, off_gnv);
526 if (unlikely(!gh))
527 goto out;
528 }
529
530 list_for_each_entry(p, head, list) {
531 if (!NAPI_GRO_CB(p)->same_flow)
532 continue;
533
534 gh2 = (struct genevehdr *)(p->data + off_gnv);
535 if (gh->opt_len != gh2->opt_len ||
536 memcmp(gh, gh2, gh_len)) {
537 NAPI_GRO_CB(p)->same_flow = 0;
538 continue;
539 }
540 }
541
542 skb_gro_pull(skb, gh_len);
543 skb_gro_postpull_rcsum(skb, gh, gh_len);
544 type = gh->proto_type;
545 if (likely(type == htons(ETH_P_TEB)))
546 return call_gro_receive(eth_gro_receive, head, skb);
547
548 ptype = gro_find_receive_by_type(type);
549 if (!ptype)
550 goto out;
551
552 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
553 flush = 0;
554
555out:
556 skb_gro_flush_final(skb, pp, flush);
557
558 return pp;
559}
560
561static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
562 int nhoff)
563{
564 struct genevehdr *gh;
565 struct packet_offload *ptype;
566 __be16 type;
567 int gh_len;
568 int err = -ENOSYS;
569
570 gh = (struct genevehdr *)(skb->data + nhoff);
571 gh_len = geneve_hlen(gh);
572 type = gh->proto_type;
573
574 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
575 if (likely(type == htons(ETH_P_TEB)))
576 return eth_gro_complete(skb, nhoff + gh_len);
577
578 ptype = gro_find_complete_by_type(type);
579 if (ptype)
580 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
581
582 skb_set_inner_mac_header(skb, nhoff + gh_len);
583
584 return err;
585}
586
587/* Create new listen socket if needed */
588static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
589 bool ipv6, bool ipv6_rx_csum)
590{
591 struct geneve_net *gn = net_generic(net, geneve_net_id);
592 struct geneve_sock *gs;
593 struct socket *sock;
594 struct udp_tunnel_sock_cfg tunnel_cfg;
595 int h;
596
597 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
598 if (!gs)
599 return ERR_PTR(-ENOMEM);
600
601 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
602 if (IS_ERR(sock)) {
603 kfree(gs);
604 return ERR_CAST(sock);
605 }
606
607 gs->sock = sock;
608 gs->refcnt = 1;
609 for (h = 0; h < VNI_HASH_SIZE; ++h)
610 INIT_HLIST_HEAD(&gs->vni_list[h]);
611
612 /* Initialize the geneve udp offloads structure */
613 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
614
615 /* Mark socket as an encapsulation socket */
616 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
617 tunnel_cfg.sk_user_data = gs;
618 tunnel_cfg.encap_type = 1;
619 tunnel_cfg.gro_receive = geneve_gro_receive;
620 tunnel_cfg.gro_complete = geneve_gro_complete;
621 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
622 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
623 tunnel_cfg.encap_destroy = NULL;
624 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
625 list_add(&gs->list, &gn->sock_list);
626 return gs;
627}
628
629static void __geneve_sock_release(struct geneve_sock *gs)
630{
631 if (!gs || --gs->refcnt)
632 return;
633
634 list_del(&gs->list);
635 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
636 udp_tunnel_sock_release(gs->sock);
637 kfree_rcu(gs, rcu);
638}
639
640static void geneve_sock_release(struct geneve_dev *geneve)
641{
642 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
643#if IS_ENABLED(CONFIG_IPV6)
644 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
645
646 rcu_assign_pointer(geneve->sock6, NULL);
647#endif
648
649 rcu_assign_pointer(geneve->sock4, NULL);
650 synchronize_net();
651
652 __geneve_sock_release(gs4);
653#if IS_ENABLED(CONFIG_IPV6)
654 __geneve_sock_release(gs6);
655#endif
656}
657
658static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
659 sa_family_t family,
660 __be16 dst_port)
661{
662 struct geneve_sock *gs;
663
664 list_for_each_entry(gs, &gn->sock_list, list) {
665 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
666 geneve_get_sk_family(gs) == family) {
667 return gs;
668 }
669 }
670 return NULL;
671}
672
673static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
674{
675 struct net *net = geneve->net;
676 struct geneve_net *gn = net_generic(net, geneve_net_id);
677 struct geneve_dev_node *node;
678 struct geneve_sock *gs;
679 __u8 vni[3];
680 __u32 hash;
681
682 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
683 if (gs) {
684 gs->refcnt++;
685 goto out;
686 }
687
688 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
689 geneve->cfg.use_udp6_rx_checksums);
690 if (IS_ERR(gs))
691 return PTR_ERR(gs);
692
693out:
694 gs->collect_md = geneve->cfg.collect_md;
695#if IS_ENABLED(CONFIG_IPV6)
696 if (ipv6) {
697 rcu_assign_pointer(geneve->sock6, gs);
698 node = &geneve->hlist6;
699 } else
700#endif
701 {
702 rcu_assign_pointer(geneve->sock4, gs);
703 node = &geneve->hlist4;
704 }
705 node->geneve = geneve;
706
707 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
708 hash = geneve_net_vni_hash(vni);
709 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
710 return 0;
711}
712
713static int geneve_open(struct net_device *dev)
714{
715 struct geneve_dev *geneve = netdev_priv(dev);
716 bool metadata = geneve->cfg.collect_md;
717 bool ipv4, ipv6;
718 int ret = 0;
719
720 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
721 ipv4 = !ipv6 || metadata;
722#if IS_ENABLED(CONFIG_IPV6)
723 if (ipv6) {
724 ret = geneve_sock_add(geneve, true);
725 if (ret < 0 && ret != -EAFNOSUPPORT)
726 ipv4 = false;
727 }
728#endif
729 if (ipv4)
730 ret = geneve_sock_add(geneve, false);
731 if (ret < 0)
732 geneve_sock_release(geneve);
733
734 return ret;
735}
736
737static int geneve_stop(struct net_device *dev)
738{
739 struct geneve_dev *geneve = netdev_priv(dev);
740
741 hlist_del_init_rcu(&geneve->hlist4.hlist);
742#if IS_ENABLED(CONFIG_IPV6)
743 hlist_del_init_rcu(&geneve->hlist6.hlist);
744#endif
745 geneve_sock_release(geneve);
746 return 0;
747}
748
749static void geneve_build_header(struct genevehdr *geneveh,
750 const struct ip_tunnel_info *info,
751 __be16 inner_proto)
752{
753 geneveh->ver = GENEVE_VER;
754 geneveh->opt_len = info->options_len / 4;
755 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
756 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
757 geneveh->rsvd1 = 0;
758 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
759 geneveh->proto_type = inner_proto;
760 geneveh->rsvd2 = 0;
761
762 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
763 ip_tunnel_info_opts_get(geneveh->options, info);
764}
765
766static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
767 const struct ip_tunnel_info *info,
768 bool xnet, int ip_hdr_len,
769 bool inner_proto_inherit)
770{
771 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
772 struct genevehdr *gnvh;
773 __be16 inner_proto;
774 int min_headroom;
775 int err;
776
777 skb_reset_mac_header(skb);
778 skb_scrub_packet(skb, xnet);
779
780 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
781 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
782 err = skb_cow_head(skb, min_headroom);
783 if (unlikely(err))
784 goto free_dst;
785
786 err = udp_tunnel_handle_offloads(skb, udp_sum);
787 if (err)
788 goto free_dst;
789
790 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
791 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
792 geneve_build_header(gnvh, info, inner_proto);
793 skb_set_inner_protocol(skb, inner_proto);
794 return 0;
795
796free_dst:
797 dst_release(dst);
798 return err;
799}
800
801static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
802 const struct ip_tunnel_info *info,
803 bool *use_cache)
804{
805 struct geneve_dev *geneve = netdev_priv(dev);
806 u8 dsfield;
807
808 dsfield = info->key.tos;
809 if (dsfield == 1 && !geneve->cfg.collect_md) {
810 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
811 *use_cache = false;
812 }
813
814 return dsfield;
815}
816
817static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
818 struct geneve_dev *geneve,
819 const struct ip_tunnel_info *info)
820{
821 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
822 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
823 const struct ip_tunnel_key *key = &info->key;
824 struct rtable *rt;
825 bool use_cache;
826 __u8 tos, ttl;
827 __be16 df = 0;
828 __be32 saddr;
829 __be16 sport;
830 int err;
831
832 if (!pskb_inet_may_pull(skb))
833 return -EINVAL;
834
835 if (!gs4)
836 return -EIO;
837
838 use_cache = ip_tunnel_dst_cache_usable(skb, info);
839 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
840 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
841
842 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
843 &info->key,
844 sport, geneve->cfg.info.key.tp_dst, tos,
845 use_cache ?
846 (struct dst_cache *)&info->dst_cache : NULL);
847 if (IS_ERR(rt))
848 return PTR_ERR(rt);
849
850 err = skb_tunnel_check_pmtu(skb, &rt->dst,
851 GENEVE_IPV4_HLEN + info->options_len,
852 netif_is_any_bridge_port(dev));
853 if (err < 0) {
854 dst_release(&rt->dst);
855 return err;
856 } else if (err) {
857 struct ip_tunnel_info *info;
858
859 info = skb_tunnel_info(skb);
860 if (info) {
861 struct ip_tunnel_info *unclone;
862
863 unclone = skb_tunnel_info_unclone(skb);
864 if (unlikely(!unclone)) {
865 dst_release(&rt->dst);
866 return -ENOMEM;
867 }
868
869 unclone->key.u.ipv4.dst = saddr;
870 unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
871 }
872
873 if (!pskb_may_pull(skb, ETH_HLEN)) {
874 dst_release(&rt->dst);
875 return -EINVAL;
876 }
877
878 skb->protocol = eth_type_trans(skb, geneve->dev);
879 __netif_rx(skb);
880 dst_release(&rt->dst);
881 return -EMSGSIZE;
882 }
883
884 tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
885 if (geneve->cfg.collect_md) {
886 ttl = key->ttl;
887
888 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
889 } else {
890 if (geneve->cfg.ttl_inherit)
891 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
892 else
893 ttl = key->ttl;
894 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
895
896 if (geneve->cfg.df == GENEVE_DF_SET) {
897 df = htons(IP_DF);
898 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
899 struct ethhdr *eth = eth_hdr(skb);
900
901 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
902 df = htons(IP_DF);
903 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
904 struct iphdr *iph = ip_hdr(skb);
905
906 if (iph->frag_off & htons(IP_DF))
907 df = htons(IP_DF);
908 }
909 }
910 }
911
912 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
913 geneve->cfg.inner_proto_inherit);
914 if (unlikely(err))
915 return err;
916
917 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
918 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
919 !net_eq(geneve->net, dev_net(geneve->dev)),
920 !(info->key.tun_flags & TUNNEL_CSUM));
921 return 0;
922}
923
924#if IS_ENABLED(CONFIG_IPV6)
925static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
926 struct geneve_dev *geneve,
927 const struct ip_tunnel_info *info)
928{
929 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
930 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
931 const struct ip_tunnel_key *key = &info->key;
932 struct dst_entry *dst = NULL;
933 struct in6_addr saddr;
934 bool use_cache;
935 __u8 prio, ttl;
936 __be16 sport;
937 int err;
938
939 if (!pskb_inet_may_pull(skb))
940 return -EINVAL;
941
942 if (!gs6)
943 return -EIO;
944
945 use_cache = ip_tunnel_dst_cache_usable(skb, info);
946 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
947 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
948
949 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
950 &saddr, key, sport,
951 geneve->cfg.info.key.tp_dst, prio,
952 use_cache ?
