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/netdev_lock.h>
22 #include <net/protocol.h>
23
24 #define GENEVE_NETDEV_VER "0.6"
25
26 #define GENEVE_N_VID (1u << 24)
27 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
28
29 #define VNI_HASH_BITS 10
30 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
31
32 static bool log_ecn_error = true;
33 module_param(log_ecn_error, bool, 0644);
34 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
35
36 #define GENEVE_VER 0
37 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
38 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
39 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
40
41 /* per-network namespace private data for this module */
42 struct geneve_net {
43 struct list_head geneve_list;
44 /* sock_list is protected by rtnl lock */
45 struct list_head sock_list;
46 };
47
48 static unsigned int geneve_net_id;
49
50 struct geneve_dev_node {
51 struct hlist_node hlist;
52 struct geneve_dev *geneve;
53 };
54
55 struct geneve_config {
56 struct ip_tunnel_info info;
57 bool collect_md;
58 bool use_udp6_rx_checksums;
59 bool ttl_inherit;
60 enum ifla_geneve_df df;
61 bool inner_proto_inherit;
62 u16 port_min;
63 u16 port_max;
64 };
65
66 /* Pseudo network device */
67 struct geneve_dev {
68 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
69 #if IS_ENABLED(CONFIG_IPV6)
70 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
71 #endif
72 struct net *net; /* netns for packet i/o */
73 struct net_device *dev; /* netdev for geneve tunnel */
74 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
75 #if IS_ENABLED(CONFIG_IPV6)
76 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
77 #endif
78 struct list_head next; /* geneve's per namespace list */
79 struct gro_cells gro_cells;
80 struct geneve_config cfg;
81 };
82
83 struct geneve_sock {
84 bool collect_md;
85 struct list_head list;
86 struct socket *sock;
87 struct rcu_head rcu;
88 int refcnt;
89 struct hlist_head vni_list[VNI_HASH_SIZE];
90 };
91
geneve_net_vni_hash(u8 vni[3])92 static inline __u32 geneve_net_vni_hash(u8 vni[3])
93 {
94 __u32 vnid;
95
96 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
97 return hash_32(vnid, VNI_HASH_BITS);
98 }
99
vni_to_tunnel_id(const __u8 * vni)100 static __be64 vni_to_tunnel_id(const __u8 *vni)
101 {
102 #ifdef __BIG_ENDIAN
103 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
104 #else
105 return (__force __be64)(((__force u64)vni[0] << 40) |
106 ((__force u64)vni[1] << 48) |
107 ((__force u64)vni[2] << 56));
108 #endif
109 }
110
111 /* Convert 64 bit tunnel ID to 24 bit VNI. */
tunnel_id_to_vni(__be64 tun_id,__u8 * vni)112 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
113 {
114 #ifdef __BIG_ENDIAN
115 vni[0] = (__force __u8)(tun_id >> 16);
116 vni[1] = (__force __u8)(tun_id >> 8);
117 vni[2] = (__force __u8)tun_id;
118 #else
119 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
120 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
121 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
122 #endif
123 }
124
eq_tun_id_and_vni(u8 * tun_id,u8 * vni)125 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
126 {
127 return !memcmp(vni, &tun_id[5], 3);
128 }
129
geneve_get_sk_family(struct geneve_sock * gs)130 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
131 {
132 return gs->sock->sk->sk_family;
133 }
134
geneve_lookup(struct geneve_sock * gs,__be32 addr,u8 vni[])135 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
136 __be32 addr, u8 vni[])
137 {
138 struct hlist_head *vni_list_head;
139 struct geneve_dev_node *node;
140 __u32 hash;
141
142 /* Find the device for this VNI */
143 hash = geneve_net_vni_hash(vni);
144 vni_list_head = &gs->vni_list[hash];
145 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
146 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
147 addr == node->geneve->cfg.info.key.u.ipv4.dst)
148 return node->geneve;
149 }
150 return NULL;
151 }
152
153 #if IS_ENABLED(CONFIG_IPV6)
geneve6_lookup(struct geneve_sock * gs,struct in6_addr addr6,u8 vni[])154 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
155 struct in6_addr addr6, u8 vni[])
156 {
157 struct hlist_head *vni_list_head;
158 struct geneve_dev_node *node;
159 __u32 hash;
160
161 /* Find the device for this VNI */
162 hash = geneve_net_vni_hash(vni);
163 vni_list_head = &gs->vni_list[hash];
164 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
165 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
166 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
167 return node->geneve;
168 }
169 return NULL;
170 }
171 #endif
172
geneve_hdr(const struct sk_buff * skb)173 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
174 {
175 return (struct genevehdr *)(udp_hdr(skb) + 1);
176 }
177
geneve_lookup_skb(struct geneve_sock * gs,struct sk_buff * skb)178 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
179 struct sk_buff *skb)
180 {
181 static u8 zero_vni[3];
182 u8 *vni;
183
184 if (geneve_get_sk_family(gs) == AF_INET) {
185 struct iphdr *iph;
186 __be32 addr;
187
188 iph = ip_hdr(skb); /* outer IP header... */
189
190 if (gs->collect_md) {
191 vni = zero_vni;
192 addr = 0;
193 } else {
194 vni = geneve_hdr(skb)->vni;
195 addr = iph->saddr;
196 }
197
198 return geneve_lookup(gs, addr, vni);
199 #if IS_ENABLED(CONFIG_IPV6)
200 } else if (geneve_get_sk_family(gs) == AF_INET6) {
201 static struct in6_addr zero_addr6;
202 struct ipv6hdr *ip6h;
203 struct in6_addr addr6;
204
205 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
206
207 if (gs->collect_md) {
208 vni = zero_vni;
209 addr6 = zero_addr6;
210 } else {
211 vni = geneve_hdr(skb)->vni;
212 addr6 = ip6h->saddr;
213 }
214
215 return geneve6_lookup(gs, addr6, vni);
216 #endif
217 }
218 return NULL;
219 }
220
221 /* geneve receive/decap routine */
geneve_rx(struct geneve_dev * geneve,struct geneve_sock * gs,struct sk_buff * skb)222 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
223 struct sk_buff *skb)
224 {
225 struct genevehdr *gnvh = geneve_hdr(skb);
226 struct metadata_dst *tun_dst = NULL;
227 unsigned int len;
228 int nh, err = 0;
229 void *oiph;
230
231 if (ip_tunnel_collect_metadata() || gs->collect_md) {
232 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
233
234 __set_bit(IP_TUNNEL_KEY_BIT, flags);
235 __assign_bit(IP_TUNNEL_OAM_BIT, flags, gnvh->oam);
236 __assign_bit(IP_TUNNEL_CRIT_OPT_BIT, flags, gnvh->critical);
237
238 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
239 vni_to_tunnel_id(gnvh->vni),
240 gnvh->opt_len * 4);
241 if (!tun_dst) {
242 dev_dstats_rx_dropped(geneve->dev);
243 goto drop;
244 }
245 /* Update tunnel dst according to Geneve options. */
246 ip_tunnel_flags_zero(flags);
247 __set_bit(IP_TUNNEL_GENEVE_OPT_BIT, flags);
248 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
249 gnvh->options, gnvh->opt_len * 4,
250 flags);
251 } else {
252 /* Drop packets w/ critical options,
253 * since we don't support any...
