?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
perf tools: Don't clone maps from parent when synthesizing forks
[linux/fpc-iii.git] / drivers / net / geneve.c
bloba0cd1c41cf5f07255283be656f1ecf074417532a
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
33 static bool log_ecn_error = true;
34 module_param(log_ecn_error, bool, 0644);
35 MODULE_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 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
40 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
41
42 /* per-network namespace private data for this module */
43 struct geneve_net {
44 struct list_head geneve_list;
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 /* Pseudo network device */
56 struct geneve_dev {
57 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
58 #if IS_ENABLED(CONFIG_IPV6)
59 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
60 #endif
61 struct net *net; /* netns for packet i/o */
62 struct net_device *dev; /* netdev for geneve tunnel */
63 struct ip_tunnel_info info;
64 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
65 #if IS_ENABLED(CONFIG_IPV6)
66 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
67 #endif
68 struct list_head next; /* geneve's per namespace list */
69 struct gro_cells gro_cells;
70 bool collect_md;
71 bool use_udp6_rx_checksums;
72 bool ttl_inherit;
73 };
74
75 struct geneve_sock {
76 bool collect_md;
77 struct list_head list;
78 struct socket *sock;
79 struct rcu_head rcu;
80 int refcnt;
81 struct hlist_head vni_list[VNI_HASH_SIZE];
82 };
83
84 static inline __u32 geneve_net_vni_hash(u8 vni[3])
85 {
86 __u32 vnid;
87
88 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
89 return hash_32(vnid, VNI_HASH_BITS);
90 }
91
92 static __be64 vni_to_tunnel_id(const __u8 *vni)
93 {
94 #ifdef __BIG_ENDIAN
95 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
96 #else
97 return (__force __be64)(((__force u64)vni[0] << 40) |
98 ((__force u64)vni[1] << 48) |
99 ((__force u64)vni[2] << 56));
100 #endif
101 }
102
103 /* Convert 64 bit tunnel ID to 24 bit VNI. */
104 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
105 {
106 #ifdef __BIG_ENDIAN
107 vni[0] = (__force __u8)(tun_id >> 16);
108 vni[1] = (__force __u8)(tun_id >> 8);
109 vni[2] = (__force __u8)tun_id;
110 #else
111 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
112 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
113 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
114 #endif
115 }
116
117 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
118 {
119 return !memcmp(vni, &tun_id[5], 3);
120 }
121
122 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
123 {
124 return gs->sock->sk->sk_family;
125 }
126
127 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
128 __be32 addr, u8 vni[])
129 {
130 struct hlist_head *vni_list_head;
131 struct geneve_dev_node *node;
132 __u32 hash;
133
134 /* Find the device for this VNI */
135 hash = geneve_net_vni_hash(vni);
136 vni_list_head = &gs->vni_list[hash];
137 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
138 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
139 addr == node->geneve->info.key.u.ipv4.dst)
140 return node->geneve;
141 }
142 return NULL;
143 }
144
145 #if IS_ENABLED(CONFIG_IPV6)
146 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
147 struct in6_addr addr6, u8 vni[])
148 {
149 struct hlist_head *vni_list_head;
150 struct geneve_dev_node *node;
151 __u32 hash;
152
153 /* Find the device for this VNI */
154 hash = geneve_net_vni_hash(vni);
155 vni_list_head = &gs->vni_list[hash];
156 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
157 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
158 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
159 return node->geneve;
160 }
161 return NULL;
162 }
163 #endif
164
165 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
166 {
167 return (struct genevehdr *)(udp_hdr(skb) + 1);
168 }
169
170 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
171 struct sk_buff *skb)
172 {
173 static u8 zero_vni[3];
174 u8 *vni;
175
176 if (geneve_get_sk_family(gs) == AF_INET) {
177 struct iphdr *iph;
178 __be32 addr;
179
180 iph = ip_hdr(skb); /* outer IP header... */
181
182 if (gs->collect_md) {
183 vni = zero_vni;
184 addr = 0;
185 } else {
186 vni = geneve_hdr(skb)->vni;
187 addr = iph->saddr;
188 }
189
190 return geneve_lookup(gs, addr, vni);
191 #if IS_ENABLED(CONFIG_IPV6)
192 } else if (geneve_get_sk_family(gs) == AF_INET6) {
193 static struct in6_addr zero_addr6;
194 struct ipv6hdr *ip6h;
195 struct in6_addr addr6;
196
197 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
198
199 if (gs->collect_md) {
200 vni = zero_vni;
201 addr6 = zero_addr6;
202 } else {
203 vni = geneve_hdr(skb)->vni;
204 addr6 = ip6h->saddr;
205 }
206
207 return geneve6_lookup(gs, addr6, vni);
208 #endif
209 }
210 return NULL;
211 }
212
213 /* geneve receive/decap routine */
214 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
215 struct sk_buff *skb)
216 {
217 struct genevehdr *gnvh = geneve_hdr(skb);
218 struct metadata_dst *tun_dst = NULL;
219 struct pcpu_sw_netstats *stats;
220 unsigned int len;
221 int err = 0;
222 void *oiph;
223
224 if (ip_tunnel_collect_metadata() || gs->collect_md) {
225 __be16 flags;
226
227 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
228 (gnvh->oam ? TUNNEL_OAM : 0) |
229 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
230
231 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
232 vni_to_tunnel_id(gnvh->vni),
233 gnvh->opt_len * 4);
234 if (!tun_dst) {
235 geneve->dev->stats.rx_dropped++;
236 goto drop;
237 }
238 /* Update tunnel dst according to Geneve options. */
239 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
240 gnvh->options, gnvh->opt_len * 4,
241 TUNNEL_GENEVE_OPT);
242 } else {
243 /* Drop packets w/ critical options,
244 * since we don't support any...