953 (struct dst_cache *)&info->dst_cache : NULL);
954 if (IS_ERR(dst))
955 return PTR_ERR(dst);
956
957 err = skb_tunnel_check_pmtu(skb, dst,
958 GENEVE_IPV6_HLEN + info->options_len,
959 netif_is_any_bridge_port(dev));
960 if (err < 0) {
961 dst_release(dst);
962 return err;
963 } else if (err) {
964 struct ip_tunnel_info *info = skb_tunnel_info(skb);
965
966 if (info) {
967 struct ip_tunnel_info *unclone;
968
969 unclone = skb_tunnel_info_unclone(skb);
970 if (unlikely(!unclone)) {
971 dst_release(dst);
972 return -ENOMEM;
973 }
974
975 unclone->key.u.ipv6.dst = saddr;
976 unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
977 }
978
979 if (!pskb_may_pull(skb, ETH_HLEN)) {
980 dst_release(dst);
981 return -EINVAL;
982 }
983
984 skb->protocol = eth_type_trans(skb, geneve->dev);
985 __netif_rx(skb);
986 dst_release(dst);
987 return -EMSGSIZE;
988 }
989
990 prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
991 if (geneve->cfg.collect_md) {
992 ttl = key->ttl;
993 } else {
994 if (geneve->cfg.ttl_inherit)
995 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
996 else
997 ttl = key->ttl;
998 ttl = ttl ? : ip6_dst_hoplimit(dst);
999 }
1000 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1001 geneve->cfg.inner_proto_inherit);
1002 if (unlikely(err))
1003 return err;
1004
1005 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1006 &saddr, &key->u.ipv6.dst, prio, ttl,
1007 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1008 !(info->key.tun_flags & TUNNEL_CSUM));
1009 return 0;
1010}
1011#endif
1012
1013static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1014{
1015 struct geneve_dev *geneve = netdev_priv(dev);
1016 struct ip_tunnel_info *info = NULL;
1017 int err;
1018
1019 if (geneve->cfg.collect_md) {
1020 info = skb_tunnel_info(skb);
1021 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1022 netdev_dbg(dev, "no tunnel metadata\n");
1023 dev_kfree_skb(skb);
1024 DEV_STATS_INC(dev, tx_dropped);
1025 return NETDEV_TX_OK;
1026 }
1027 } else {
1028 info = &geneve->cfg.info;
1029 }
1030
1031 rcu_read_lock();
1032#if IS_ENABLED(CONFIG_IPV6)
1033 if (info->mode & IP_TUNNEL_INFO_IPV6)
1034 err = geneve6_xmit_skb(skb, dev, geneve, info);
1035 else
1036#endif
1037 err = geneve_xmit_skb(skb, dev, geneve, info);
1038 rcu_read_unlock();
1039
1040 if (likely(!err))
1041 return NETDEV_TX_OK;
1042
1043 if (err != -EMSGSIZE)
1044 dev_kfree_skb(skb);
1045
1046 if (err == -ELOOP)
1047 DEV_STATS_INC(dev, collisions);
1048 else if (err == -ENETUNREACH)
1049 DEV_STATS_INC(dev, tx_carrier_errors);
1050
1051 DEV_STATS_INC(dev, tx_errors);
1052 return NETDEV_TX_OK;
1053}
1054
1055static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1056{
1057 if (new_mtu > dev->max_mtu)
1058 new_mtu = dev->max_mtu;
1059 else if (new_mtu < dev->min_mtu)
1060 new_mtu = dev->min_mtu;
1061
1062 dev->mtu = new_mtu;
1063 return 0;
1064}
1065
1066static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1067{
1068 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1069 struct geneve_dev *geneve = netdev_priv(dev);
1070 __be16 sport;
1071
1072 if (ip_tunnel_info_af(info) == AF_INET) {
1073 struct rtable *rt;
1074 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1075 bool use_cache;
1076 __be32 saddr;
1077 u8 tos;
1078
1079 if (!gs4)
1080 return -EIO;
1081
1082 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1083 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1084 sport = udp_flow_src_port(geneve->net, skb,
1085 1, USHRT_MAX, true);
1086
1087 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1088 &info->key,
1089 sport, geneve->cfg.info.key.tp_dst,
1090 tos,
1091 use_cache ? &info->dst_cache : NULL);
1092 if (IS_ERR(rt))
1093 return PTR_ERR(rt);
1094
1095 ip_rt_put(rt);
1096 info->key.u.ipv4.src = saddr;
1097#if IS_ENABLED(CONFIG_IPV6)
1098 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1099 struct dst_entry *dst;
1100 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1101 struct in6_addr saddr;
1102 bool use_cache;
1103 u8 prio;
1104
1105 if (!gs6)
1106 return -EIO;
1107
1108 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1109 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1110 sport = udp_flow_src_port(geneve->net, skb,
1111 1, USHRT_MAX, true);
1112
1113 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1114 &saddr, &info->key, sport,
1115 geneve->cfg.info.key.tp_dst, prio,
1116 use_cache ? &info->dst_cache : NULL);
1117 if (IS_ERR(dst))
1118 return PTR_ERR(dst);
1119
1120 dst_release(dst);
1121 info->key.u.ipv6.src = saddr;
1122#endif
1123 } else {
1124 return -EINVAL;
1125 }
1126
1127 info->key.tp_src = sport;
1128 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1129 return 0;
1130}
1131
1132static const struct net_device_ops geneve_netdev_ops = {
1133 .ndo_init = geneve_init,
1134 .ndo_uninit = geneve_uninit,
1135 .ndo_open = geneve_open,
1136 .ndo_stop = geneve_stop,
1137 .ndo_start_xmit = geneve_xmit,
1138 .ndo_get_stats64 = dev_get_tstats64,
1139 .ndo_change_mtu = geneve_change_mtu,
1140 .ndo_validate_addr = eth_validate_addr,
1141 .ndo_set_mac_address = eth_mac_addr,
1142 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1143};
1144
1145static void geneve_get_drvinfo(struct net_device *dev,
1146 struct ethtool_drvinfo *drvinfo)
1147{
1148 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1149 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1150}
1151
1152static const struct ethtool_ops geneve_ethtool_ops = {
1153 .get_drvinfo = geneve_get_drvinfo,
1154 .get_link = ethtool_op_get_link,
1155};
1156
1157/* Info for udev, that this is a virtual tunnel endpoint */
1158static struct device_type geneve_type = {
1159 .name = "geneve",
1160};
1161
1162/* Calls the ndo_udp_tunnel_add of the caller in order to
1163 * supply the listening GENEVE udp ports. Callers are expected
1164 * to implement the ndo_udp_tunnel_add.
1165 */
1166static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1167{
1168 struct net *net = dev_net(dev);
1169 struct geneve_net *gn = net_generic(net, geneve_net_id);
1170 struct geneve_sock *gs;
1171
1172 rcu_read_lock();
1173 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1174 if (push) {
1175 udp_tunnel_push_rx_port(dev, gs->sock,
1176 UDP_TUNNEL_TYPE_GENEVE);
1177 } else {
1178 udp_tunnel_drop_rx_port(dev, gs->sock,
1179 UDP_TUNNEL_TYPE_GENEVE);
1180 }
1181 }
1182 rcu_read_unlock();
1183}
1184
1185/* Initialize the device structure. */
1186static void geneve_setup(struct net_device *dev)
1187{
1188 ether_setup(dev);
1189
1190 dev->netdev_ops = &geneve_netdev_ops;
1191 dev->ethtool_ops = &geneve_ethtool_ops;
1192 dev->needs_free_netdev = true;
1193
1194 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1195
1196 dev->features |= NETIF_F_LLTX;
1197 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1198 dev->features |= NETIF_F_RXCSUM;
1199 dev->features |= NETIF_F_GSO_SOFTWARE;
1200
1201 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1202 dev->hw_features |= NETIF_F_RXCSUM;
1203 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1204
1205 /* MTU range: 68 - (something less than 65535) */
1206 dev->min_mtu = ETH_MIN_MTU;
1207 /* The max_mtu calculation does not take account of GENEVE
1208 * options, to avoid excluding potentially valid
1209 * configurations. This will be further reduced by IPvX hdr size.