254 */
255 if (gnvh->critical) {
256 DEV_STATS_INC(geneve->dev, rx_frame_errors);
257 DEV_STATS_INC(geneve->dev, rx_errors);
258 goto drop;
259 }
260 }
261
262 if (tun_dst)
263 skb_dst_set(skb, &tun_dst->dst);
264
265 if (gnvh->proto_type == htons(ETH_P_TEB)) {
266 skb_reset_mac_header(skb);
267 skb->protocol = eth_type_trans(skb, geneve->dev);
268 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
269
270 /* Ignore packet loops (and multicast echo) */
271 if (ether_addr_equal(eth_hdr(skb)->h_source,
272 geneve->dev->dev_addr)) {
273 DEV_STATS_INC(geneve->dev, rx_errors);
274 goto drop;
275 }
276 } else {
277 skb_reset_mac_header(skb);
278 skb->dev = geneve->dev;
279 skb->pkt_type = PACKET_HOST;
280 }
281
282 /* Save offset of outer header relative to skb->head,
283 * because we are going to reset the network header to the inner header
284 * and might change skb->head.
285 */
286 nh = skb_network_header(skb) - skb->head;
287
288 skb_reset_network_header(skb);
289
290 if (!pskb_inet_may_pull(skb)) {
291 DEV_STATS_INC(geneve->dev, rx_length_errors);
292 DEV_STATS_INC(geneve->dev, rx_errors);
293 goto drop;
294 }
295
296 /* Get the outer header. */
297 oiph = skb->head + nh;
298
299 if (geneve_get_sk_family(gs) == AF_INET)
300 err = IP_ECN_decapsulate(oiph, skb);
301 #if IS_ENABLED(CONFIG_IPV6)
302 else
303 err = IP6_ECN_decapsulate(oiph, skb);
304 #endif
305
306 if (unlikely(err)) {
307 if (log_ecn_error) {
308 if (geneve_get_sk_family(gs) == AF_INET)
309 net_info_ratelimited("non-ECT from %pI4 "
310 "with TOS=%#x\n",
311 &((struct iphdr *)oiph)->saddr,
312 ((struct iphdr *)oiph)->tos);
313 #if IS_ENABLED(CONFIG_IPV6)
314 else
315 net_info_ratelimited("non-ECT from %pI6\n",
316 &((struct ipv6hdr *)oiph)->saddr);
317 #endif
318 }
319 if (err > 1) {
320 DEV_STATS_INC(geneve->dev, rx_frame_errors);
321 DEV_STATS_INC(geneve->dev, rx_errors);
322 goto drop;
323 }
324 }
325
326 len = skb->len;
327 err = gro_cells_receive(&geneve->gro_cells, skb);
328 if (likely(err == NET_RX_SUCCESS))
329 dev_dstats_rx_add(geneve->dev, len);
330
331 return;
332 drop:
333 /* Consume bad packet */
334 kfree_skb(skb);
335 }
336
337 /* Setup stats when device is created */
geneve_init(struct net_device * dev)338 static int geneve_init(struct net_device *dev)
339 {
340 struct geneve_dev *geneve = netdev_priv(dev);
341 int err;
342
343 err = gro_cells_init(&geneve->gro_cells, dev);
344 if (err)
345 return err;
346
347 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
348 if (err) {
349 gro_cells_destroy(&geneve->gro_cells);
350 return err;
351 }
352 netdev_lockdep_set_classes(dev);
353 return 0;
354 }
355
geneve_uninit(struct net_device * dev)356 static void geneve_uninit(struct net_device *dev)
357 {
358 struct geneve_dev *geneve = netdev_priv(dev);
359
360 dst_cache_destroy(&geneve->cfg.info.dst_cache);
361 gro_cells_destroy(&geneve->gro_cells);
362 }
363
364 /* Callback from net/ipv4/udp.c to receive packets */
geneve_udp_encap_recv(struct sock * sk,struct sk_buff * skb)365 static 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_dstats_rx_dropped(geneve->dev);
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_dstats_rx_dropped(geneve->dev);
402 goto drop;
403 }
404
405 geneve_rx(geneve, gs, skb);
406 return 0;
407
408 drop:
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 */
geneve_udp_encap_err_lookup(struct sock * sk,struct sk_buff * skb)415 static 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
geneve_create_sock(struct net * net,bool ipv6,__be16 port,bool ipv6_rx_csum)467 static 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
geneve_hlen(struct genevehdr * gh)496 static int geneve_hlen(struct genevehdr *gh)
497 {
498 return sizeof(*gh) + gh->opt_len * 4;
499 }
500
geneve_gro_receive(struct sock * sk,struct list_head * head,struct sk_buff * skb)501 static 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_may_pull(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
555 out:
556 skb_gro_flush_final(skb, pp, flush);
557
558 return pp;
559 }
560
geneve_gro_complete(struct sock * sk,struct sk_buff * skb,int nhoff)561 static 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 */
geneve_socket_create(struct net * net,__be16 port,bool ipv6,bool ipv6_rx_csum)588 static 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
__geneve_sock_release(struct geneve_sock * gs)629 static 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
geneve_sock_release(struct geneve_dev * geneve)640 static 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
geneve_find_sock(struct geneve_net * gn,sa_family_t family,__be16 dst_port)658 static 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
geneve_sock_add(struct geneve_dev * geneve,bool ipv6)673 static 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
693 out:
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
geneve_open(struct net_device * dev)713 static 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
geneve_stop(struct net_device * dev)737 static 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
geneve_build_header(struct genevehdr * geneveh,const struct ip_tunnel_info * info,__be16 inner_proto)749 static 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 = test_bit(IP_TUNNEL_OAM_BIT, info->key.tun_flags);
756 geneveh->critical = test_bit(IP_TUNNEL_CRIT_OPT_BIT,
757 info->key.tun_flags);
758 geneveh->rsvd1 = 0;
759 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
760 geneveh->proto_type = inner_proto;
761 geneveh->rsvd2 = 0;
762
763 if (test_bit(IP_TUNNEL_GENEVE_OPT_BIT, info->key.tun_flags))
764 ip_tunnel_info_opts_get(geneveh->options, info);
765 }
766
geneve_build_skb(struct dst_entry * dst,struct sk_buff * skb,const struct ip_tunnel_info * info,bool xnet,int ip_hdr_len,bool inner_proto_inherit)767 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