245 */
246 if (gnvh->critical) {
247 geneve->dev->stats.rx_frame_errors++;
248 geneve->dev->stats.rx_errors++;
249 goto drop;
250 }
251 }
252
253 skb_reset_mac_header(skb);
254 skb->protocol = eth_type_trans(skb, geneve->dev);
255 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
256
257 if (tun_dst)
258 skb_dst_set(skb, &tun_dst->dst);
259
260 /* Ignore packet loops (and multicast echo) */
261 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
262 geneve->dev->stats.rx_errors++;
263 goto drop;
264 }
265
266 oiph = skb_network_header(skb);
267 skb_reset_network_header(skb);
268
269 if (geneve_get_sk_family(gs) == AF_INET)
270 err = IP_ECN_decapsulate(oiph, skb);
271 #if IS_ENABLED(CONFIG_IPV6)
272 else
273 err = IP6_ECN_decapsulate(oiph, skb);
274 #endif
275
276 if (unlikely(err)) {
277 if (log_ecn_error) {
278 if (geneve_get_sk_family(gs) == AF_INET)
279 net_info_ratelimited("non-ECT from %pI4 "
280 "with TOS=%#x\n",
281 &((struct iphdr *)oiph)->saddr,
282 ((struct iphdr *)oiph)->tos);
283 #if IS_ENABLED(CONFIG_IPV6)
284 else
285 net_info_ratelimited("non-ECT from %pI6\n",
286 &((struct ipv6hdr *)oiph)->saddr);
287 #endif
288 }
289 if (err > 1) {
290 ++geneve->dev->stats.rx_frame_errors;
291 ++geneve->dev->stats.rx_errors;
292 goto drop;
293 }
294 }
295
296 len = skb->len;
297 err = gro_cells_receive(&geneve->gro_cells, skb);
298 if (likely(err == NET_RX_SUCCESS)) {
299 stats = this_cpu_ptr(geneve->dev->tstats);
300 u64_stats_update_begin(&stats->syncp);
301 stats->rx_packets++;
302 stats->rx_bytes += len;
303 u64_stats_update_end(&stats->syncp);
304 }
305 return;
306 drop:
307 /* Consume bad packet */
308 kfree_skb(skb);
309 }
310
311 /* Setup stats when device is created */
312 static int geneve_init(struct net_device *dev)
313 {
314 struct geneve_dev *geneve = netdev_priv(dev);
315 int err;
316
317 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
318 if (!dev->tstats)
319 return -ENOMEM;
320
321 err = gro_cells_init(&geneve->gro_cells, dev);
322 if (err) {
323 free_percpu(dev->tstats);
324 return err;
325 }
326
327 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
328 if (err) {
329 free_percpu(dev->tstats);
330 gro_cells_destroy(&geneve->gro_cells);
331 return err;
332 }
333 return 0;
334 }
335
336 static void geneve_uninit(struct net_device *dev)
337 {
338 struct geneve_dev *geneve = netdev_priv(dev);
339
340 dst_cache_destroy(&geneve->info.dst_cache);
341 gro_cells_destroy(&geneve->gro_cells);
342 free_percpu(dev->tstats);
343 }
344
345 /* Callback from net/ipv4/udp.c to receive packets */
346 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
347 {
348 struct genevehdr *geneveh;
349 struct geneve_dev *geneve;
350 struct geneve_sock *gs;
351 int opts_len;
352
353 /* Need UDP and Geneve header to be present */
354 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
355 goto drop;
356
357 /* Return packets with reserved bits set */
358 geneveh = geneve_hdr(skb);
359 if (unlikely(geneveh->ver != GENEVE_VER))
360 goto drop;
361
362 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
363 goto drop;
364
365 gs = rcu_dereference_sk_user_data(sk);
366 if (!gs)
367 goto drop;
368
369 geneve = geneve_lookup_skb(gs, skb);
370 if (!geneve)
371 goto drop;
372
373 opts_len = geneveh->opt_len * 4;
374 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
375 htons(ETH_P_TEB),
376 !net_eq(geneve->net, dev_net(geneve->dev)))) {
377 geneve->dev->stats.rx_dropped++;
378 goto drop;
379 }
380
381 geneve_rx(geneve, gs, skb);
382 return 0;
383
384 drop:
385 /* Consume bad packet */
386 kfree_skb(skb);
387 return 0;
388 }
389
390 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
391 __be16 port, bool ipv6_rx_csum)
392 {
393 struct socket *sock;
394 struct udp_port_cfg udp_conf;
395 int err;
396
397 memset(&udp_conf, 0, sizeof(udp_conf));
398
399 if (ipv6) {
400 udp_conf.family = AF_INET6;
401 udp_conf.ipv6_v6only = 1;
402 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
403 } else {
404 udp_conf.family = AF_INET;
405 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
406 }
407
408 udp_conf.local_udp_port = port;
409
410 /* Open UDP socket */
411 err = udp_sock_create(net, &udp_conf, &sock);
412 if (err < 0)
413 return ERR_PTR(err);
414
415 return sock;
416 }
417
418 static int geneve_hlen(struct genevehdr *gh)
419 {
420 return sizeof(*gh) + gh->opt_len * 4;
421 }
422
423 static struct sk_buff *geneve_gro_receive(struct sock *sk,
424 struct list_head *head,
425 struct sk_buff *skb)
426 {
427 struct sk_buff *pp = NULL;