1210 */
1211 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1212
1213 netif_keep_dst(dev);
1214 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1215 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1216 eth_hw_addr_random(dev);
1217}
1218
1219static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1220 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1221 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1222 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1223 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1224 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1225 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1226 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1227 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1228 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1229 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1230 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1231 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1232 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1233 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1234 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG },
1235};
1236
1237static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1238 struct netlink_ext_ack *extack)
1239{
1240 if (tb[IFLA_ADDRESS]) {
1241 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1242 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1243 "Provided link layer address is not Ethernet");
1244 return -EINVAL;
1245 }
1246
1247 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1248 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1249 "Provided Ethernet address is not unicast");
1250 return -EADDRNOTAVAIL;
1251 }
1252 }
1253
1254 if (!data) {
1255 NL_SET_ERR_MSG(extack,
1256 "Not enough attributes provided to perform the operation");
1257 return -EINVAL;
1258 }
1259
1260 if (data[IFLA_GENEVE_ID]) {
1261 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1262
1263 if (vni >= GENEVE_N_VID) {
1264 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1265 "Geneve ID must be lower than 16777216");
1266 return -ERANGE;
1267 }
1268 }
1269
1270 if (data[IFLA_GENEVE_DF]) {
1271 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1272
1273 if (df < 0 || df > GENEVE_DF_MAX) {
1274 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1275 "Invalid DF attribute");
1276 return -EINVAL;
1277 }
1278 }
1279
1280 return 0;
1281}
1282
1283static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1284 const struct ip_tunnel_info *info,
1285 bool *tun_on_same_port,
1286 bool *tun_collect_md)
1287{
1288 struct geneve_dev *geneve, *t = NULL;
1289
1290 *tun_on_same_port = false;
1291 *tun_collect_md = false;
1292 list_for_each_entry(geneve, &gn->geneve_list, next) {
1293 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1294 *tun_collect_md = geneve->cfg.collect_md;
1295 *tun_on_same_port = true;
1296 }
1297 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1298 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1299 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1300 t = geneve;
1301 }
1302 return t;
1303}
1304
1305static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1306{
1307 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1308 info->key.ttl || info->key.label || info->key.tp_src ||
1309 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1310}
1311
1312static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1313 struct ip_tunnel_info *b)
1314{
1315 if (ip_tunnel_info_af(a) == AF_INET)
1316 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1317 else
1318 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1319}
1320
1321static int geneve_configure(struct net *net, struct net_device *dev,
1322 struct netlink_ext_ack *extack,
1323 const struct geneve_config *cfg)
1324{
1325 struct geneve_net *gn = net_generic(net, geneve_net_id);
1326 struct geneve_dev *t, *geneve = netdev_priv(dev);
1327 const struct ip_tunnel_info *info = &cfg->info;
1328 bool tun_collect_md, tun_on_same_port;
1329 int err, encap_len;
1330
1331 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1332 NL_SET_ERR_MSG(extack,
1333 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1334 return -EINVAL;
1335 }
1336
1337 geneve->net = net;
1338 geneve->dev = dev;
1339
1340 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1341 if (t)
1342 return -EBUSY;
1343
1344 /* make enough headroom for basic scenario */
1345 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1346 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1347 encap_len += sizeof(struct iphdr);
1348 dev->max_mtu -= sizeof(struct iphdr);
1349 } else {
1350 encap_len += sizeof(struct ipv6hdr);
1351 dev->max_mtu -= sizeof(struct ipv6hdr);
1352 }
1353 dev->needed_headroom = encap_len + ETH_HLEN;
1354
1355 if (cfg->collect_md) {
1356 if (tun_on_same_port) {
1357 NL_SET_ERR_MSG(extack,
1358 "There can be only one externally controlled device on a destination port");
1359 return -EPERM;
1360 }
1361 } else {
1362 if (tun_collect_md) {
1363 NL_SET_ERR_MSG(extack,
1364 "There already exists an externally controlled device on this destination port");
1365 return -EPERM;
1366 }
1367 }
1368
1369 dst_cache_reset(&geneve->cfg.info.dst_cache);
1370 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1371
1372 if (geneve->cfg.inner_proto_inherit) {
1373 dev->header_ops = NULL;
1374 dev->type = ARPHRD_NONE;
1375 dev->hard_header_len = 0;
1376 dev->addr_len = 0;
1377 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1378 }
1379
1380 err = register_netdevice(dev);
1381 if (err)
1382 return err;
1383
1384 list_add(&geneve->next, &gn->geneve_list);
1385 return 0;
1386}
1387
1388static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1389{
1390 memset(info, 0, sizeof(*info));
1391 info->key.tp_dst = htons(dst_port);
1392}
1393
1394static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1395 struct netlink_ext_ack *extack,
1396 struct geneve_config *cfg, bool changelink)
1397{
1398 struct ip_tunnel_info *info = &cfg->info;
1399 int attrtype;
1400
1401 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1402 NL_SET_ERR_MSG(extack,
1403 "Cannot specify both IPv4 and IPv6 Remote addresses");
1404 return -EINVAL;
1405 }
1406
1407 if (data[IFLA_GENEVE_REMOTE]) {
1408 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1409 attrtype = IFLA_GENEVE_REMOTE;
1410 goto change_notsup;
1411 }
1412
1413 info->key.u.ipv4.dst =
1414 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1415
1416 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1417 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1418 "Remote IPv4 address cannot be Multicast");
1419 return -EINVAL;
1420 }
1421 }
1422
1423 if (data[IFLA_GENEVE_REMOTE6]) {
1424#if IS_ENABLED(CONFIG_IPV6)
1425 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1426 attrtype = IFLA_GENEVE_REMOTE6;
1427 goto change_notsup;
1428 }
1429
1430 info->mode = IP_TUNNEL_INFO_IPV6;
1431 info->key.u.ipv6.dst =
1432 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1433
1434 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1435 IPV6_ADDR_LINKLOCAL) {
1436 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1437 "Remote IPv6 address cannot be link-local");
1438 return -EINVAL;
1439 }
1440 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1441 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1442 "Remote IPv6 address cannot be Multicast");
1443 return -EINVAL;
1444 }
1445 info->key.tun_flags |= TUNNEL_CSUM;
1446 cfg->use_udp6_rx_checksums = true;
1447#else
1448 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1449 "IPv6 support not enabled in the kernel");
1450 return -EPFNOSUPPORT;
1451#endif
1452 }
1453
1454 if (data[IFLA_GENEVE_ID]) {
1455 __u32 vni;
1456 __u8 tvni[3];
1457 __be64 tunid;
1458
1459 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1460 tvni[0] = (vni & 0x00ff0000) >> 16;
1461 tvni[1] = (vni & 0x0000ff00) >> 8;
1462 tvni[2] = vni & 0x000000ff;
1463
1464 tunid = vni_to_tunnel_id(tvni);
1465 if (changelink && (tunid != info->key.tun_id)) {
1466 attrtype = IFLA_GENEVE_ID;
1467 goto change_notsup;
1468 }
1469 info->key.tun_id = tunid;
1470 }
1471
1472 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1473 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1474 cfg->ttl_inherit = true;
1475 else
1476 cfg->ttl_inherit = false;
1477 } else if (data[IFLA_GENEVE_TTL]) {
1478 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1479 cfg->ttl_inherit = false;
1480 }
1481
1482 if (data[IFLA_GENEVE_TOS])
1483 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1484
1485 if (data[IFLA_GENEVE_DF])
1486 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1487
1488 if (data[IFLA_GENEVE_LABEL]) {
1489 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1490 IPV6_FLOWLABEL_MASK;
1491 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1492 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1493 "Label attribute only applies for IPv6 Geneve devices");
1494 return -EINVAL;
1495 }
1496 }
1497
1498 if (data[IFLA_GENEVE_PORT]) {
1499 if (changelink) {
1500 attrtype = IFLA_GENEVE_PORT;
1501 goto change_notsup;
1502 }
1503 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1504 }
1505
1506 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1507 if (changelink) {
1508 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1509 goto change_notsup;
1510 }
1511 cfg->collect_md = true;
1512 }
1513
1514 if (data[IFLA_GENEVE_UDP_CSUM]) {
1515 if (changelink) {
1516 attrtype = IFLA_GENEVE_UDP_CSUM;
1517 goto change_notsup;
1518 }
1519 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1520 info->key.tun_flags |= TUNNEL_CSUM;
1521 }
1522
1523 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1524#if IS_ENABLED(CONFIG_IPV6)
1525 if (changelink) {
1526 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1527 goto change_notsup;
1528 }
1529 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1530 info->key.tun_flags &= ~TUNNEL_CSUM;
1531#else
1532 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1533 "IPv6 support not enabled in the kernel");
1534 return -EPFNOSUPPORT;
1535#endif
1536 }
1537
1538 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1539#if IS_ENABLED(CONFIG_IPV6)
1540 if (changelink) {
1541 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1542 goto change_notsup;
1543 }
1544 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1545 cfg->use_udp6_rx_checksums = false;
1546#else
1547 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1548 "IPv6 support not enabled in the kernel");
1549 return -EPFNOSUPPORT;
1550#endif
1551 }
1552
1553 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1554 if (changelink) {
1555 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1556 goto change_notsup;
1557 }
1558 cfg->inner_proto_inherit = true;
1559 }
1560
1561 return 0;
1562change_notsup:
1563 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1564 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1565 return -EOPNOTSUPP;
1566}
1567
1568static void geneve_link_config(struct net_device *dev,
1569 struct ip_tunnel_info *info, struct nlattr *tb[])
1570{
1571 struct geneve_dev *geneve = netdev_priv(dev);
1572 int ldev_mtu = 0;
1573
1574 if (tb[IFLA_MTU]) {
1575 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1576 return;
1577 }
1578
1579 switch (ip_tunnel_info_af(info)) {
1580 case AF_INET: {
1581 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1582 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1583
1584 if (!IS_ERR(rt) && rt->dst.dev) {
1585 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1586 ip_rt_put(rt);
1587 }
1588 break;
1589 }
1590#if IS_ENABLED(CONFIG_IPV6)
1591 case AF_INET6: {
1592 struct rt6_info *rt;
1593
1594 if (!__in6_dev_get(dev))
1595 break;
1596
1597 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1598 NULL, 0);
1599
1600 if (rt && rt->dst.dev)
1601 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1602 ip6_rt_put(rt);
1603 break;
1604 }
1605#endif
1606 }
1607
1608 if (ldev_mtu <= 0)
1609 return;
1610
1611 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1612}
1613
1614static int geneve_newlink(struct net *net, struct net_device *dev,
1615 struct nlattr *tb[], struct nlattr *data[],
1616 struct netlink_ext_ack *extack)
1617{
1618 struct geneve_config cfg = {
1619 .df = GENEVE_DF_UNSET,
1620 .use_udp6_rx_checksums = false,
1621 .ttl_inherit = false,
1622 .collect_md = false,
1623 };
1624 int err;
1625
1626 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1627 err = geneve_nl2info(tb, data, extack, &cfg, false);
1628 if (err)
1629 return err;
1630
1631 err = geneve_configure(net, dev, extack, &cfg);
1632 if (err)
1633 return err;
1634
1635 geneve_link_config(dev, &cfg.info, tb);
1636
1637 return 0;
1638}
1639
1640/* Quiesces the geneve device data path for both TX and RX.
1641 *
1642 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1643 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1644 * to complete for the existing set of in-flight packets to be transmitted,
1645 * then we would have quiesced the transmit data path. All the future packets
1646 * will get dropped until we unquiesce the data path.
1647 *
1648 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1649 * if we set that to NULL under RCU and wait for synchronize_net() to
1650 * complete, then we would have quiesced the receive data path.
1651 */
1652static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1653 struct geneve_sock **gs6)
1654{
1655 *gs4 = rtnl_dereference(geneve->sock4);
1656 rcu_assign_pointer(geneve->sock4, NULL);
1657 if (*gs4)
1658 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1659#if IS_ENABLED(CONFIG_IPV6)
1660 *gs6 = rtnl_dereference(geneve->sock6);
1661 rcu_assign_pointer(geneve->sock6, NULL);
1662 if (*gs6)
1663 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1664#else
1665 *gs6 = NULL;
1666#endif
1667 synchronize_net();
1668}
1669
1670/* Resumes the geneve device data path for both TX and RX. */
1671static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1672 struct geneve_sock __maybe_unused *gs6)
1673{
1674 rcu_assign_pointer(geneve->sock4, gs4);
1675 if (gs4)
1676 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1677#if IS_ENABLED(CONFIG_IPV6)
1678 rcu_assign_pointer(geneve->sock6, gs6);
1679 if (gs6)
1680 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1681#endif
1682 synchronize_net();
1683}
1684
1685static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1686 struct nlattr *data[],
1687 struct netlink_ext_ack *extack)
1688{
1689 struct geneve_dev *geneve = netdev_priv(dev);
1690 struct geneve_sock *gs4, *gs6;
1691 struct geneve_config cfg;
1692 int err;
1693
1694 /* If the geneve device is configured for metadata (or externally
1695 * controlled, for example, OVS), then nothing can be changed.