768 const struct ip_tunnel_info *info,
769 bool xnet, int ip_hdr_len,
770 bool inner_proto_inherit)
771 {
772 bool udp_sum = test_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
773 struct genevehdr *gnvh;
774 __be16 inner_proto;
775 int min_headroom;
776 int err;
777
778 skb_reset_mac_header(skb);
779 skb_scrub_packet(skb, xnet);
780
781 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
782 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
783 err = skb_cow_head(skb, min_headroom);
784 if (unlikely(err))
785 goto free_dst;
786
787 err = udp_tunnel_handle_offloads(skb, udp_sum);
788 if (err)
789 goto free_dst;
790
791 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
792 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
793 geneve_build_header(gnvh, info, inner_proto);
794 skb_set_inner_protocol(skb, inner_proto);
795 return 0;
796
797 free_dst:
798 dst_release(dst);
799 return err;
800 }
801
geneve_get_dsfield(struct sk_buff * skb,struct net_device * dev,const struct ip_tunnel_info * info,bool * use_cache)802 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
803 const struct ip_tunnel_info *info,
804 bool *use_cache)
805 {
806 struct geneve_dev *geneve = netdev_priv(dev);
807 u8 dsfield;
808
809 dsfield = info->key.tos;
810 if (dsfield == 1 && !geneve->cfg.collect_md) {
811 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
812 *use_cache = false;
813 }
814
815 return dsfield;
816 }
817
geneve_xmit_skb(struct sk_buff * skb,struct net_device * dev,struct geneve_dev * geneve,const struct ip_tunnel_info * info)818 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
819 struct geneve_dev *geneve,
820 const struct ip_tunnel_info *info)
821 {
822 bool inner_proto_inherit = geneve->cfg.inner_proto_inherit;
823 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
824 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
825 const struct ip_tunnel_key *key = &info->key;
826 struct rtable *rt;
827 bool use_cache;
828 __u8 tos, ttl;
829 __be16 df = 0;
830 __be32 saddr;
831 __be16 sport;
832 int err;
833
834 if (skb_vlan_inet_prepare(skb, inner_proto_inherit))
835 return -EINVAL;
836
837 if (!gs4)
838 return -EIO;
839
840 use_cache = ip_tunnel_dst_cache_usable(skb, info);
841 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
842 sport = udp_flow_src_port(geneve->net, skb,
843 geneve->cfg.port_min,
844 geneve->cfg.port_max, true);
845
846 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
847 &info->key,
848 sport, geneve->cfg.info.key.tp_dst, tos,
849 use_cache ?
850 (struct dst_cache *)&info->dst_cache : NULL);
851 if (IS_ERR(rt))
852 return PTR_ERR(rt);
853
854 err = skb_tunnel_check_pmtu(skb, &rt->dst,
855 GENEVE_IPV4_HLEN + info->options_len,
856 netif_is_any_bridge_port(dev));
857 if (err < 0) {
858 dst_release(&rt->dst);
859 return err;
860 } else if (err) {
861 struct ip_tunnel_info *info;
862
863 info = skb_tunnel_info(skb);
864 if (info) {
865 struct ip_tunnel_info *unclone;
866
867 unclone = skb_tunnel_info_unclone(skb);
868 if (unlikely(!unclone)) {
869 dst_release(&rt->dst);
870 return -ENOMEM;
871 }
872
873 unclone->key.u.ipv4.dst = saddr;
874 unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
875 }
876
877 if (!pskb_may_pull(skb, ETH_HLEN)) {
878 dst_release(&rt->dst);
879 return -EINVAL;
880 }
881
882 skb->protocol = eth_type_trans(skb, geneve->dev);
883 __netif_rx(skb);
884 dst_release(&rt->dst);
885 return -EMSGSIZE;
886 }
887
888 tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
889 if (geneve->cfg.collect_md) {
890 ttl = key->ttl;
891
892 df = test_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, key->tun_flags) ?
893 htons(IP_DF) : 0;
894 } else {
895 if (geneve->cfg.ttl_inherit)
896 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
897 else
898 ttl = key->ttl;
899 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
900
901 if (geneve->cfg.df == GENEVE_DF_SET) {
902 df = htons(IP_DF);
903 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
904 struct ethhdr *eth = skb_eth_hdr(skb);
905
906 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
907 df = htons(IP_DF);
908 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
909 struct iphdr *iph = ip_hdr(skb);
910
911 if (iph->frag_off & htons(IP_DF))
912 df = htons(IP_DF);
913 }
914 }
915 }
916
917 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
918 inner_proto_inherit);
919 if (unlikely(err))
920 return err;
921
922 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
923 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
924 !net_eq(geneve->net, dev_net(geneve->dev)),
925 !test_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags),
926 0);
927 return 0;
928 }
929
930 #if IS_ENABLED(CONFIG_IPV6)
geneve6_xmit_skb(struct sk_buff * skb,struct net_device * dev,struct geneve_dev * geneve,const struct ip_tunnel_info * info)931 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
932 struct geneve_dev *geneve,
933 const struct ip_tunnel_info *info)
934 {
935 bool inner_proto_inherit = geneve->cfg.inner_proto_inherit;
936 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
937 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
938 const struct ip_tunnel_key *key = &info->key;
939 struct dst_entry *dst = NULL;
940 struct in6_addr saddr;
941 bool use_cache;
942 __u8 prio, ttl;
943 __be16 sport;
944 int err;
945
946 if (skb_vlan_inet_prepare(skb, inner_proto_inherit))
947 return -EINVAL;
948
949 if (!gs6)
950 return -EIO;
951
952 use_cache = ip_tunnel_dst_cache_usable(skb, info);
953 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
954 sport = udp_flow_src_port(geneve->net, skb,
955 geneve->cfg.port_min,
956 geneve->cfg.port_max, true);
957
958 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
959 &saddr, key, sport,
960 geneve->cfg.info.key.tp_dst, prio,
961 use_cache ?