428 struct sk_buff *p;
429 struct genevehdr *gh, *gh2;
430 unsigned int hlen, gh_len, off_gnv;
431 const struct packet_offload *ptype;
432 __be16 type;
433 int flush = 1;
434
435 off_gnv = skb_gro_offset(skb);
436 hlen = off_gnv + sizeof(*gh);
437 gh = skb_gro_header_fast(skb, off_gnv);
438 if (skb_gro_header_hard(skb, hlen)) {
439 gh = skb_gro_header_slow(skb, hlen, off_gnv);
440 if (unlikely(!gh))
441 goto out;
442 }
443
444 if (gh->ver != GENEVE_VER || gh->oam)
445 goto out;
446 gh_len = geneve_hlen(gh);
447
448 hlen = off_gnv + gh_len;
449 if (skb_gro_header_hard(skb, hlen)) {
450 gh = skb_gro_header_slow(skb, hlen, off_gnv);
451 if (unlikely(!gh))
452 goto out;
453 }
454
455 list_for_each_entry(p, head, list) {
456 if (!NAPI_GRO_CB(p)->same_flow)
457 continue;
458
459 gh2 = (struct genevehdr *)(p->data + off_gnv);
460 if (gh->opt_len != gh2->opt_len ||
461 memcmp(gh, gh2, gh_len)) {
462 NAPI_GRO_CB(p)->same_flow = 0;
463 continue;
464 }
465 }
466
467 type = gh->proto_type;
468
469 rcu_read_lock();
470 ptype = gro_find_receive_by_type(type);
471 if (!ptype)
472 goto out_unlock;
473
474 skb_gro_pull(skb, gh_len);
475 skb_gro_postpull_rcsum(skb, gh, gh_len);
476 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
477 flush = 0;
478
479 out_unlock:
480 rcu_read_unlock();
481 out:
482 skb_gro_flush_final(skb, pp, flush);
483
484 return pp;
485 }
486
487 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
488 int nhoff)
489 {
490 struct genevehdr *gh;
491 struct packet_offload *ptype;
492 __be16 type;
493 int gh_len;
494 int err = -ENOSYS;
495
496 gh = (struct genevehdr *)(skb->data + nhoff);
497 gh_len = geneve_hlen(gh);
498 type = gh->proto_type;
499
500 rcu_read_lock();
501 ptype = gro_find_complete_by_type(type);
502 if (ptype)
503 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
504
505 rcu_read_unlock();
506
507 skb_set_inner_mac_header(skb, nhoff + gh_len);
508
509 return err;
510 }
511
512 /* Create new listen socket if needed */
513 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
514 bool ipv6, bool ipv6_rx_csum)
515 {
516 struct geneve_net *gn = net_generic(net, geneve_net_id);
517 struct geneve_sock *gs;
518 struct socket *sock;
519 struct udp_tunnel_sock_cfg tunnel_cfg;
520 int h;
521
522 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
523 if (!gs)
524 return ERR_PTR(-ENOMEM);
525
526 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
527 if (IS_ERR(sock)) {
528 kfree(gs);
529 return ERR_CAST(sock);
530 }
531
532 gs->sock = sock;
533 gs->refcnt = 1;
534 for (h = 0; h < VNI_HASH_SIZE; ++h)
535 INIT_HLIST_HEAD(&gs->vni_list[h]);
536
537 /* Initialize the geneve udp offloads structure */
538 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
539
540 /* Mark socket as an encapsulation socket */
541 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
542 tunnel_cfg.sk_user_data = gs;
543 tunnel_cfg.encap_type = 1;
544 tunnel_cfg.gro_receive = geneve_gro_receive;
545 tunnel_cfg.gro_complete = geneve_gro_complete;
546 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
547 tunnel_cfg.encap_destroy = NULL;
548 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
549 list_add(&gs->list, &gn->sock_list);
550 return gs;
551 }
552
553 static void __geneve_sock_release(struct geneve_sock *gs)
554 {
555 if (!gs || --gs->refcnt)
556 return;
557
558 list_del(&gs->list);
559 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
560 udp_tunnel_sock_release(gs->sock);
561 kfree_rcu(gs, rcu);
562 }
563
564 static void geneve_sock_release(struct geneve_dev *geneve)
565 {
566 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
567 #if IS_ENABLED(CONFIG_IPV6)
568 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
569
570 rcu_assign_pointer(geneve->sock6, NULL);
571 #endif
572
573 rcu_assign_pointer(geneve->sock4, NULL);
574 synchronize_net();
575
576 __geneve_sock_release(gs4);
577 #if IS_ENABLED(CONFIG_IPV6)
578 __geneve_sock_release(gs6);
579 #endif
580 }
581
582 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
583 sa_family_t family,
584 __be16 dst_port)
585 {
586 struct geneve_sock *gs;
587
588 list_for_each_entry(gs, &gn->sock_list, list) {
589 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
590 geneve_get_sk_family(gs) == family) {
591 return gs;
592 }
593 }
594 return NULL;
595 }
596
597 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
598 {
599 struct net *net = geneve->net;
600 struct geneve_net *gn = net_generic(net, geneve_net_id);
601 struct geneve_dev_node *node;