1696 */
1697 if (geneve->cfg.collect_md)
1698 return -EOPNOTSUPP;
1699
1700 /* Start with the existing info. */
1701 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1702 err = geneve_nl2info(tb, data, extack, &cfg, true);
1703 if (err)
1704 return err;
1705
1706 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1707 dst_cache_reset(&cfg.info.dst_cache);
1708 geneve_link_config(dev, &cfg.info, tb);
1709 }
1710
1711 geneve_quiesce(geneve, &gs4, &gs6);
1712 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1713 geneve_unquiesce(geneve, gs4, gs6);
1714
1715 return 0;
1716}
1717
1718static void geneve_dellink(struct net_device *dev, struct list_head *head)
1719{
1720 struct geneve_dev *geneve = netdev_priv(dev);
1721
1722 list_del(&geneve->next);
1723 unregister_netdevice_queue(dev, head);
1724}
1725
1726static size_t geneve_get_size(const struct net_device *dev)
1727{
1728 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1729 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1730 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1731 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1732 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1733 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1734 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1735 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1736 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1737 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1738 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1739 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1740 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1741 0;
1742}
1743
1744static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1745{
1746 struct geneve_dev *geneve = netdev_priv(dev);
1747 struct ip_tunnel_info *info = &geneve->cfg.info;
1748 bool ttl_inherit = geneve->cfg.ttl_inherit;
1749 bool metadata = geneve->cfg.collect_md;
1750 __u8 tmp_vni[3];
1751 __u32 vni;
1752
1753 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1754 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1755 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1756 goto nla_put_failure;
1757
1758 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1759 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1760 info->key.u.ipv4.dst))
1761 goto nla_put_failure;
1762 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1763 !!(info->key.tun_flags & TUNNEL_CSUM)))
1764 goto nla_put_failure;
1765
1766#if IS_ENABLED(CONFIG_IPV6)
1767 } else if (!metadata) {
1768 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1769 &info->key.u.ipv6.dst))
1770 goto nla_put_failure;
1771 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1772 !(info->key.tun_flags & TUNNEL_CSUM)))
1773 goto nla_put_failure;
1774#endif
1775 }
1776
1777 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1778 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1779 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1780 goto nla_put_failure;
1781
1782 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1783 goto nla_put_failure;
1784
1785 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1786 goto nla_put_failure;
1787
1788 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1789 goto nla_put_failure;
1790
1791#if IS_ENABLED(CONFIG_IPV6)
1792 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1793 !geneve->cfg.use_udp6_rx_checksums))
1794 goto nla_put_failure;
1795#endif
1796
1797 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1798 goto nla_put_failure;
1799
1800 if (geneve->cfg.inner_proto_inherit &&
1801 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1802 goto nla_put_failure;
1803
1804 return 0;
1805
1806nla_put_failure:
1807 return -EMSGSIZE;
1808}
1809
1810static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1811 .kind = "geneve",
1812 .maxtype = IFLA_GENEVE_MAX,
1813 .policy = geneve_policy,
1814 .priv_size = sizeof(struct geneve_dev),
1815 .setup = geneve_setup,
1816 .validate = geneve_validate,
1817 .newlink = geneve_newlink,
1818 .changelink = geneve_changelink,
1819 .dellink = geneve_dellink,
1820 .get_size = geneve_get_size,
1821 .fill_info = geneve_fill_info,
1822};
1823
1824struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1825 u8 name_assign_type, u16 dst_port)
1826{
1827 struct nlattr *tb[IFLA_MAX + 1];
1828 struct net_device *dev;
1829 LIST_HEAD(list_kill);
1830 int err;
1831 struct geneve_config cfg = {
1832 .df = GENEVE_DF_UNSET,
1833 .use_udp6_rx_checksums = true,
1834 .ttl_inherit = false,
1835 .collect_md = true,
1836 };
1837
1838 memset(tb, 0, sizeof(tb));
1839 dev = rtnl_create_link(net, name, name_assign_type,
1840 &geneve_link_ops, tb, NULL);
1841 if (IS_ERR(dev))
1842 return dev;
1843
1844 init_tnl_info(&cfg.info, dst_port);
1845 err = geneve_configure(net, dev, NULL, &cfg);
1846 if (err) {
1847 free_netdev(dev);
1848 return ERR_PTR(err);
1849 }
1850
1851 /* openvswitch users expect packet sizes to be unrestricted,
1852 * so set the largest MTU we can.
1853 */
1854 err = geneve_change_mtu(dev, IP_MAX_MTU);
1855 if (err)
1856 goto err;
1857
1858 err = rtnl_configure_link(dev, NULL, 0, NULL);
1859 if (err < 0)
1860 goto err;
1861
1862 return dev;
1863err:
1864 geneve_dellink(dev, &list_kill);
1865 unregister_netdevice_many(&list_kill);
1866 return ERR_PTR(err);
1867}
1868EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1869
1870static int geneve_netdevice_event(struct notifier_block *unused,
1871 unsigned long event, void *ptr)
1872{
1873 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1874
1875 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1876 geneve_offload_rx_ports(dev, true);
1877 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1878 geneve_offload_rx_ports(dev, false);
1879
1880 return NOTIFY_DONE;
1881}
1882
1883static struct notifier_block geneve_notifier_block __read_mostly = {
1884 .notifier_call = geneve_netdevice_event,
1885};
1886
1887static __net_init int geneve_init_net(struct net *net)
1888{
1889 struct geneve_net *gn = net_generic(net, geneve_net_id);
1890
1891 INIT_LIST_HEAD(&gn->geneve_list);
1892 INIT_LIST_HEAD(&gn->sock_list);
1893 return 0;
1894}
1895
1896static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1897{
1898 struct geneve_net *gn = net_generic(net, geneve_net_id);
1899 struct geneve_dev *geneve, *next;
1900 struct net_device *dev, *aux;
1901
1902 /* gather any geneve devices that were moved into this ns */
1903 for_each_netdev_safe(net, dev, aux)
1904 if (dev->rtnl_link_ops == &geneve_link_ops)
1905 unregister_netdevice_queue(dev, head);
1906
1907 /* now gather any other geneve devices that were created in this ns */
1908 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1909 /* If geneve->dev is in the same netns, it was already added
1910 * to the list by the previous loop.
1911 */
1912 if (!net_eq(dev_net(geneve->dev), net))
1913 unregister_netdevice_queue(geneve->dev, head);
1914 }
1915}
1916
1917static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1918{
1919 struct net *net;
1920 LIST_HEAD(list);
1921
1922 rtnl_lock();
1923 list_for_each_entry(net, net_list, exit_list)
1924 geneve_destroy_tunnels(net, &list);
1925
1926 /* unregister the devices gathered above */
1927 unregister_netdevice_many(&list);
1928 rtnl_unlock();
1929
1930 list_for_each_entry(net, net_list, exit_list) {
1931 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1932
1933 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1934 }
1935}
1936
1937static struct pernet_operations geneve_net_ops = {
1938 .init = geneve_init_net,
1939 .exit_batch = geneve_exit_batch_net,
1940 .id = &geneve_net_id,
1941 .size = sizeof(struct geneve_net),
1942};
1943
1944static int __init geneve_init_module(void)
1945{
1946 int rc;
1947
1948 rc = register_pernet_subsys(&geneve_net_ops);
1949 if (rc)
1950 goto out1;
1951
1952 rc = register_netdevice_notifier(&geneve_notifier_block);
1953 if (rc)
1954 goto out2;
1955
1956 rc = rtnl_link_register(&geneve_link_ops);
1957 if (rc)
1958 goto out3;
1959
1960 return 0;
1961out3:
1962 unregister_netdevice_notifier(&geneve_notifier_block);
1963out2:
1964 unregister_pernet_subsys(&geneve_net_ops);
1965out1:
1966 return rc;
1967}
1968late_initcall(geneve_init_module);
1969
1970static void __exit geneve_cleanup_module(void)
1971{
1972 rtnl_link_unregister(&geneve_link_ops);
1973 unregister_netdevice_notifier(&geneve_notifier_block);
1974 unregister_pernet_subsys(&geneve_net_ops);
1975}
1976module_exit(geneve_cleanup_module);
1977
1978MODULE_LICENSE("GPL");
1979MODULE_VERSION(GENEVE_NETDEV_VER);
1980MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1981MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1982MODULE_ALIAS_RTNL_LINK("geneve");
1/*
2 * GENEVE: Generic Network Virtualization Encapsulation
3 *
4 * Copyright (c) 2015 Red Hat, Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 */
10
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/etherdevice.h>
16#include <linux/hash.h>
17#include <net/dst_metadata.h>
18#include <net/gro_cells.h>
19#include <net/rtnetlink.h>
20#include <net/geneve.h>
21#include <net/protocol.h>
22
23#define GENEVE_NETDEV_VER "0.6"
24
25#define GENEVE_UDP_PORT 6081
26
27#define GENEVE_N_VID (1u << 24)
28#define GENEVE_VID_MASK (GENEVE_N_VID - 1)
29
30#define VNI_HASH_BITS 10
31#define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
32
33static bool log_ecn_error = true;
34module_param(log_ecn_error, bool, 0644);
35MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
36
37#define GENEVE_VER 0
38#define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
39
40/* per-network namespace private data for this module */
41struct geneve_net {
42 struct list_head geneve_list;
43 struct list_head sock_list;
44};
45
46static unsigned int geneve_net_id;
47
48struct geneve_dev_node {
49 struct hlist_node hlist;
50 struct geneve_dev *geneve;
51};
52
53/* Pseudo network device */
54struct geneve_dev {
55 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
56#if IS_ENABLED(CONFIG_IPV6)
57 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
58#endif
59 struct net *net; /* netns for packet i/o */
60 struct net_device *dev; /* netdev for geneve tunnel */
61 struct ip_tunnel_info info;
62 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
63#if IS_ENABLED(CONFIG_IPV6)
64 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
65#endif
66 struct list_head next; /* geneve's per namespace list */
67 struct gro_cells gro_cells;
68 bool collect_md;
69 bool use_udp6_rx_checksums;
70};
71
72struct geneve_sock {
73 bool collect_md;
74 struct list_head list;
75 struct socket *sock;
76 struct rcu_head rcu;
77 int refcnt;
78 struct hlist_head vni_list[VNI_HASH_SIZE];
79};
80
81static inline __u32 geneve_net_vni_hash(u8 vni[3])
82{
83 __u32 vnid;
84
85 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
86 return hash_32(vnid, VNI_HASH_BITS);
87}
88
89static __be64 vni_to_tunnel_id(const __u8 *vni)
90{
91#ifdef __BIG_ENDIAN
92 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
93#else
94 return (__force __be64)(((__force u64)vni[0] << 40) |
95 ((__force u64)vni[1] << 48) |
96 ((__force u64)vni[2] << 56));
97#endif
98}
99
100/* Convert 64 bit tunnel ID to 24 bit VNI. */
101static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
102{
103#ifdef __BIG_ENDIAN
104 vni[0] = (__force __u8)(tun_id >> 16);
105 vni[1] = (__force __u8)(tun_id >> 8);
106 vni[2] = (__force __u8)tun_id;
107#else
108 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
109 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
110 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
111#endif
112}
113
114static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
115{
116 return !memcmp(vni, &tun_id[5], 3);
117}
118
119static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
120{
121 return gs->sock->sk->sk_family;
122}
123
124static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
125 __be32 addr, u8 vni[])
126{
127 struct hlist_head *vni_list_head;
128 struct geneve_dev_node *node;
129 __u32 hash;
130
131 /* Find the device for this VNI */
132 hash = geneve_net_vni_hash(vni);
133 vni_list_head = &gs->vni_list[hash];
134 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
135 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
136 addr == node->geneve->info.key.u.ipv4.dst)
137 return node->geneve;
138 }
139 return NULL;
140}
141
142#if IS_ENABLED(CONFIG_IPV6)
143static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
144 struct in6_addr addr6, u8 vni[])
145{
146 struct hlist_head *vni_list_head;
147 struct geneve_dev_node *node;
148 __u32 hash;
149
150 /* Find the device for this VNI */
151 hash = geneve_net_vni_hash(vni);
152 vni_list_head = &gs->vni_list[hash];
153 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
154 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
155 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
156 return node->geneve;
157 }
158 return NULL;
159}
160#endif
161
162static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
163{
164 return (struct genevehdr *)(udp_hdr(skb) + 1);
165}
166
167static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
168 struct sk_buff *skb)
169{
170 static u8 zero_vni[3];
171 u8 *vni;
172
173 if (geneve_get_sk_family(gs) == AF_INET) {
174 struct iphdr *iph;
175 __be32 addr;
176
177 iph = ip_hdr(skb); /* outer IP header... */
178
179 if (gs->collect_md) {
180 vni = zero_vni;
181 addr = 0;
182 } else {
183 vni = geneve_hdr(skb)->vni;
184 addr = iph->saddr;
185 }
186
187 return geneve_lookup(gs, addr, vni);
188#if IS_ENABLED(CONFIG_IPV6)
189 } else if (geneve_get_sk_family(gs) == AF_INET6) {
190 static struct in6_addr zero_addr6;
191 struct ipv6hdr *ip6h;
192 struct in6_addr addr6;
193
194 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
195
196 if (gs->collect_md) {
197 vni = zero_vni;
198 addr6 = zero_addr6;
199 } else {
200 vni = geneve_hdr(skb)->vni;
201 addr6 = ip6h->saddr;
202 }
203
204 return geneve6_lookup(gs, addr6, vni);
205#endif
206 }
207 return NULL;
208}
209
210/* geneve receive/decap routine */
211static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
212 struct sk_buff *skb)
213{
214 struct genevehdr *gnvh = geneve_hdr(skb);
215 struct metadata_dst *tun_dst = NULL;
216 struct pcpu_sw_netstats *stats;
217 unsigned int len;
218 int err = 0;
219 void *oiph;
220
221 if (ip_tunnel_collect_metadata() || gs->collect_md) {
222 __be16 flags;
223
224 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
225 (gnvh->oam ? TUNNEL_OAM : 0) |
226 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
227
228 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
229 vni_to_tunnel_id(gnvh->vni),
230 gnvh->opt_len * 4);
231 if (!tun_dst) {
232 geneve->dev->stats.rx_dropped++;
233 goto drop;
234 }
235 /* Update tunnel dst according to Geneve options. */
236 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
237 gnvh->options, gnvh->opt_len * 4);
238 } else {
239 /* Drop packets w/ critical options,
240 * since we don't support any...