962 (struct dst_cache *)&info->dst_cache : NULL);
963 if (IS_ERR(dst))
964 return PTR_ERR(dst);
965
966 err = skb_tunnel_check_pmtu(skb, dst,
967 GENEVE_IPV6_HLEN + info->options_len,
968 netif_is_any_bridge_port(dev));
969 if (err < 0) {
970 dst_release(dst);
971 return err;
972 } else if (err) {
973 struct ip_tunnel_info *info = skb_tunnel_info(skb);
974
975 if (info) {
976 struct ip_tunnel_info *unclone;
977
978 unclone = skb_tunnel_info_unclone(skb);
979 if (unlikely(!unclone)) {
980 dst_release(dst);
981 return -ENOMEM;
982 }
983
984 unclone->key.u.ipv6.dst = saddr;
985 unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
986 }
987
988 if (!pskb_may_pull(skb, ETH_HLEN)) {
989 dst_release(dst);
990 return -EINVAL;
991 }
992
993 skb->protocol = eth_type_trans(skb, geneve->dev);
994 __netif_rx(skb);
995 dst_release(dst);
996 return -EMSGSIZE;
997 }
998
999 prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
1000 if (geneve->cfg.collect_md) {
1001 ttl = key->ttl;
1002 } else {
1003 if (geneve->cfg.ttl_inherit)
1004 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1005 else
1006 ttl = key->ttl;
1007 ttl = ttl ? : ip6_dst_hoplimit(dst);
1008 }
1009 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1010 inner_proto_inherit);
1011 if (unlikely(err))
1012 return err;
1013
1014 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1015 &saddr, &key->u.ipv6.dst, prio, ttl,
1016 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1017 !test_bit(IP_TUNNEL_CSUM_BIT,
1018 info->key.tun_flags),
1019 0);
1020 return 0;
1021 }
1022 #endif
1023
geneve_xmit(struct sk_buff * skb,struct net_device * dev)1024 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1025 {
1026 struct geneve_dev *geneve = netdev_priv(dev);
1027 struct ip_tunnel_info *info = NULL;
1028 int err;
1029
1030 if (geneve->cfg.collect_md) {
1031 info = skb_tunnel_info(skb);
1032 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1033 netdev_dbg(dev, "no tunnel metadata\n");
1034 dev_kfree_skb(skb);
1035 dev_dstats_tx_dropped(dev);
1036 return NETDEV_TX_OK;
1037 }
1038 } else {
1039 info = &geneve->cfg.info;
1040 }
1041
1042 rcu_read_lock();
1043 #if IS_ENABLED(CONFIG_IPV6)
1044 if (info->mode & IP_TUNNEL_INFO_IPV6)
1045 err = geneve6_xmit_skb(skb, dev, geneve, info);
1046 else
1047 #endif
1048 err = geneve_xmit_skb(skb, dev, geneve, info);
1049 rcu_read_unlock();
1050
1051 if (likely(!err))
1052 return NETDEV_TX_OK;
1053
1054 if (err != -EMSGSIZE)
1055 dev_kfree_skb(skb);
1056
1057 if (err == -ELOOP)
1058 DEV_STATS_INC(dev, collisions);
1059 else if (err == -ENETUNREACH)
1060 DEV_STATS_INC(dev, tx_carrier_errors);
1061
1062 DEV_STATS_INC(dev, tx_errors);
1063 return NETDEV_TX_OK;
1064 }
1065
geneve_change_mtu(struct net_device * dev,int new_mtu)1066 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1067 {
1068 if (new_mtu > dev->max_mtu)
1069 new_mtu = dev->max_mtu;
1070 else if (new_mtu < dev->min_mtu)
1071 new_mtu = dev->min_mtu;
1072
1073 WRITE_ONCE(dev->mtu, new_mtu);
1074 return 0;
1075 }
1076
geneve_fill_metadata_dst(struct net_device * dev,struct sk_buff * skb)1077 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1078 {
1079 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1080 struct geneve_dev *geneve = netdev_priv(dev);
1081 __be16 sport;
1082
1083 if (ip_tunnel_info_af(info) == AF_INET) {
1084 struct rtable *rt;
1085 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1086 bool use_cache;
1087 __be32 saddr;
1088 u8 tos;
1089
1090 if (!gs4)
1091 return -EIO;
1092
1093 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1094 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1095 sport = udp_flow_src_port(geneve->net, skb,
1096 geneve->cfg.port_min,
1097 geneve->cfg.port_max, true);
1098
1099 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1100 &info->key,
1101 sport, geneve->cfg.info.key.tp_dst,
1102 tos,
1103 use_cache ? &info->dst_cache : NULL);
1104 if (IS_ERR(rt))
1105 return PTR_ERR(rt);
1106
1107 ip_rt_put(rt);
1108 info->key.u.ipv4.src = saddr;
1109 #if IS_ENABLED(CONFIG_IPV6)
1110 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1111 struct dst_entry *dst;
1112 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1113 struct in6_addr saddr;
1114 bool use_cache;
1115 u8 prio;
1116
1117 if (!gs6)
1118 return -EIO;
1119
1120 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1121 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1122 sport = udp_flow_src_port(geneve->net, skb,
1123 geneve->cfg.port_min,
1124 geneve->cfg.port_max, true);
1125
1126 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1127 &saddr, &info->key, sport,
1128 geneve->cfg.info.key.tp_dst, prio,
1129 use_cache ? &info->dst_cache : NULL);
1130 if (IS_ERR(dst))
1131 return PTR_ERR(dst);
1132
1133 dst_release(dst);
1134 info->key.u.ipv6.src = saddr;
1135 #endif
1136 } else {
1137 return -EINVAL;
1138 }
1139
1140 info->key.tp_src = sport;
1141 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1142 return 0;
1143 }
1144
1145 static const struct net_device_ops geneve_netdev_ops = {
1146 .ndo_init = geneve_init,
1147 .ndo_uninit = geneve_uninit,
1148 .ndo_open = geneve_open,
1149 .ndo_stop = geneve_stop,
1150 .ndo_start_xmit = geneve_xmit,
1151 .ndo_change_mtu = geneve_change_mtu,
1152 .ndo_validate_addr = eth_validate_addr,
1153 .ndo_set_mac_address = eth_mac_addr,
1154 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1155 };
1156
geneve_get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * drvinfo)1157 static void geneve_get_drvinfo(struct net_device *dev,
1158 struct ethtool_drvinfo *drvinfo)
1159 {
1160 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1161 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1162 }
1163
1164 static const struct ethtool_ops geneve_ethtool_ops = {
1165 .get_drvinfo = geneve_get_drvinfo,
1166 .get_link = ethtool_op_get_link,
1167 };
1168
1169 /* Info for udev, that this is a virtual tunnel endpoint */
1170 static const struct device_type geneve_type = {
1171 .name = "geneve",
1172 };
1173
1174 /* Calls the ndo_udp_tunnel_add of the caller in order to
1175 * supply the listening GENEVE udp ports. Callers are expected
1176 * to implement the ndo_udp_tunnel_add.