602 struct geneve_sock *gs;
603 __u8 vni[3];
604 __u32 hash;
605
606 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
607 if (gs) {
608 gs->refcnt++;
609 goto out;
610 }
611
612 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
613 geneve->use_udp6_rx_checksums);
614 if (IS_ERR(gs))
615 return PTR_ERR(gs);
616
617 out:
618 gs->collect_md = geneve->collect_md;
619 #if IS_ENABLED(CONFIG_IPV6)
620 if (ipv6) {
621 rcu_assign_pointer(geneve->sock6, gs);
622 node = &geneve->hlist6;
623 } else
624 #endif
625 {
626 rcu_assign_pointer(geneve->sock4, gs);
627 node = &geneve->hlist4;
628 }
629 node->geneve = geneve;
630
631 tunnel_id_to_vni(geneve->info.key.tun_id, vni);
632 hash = geneve_net_vni_hash(vni);
633 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
634 return 0;
635 }
636
637 static int geneve_open(struct net_device *dev)
638 {
639 struct geneve_dev *geneve = netdev_priv(dev);
640 bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6);
641 bool metadata = geneve->collect_md;
642 int ret = 0;
643
644 #if IS_ENABLED(CONFIG_IPV6)
645 if (ipv6 || metadata)
646 ret = geneve_sock_add(geneve, true);
647 #endif
648 if (!ret && (!ipv6 || metadata))
649 ret = geneve_sock_add(geneve, false);
650 if (ret < 0)
651 geneve_sock_release(geneve);
652
653 return ret;
654 }
655
656 static int geneve_stop(struct net_device *dev)
657 {
658 struct geneve_dev *geneve = netdev_priv(dev);
659
660 hlist_del_init_rcu(&geneve->hlist4.hlist);
661 #if IS_ENABLED(CONFIG_IPV6)
662 hlist_del_init_rcu(&geneve->hlist6.hlist);
663 #endif
664 geneve_sock_release(geneve);
665 return 0;
666 }
667
668 static void geneve_build_header(struct genevehdr *geneveh,
669 const struct ip_tunnel_info *info)
670 {
671 geneveh->ver = GENEVE_VER;
672 geneveh->opt_len = info->options_len / 4;
673 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
674 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
675 geneveh->rsvd1 = 0;
676 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
677 geneveh->proto_type = htons(ETH_P_TEB);
678 geneveh->rsvd2 = 0;
679
680 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
681 ip_tunnel_info_opts_get(geneveh->options, info);
682 }
683
684 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
685 const struct ip_tunnel_info *info,
686 bool xnet, int ip_hdr_len)
687 {
688 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
689 struct genevehdr *gnvh;
690 int min_headroom;
691 int err;
692
693 skb_reset_mac_header(skb);
694 skb_scrub_packet(skb, xnet);
695
696 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
697 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
698 err = skb_cow_head(skb, min_headroom);
699 if (unlikely(err))
700 goto free_dst;
701
702 err = udp_tunnel_handle_offloads(skb, udp_sum);
703 if (err)
704 goto free_dst;
705
706 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
707 geneve_build_header(gnvh, info);
708 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
709 return 0;
710
711 free_dst:
712 dst_release(dst);
713 return err;
714 }
715
716 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
717 struct net_device *dev,
718 struct geneve_sock *gs4,
719 struct flowi4 *fl4,
720 const struct ip_tunnel_info *info)
721 {
722 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
723 struct geneve_dev *geneve = netdev_priv(dev);
724 struct dst_cache *dst_cache;
725 struct rtable *rt = NULL;
726 __u8 tos;
727
728 if (!gs4)
729 return ERR_PTR(-EIO);
730
731 memset(fl4, 0, sizeof(*fl4));
732 fl4->flowi4_mark = skb->mark;
733 fl4->flowi4_proto = IPPROTO_UDP;
734 fl4->daddr = info->key.u.ipv4.dst;
735 fl4->saddr = info->key.u.ipv4.src;
736
737 tos = info->key.tos;
738 if ((tos == 1) && !geneve->collect_md) {
739 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
740 use_cache = false;
741 }
742 fl4->flowi4_tos = RT_TOS(tos);
743
744 dst_cache = (struct dst_cache *)&info->dst_cache;
745 if (use_cache) {
746 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
747 if (rt)
748 return rt;
749 }
750 rt = ip_route_output_key(geneve->net, fl4);
751 if (IS_ERR(rt)) {
752 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
753 return ERR_PTR(-ENETUNREACH);
754 }
755 if (rt->dst.dev == dev) { /* is this necessary? */
756 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
757 ip_rt_put(rt);
758 return ERR_PTR(-ELOOP);
759 }
760 if (use_cache)
761 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
762 return rt;
763 }
764
765 #if IS_ENABLED(CONFIG_IPV6)
766 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