241 */
242 if (gnvh->critical) {
243 geneve->dev->stats.rx_frame_errors++;
244 geneve->dev->stats.rx_errors++;
245 goto drop;
246 }
247 }
248
249 skb_reset_mac_header(skb);
250 skb->protocol = eth_type_trans(skb, geneve->dev);
251 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
252
253 if (tun_dst)
254 skb_dst_set(skb, &tun_dst->dst);
255
256 /* Ignore packet loops (and multicast echo) */
257 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
258 geneve->dev->stats.rx_errors++;
259 goto drop;
260 }
261
262 oiph = skb_network_header(skb);
263 skb_reset_network_header(skb);
264
265 if (geneve_get_sk_family(gs) == AF_INET)
266 err = IP_ECN_decapsulate(oiph, skb);
267#if IS_ENABLED(CONFIG_IPV6)
268 else
269 err = IP6_ECN_decapsulate(oiph, skb);
270#endif
271
272 if (unlikely(err)) {
273 if (log_ecn_error) {
274 if (geneve_get_sk_family(gs) == AF_INET)
275 net_info_ratelimited("non-ECT from %pI4 "
276 "with TOS=%#x\n",
277 &((struct iphdr *)oiph)->saddr,
278 ((struct iphdr *)oiph)->tos);
279#if IS_ENABLED(CONFIG_IPV6)
280 else
281 net_info_ratelimited("non-ECT from %pI6\n",
282 &((struct ipv6hdr *)oiph)->saddr);
283#endif
284 }
285 if (err > 1) {
286 ++geneve->dev->stats.rx_frame_errors;
287 ++geneve->dev->stats.rx_errors;
288 goto drop;
289 }
290 }
291
292 len = skb->len;
293 err = gro_cells_receive(&geneve->gro_cells, skb);
294 if (likely(err == NET_RX_SUCCESS)) {
295 stats = this_cpu_ptr(geneve->dev->tstats);
296 u64_stats_update_begin(&stats->syncp);
297 stats->rx_packets++;
298 stats->rx_bytes += len;
299 u64_stats_update_end(&stats->syncp);
300 }
301 return;
302drop:
303 /* Consume bad packet */
304 kfree_skb(skb);
305}
306
307/* Setup stats when device is created */
308static int geneve_init(struct net_device *dev)
309{
310 struct geneve_dev *geneve = netdev_priv(dev);
311 int err;
312
313 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
314 if (!dev->tstats)
315 return -ENOMEM;
316
317 err = gro_cells_init(&geneve->gro_cells, dev);
318 if (err) {
319 free_percpu(dev->tstats);
320 return err;
321 }
322
323 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
324 if (err) {
325 free_percpu(dev->tstats);
326 gro_cells_destroy(&geneve->gro_cells);
327 return err;
328 }
329 return 0;
330}
331
332static void geneve_uninit(struct net_device *dev)
333{
334 struct geneve_dev *geneve = netdev_priv(dev);
335
336 dst_cache_destroy(&geneve->info.dst_cache);
337 gro_cells_destroy(&geneve->gro_cells);
338 free_percpu(dev->tstats);
339}
340
341/* Callback from net/ipv4/udp.c to receive packets */
342static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
343{
344 struct genevehdr *geneveh;
345 struct geneve_dev *geneve;
346 struct geneve_sock *gs;
347 int opts_len;
348
349 /* Need UDP and Geneve header to be present */
350 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
351 goto drop;
352
353 /* Return packets with reserved bits set */
354 geneveh = geneve_hdr(skb);
355 if (unlikely(geneveh->ver != GENEVE_VER))
356 goto drop;
357
358 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
359 goto drop;
360
361 gs = rcu_dereference_sk_user_data(sk);
362 if (!gs)
363 goto drop;
364
365 geneve = geneve_lookup_skb(gs, skb);
366 if (!geneve)
367 goto drop;
368
369 opts_len = geneveh->opt_len * 4;
370 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
371 htons(ETH_P_TEB),
372 !net_eq(geneve->net, dev_net(geneve->dev)))) {
373 geneve->dev->stats.rx_dropped++;
374 goto drop;
375 }
376
377 geneve_rx(geneve, gs, skb);
378 return 0;
379
380drop:
381 /* Consume bad packet */
382 kfree_skb(skb);
383 return 0;
384}
385
386static struct socket *geneve_create_sock(struct net *net, bool ipv6,
387 __be16 port, bool ipv6_rx_csum)
388{
389 struct socket *sock;
390 struct udp_port_cfg udp_conf;
391 int err;
392
393 memset(&udp_conf, 0, sizeof(udp_conf));
394
395 if (ipv6) {
396 udp_conf.family = AF_INET6;
397 udp_conf.ipv6_v6only = 1;
398 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
399 } else {
400 udp_conf.family = AF_INET;
401 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
402 }
403
404 udp_conf.local_udp_port = port;
405
406 /* Open UDP socket */
407 err = udp_sock_create(net, &udp_conf, &sock);
408 if (err < 0)
409 return ERR_PTR(err);
410
411 return sock;
412}
413
414static int geneve_hlen(struct genevehdr *gh)
415{
416 return sizeof(*gh) + gh->opt_len * 4;
417}
418
419static struct sk_buff **geneve_gro_receive(struct sock *sk,
420 struct sk_buff **head,
421 struct sk_buff *skb)
422{
423 struct sk_buff *p, **pp = NULL;
424 struct genevehdr *gh, *gh2;
425 unsigned int hlen, gh_len, off_gnv;
426 const struct packet_offload *ptype;
427 __be16 type;
428 int flush = 1;
429
430 off_gnv = skb_gro_offset(skb);
431 hlen = off_gnv + sizeof(*gh);
432 gh = skb_gro_header_fast(skb, off_gnv);
433 if (skb_gro_header_hard(skb, hlen)) {
434 gh = skb_gro_header_slow(skb, hlen, off_gnv);
435 if (unlikely(!gh))
436 goto out;
437 }
438
439 if (gh->ver != GENEVE_VER || gh->oam)
440 goto out;
441 gh_len = geneve_hlen(gh);
442
443 hlen = off_gnv + gh_len;
444 if (skb_gro_header_hard(skb, hlen)) {
445 gh = skb_gro_header_slow(skb, hlen, off_gnv);
446 if (unlikely(!gh))
447 goto out;
448 }
449
450 for (p = *head; p; p = p->next) {
451 if (!NAPI_GRO_CB(p)->same_flow)
452 continue;
453
454 gh2 = (struct genevehdr *)(p->data + off_gnv);
455 if (gh->opt_len != gh2->opt_len ||
456 memcmp(gh, gh2, gh_len)) {
457 NAPI_GRO_CB(p)->same_flow = 0;
458 continue;
459 }
460 }
461
462 type = gh->proto_type;
463
464 rcu_read_lock();
465 ptype = gro_find_receive_by_type(type);
466 if (!ptype)
467 goto out_unlock;
468
469 skb_gro_pull(skb, gh_len);
470 skb_gro_postpull_rcsum(skb, gh, gh_len);
471 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
472 flush = 0;
473
474out_unlock:
475 rcu_read_unlock();
476out:
477 NAPI_GRO_CB(skb)->flush |= flush;
478
479 return pp;
480}
481
482static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
483 int nhoff)
484{
485 struct genevehdr *gh;
486 struct packet_offload *ptype;
487 __be16 type;
488 int gh_len;
489 int err = -ENOSYS;
490
491 gh = (struct genevehdr *)(skb->data + nhoff);
492 gh_len = geneve_hlen(gh);
493 type = gh->proto_type;
494
495 rcu_read_lock();
496 ptype = gro_find_complete_by_type(type);
497 if (ptype)
498 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
499
500 rcu_read_unlock();
501
502 skb_set_inner_mac_header(skb, nhoff + gh_len);
503
504 return err;
505}
506
507/* Create new listen socket if needed */
508static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
509 bool ipv6, bool ipv6_rx_csum)
510{
511 struct geneve_net *gn = net_generic(net, geneve_net_id);
512 struct geneve_sock *gs;
513 struct socket *sock;
514 struct udp_tunnel_sock_cfg tunnel_cfg;
515 int h;
516
517 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
518 if (!gs)
519 return ERR_PTR(-ENOMEM);
520
521 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
522 if (IS_ERR(sock)) {
523 kfree(gs);
524 return ERR_CAST(sock);
525 }
526
527 gs->sock = sock;
528 gs->refcnt = 1;
529 for (h = 0; h < VNI_HASH_SIZE; ++h)
530 INIT_HLIST_HEAD(&gs->vni_list[h]);
531
532 /* Initialize the geneve udp offloads structure */
533 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
534
535 /* Mark socket as an encapsulation socket */
536 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
537 tunnel_cfg.sk_user_data = gs;
538 tunnel_cfg.encap_type = 1;
539 tunnel_cfg.gro_receive = geneve_gro_receive;
540 tunnel_cfg.gro_complete = geneve_gro_complete;
541 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
542 tunnel_cfg.encap_destroy = NULL;
543 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
544 list_add(&gs->list, &gn->sock_list);
545 return gs;
546}
547
548static void __geneve_sock_release(struct geneve_sock *gs)
549{
550 if (!gs || --gs->refcnt)
551 return;
552
553 list_del(&gs->list);
554 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
555 udp_tunnel_sock_release(gs->sock);
556 kfree_rcu(gs, rcu);
557}
558
559static void geneve_sock_release(struct geneve_dev *geneve)
560{
561 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
562#if IS_ENABLED(CONFIG_IPV6)
563 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
564
565 rcu_assign_pointer(geneve->sock6, NULL);
566#endif
567
568 rcu_assign_pointer(geneve->sock4, NULL);
569 synchronize_net();
570
571 __geneve_sock_release(gs4);
572#if IS_ENABLED(CONFIG_IPV6)
573 __geneve_sock_release(gs6);
574#endif
575}
576
577static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
578 sa_family_t family,
579 __be16 dst_port)
580{
581 struct geneve_sock *gs;
582
583 list_for_each_entry(gs, &gn->sock_list, list) {
584 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
585 geneve_get_sk_family(gs) == family) {
586 return gs;
587 }
588 }
589 return NULL;
590}
591
592static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
593{
594 struct net *net = geneve->net;
595 struct geneve_net *gn = net_generic(net, geneve_net_id);
596 struct geneve_dev_node *node;
597 struct geneve_sock *gs;
598 __u8 vni[3];
599 __u32 hash;
600
601 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
602 if (gs) {
603 gs->refcnt++;
604 goto out;
605 }
606
607 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
608 geneve->use_udp6_rx_checksums);
609 if (IS_ERR(gs))
610 return PTR_ERR(gs);
611
612out:
613 gs->collect_md = geneve->collect_md;
614#if IS_ENABLED(CONFIG_IPV6)