1177 */
geneve_offload_rx_ports(struct net_device * dev,bool push)1178 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1179 {
1180 struct net *net = dev_net(dev);
1181 struct geneve_net *gn = net_generic(net, geneve_net_id);
1182 struct geneve_sock *gs;
1183
1184 ASSERT_RTNL();
1185
1186 list_for_each_entry(gs, &gn->sock_list, list) {
1187 if (push) {
1188 udp_tunnel_push_rx_port(dev, gs->sock,
1189 UDP_TUNNEL_TYPE_GENEVE);
1190 } else {
1191 udp_tunnel_drop_rx_port(dev, gs->sock,
1192 UDP_TUNNEL_TYPE_GENEVE);
1193 }
1194 }
1195 }
1196
1197 /* Initialize the device structure. */
geneve_setup(struct net_device * dev)1198 static void geneve_setup(struct net_device *dev)
1199 {
1200 ether_setup(dev);
1201
1202 dev->netdev_ops = &geneve_netdev_ops;
1203 dev->ethtool_ops = &geneve_ethtool_ops;
1204 dev->needs_free_netdev = true;
1205
1206 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1207
1208 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1209 dev->features |= NETIF_F_RXCSUM;
1210 dev->features |= NETIF_F_GSO_SOFTWARE;
1211
1212 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1213 dev->hw_features |= NETIF_F_RXCSUM;
1214 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1215
1216 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS;
1217 /* MTU range: 68 - (something less than 65535) */
1218 dev->min_mtu = ETH_MIN_MTU;
1219 /* The max_mtu calculation does not take account of GENEVE
1220 * options, to avoid excluding potentially valid
1221 * configurations. This will be further reduced by IPvX hdr size.
1222 */
1223 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1224
1225 netif_keep_dst(dev);
1226 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1227 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1228 dev->lltx = true;
1229 eth_hw_addr_random(dev);
1230 }
1231
1232 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1233 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1234 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1235 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1236 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1237 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1238 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1239 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1240 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1241 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1242 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1243 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1244 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1245 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1246 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1247 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG },
1248 [IFLA_GENEVE_PORT_RANGE] = NLA_POLICY_EXACT_LEN(sizeof(struct ifla_geneve_port_range)),
1249 };
1250
geneve_validate(struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack)1251 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1252 struct netlink_ext_ack *extack)
1253 {
1254 if (tb[IFLA_ADDRESS]) {
1255 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1256 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1257 "Provided link layer address is not Ethernet");
1258 return -EINVAL;
1259 }
1260
1261 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1262 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1263 "Provided Ethernet address is not unicast");
1264 return -EADDRNOTAVAIL;
1265 }
1266 }
1267
1268 if (!data) {
1269 NL_SET_ERR_MSG(extack,
1270 "Not enough attributes provided to perform the operation");
1271 return -EINVAL;
1272 }
1273
1274 if (data[IFLA_GENEVE_ID]) {
1275 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1276
1277 if (vni >= GENEVE_N_VID) {
1278 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1279 "Geneve ID must be lower than 16777216");
1280 return -ERANGE;
1281 }
1282 }
1283
1284 if (data[IFLA_GENEVE_DF]) {
1285 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1286
1287 if (df < 0 || df > GENEVE_DF_MAX) {
1288 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1289 "Invalid DF attribute");
1290 return -EINVAL;
1291 }
1292 }
1293
1294 if (data[IFLA_GENEVE_PORT_RANGE]) {
1295 const struct ifla_geneve_port_range *p;
1296
1297 p = nla_data(data[IFLA_GENEVE_PORT_RANGE]);
1298 if (ntohs(p->high) < ntohs(p->low)) {
1299 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_PORT_RANGE],
1300 "Invalid source port range");
1301 return -EINVAL;
1302 }
1303 }
1304
1305 return 0;
1306 }
1307
geneve_find_dev(struct geneve_net * gn,const struct ip_tunnel_info * info,bool * tun_on_same_port,bool * tun_collect_md)1308 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1309 const struct ip_tunnel_info *info,
1310 bool *tun_on_same_port,
1311 bool *tun_collect_md)
1312 {
1313 struct geneve_dev *geneve, *t = NULL;
1314
1315 *tun_on_same_port = false;
1316 *tun_collect_md = false;
1317 list_for_each_entry(geneve, &gn->geneve_list, next) {
1318 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1319 *tun_collect_md = geneve->cfg.collect_md;
1320 *tun_on_same_port = true;
1321 }
1322 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1323 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1324 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1325 t = geneve;
1326 }
1327 return t;
1328 }
1329
is_tnl_info_zero(const struct ip_tunnel_info * info)1330 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1331 {
1332 return !(info->key.tun_id || info->key.tos ||
1333 !ip_tunnel_flags_empty(info->key.tun_flags) ||
1334 info->key.ttl || info->key.label || info->key.tp_src ||
1335 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1336 }
1337
geneve_dst_addr_equal(struct ip_tunnel_info * a,struct ip_tunnel_info * b)1338 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1339 struct ip_tunnel_info *b)
1340 {
1341 if (ip_tunnel_info_af(a) == AF_INET)
1342 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1343 else
1344 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1345 }
1346
geneve_configure(struct net * net,struct net_device * dev,struct netlink_ext_ack * extack,const struct geneve_config * cfg)1347 static int geneve_configure(struct net *net, struct net_device *dev,
1348 struct netlink_ext_ack *extack,
1349 const struct geneve_config *cfg)
1350 {
1351 struct geneve_net *gn = net_generic(net, geneve_net_id);
1352 struct geneve_dev *t, *geneve = netdev_priv(dev);
1353 const struct ip_tunnel_info *info = &cfg->info;
1354 bool tun_collect_md, tun_on_same_port;
1355 int err, encap_len;
1356
1357 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1358 NL_SET_ERR_MSG(extack,
1359 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1360 return -EINVAL;
1361 }
1362
1363 geneve->net = net;
1364 geneve->dev = dev;
1365
1366 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1367 if (t)
1368 return -EBUSY;