767 struct net_device *dev,
768 struct geneve_sock *gs6,
769 struct flowi6 *fl6,
770 const struct ip_tunnel_info *info)
771 {
772 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
773 struct geneve_dev *geneve = netdev_priv(dev);
774 struct dst_entry *dst = NULL;
775 struct dst_cache *dst_cache;
776 __u8 prio;
777
778 if (!gs6)
779 return ERR_PTR(-EIO);
780
781 memset(fl6, 0, sizeof(*fl6));
782 fl6->flowi6_mark = skb->mark;
783 fl6->flowi6_proto = IPPROTO_UDP;
784 fl6->daddr = info->key.u.ipv6.dst;
785 fl6->saddr = info->key.u.ipv6.src;
786 prio = info->key.tos;
787 if ((prio == 1) && !geneve->collect_md) {
788 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
789 use_cache = false;
790 }
791
792 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
793 info->key.label);
794 dst_cache = (struct dst_cache *)&info->dst_cache;
795 if (use_cache) {
796 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
797 if (dst)
798 return dst;
799 }
800 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
801 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
802 return ERR_PTR(-ENETUNREACH);
803 }
804 if (dst->dev == dev) { /* is this necessary? */
805 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
806 dst_release(dst);
807 return ERR_PTR(-ELOOP);
808 }
809
810 if (use_cache)
811 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
812 return dst;
813 }
814 #endif
815
816 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
817 struct geneve_dev *geneve,
818 const struct ip_tunnel_info *info)
819 {
820 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
821 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
822 const struct ip_tunnel_key *key = &info->key;
823 struct rtable *rt;
824 struct flowi4 fl4;
825 __u8 tos, ttl;
826 __be16 sport;
827 __be16 df;
828 int err;
829
830 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
831 if (IS_ERR(rt))
832 return PTR_ERR(rt);
833
834 skb_tunnel_check_pmtu(skb, &rt->dst,
835 GENEVE_IPV4_HLEN + info->options_len);
836
837 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
838 if (geneve->collect_md) {
839 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
840 ttl = key->ttl;
841 } else {
842 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
843 if (geneve->ttl_inherit)
844 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
845 else
846 ttl = key->ttl;
847 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
848 }
849 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
850
851 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
852 if (unlikely(err))
853 return err;
854
855 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
856 tos, ttl, df, sport, geneve->info.key.tp_dst,
857 !net_eq(geneve->net, dev_net(geneve->dev)),
858 !(info->key.tun_flags & TUNNEL_CSUM));
859 return 0;
860 }
861
862 #if IS_ENABLED(CONFIG_IPV6)
863 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
864 struct geneve_dev *geneve,
865 const struct ip_tunnel_info *info)
866 {
867 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
868 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
869 const struct ip_tunnel_key *key = &info->key;
870 struct dst_entry *dst = NULL;
871 struct flowi6 fl6;
872 __u8 prio, ttl;
873 __be16 sport;
874 int err;
875
876 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
877 if (IS_ERR(dst))
878 return PTR_ERR(dst);
879
880 skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
881
882 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
883 if (geneve->collect_md) {
884 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
885 ttl = key->ttl;
886 } else {
887 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
888 ip_hdr(skb), skb);
889 if (geneve->ttl_inherit)
890 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
891 else
892 ttl = key->ttl;
893 ttl = ttl ? : ip6_dst_hoplimit(dst);
894 }
895 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
896 if (unlikely(err))
897 return err;
898
899 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
900 &fl6.saddr, &fl6.daddr, prio, ttl,
901 info->key.label, sport, geneve->info.key.tp_dst,
902 !(info->key.tun_flags & TUNNEL_CSUM));
903 return 0;
904 }
905 #endif
906
907 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
908 {
909 struct geneve_dev *geneve = netdev_priv(dev);
910 struct ip_tunnel_info *info = NULL;
911 int err;
912
913 if (geneve->collect_md) {
914 info = skb_tunnel_info(skb);
915 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
916 err = -EINVAL;
917 netdev_dbg(dev, "no tunnel metadata\n");