615 if (ipv6) {
616 rcu_assign_pointer(geneve->sock6, gs);
617 node = &geneve->hlist6;
618 } else
619#endif
620 {
621 rcu_assign_pointer(geneve->sock4, gs);
622 node = &geneve->hlist4;
623 }
624 node->geneve = geneve;
625
626 tunnel_id_to_vni(geneve->info.key.tun_id, vni);
627 hash = geneve_net_vni_hash(vni);
628 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
629 return 0;
630}
631
632static int geneve_open(struct net_device *dev)
633{
634 struct geneve_dev *geneve = netdev_priv(dev);
635 bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6);
636 bool metadata = geneve->collect_md;
637 int ret = 0;
638
639#if IS_ENABLED(CONFIG_IPV6)
640 if (ipv6 || metadata)
641 ret = geneve_sock_add(geneve, true);
642#endif
643 if (!ret && (!ipv6 || metadata))
644 ret = geneve_sock_add(geneve, false);
645 if (ret < 0)
646 geneve_sock_release(geneve);
647
648 return ret;
649}
650
651static int geneve_stop(struct net_device *dev)
652{
653 struct geneve_dev *geneve = netdev_priv(dev);
654
655 hlist_del_init_rcu(&geneve->hlist4.hlist);
656#if IS_ENABLED(CONFIG_IPV6)
657 hlist_del_init_rcu(&geneve->hlist6.hlist);
658#endif
659 geneve_sock_release(geneve);
660 return 0;
661}
662
663static void geneve_build_header(struct genevehdr *geneveh,
664 const struct ip_tunnel_info *info)
665{
666 geneveh->ver = GENEVE_VER;
667 geneveh->opt_len = info->options_len / 4;
668 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
669 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
670 geneveh->rsvd1 = 0;
671 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
672 geneveh->proto_type = htons(ETH_P_TEB);
673 geneveh->rsvd2 = 0;
674
675 ip_tunnel_info_opts_get(geneveh->options, info);
676}
677
678static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
679 const struct ip_tunnel_info *info,
680 bool xnet, int ip_hdr_len)
681{
682 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
683 struct genevehdr *gnvh;
684 int min_headroom;
685 int err;
686
687 skb_reset_mac_header(skb);
688 skb_scrub_packet(skb, xnet);
689
690 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
691 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
692 err = skb_cow_head(skb, min_headroom);
693 if (unlikely(err))
694 goto free_dst;
695
696 err = udp_tunnel_handle_offloads(skb, udp_sum);
697 if (err)
698 goto free_dst;
699
700 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
701 geneve_build_header(gnvh, info);
702 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
703 return 0;
704
705free_dst:
706 dst_release(dst);
707 return err;
708}
709
710static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
711 struct net_device *dev,
712 struct geneve_sock *gs4,
713 struct flowi4 *fl4,
714 const struct ip_tunnel_info *info)
715{
716 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
717 struct geneve_dev *geneve = netdev_priv(dev);
718 struct dst_cache *dst_cache;
719 struct rtable *rt = NULL;
720 __u8 tos;
721
722 if (!gs4)
723 return ERR_PTR(-EIO);
724
725 memset(fl4, 0, sizeof(*fl4));
726 fl4->flowi4_mark = skb->mark;
727 fl4->flowi4_proto = IPPROTO_UDP;
728 fl4->daddr = info->key.u.ipv4.dst;
729 fl4->saddr = info->key.u.ipv4.src;
730
731 tos = info->key.tos;
732 if ((tos == 1) && !geneve->collect_md) {
733 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
734 use_cache = false;
735 }
736 fl4->flowi4_tos = RT_TOS(tos);
737
738 dst_cache = (struct dst_cache *)&info->dst_cache;
739 if (use_cache) {
740 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
741 if (rt)
742 return rt;
743 }
744 rt = ip_route_output_key(geneve->net, fl4);
745 if (IS_ERR(rt)) {
746 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
747 return ERR_PTR(-ENETUNREACH);
748 }
749 if (rt->dst.dev == dev) { /* is this necessary? */
750 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
751 ip_rt_put(rt);
752 return ERR_PTR(-ELOOP);
753 }
754 if (use_cache)
755 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
756 return rt;
757}
758
759#if IS_ENABLED(CONFIG_IPV6)
760static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
761 struct net_device *dev,
762 struct geneve_sock *gs6,
763 struct flowi6 *fl6,
764 const struct ip_tunnel_info *info)
765{
766 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
767 struct geneve_dev *geneve = netdev_priv(dev);
768 struct dst_entry *dst = NULL;
769 struct dst_cache *dst_cache;
770 __u8 prio;
771
772 if (!gs6)
773 return ERR_PTR(-EIO);
774
775 memset(fl6, 0, sizeof(*fl6));
776 fl6->flowi6_mark = skb->mark;
777 fl6->flowi6_proto = IPPROTO_UDP;
778 fl6->daddr = info->key.u.ipv6.dst;
779 fl6->saddr = info->key.u.ipv6.src;
780 prio = info->key.tos;
781 if ((prio == 1) && !geneve->collect_md) {
782 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
783 use_cache = false;
784 }
785
786 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
787 info->key.label);
788 dst_cache = (struct dst_cache *)&info->dst_cache;
789 if (use_cache) {
790 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
791 if (dst)
792 return dst;
793 }
794 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
795 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
796 return ERR_PTR(-ENETUNREACH);
797 }
798 if (dst->dev == dev) { /* is this necessary? */
799 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
800 dst_release(dst);
801 return ERR_PTR(-ELOOP);
802 }
803
804 if (use_cache)
805 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
806 return dst;
807}
808#endif
809
810static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
811 struct geneve_dev *geneve,
812 const struct ip_tunnel_info *info)
813{
814 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
815 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
816 const struct ip_tunnel_key *key = &info->key;
817 struct rtable *rt;
818 struct flowi4 fl4;
819 __u8 tos, ttl;
820 __be16 sport;
821 __be16 df;
822 int err;
823
824 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
825 if (IS_ERR(rt))
826 return PTR_ERR(rt);
827
828 if (skb_dst(skb)) {
829 int mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr) -
830 GENEVE_BASE_HLEN - info->options_len - 14;
831
832 skb_dst_update_pmtu(skb, mtu);
833 }
834
835 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
836 if (geneve->collect_md) {
837 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
838 ttl = key->ttl;
839 } else {
840 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
841 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
842 }
843 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
844
845 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
846 if (unlikely(err))
847 return err;
848
849 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
850 tos, ttl, df, sport, geneve->info.key.tp_dst,
851 !net_eq(geneve->net, dev_net(geneve->dev)),
852 !(info->key.tun_flags & TUNNEL_CSUM));
853 return 0;
854}
855
856#if IS_ENABLED(CONFIG_IPV6)
857static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
858 struct geneve_dev *geneve,
859 const struct ip_tunnel_info *info)
860{
861 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
862 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
863 const struct ip_tunnel_key *key = &info->key;
864 struct dst_entry *dst = NULL;
865 struct flowi6 fl6;
866 __u8 prio, ttl;
867 __be16 sport;
868 int err;
869
870 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
871 if (IS_ERR(dst))
872 return PTR_ERR(dst);
873
874 if (skb_dst(skb)) {
875 int mtu = dst_mtu(dst) - sizeof(struct ipv6hdr) -
876 GENEVE_BASE_HLEN - info->options_len - 14;
877
878 skb_dst_update_pmtu(skb, mtu);
879 }
880
881 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
882 if (geneve->collect_md) {
883 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
884 ttl = key->ttl;
885 } else {
886 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
887 ip_hdr(skb), skb);
888 ttl = key->ttl ? : ip6_dst_hoplimit(dst);
889 }
890 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
891 if (unlikely(err))
892 return err;
893
894 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
895 &fl6.saddr, &fl6.daddr, prio, ttl,
896 info->key.label, sport, geneve->info.key.tp_dst,
897 !(info->key.tun_flags & TUNNEL_CSUM));
898 return 0;
899}
900#endif
901
902static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
903{
904 struct geneve_dev *geneve = netdev_priv(dev);
905 struct ip_tunnel_info *info = NULL;
906 int err;
907
908 if (geneve->collect_md) {
909 info = skb_tunnel_info(skb);
910 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
911 err = -EINVAL;
912 netdev_dbg(dev, "no tunnel metadata\n");
913 goto tx_error;
914 }
915 } else {
916 info = &geneve->info;
917 }
918
919 rcu_read_lock();
920#if IS_ENABLED(CONFIG_IPV6)
921 if (info->mode & IP_TUNNEL_INFO_IPV6)
922 err = geneve6_xmit_skb(skb, dev, geneve, info);
923 else
924#endif
925 err = geneve_xmit_skb(skb, dev, geneve, info);
926 rcu_read_unlock();
927
928 if (likely(!err))
929 return NETDEV_TX_OK;
930tx_error:
931 dev_kfree_skb(skb);
932
933 if (err == -ELOOP)
934 dev->stats.collisions++;
935 else if (err == -ENETUNREACH)
936 dev->stats.tx_carrier_errors++;
937
938 dev->stats.tx_errors++;
939 return NETDEV_TX_OK;
940}
941
942static int geneve_change_mtu(struct net_device *dev, int new_mtu)
943{
944 /* Only possible if called internally, ndo_change_mtu path's new_mtu
945 * is guaranteed to be between dev->min_mtu and dev->max_mtu.