1369
1370 /* make enough headroom for basic scenario */
1371 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1372 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1373 encap_len += sizeof(struct iphdr);
1374 dev->max_mtu -= sizeof(struct iphdr);
1375 } else {
1376 encap_len += sizeof(struct ipv6hdr);
1377 dev->max_mtu -= sizeof(struct ipv6hdr);
1378 }
1379 dev->needed_headroom = encap_len + ETH_HLEN;
1380
1381 if (cfg->collect_md) {
1382 if (tun_on_same_port) {
1383 NL_SET_ERR_MSG(extack,
1384 "There can be only one externally controlled device on a destination port");
1385 return -EPERM;
1386 }
1387 } else {
1388 if (tun_collect_md) {
1389 NL_SET_ERR_MSG(extack,
1390 "There already exists an externally controlled device on this destination port");
1391 return -EPERM;
1392 }
1393 }
1394
1395 dst_cache_reset(&geneve->cfg.info.dst_cache);
1396 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1397
1398 if (geneve->cfg.inner_proto_inherit) {
1399 dev->header_ops = NULL;
1400 dev->type = ARPHRD_NONE;
1401 dev->hard_header_len = 0;
1402 dev->addr_len = 0;
1403 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1404 }
1405
1406 err = register_netdevice(dev);
1407 if (err)
1408 return err;
1409
1410 list_add(&geneve->next, &gn->geneve_list);
1411 return 0;
1412 }
1413
init_tnl_info(struct ip_tunnel_info * info,__u16 dst_port)1414 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1415 {
1416 memset(info, 0, sizeof(*info));
1417 info->key.tp_dst = htons(dst_port);
1418 }
1419
geneve_nl2info(struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack,struct geneve_config * cfg,bool changelink)1420 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1421 struct netlink_ext_ack *extack,
1422 struct geneve_config *cfg, bool changelink)
1423 {
1424 struct ip_tunnel_info *info = &cfg->info;
1425 int attrtype;
1426
1427 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1428 NL_SET_ERR_MSG(extack,
1429 "Cannot specify both IPv4 and IPv6 Remote addresses");
1430 return -EINVAL;
1431 }
1432
1433 if (data[IFLA_GENEVE_REMOTE]) {
1434 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1435 attrtype = IFLA_GENEVE_REMOTE;
1436 goto change_notsup;
1437 }
1438
1439 info->key.u.ipv4.dst =
1440 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1441
1442 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1443 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1444 "Remote IPv4 address cannot be Multicast");
1445 return -EINVAL;
1446 }
1447 }
1448
1449 if (data[IFLA_GENEVE_REMOTE6]) {
1450 #if IS_ENABLED(CONFIG_IPV6)
1451 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1452 attrtype = IFLA_GENEVE_REMOTE6;
1453 goto change_notsup;
1454 }
1455
1456 info->mode = IP_TUNNEL_INFO_IPV6;
1457 info->key.u.ipv6.dst =
1458 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1459
1460 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1461 IPV6_ADDR_LINKLOCAL) {
1462 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1463 "Remote IPv6 address cannot be link-local");
1464 return -EINVAL;
1465 }
1466 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1467 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1468 "Remote IPv6 address cannot be Multicast");
1469 return -EINVAL;
1470 }
1471 __set_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
1472 cfg->use_udp6_rx_checksums = true;
1473 #else
1474 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1475 "IPv6 support not enabled in the kernel");
1476 return -EPFNOSUPPORT;
1477 #endif
1478 }
1479
1480 if (data[IFLA_GENEVE_ID]) {
1481 __u32 vni;
1482 __u8 tvni[3];
1483 __be64 tunid;
1484
1485 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1486 tvni[0] = (vni & 0x00ff0000) >> 16;
1487 tvni[1] = (vni & 0x0000ff00) >> 8;
1488 tvni[2] = vni & 0x000000ff;
1489
1490 tunid = vni_to_tunnel_id(tvni);
1491 if (changelink && (tunid != info->key.tun_id)) {
1492 attrtype = IFLA_GENEVE_ID;
1493 goto change_notsup;
1494 }
1495 info->key.tun_id = tunid;
1496 }
1497
1498 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1499 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1500 cfg->ttl_inherit = true;
1501 else
1502 cfg->ttl_inherit = false;
1503 } else if (data[IFLA_GENEVE_TTL]) {
1504 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1505 cfg->ttl_inherit = false;
1506 }
1507
1508 if (data[IFLA_GENEVE_TOS])
1509 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1510
1511 if (data[IFLA_GENEVE_DF])
1512 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1513
1514 if (data[IFLA_GENEVE_LABEL]) {
1515 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1516 IPV6_FLOWLABEL_MASK;
1517 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1518 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1519 "Label attribute only applies for IPv6 Geneve devices");
1520 return -EINVAL;
1521 }
1522 }
1523
1524 if (data[IFLA_GENEVE_PORT]) {
1525 if (changelink) {
1526 attrtype = IFLA_GENEVE_PORT;
1527 goto change_notsup;
1528 }
1529 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1530 }
1531
1532 if (data[IFLA_GENEVE_PORT_RANGE]) {
1533 const struct ifla_geneve_port_range *p;
1534
1535 if (changelink) {
1536 attrtype = IFLA_GENEVE_PORT_RANGE;
1537 goto change_notsup;
1538 }
1539 p = nla_data(data[IFLA_GENEVE_PORT_RANGE]);
1540 cfg->port_min = ntohs(p->low);
1541 cfg->port_max = ntohs(p->high);
1542 }
1543
1544 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1545 if (changelink) {
1546 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1547 goto change_notsup;
1548 }
1549 cfg->collect_md = true;
1550 }
1551
1552 if (data[IFLA_GENEVE_UDP_CSUM]) {
1553 if (changelink) {
1554 attrtype = IFLA_GENEVE_UDP_CSUM;
1555 goto change_notsup;
1556 }
1557 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1558 __set_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
1559 }
1560
1561 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1562 #if IS_ENABLED(CONFIG_IPV6)
1563 if (changelink) {
1564 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1565 goto change_notsup;
1566 }
1567 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1568 __clear_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags);
1569 #else
1570 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1571 "IPv6 support not enabled in the kernel");
1572 return -EPFNOSUPPORT;
1573 #endif
1574 }
1575
1576 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1577 #if IS_ENABLED(CONFIG_IPV6)
1578 if (changelink) {
1579 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1580 goto change_notsup;
1581 }
1582 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1583 cfg->use_udp6_rx_checksums = false;
1584 #else
1585 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1586 "IPv6 support not enabled in the kernel");
1587 return -EPFNOSUPPORT;
1588 #endif
1589 }
1590
1591 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1592 if (changelink) {