918 goto tx_error;
919 }
920 } else {
921 info = &geneve->info;
922 }
923
924 rcu_read_lock();
925 #if IS_ENABLED(CONFIG_IPV6)
926 if (info->mode & IP_TUNNEL_INFO_IPV6)
927 err = geneve6_xmit_skb(skb, dev, geneve, info);
928 else
929 #endif
930 err = geneve_xmit_skb(skb, dev, geneve, info);
931 rcu_read_unlock();
932
933 if (likely(!err))
934 return NETDEV_TX_OK;
935 tx_error:
936 dev_kfree_skb(skb);
937
938 if (err == -ELOOP)
939 dev->stats.collisions++;
940 else if (err == -ENETUNREACH)
941 dev->stats.tx_carrier_errors++;
942
943 dev->stats.tx_errors++;
944 return NETDEV_TX_OK;
945 }
946
947 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
948 {
949 if (new_mtu > dev->max_mtu)
950 new_mtu = dev->max_mtu;
951 else if (new_mtu < dev->min_mtu)
952 new_mtu = dev->min_mtu;
953
954 dev->mtu = new_mtu;
955 return 0;
956 }
957
958 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
959 {
960 struct ip_tunnel_info *info = skb_tunnel_info(skb);
961 struct geneve_dev *geneve = netdev_priv(dev);
962
963 if (ip_tunnel_info_af(info) == AF_INET) {
964 struct rtable *rt;
965 struct flowi4 fl4;
966 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
967
968 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
969 if (IS_ERR(rt))
970 return PTR_ERR(rt);
971
972 ip_rt_put(rt);
973 info->key.u.ipv4.src = fl4.saddr;
974 #if IS_ENABLED(CONFIG_IPV6)
975 } else if (ip_tunnel_info_af(info) == AF_INET6) {
976 struct dst_entry *dst;
977 struct flowi6 fl6;
978 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
979
980 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
981 if (IS_ERR(dst))
982 return PTR_ERR(dst);
983
984 dst_release(dst);
985 info->key.u.ipv6.src = fl6.saddr;
986 #endif
987 } else {
988 return -EINVAL;
989 }
990
991 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
992 1, USHRT_MAX, true);
993 info->key.tp_dst = geneve->info.key.tp_dst;
994 return 0;
995 }
996
997 static const struct net_device_ops geneve_netdev_ops = {
998 .ndo_init = geneve_init,
999 .ndo_uninit = geneve_uninit,
1000 .ndo_open = geneve_open,
1001 .ndo_stop = geneve_stop,
1002 .ndo_start_xmit = geneve_xmit,
1003 .ndo_get_stats64 = ip_tunnel_get_stats64,
1004 .ndo_change_mtu = geneve_change_mtu,
1005 .ndo_validate_addr = eth_validate_addr,
1006 .ndo_set_mac_address = eth_mac_addr,
1007 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1008 };
1009
1010 static void geneve_get_drvinfo(struct net_device *dev,
1011 struct ethtool_drvinfo *drvinfo)
1012 {
1013 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1014 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1015 }
1016
1017 static const struct ethtool_ops geneve_ethtool_ops = {
1018 .get_drvinfo = geneve_get_drvinfo,
1019 .get_link = ethtool_op_get_link,
1020 };
1021
1022 /* Info for udev, that this is a virtual tunnel endpoint */
1023 static struct device_type geneve_type = {
1024 .name = "geneve",
1025 };
1026
1027 /* Calls the ndo_udp_tunnel_add of the caller in order to
1028 * supply the listening GENEVE udp ports. Callers are expected
1029 * to implement the ndo_udp_tunnel_add.
1030 */
1031 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1032 {
1033 struct net *net = dev_net(dev);
1034 struct geneve_net *gn = net_generic(net, geneve_net_id);
1035 struct geneve_sock *gs;
1036
1037 rcu_read_lock();
1038 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1039 if (push) {
1040 udp_tunnel_push_rx_port(dev, gs->sock,
1041 UDP_TUNNEL_TYPE_GENEVE);
1042 } else {
1043 udp_tunnel_drop_rx_port(dev, gs->sock,
1044 UDP_TUNNEL_TYPE_GENEVE);
1045 }
1046 }
1047 rcu_read_unlock();
1048 }
1049
1050 /* Initialize the device structure. */
1051 static void geneve_setup(struct net_device *dev)
1052 {
1053 ether_setup(dev);
1054
1055 dev->netdev_ops = &geneve_netdev_ops;
1056 dev->ethtool_ops = &geneve_ethtool_ops;
1057 dev->needs_free_netdev = true;
1058
1059 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1060
1061 dev->features |= NETIF_F_LLTX;
1062 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1063 dev->features |= NETIF_F_RXCSUM;
1064 dev->features |= NETIF_F_GSO_SOFTWARE;
1065
1066 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1067 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1068
1069 /* MTU range: 68 - (something less than 65535) */
1070 dev->min_mtu = ETH_MIN_MTU;
1071 /* The max_mtu calculation does not take account of GENEVE
1072 * options, to avoid excluding potentially valid
1073 * configurations. This will be further reduced by IPvX hdr size.