946 */
947 if (new_mtu > dev->max_mtu)
948 new_mtu = dev->max_mtu;
949
950 dev->mtu = new_mtu;
951 return 0;
952}
953
954static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
955{
956 struct ip_tunnel_info *info = skb_tunnel_info(skb);
957 struct geneve_dev *geneve = netdev_priv(dev);
958
959 if (ip_tunnel_info_af(info) == AF_INET) {
960 struct rtable *rt;
961 struct flowi4 fl4;
962 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
963
964 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
965 if (IS_ERR(rt))
966 return PTR_ERR(rt);
967
968 ip_rt_put(rt);
969 info->key.u.ipv4.src = fl4.saddr;
970#if IS_ENABLED(CONFIG_IPV6)
971 } else if (ip_tunnel_info_af(info) == AF_INET6) {
972 struct dst_entry *dst;
973 struct flowi6 fl6;
974 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
975
976 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
977 if (IS_ERR(dst))
978 return PTR_ERR(dst);
979
980 dst_release(dst);
981 info->key.u.ipv6.src = fl6.saddr;
982#endif
983 } else {
984 return -EINVAL;
985 }
986
987 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
988 1, USHRT_MAX, true);
989 info->key.tp_dst = geneve->info.key.tp_dst;
990 return 0;
991}
992
993static const struct net_device_ops geneve_netdev_ops = {
994 .ndo_init = geneve_init,
995 .ndo_uninit = geneve_uninit,
996 .ndo_open = geneve_open,
997 .ndo_stop = geneve_stop,
998 .ndo_start_xmit = geneve_xmit,
999 .ndo_get_stats64 = ip_tunnel_get_stats64,
1000 .ndo_change_mtu = geneve_change_mtu,
1001 .ndo_validate_addr = eth_validate_addr,
1002 .ndo_set_mac_address = eth_mac_addr,
1003 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1004};
1005
1006static void geneve_get_drvinfo(struct net_device *dev,
1007 struct ethtool_drvinfo *drvinfo)
1008{
1009 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1010 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1011}
1012
1013static const struct ethtool_ops geneve_ethtool_ops = {
1014 .get_drvinfo = geneve_get_drvinfo,
1015 .get_link = ethtool_op_get_link,
1016};
1017
1018/* Info for udev, that this is a virtual tunnel endpoint */
1019static struct device_type geneve_type = {
1020 .name = "geneve",
1021};
1022
1023/* Calls the ndo_udp_tunnel_add of the caller in order to
1024 * supply the listening GENEVE udp ports. Callers are expected
1025 * to implement the ndo_udp_tunnel_add.
1026 */
1027static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1028{
1029 struct net *net = dev_net(dev);
1030 struct geneve_net *gn = net_generic(net, geneve_net_id);
1031 struct geneve_sock *gs;
1032
1033 rcu_read_lock();
1034 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1035 if (push) {
1036 udp_tunnel_push_rx_port(dev, gs->sock,
1037 UDP_TUNNEL_TYPE_GENEVE);
1038 } else {
1039 udp_tunnel_drop_rx_port(dev, gs->sock,
1040 UDP_TUNNEL_TYPE_GENEVE);
1041 }
1042 }
1043 rcu_read_unlock();
1044}
1045
1046/* Initialize the device structure. */
1047static void geneve_setup(struct net_device *dev)
1048{
1049 ether_setup(dev);
1050
1051 dev->netdev_ops = &geneve_netdev_ops;
1052 dev->ethtool_ops = &geneve_ethtool_ops;
1053 dev->needs_free_netdev = true;
1054
1055 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1056
1057 dev->features |= NETIF_F_LLTX;
1058 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1059 dev->features |= NETIF_F_RXCSUM;
1060 dev->features |= NETIF_F_GSO_SOFTWARE;
1061
1062 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1063 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1064
1065 /* MTU range: 68 - (something less than 65535) */
1066 dev->min_mtu = ETH_MIN_MTU;
1067 /* The max_mtu calculation does not take account of GENEVE
1068 * options, to avoid excluding potentially valid
1069 * configurations. This will be further reduced by IPvX hdr size.
1070 */
1071 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1072
1073 netif_keep_dst(dev);
1074 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1075 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1076 eth_hw_addr_random(dev);
1077}
1078
1079static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1080 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1081 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1082 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1083 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1084 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1085 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1086 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1087 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1088 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1089 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1090 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1091};
1092
1093static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1094 struct netlink_ext_ack *extack)
1095{
1096 if (tb[IFLA_ADDRESS]) {
1097 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1098 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1099 "Provided link layer address is not Ethernet");
1100 return -EINVAL;
1101 }
1102
1103 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1104 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1105 "Provided Ethernet address is not unicast");
1106 return -EADDRNOTAVAIL;
1107 }
1108 }
1109
1110 if (!data) {
1111 NL_SET_ERR_MSG(extack,
1112 "Not enough attributes provided to perform the operation");
1113 return -EINVAL;
1114 }
1115
1116 if (data[IFLA_GENEVE_ID]) {
1117 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1118
1119 if (vni >= GENEVE_N_VID) {
1120 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1121 "Geneve ID must be lower than 16777216");
1122 return -ERANGE;
1123 }
1124 }
1125
1126 return 0;
1127}
1128
1129static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1130 const struct ip_tunnel_info *info,
1131 bool *tun_on_same_port,
1132 bool *tun_collect_md)
1133{
1134 struct geneve_dev *geneve, *t = NULL;
1135
1136 *tun_on_same_port = false;
1137 *tun_collect_md = false;
1138 list_for_each_entry(geneve, &gn->geneve_list, next) {
1139 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1140 *tun_collect_md = geneve->collect_md;
1141 *tun_on_same_port = true;
1142 }
1143 if (info->key.tun_id == geneve->info.key.tun_id &&
1144 info->key.tp_dst == geneve->info.key.tp_dst &&
1145 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1146 t = geneve;
1147 }
1148 return t;
1149}
1150
1151static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1152{
1153 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1154 info->key.ttl || info->key.label || info->key.tp_src ||
1155 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1156}
1157
1158static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1159 struct ip_tunnel_info *b)
1160{
1161 if (ip_tunnel_info_af(a) == AF_INET)
1162 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1163 else
1164 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1165}
1166
1167static int geneve_configure(struct net *net, struct net_device *dev,
1168 struct netlink_ext_ack *extack,
1169 const struct ip_tunnel_info *info,
1170 bool metadata, bool ipv6_rx_csum)
1171{
1172 struct geneve_net *gn = net_generic(net, geneve_net_id);
1173 struct geneve_dev *t, *geneve = netdev_priv(dev);
1174 bool tun_collect_md, tun_on_same_port;
1175 int err, encap_len;
1176
1177 if (metadata && !is_tnl_info_zero(info)) {
1178 NL_SET_ERR_MSG(extack,
1179 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1180 return -EINVAL;
1181 }
1182
1183 geneve->net = net;
1184 geneve->dev = dev;
1185
1186 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1187 if (t)
1188 return -EBUSY;
1189
1190 /* make enough headroom for basic scenario */
1191 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1192 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1193 encap_len += sizeof(struct iphdr);
1194 dev->max_mtu -= sizeof(struct iphdr);
1195 } else {
1196 encap_len += sizeof(struct ipv6hdr);
1197 dev->max_mtu -= sizeof(struct ipv6hdr);
1198 }
1199 dev->needed_headroom = encap_len + ETH_HLEN;
1200
1201 if (metadata) {
1202 if (tun_on_same_port) {
1203 NL_SET_ERR_MSG(extack,
1204 "There can be only one externally controlled device on a destination port");
1205 return -EPERM;
1206 }
1207 } else {
1208 if (tun_collect_md) {
1209 NL_SET_ERR_MSG(extack,
1210 "There already exists an externally controlled device on this destination port");
1211 return -EPERM;
1212 }
1213 }
1214
1215 dst_cache_reset(&geneve->info.dst_cache);
1216 geneve->info = *info;
1217 geneve->collect_md = metadata;
1218 geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1219
1220 err = register_netdevice(dev);
1221 if (err)
1222 return err;
1223
1224 list_add(&geneve->next, &gn->geneve_list);
1225 return 0;
1226}
1227
1228static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1229{
1230 memset(info, 0, sizeof(*info));
1231 info->key.tp_dst = htons(dst_port);
1232}
1233
1234static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1235 struct netlink_ext_ack *extack,
1236 struct ip_tunnel_info *info, bool *metadata,
1237 bool *use_udp6_rx_checksums, bool changelink)
1238{
1239 int attrtype;
1240
1241 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1242 NL_SET_ERR_MSG(extack,
1243 "Cannot specify both IPv4 and IPv6 Remote addresses");
1244 return -EINVAL;
1245 }
1246
1247 if (data[IFLA_GENEVE_REMOTE]) {
1248 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1249 attrtype = IFLA_GENEVE_REMOTE;
1250 goto change_notsup;
1251 }
1252
1253 info->key.u.ipv4.dst =
1254 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1255
1256 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1257 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1258 "Remote IPv4 address cannot be Multicast");
1259 return -EINVAL;
1260 }
1261 }
1262
1263 if (data[IFLA_GENEVE_REMOTE6]) {
1264 #if IS_ENABLED(CONFIG_IPV6)
1265 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1266 attrtype = IFLA_GENEVE_REMOTE6;
1267 goto change_notsup;
1268 }
1269
1270 info->mode = IP_TUNNEL_INFO_IPV6;
1271 info->key.u.ipv6.dst =
1272 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1273
1274 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1275 IPV6_ADDR_LINKLOCAL) {
1276 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1277 "Remote IPv6 address cannot be link-local");
1278 return -EINVAL;
1279 }
1280 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1281 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1282 "Remote IPv6 address cannot be Multicast");
1283 return -EINVAL;
1284 }
1285 info->key.tun_flags |= TUNNEL_CSUM;
1286 *use_udp6_rx_checksums = true;
1287#else
1288 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1289 "IPv6 support not enabled in the kernel");
1290 return -EPFNOSUPPORT;
1291#endif
1292 }
1293
1294 if (data[IFLA_GENEVE_ID]) {
1295 __u32 vni;
1296 __u8 tvni[3];
1297 __be64 tunid;
1298
1299 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1300 tvni[0] = (vni & 0x00ff0000) >> 16;
1301 tvni[1] = (vni & 0x0000ff00) >> 8;
1302 tvni[2] = vni & 0x000000ff;
1303
1304 tunid = vni_to_tunnel_id(tvni);
1305 if (changelink && (tunid != info->key.tun_id)) {
1306 attrtype = IFLA_GENEVE_ID;
1307 goto change_notsup;
1308 }
1309 info->key.tun_id = tunid;
1310 }
1311
1312 if (data[IFLA_GENEVE_TTL])
1313 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1314
1315 if (data[IFLA_GENEVE_TOS])
1316 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1317
1318 if (data[IFLA_GENEVE_LABEL]) {
1319 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1320 IPV6_FLOWLABEL_MASK;
1321 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1322 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1323 "Label attribute only applies for IPv6 Geneve devices");
1324 return -EINVAL;
1325 }
1326 }
1327
1328 if (data[IFLA_GENEVE_PORT]) {
1329 if (changelink) {
1330 attrtype = IFLA_GENEVE_PORT;
1331 goto change_notsup;
1332 }
1333 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1334 }
1335
1336 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1337 if (changelink) {
1338 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1339 goto change_notsup;
1340 }
1341 *metadata = true;
1342 }
1343
1344 if (data[IFLA_GENEVE_UDP_CSUM]) {
1345 if (changelink) {
1346 attrtype = IFLA_GENEVE_UDP_CSUM;
1347 goto change_notsup;
1348 }
1349 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1350 info->key.tun_flags |= TUNNEL_CSUM;
1351 }
1352
1353 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1354#if IS_ENABLED(CONFIG_IPV6)
1355 if (changelink) {
1356 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1357 goto change_notsup;
1358 }
1359 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1360 info->key.tun_flags &= ~TUNNEL_CSUM;
1361#else
1362 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1363 "IPv6 support not enabled in the kernel");
1364 return -EPFNOSUPPORT;
1365#endif
1366 }
1367
1368 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1369#if IS_ENABLED(CONFIG_IPV6)
1370 if (changelink) {
1371 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1372 goto change_notsup;
1373 }
1374 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1375 *use_udp6_rx_checksums = false;
1376#else
1377 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1378 "IPv6 support not enabled in the kernel");
1379 return -EPFNOSUPPORT;
1380#endif
1381 }
1382
1383 return 0;
1384change_notsup:
1385 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1386 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1387 return -EOPNOTSUPP;
1388}
1389
1390static int geneve_newlink(struct net *net, struct net_device *dev,
1391 struct nlattr *tb[], struct nlattr *data[],
1392 struct netlink_ext_ack *extack)
1393{
1394 bool use_udp6_rx_checksums = false;
1395 struct ip_tunnel_info info;
1396 bool metadata = false;
1397 int err;
1398
1399 init_tnl_info(&info, GENEVE_UDP_PORT);
1400 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1401 &use_udp6_rx_checksums, false);
1402 if (err)
1403 return err;
1404
1405 return geneve_configure(net, dev, extack, &info, metadata,
1406 use_udp6_rx_checksums);
1407}
1408
1409/* Quiesces the geneve device data path for both TX and RX.