1593 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1594 goto change_notsup;
1595 }
1596 cfg->inner_proto_inherit = true;
1597 }
1598
1599 return 0;
1600 change_notsup:
1601 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1602 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1603 return -EOPNOTSUPP;
1604 }
1605
geneve_link_config(struct net_device * dev,struct ip_tunnel_info * info,struct nlattr * tb[])1606 static void geneve_link_config(struct net_device *dev,
1607 struct ip_tunnel_info *info, struct nlattr *tb[])
1608 {
1609 struct geneve_dev *geneve = netdev_priv(dev);
1610 int ldev_mtu = 0;
1611
1612 if (tb[IFLA_MTU]) {
1613 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1614 return;
1615 }
1616
1617 switch (ip_tunnel_info_af(info)) {
1618 case AF_INET: {
1619 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1620 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1621
1622 if (!IS_ERR(rt) && rt->dst.dev) {
1623 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1624 ip_rt_put(rt);
1625 }
1626 break;
1627 }
1628 #if IS_ENABLED(CONFIG_IPV6)
1629 case AF_INET6: {
1630 struct rt6_info *rt;
1631
1632 if (!__in6_dev_get(dev))
1633 break;
1634
1635 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1636 NULL, 0);
1637
1638 if (rt && rt->dst.dev)
1639 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1640 ip6_rt_put(rt);
1641 break;
1642 }
1643 #endif
1644 }
1645
1646 if (ldev_mtu <= 0)
1647 return;
1648
1649 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1650 }
1651
geneve_newlink(struct net_device * dev,struct rtnl_newlink_params * params,struct netlink_ext_ack * extack)1652 static int geneve_newlink(struct net_device *dev,
1653 struct rtnl_newlink_params *params,
1654 struct netlink_ext_ack *extack)
1655 {
1656 struct net *link_net = rtnl_newlink_link_net(params);
1657 struct nlattr **data = params->data;
1658 struct nlattr **tb = params->tb;
1659 struct geneve_config cfg = {
1660 .df = GENEVE_DF_UNSET,
1661 .use_udp6_rx_checksums = false,
1662 .ttl_inherit = false,
1663 .collect_md = false,
1664 .port_min = 1,
1665 .port_max = USHRT_MAX,
1666 };
1667 int err;
1668
1669 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1670 err = geneve_nl2info(tb, data, extack, &cfg, false);
1671 if (err)
1672 return err;
1673
1674 err = geneve_configure(link_net, dev, extack, &cfg);
1675 if (err)
1676 return err;
1677
1678 geneve_link_config(dev, &cfg.info, tb);
1679
1680 return 0;
1681 }
1682
1683 /* Quiesces the geneve device data path for both TX and RX.
1684 *
1685 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1686 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1687 * to complete for the existing set of in-flight packets to be transmitted,
1688 * then we would have quiesced the transmit data path. All the future packets
1689 * will get dropped until we unquiesce the data path.
1690 *
1691 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1692 * if we set that to NULL under RCU and wait for synchronize_net() to
1693 * complete, then we would have quiesced the receive data path.
1694 */
geneve_quiesce(struct geneve_dev * geneve,struct geneve_sock ** gs4,struct geneve_sock ** gs6)1695 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1696 struct geneve_sock **gs6)
1697 {
1698 *gs4 = rtnl_dereference(geneve->sock4);
1699 rcu_assign_pointer(geneve->sock4, NULL);
1700 if (*gs4)
1701 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1702 #if IS_ENABLED(CONFIG_IPV6)
1703 *gs6 = rtnl_dereference(geneve->sock6);
1704 rcu_assign_pointer(geneve->sock6, NULL);
1705 if (*gs6)
1706 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1707 #else
1708 *gs6 = NULL;
1709 #endif
1710 synchronize_net();
1711 }
1712
1713 /* Resumes the geneve device data path for both TX and RX. */
geneve_unquiesce(struct geneve_dev * geneve,struct geneve_sock * gs4,struct geneve_sock __maybe_unused * gs6)1714 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1715 struct geneve_sock __maybe_unused *gs6)
1716 {
1717 rcu_assign_pointer(geneve->sock4, gs4);
1718 if (gs4)
1719 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1720 #if IS_ENABLED(CONFIG_IPV6)
1721 rcu_assign_pointer(geneve->sock6, gs6);
1722 if (gs6)
1723 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1724 #endif
1725 synchronize_net();
1726 }
1727
geneve_changelink(struct net_device * dev,struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack)1728 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1729 struct nlattr *data[],
1730 struct netlink_ext_ack *extack)
1731 {
1732 struct geneve_dev *geneve = netdev_priv(dev);
1733 struct geneve_sock *gs4, *gs6;
1734 struct geneve_config cfg;
1735 int err;
1736
1737 /* If the geneve device is configured for metadata (or externally
1738 * controlled, for example, OVS), then nothing can be changed.
1739 */
1740 if (geneve->cfg.collect_md)
1741 return -EOPNOTSUPP;
1742
1743 /* Start with the existing info. */
1744 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1745 err = geneve_nl2info(tb, data, extack, &cfg, true);
1746 if (err)
1747 return err;
1748
1749 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1750 dst_cache_reset(&cfg.info.dst_cache);
1751 geneve_link_config(dev, &cfg.info, tb);
1752 }
1753
1754 geneve_quiesce(geneve, &gs4, &gs6);
1755 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1756 geneve_unquiesce(geneve, gs4, gs6);
1757
1758 return 0;
1759 }
1760
geneve_dellink(struct net_device * dev,struct list_head * head)1761 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1762 {
1763 struct geneve_dev *geneve = netdev_priv(dev);
1764
1765 list_del(&geneve->next);
1766 unregister_netdevice_queue(dev, head);
1767 }
1768
geneve_get_size(const struct net_device * dev)1769 static size_t geneve_get_size(const struct net_device *dev)
1770 {
1771 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1772 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1773 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1774 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1775 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1776 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1777 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1778 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1779 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1780 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1781 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1782 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1783 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1784 nla_total_size(sizeof(struct ifla_geneve_port_range)) + /* IFLA_GENEVE_PORT_RANGE */
1785 0;
1786 }
1787
geneve_fill_info(struct sk_buff * skb,const struct net_device * dev)1788 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1789 {
1790 struct geneve_dev *geneve = netdev_priv(dev);
1791 struct ip_tunnel_info *info = &geneve->cfg.info;