1074 */
1075 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1076
1077 netif_keep_dst(dev);
1078 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1079 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1080 eth_hw_addr_random(dev);
1081 }
1082
1083 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1084 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1085 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1086 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1087 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1088 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1089 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1090 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1091 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1092 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1093 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1094 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1095 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1096 };
1097
1098 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1099 struct netlink_ext_ack *extack)
1100 {
1101 if (tb[IFLA_ADDRESS]) {
1102 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1103 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1104 "Provided link layer address is not Ethernet");
1105 return -EINVAL;
1106 }
1107
1108 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1109 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1110 "Provided Ethernet address is not unicast");
1111 return -EADDRNOTAVAIL;
1112 }
1113 }
1114
1115 if (!data) {
1116 NL_SET_ERR_MSG(extack,
1117 "Not enough attributes provided to perform the operation");
1118 return -EINVAL;
1119 }
1120
1121 if (data[IFLA_GENEVE_ID]) {
1122 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1123
1124 if (vni >= GENEVE_N_VID) {
1125 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1126 "Geneve ID must be lower than 16777216");
1127 return -ERANGE;
1128 }
1129 }
1130
1131 return 0;
1132 }
1133
1134 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1135 const struct ip_tunnel_info *info,
1136 bool *tun_on_same_port,
1137 bool *tun_collect_md)
1138 {
1139 struct geneve_dev *geneve, *t = NULL;
1140
1141 *tun_on_same_port = false;
1142 *tun_collect_md = false;
1143 list_for_each_entry(geneve, &gn->geneve_list, next) {
1144 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1145 *tun_collect_md = geneve->collect_md;
1146 *tun_on_same_port = true;
1147 }
1148 if (info->key.tun_id == geneve->info.key.tun_id &&
1149 info->key.tp_dst == geneve->info.key.tp_dst &&
1150 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1151 t = geneve;
1152 }
1153 return t;
1154 }
1155
1156 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1157 {
1158 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1159 info->key.ttl || info->key.label || info->key.tp_src ||
1160 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1161 }
1162
1163 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1164 struct ip_tunnel_info *b)
1165 {
1166 if (ip_tunnel_info_af(a) == AF_INET)
1167 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1168 else
1169 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1170 }
1171
1172 static int geneve_configure(struct net *net, struct net_device *dev,
1173 struct netlink_ext_ack *extack,
1174 const struct ip_tunnel_info *info,
1175 bool metadata, bool ipv6_rx_csum,
1176 bool ttl_inherit)
1177 {
1178 struct geneve_net *gn = net_generic(net, geneve_net_id);
1179 struct geneve_dev *t, *geneve = netdev_priv(dev);
1180 bool tun_collect_md, tun_on_same_port;
1181 int err, encap_len;
1182
1183 if (metadata && !is_tnl_info_zero(info)) {
1184 NL_SET_ERR_MSG(extack,
1185 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1186 return -EINVAL;
1187 }
1188
1189 geneve->net = net;
1190 geneve->dev = dev;
1191
1192 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1193 if (t)
1194 return -EBUSY;
1195
1196 /* make enough headroom for basic scenario */
1197 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1198 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1199 encap_len += sizeof(struct iphdr);
1200 dev->max_mtu -= sizeof(struct iphdr);
1201 } else {
1202 encap_len += sizeof(struct ipv6hdr);
1203 dev->max_mtu -= sizeof(struct ipv6hdr);
1204 }
1205 dev->needed_headroom = encap_len + ETH_HLEN;
1206
1207 if (metadata) {
1208 if (tun_on_same_port) {
1209 NL_SET_ERR_MSG(extack,
1210 "There can be only one externally controlled device on a destination port");
1211 return -EPERM;
1212 }
1213 } else {
1214 if (tun_collect_md) {
1215 NL_SET_ERR_MSG(extack,
1216 "There already exists an externally controlled device on this destination port");
1217 return -EPERM;
1218 }
1219 }
1220
1221 dst_cache_reset(&geneve->info.dst_cache);
1222 geneve->info = *info;
1223 geneve->collect_md = metadata;
1224 geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1225 geneve->ttl_inherit = ttl_inherit;
1226
1227 err = register_netdevice(dev);
1228 if (err)
1229 return err;
1230
1231 list_add(&geneve->next, &gn->geneve_list);
1232 return 0;
1233 }
1234
1235 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1236 {
1237 memset(info, 0, sizeof(*info));
1238 info->key.tp_dst = htons(dst_port);
1239 }
1240
1241 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1242 struct netlink_ext_ack *extack,
1243 struct ip_tunnel_info *info, bool *metadata,
1244 bool *use_udp6_rx_checksums, bool *ttl_inherit,
1245 bool changelink)
1246 {
1247 int attrtype;
1248
1249 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1250 NL_SET_ERR_MSG(extack,
1251 "Cannot specify both IPv4 and IPv6 Remote addresses");
1252 return -EINVAL;
1253 }
1254
1255 if (data[IFLA_GENEVE_REMOTE]) {
1256 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1257 attrtype = IFLA_GENEVE_REMOTE;
1258 goto change_notsup;
1259 }
1260
1261 info->key.u.ipv4.dst =
1262 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1263
1264 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1265 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1266 "Remote IPv4 address cannot be Multicast");
1267 return -EINVAL;
1268 }
1269 }
1270
1271 if (data[IFLA_GENEVE_REMOTE6]) {
1272 #if IS_ENABLED(CONFIG_IPV6)
1273 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1274 attrtype = IFLA_GENEVE_REMOTE6;
1275 goto change_notsup;
1276 }
1277
1278 info->mode = IP_TUNNEL_INFO_IPV6;
1279 info->key.u.ipv6.dst =
1280 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1281