1410 *
1411 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1412 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1413 * to complete for the existing set of in-flight packets to be transmitted,
1414 * then we would have quiesced the transmit data path. All the future packets
1415 * will get dropped until we unquiesce the data path.
1416 *
1417 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1418 * if we set that to NULL under RCU and wait for synchronize_net() to
1419 * complete, then we would have quiesced the receive data path.
1420 */
1421static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1422 struct geneve_sock **gs6)
1423{
1424 *gs4 = rtnl_dereference(geneve->sock4);
1425 rcu_assign_pointer(geneve->sock4, NULL);
1426 if (*gs4)
1427 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1428#if IS_ENABLED(CONFIG_IPV6)
1429 *gs6 = rtnl_dereference(geneve->sock6);
1430 rcu_assign_pointer(geneve->sock6, NULL);
1431 if (*gs6)
1432 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1433#else
1434 *gs6 = NULL;
1435#endif
1436 synchronize_net();
1437}
1438
1439/* Resumes the geneve device data path for both TX and RX. */
1440static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1441 struct geneve_sock __maybe_unused *gs6)
1442{
1443 rcu_assign_pointer(geneve->sock4, gs4);
1444 if (gs4)
1445 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1446#if IS_ENABLED(CONFIG_IPV6)
1447 rcu_assign_pointer(geneve->sock6, gs6);
1448 if (gs6)
1449 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1450#endif
1451 synchronize_net();
1452}
1453
1454static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1455 struct nlattr *data[],
1456 struct netlink_ext_ack *extack)
1457{
1458 struct geneve_dev *geneve = netdev_priv(dev);
1459 struct geneve_sock *gs4, *gs6;
1460 struct ip_tunnel_info info;
1461 bool metadata;
1462 bool use_udp6_rx_checksums;
1463 int err;
1464
1465 /* If the geneve device is configured for metadata (or externally
1466 * controlled, for example, OVS), then nothing can be changed.
1467 */
1468 if (geneve->collect_md)
1469 return -EOPNOTSUPP;
1470
1471 /* Start with the existing info. */
1472 memcpy(&info, &geneve->info, sizeof(info));
1473 metadata = geneve->collect_md;
1474 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1475 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1476 &use_udp6_rx_checksums, true);
1477 if (err)
1478 return err;
1479
1480 if (!geneve_dst_addr_equal(&geneve->info, &info))
1481 dst_cache_reset(&info.dst_cache);
1482
1483 geneve_quiesce(geneve, &gs4, &gs6);
1484 geneve->info = info;
1485 geneve->collect_md = metadata;
1486 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1487 geneve_unquiesce(geneve, gs4, gs6);
1488
1489 return 0;
1490}
1491
1492static void geneve_dellink(struct net_device *dev, struct list_head *head)
1493{
1494 struct geneve_dev *geneve = netdev_priv(dev);
1495
1496 list_del(&geneve->next);
1497 unregister_netdevice_queue(dev, head);
1498}
1499
1500static size_t geneve_get_size(const struct net_device *dev)
1501{
1502 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1503 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1504 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1505 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1506 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1507 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1508 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1509 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1510 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1511 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1512 0;
1513}
1514
1515static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1516{
1517 struct geneve_dev *geneve = netdev_priv(dev);
1518 struct ip_tunnel_info *info = &geneve->info;
1519 bool metadata = geneve->collect_md;
1520 __u8 tmp_vni[3];
1521 __u32 vni;
1522
1523 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1524 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1525 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1526 goto nla_put_failure;
1527
1528 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1529 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1530 info->key.u.ipv4.dst))
1531 goto nla_put_failure;
1532 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1533 !!(info->key.tun_flags & TUNNEL_CSUM)))
1534 goto nla_put_failure;
1535
1536#if IS_ENABLED(CONFIG_IPV6)
1537 } else if (!metadata) {
1538 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1539 &info->key.u.ipv6.dst))
1540 goto nla_put_failure;
1541 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1542 !(info->key.tun_flags & TUNNEL_CSUM)))
1543 goto nla_put_failure;
1544#endif
1545 }
1546
1547 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1548 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1549 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1550 goto nla_put_failure;
1551
1552 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1553 goto nla_put_failure;
1554
1555 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1556 goto nla_put_failure;
1557
1558#if IS_ENABLED(CONFIG_IPV6)
1559 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1560 !geneve->use_udp6_rx_checksums))
1561 goto nla_put_failure;
1562#endif
1563
1564 return 0;
1565
1566nla_put_failure:
1567 return -EMSGSIZE;
1568}
1569
1570static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1571 .kind = "geneve",
1572 .maxtype = IFLA_GENEVE_MAX,
1573 .policy = geneve_policy,
1574 .priv_size = sizeof(struct geneve_dev),
1575 .setup = geneve_setup,
1576 .validate = geneve_validate,
1577 .newlink = geneve_newlink,
1578 .changelink = geneve_changelink,
1579 .dellink = geneve_dellink,
1580 .get_size = geneve_get_size,
1581 .fill_info = geneve_fill_info,
1582};
1583
1584struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1585 u8 name_assign_type, u16 dst_port)
1586{
1587 struct nlattr *tb[IFLA_MAX + 1];
1588 struct ip_tunnel_info info;
1589 struct net_device *dev;
1590 LIST_HEAD(list_kill);
1591 int err;
1592
1593 memset(tb, 0, sizeof(tb));
1594 dev = rtnl_create_link(net, name, name_assign_type,
1595 &geneve_link_ops, tb);
1596 if (IS_ERR(dev))
1597 return dev;
1598
1599 init_tnl_info(&info, dst_port);
1600 err = geneve_configure(net, dev, NULL, &info, true, true);
1601 if (err) {
1602 free_netdev(dev);
1603 return ERR_PTR(err);
1604 }
1605
1606 /* openvswitch users expect packet sizes to be unrestricted,
1607 * so set the largest MTU we can.
1608 */
1609 err = geneve_change_mtu(dev, IP_MAX_MTU);
1610 if (err)
1611 goto err;
1612
1613 err = rtnl_configure_link(dev, NULL);
1614 if (err < 0)
1615 goto err;
1616
1617 return dev;
1618err:
1619 geneve_dellink(dev, &list_kill);
1620 unregister_netdevice_many(&list_kill);
1621 return ERR_PTR(err);
1622}
1623EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1624
1625static int geneve_netdevice_event(struct notifier_block *unused,
1626 unsigned long event, void *ptr)
1627{
1628 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1629
1630 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1631 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1632 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1633 } else if (event == NETDEV_UNREGISTER) {
1634 geneve_offload_rx_ports(dev, false);
1635 } else if (event == NETDEV_REGISTER) {
1636 geneve_offload_rx_ports(dev, true);
1637 }
1638
1639 return NOTIFY_DONE;
1640}
1641
1642static struct notifier_block geneve_notifier_block __read_mostly = {
1643 .notifier_call = geneve_netdevice_event,
1644};
1645
1646static __net_init int geneve_init_net(struct net *net)
1647{
1648 struct geneve_net *gn = net_generic(net, geneve_net_id);
1649
1650 INIT_LIST_HEAD(&gn->geneve_list);
1651 INIT_LIST_HEAD(&gn->sock_list);
1652 return 0;
1653}
1654
1655static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1656{
1657 struct geneve_net *gn = net_generic(net, geneve_net_id);
1658 struct geneve_dev *geneve, *next;
1659 struct net_device *dev, *aux;
1660
1661 /* gather any geneve devices that were moved into this ns */
1662 for_each_netdev_safe(net, dev, aux)
1663 if (dev->rtnl_link_ops == &geneve_link_ops)
1664 unregister_netdevice_queue(dev, head);
1665
1666 /* now gather any other geneve devices that were created in this ns */
1667 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1668 /* If geneve->dev is in the same netns, it was already added
1669 * to the list by the previous loop.
1670 */
1671 if (!net_eq(dev_net(geneve->dev), net))
1672 unregister_netdevice_queue(geneve->dev, head);
1673 }
1674
1675 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1676}
1677
1678static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1679{
1680 struct net *net;
1681 LIST_HEAD(list);
1682
1683 rtnl_lock();
1684 list_for_each_entry(net, net_list, exit_list)
1685 geneve_destroy_tunnels(net, &list);
1686
1687 /* unregister the devices gathered above */
1688 unregister_netdevice_many(&list);
1689 rtnl_unlock();
1690}
1691
1692static struct pernet_operations geneve_net_ops = {
1693 .init = geneve_init_net,
1694 .exit_batch = geneve_exit_batch_net,
1695 .id = &geneve_net_id,
1696 .size = sizeof(struct geneve_net),
1697};
1698
1699static int __init geneve_init_module(void)
1700{
1701 int rc;
1702
1703 rc = register_pernet_subsys(&geneve_net_ops);
1704 if (rc)
1705 goto out1;
1706
1707 rc = register_netdevice_notifier(&geneve_notifier_block);
1708 if (rc)
1709 goto out2;
1710
1711 rc = rtnl_link_register(&geneve_link_ops);
1712 if (rc)
1713 goto out3;
1714
1715 return 0;
1716out3:
1717 unregister_netdevice_notifier(&geneve_notifier_block);
1718out2:
1719 unregister_pernet_subsys(&geneve_net_ops);
1720out1:
1721 return rc;
1722}
1723late_initcall(geneve_init_module);
1724
1725static void __exit geneve_cleanup_module(void)
1726{
1727 rtnl_link_unregister(&geneve_link_ops);
1728 unregister_netdevice_notifier(&geneve_notifier_block);
1729 unregister_pernet_subsys(&geneve_net_ops);
1730}
1731module_exit(geneve_cleanup_module);
1732
1733MODULE_LICENSE("GPL");
1734MODULE_VERSION(GENEVE_NETDEV_VER);
1735MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1736MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1737MODULE_ALIAS_RTNL_LINK("geneve");