1792 bool ttl_inherit = geneve->cfg.ttl_inherit;
1793 bool metadata = geneve->cfg.collect_md;
1794 struct ifla_geneve_port_range ports = {
1795 .low = htons(geneve->cfg.port_min),
1796 .high = htons(geneve->cfg.port_max),
1797 };
1798 __u8 tmp_vni[3];
1799 __u32 vni;
1800
1801 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1802 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1803 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1804 goto nla_put_failure;
1805
1806 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1807 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1808 info->key.u.ipv4.dst))
1809 goto nla_put_failure;
1810 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1811 test_bit(IP_TUNNEL_CSUM_BIT,
1812 info->key.tun_flags)))
1813 goto nla_put_failure;
1814
1815 #if IS_ENABLED(CONFIG_IPV6)
1816 } else if (!metadata) {
1817 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1818 &info->key.u.ipv6.dst))
1819 goto nla_put_failure;
1820 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1821 !test_bit(IP_TUNNEL_CSUM_BIT,
1822 info->key.tun_flags)))
1823 goto nla_put_failure;
1824 #endif
1825 }
1826
1827 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1828 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1829 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1830 goto nla_put_failure;
1831
1832 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1833 goto nla_put_failure;
1834
1835 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1836 goto nla_put_failure;
1837
1838 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1839 goto nla_put_failure;
1840
1841 #if IS_ENABLED(CONFIG_IPV6)
1842 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1843 !geneve->cfg.use_udp6_rx_checksums))
1844 goto nla_put_failure;
1845 #endif
1846
1847 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1848 goto nla_put_failure;
1849
1850 if (geneve->cfg.inner_proto_inherit &&
1851 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1852 goto nla_put_failure;
1853
1854 if (nla_put(skb, IFLA_GENEVE_PORT_RANGE, sizeof(ports), &ports))
1855 goto nla_put_failure;
1856
1857 return 0;
1858
1859 nla_put_failure:
1860 return -EMSGSIZE;
1861 }
1862
1863 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1864 .kind = "geneve",
1865 .maxtype = IFLA_GENEVE_MAX,
1866 .policy = geneve_policy,
1867 .priv_size = sizeof(struct geneve_dev),
1868 .setup = geneve_setup,
1869 .validate = geneve_validate,
1870 .newlink = geneve_newlink,
1871 .changelink = geneve_changelink,
1872 .dellink = geneve_dellink,
1873 .get_size = geneve_get_size,
1874 .fill_info = geneve_fill_info,
1875 };
1876
geneve_dev_create_fb(struct net * net,const char * name,u8 name_assign_type,u16 dst_port)1877 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1878 u8 name_assign_type, u16 dst_port)
1879 {
1880 struct nlattr *tb[IFLA_MAX + 1];
1881 struct net_device *dev;
1882 LIST_HEAD(list_kill);
1883 int err;
1884 struct geneve_config cfg = {
1885 .df = GENEVE_DF_UNSET,
1886 .use_udp6_rx_checksums = true,
1887 .ttl_inherit = false,
1888 .collect_md = true,
1889 .port_min = 1,
1890 .port_max = USHRT_MAX,
1891 };
1892
1893 memset(tb, 0, sizeof(tb));
1894 dev = rtnl_create_link(net, name, name_assign_type,
1895 &geneve_link_ops, tb, NULL);
1896 if (IS_ERR(dev))
1897 return dev;
1898
1899 init_tnl_info(&cfg.info, dst_port);
1900 err = geneve_configure(net, dev, NULL, &cfg);
1901 if (err) {
1902 free_netdev(dev);
1903 return ERR_PTR(err);
1904 }
1905
1906 /* openvswitch users expect packet sizes to be unrestricted,
1907 * so set the largest MTU we can.
1908 */
1909 err = geneve_change_mtu(dev, IP_MAX_MTU);
1910 if (err)
1911 goto err;
1912
1913 err = rtnl_configure_link(dev, NULL, 0, NULL);
1914 if (err < 0)
1915 goto err;
1916
1917 return dev;
1918 err:
1919 geneve_dellink(dev, &list_kill);
1920 unregister_netdevice_many(&list_kill);
1921 return ERR_PTR(err);
1922 }
1923 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1924
geneve_netdevice_event(struct notifier_block * unused,unsigned long event,void * ptr)1925 static int geneve_netdevice_event(struct notifier_block *unused,
1926 unsigned long event, void *ptr)
1927 {
1928 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1929
1930 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1931 geneve_offload_rx_ports(dev, true);
1932 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1933 geneve_offload_rx_ports(dev, false);
1934
1935 return NOTIFY_DONE;
1936 }
1937
1938 static struct notifier_block geneve_notifier_block __read_mostly = {
1939 .notifier_call = geneve_netdevice_event,
1940 };
1941
geneve_init_net(struct net * net)1942 static __net_init int geneve_init_net(struct net *net)
1943 {
1944 struct geneve_net *gn = net_generic(net, geneve_net_id);
1945
1946 INIT_LIST_HEAD(&gn->geneve_list);
1947 INIT_LIST_HEAD(&gn->sock_list);
1948 return 0;
1949 }
1950
geneve_exit_rtnl_net(struct net * net,struct list_head * dev_to_kill)1951 static void __net_exit geneve_exit_rtnl_net(struct net *net,
1952 struct list_head *dev_to_kill)
1953 {
1954 struct geneve_net *gn = net_generic(net, geneve_net_id);
1955 struct geneve_dev *geneve, *next;
1956
1957 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next)
1958 geneve_dellink(geneve->dev, dev_to_kill);
1959 }
1960
geneve_exit_net(struct net * net)1961 static void __net_exit geneve_exit_net(struct net *net)
1962 {
1963 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1964
1965 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1966 }
1967
1968 static struct pernet_operations geneve_net_ops = {
1969 .init = geneve_init_net,
1970 .exit_rtnl = geneve_exit_rtnl_net,
1971 .exit = geneve_exit_net,
1972 .id = &geneve_net_id,
1973 .size = sizeof(struct geneve_net),
1974 };
1975
geneve_init_module(void)1976 static int __init geneve_init_module(void)
1977 {
1978 int rc;
1979
1980 rc = register_pernet_subsys(&geneve_net_ops);
1981 if (rc)
1982 goto out1;
1983
1984 rc = register_netdevice_notifier(&geneve_notifier_block);
1985 if (rc)
1986 goto out2;
1987
1988 rc = rtnl_link_register(&geneve_link_ops);
1989 if (rc)
1990 goto out3;
1991
1992 return 0;
1993 out3:
1994 unregister_netdevice_notifier(&geneve_notifier_block);
1995 out2:
1996 unregister_pernet_subsys(&geneve_net_ops);
1997 out1:
1998 return rc;
1999 }
2000 late_initcall(geneve_init_module);
2001
geneve_cleanup_module(void)2002 static void __exit geneve_cleanup_module(void)
2003 {
2004 rtnl_link_unregister(&geneve_link_ops);
2005 unregister_netdevice_notifier(&geneve_notifier_block);
2006 unregister_pernet_subsys(&geneve_net_ops);
2007 }
2008 module_exit(geneve_cleanup_module);
2009
2010 MODULE_LICENSE("GPL");
2011 MODULE_VERSION(GENEVE_NETDEV_VER);
2012 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2013 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2014 MODULE_ALIAS_RTNL_LINK("geneve");
2015