1282 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1283 IPV6_ADDR_LINKLOCAL) {
1284 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1285 "Remote IPv6 address cannot be link-local");
1286 return -EINVAL;
1287 }
1288 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1289 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1290 "Remote IPv6 address cannot be Multicast");
1291 return -EINVAL;
1292 }
1293 info->key.tun_flags |= TUNNEL_CSUM;
1294 *use_udp6_rx_checksums = true;
1295 #else
1296 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1297 "IPv6 support not enabled in the kernel");
1298 return -EPFNOSUPPORT;
1299 #endif
1300 }
1301
1302 if (data[IFLA_GENEVE_ID]) {
1303 __u32 vni;
1304 __u8 tvni[3];
1305 __be64 tunid;
1306
1307 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1308 tvni[0] = (vni & 0x00ff0000) >> 16;
1309 tvni[1] = (vni & 0x0000ff00) >> 8;
1310 tvni[2] = vni & 0x000000ff;
1311
1312 tunid = vni_to_tunnel_id(tvni);
1313 if (changelink && (tunid != info->key.tun_id)) {
1314 attrtype = IFLA_GENEVE_ID;
1315 goto change_notsup;
1316 }
1317 info->key.tun_id = tunid;
1318 }
1319
1320 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1321 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1322 *ttl_inherit = true;
1323 else
1324 *ttl_inherit = false;
1325 } else if (data[IFLA_GENEVE_TTL]) {
1326 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1327 *ttl_inherit = false;
1328 }
1329
1330 if (data[IFLA_GENEVE_TOS])
1331 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1332
1333 if (data[IFLA_GENEVE_LABEL]) {
1334 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1335 IPV6_FLOWLABEL_MASK;
1336 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1337 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1338 "Label attribute only applies for IPv6 Geneve devices");
1339 return -EINVAL;
1340 }
1341 }
1342
1343 if (data[IFLA_GENEVE_PORT]) {
1344 if (changelink) {
1345 attrtype = IFLA_GENEVE_PORT;
1346 goto change_notsup;
1347 }
1348 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1349 }
1350
1351 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1352 if (changelink) {
1353 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1354 goto change_notsup;
1355 }
1356 *metadata = true;
1357 }
1358
1359 if (data[IFLA_GENEVE_UDP_CSUM]) {
1360 if (changelink) {
1361 attrtype = IFLA_GENEVE_UDP_CSUM;
1362 goto change_notsup;
1363 }
1364 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1365 info->key.tun_flags |= TUNNEL_CSUM;
1366 }
1367
1368 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1369 #if IS_ENABLED(CONFIG_IPV6)
1370 if (changelink) {
1371 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1372 goto change_notsup;
1373 }
1374 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1375 info->key.tun_flags &= ~TUNNEL_CSUM;
1376 #else
1377 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1378 "IPv6 support not enabled in the kernel");
1379 return -EPFNOSUPPORT;
1380 #endif
1381 }
1382
1383 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1384 #if IS_ENABLED(CONFIG_IPV6)
1385 if (changelink) {
1386 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1387 goto change_notsup;
1388 }
1389 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1390 *use_udp6_rx_checksums = false;
1391 #else
1392 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1393 "IPv6 support not enabled in the kernel");
1394 return -EPFNOSUPPORT;
1395 #endif
1396 }
1397
1398 return 0;
1399 change_notsup:
1400 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1401 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1402 return -EOPNOTSUPP;
1403 }
1404
1405 static void geneve_link_config(struct net_device *dev,
1406 struct ip_tunnel_info *info, struct nlattr *tb[])
1407 {
1408 struct geneve_dev *geneve = netdev_priv(dev);
1409 int ldev_mtu = 0;
1410
1411 if (tb[IFLA_MTU]) {
1412 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1413 return;
1414 }
1415
1416 switch (ip_tunnel_info_af(info)) {
1417 case AF_INET: {
1418 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1419 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1420
1421 if (!IS_ERR(rt) && rt->dst.dev) {
1422 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1423 ip_rt_put(rt);
1424 }
1425 break;
1426 }
1427 #if IS_ENABLED(CONFIG_IPV6)
1428 case AF_INET6: {
1429 struct rt6_info *rt = rt6_lookup(geneve->net,
1430 &info->key.u.ipv6.dst, NULL, 0,
1431 NULL, 0);
1432
1433 if (rt && rt->dst.dev)
1434 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1435 ip6_rt_put(rt);
1436 break;
1437 }
1438 #endif
1439 }
1440
1441 if (ldev_mtu <= 0)
1442 return;
1443
1444 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1445 }
1446
1447 static int geneve_newlink(struct net *net, struct net_device *dev,
1448 struct nlattr *tb[], struct nlattr *data[],
1449 struct netlink_ext_ack *extack)
1450 {
1451 bool use_udp6_rx_checksums = false;
1452 struct ip_tunnel_info info;
1453 bool ttl_inherit = false;
1454 bool metadata = false;
1455 int err;
1456
1457 init_tnl_info(&info, GENEVE_UDP_PORT);
1458 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1459 &use_udp6_rx_checksums, &ttl_inherit, false);
1460 if (err)
1461 return err;
1462
1463 err = geneve_configure(net, dev, extack, &info, metadata,
1464 use_udp6_rx_checksums, ttl_inherit);
1465 if (err)
1466 return err;
1467
1468 geneve_link_config(dev, &info, tb);
1469
1470 return 0;
1471 }
1472
1473 /* Quiesces the geneve device data path for both TX and RX.
1474 *
1475 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1476 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1477 * to complete for the existing set of in-flight packets to be transmitted,
1478 * then we would have quiesced the transmit data path. All the future packets
1479 * will get dropped until we unquiesce the data path.
1480 *
1481 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1482 * if we set that to NULL under RCU and wait for synchronize_net() to
1483 * complete, then we would have quiesced the receive data path.
1484 */
1485 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1486 struct geneve_sock **gs6)
1487 {
1488 *gs4 = rtnl_dereference(geneve->sock4);
1489 rcu_assign_pointer(geneve->sock4, NULL);
1490 if (*gs4)
1491 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1492 #if IS_ENABLED(CONFIG_IPV6)
1493 *gs6 = rtnl_dereference(geneve->sock6);
1494 rcu_assign_pointer(geneve->sock6, NULL);
1495 if (*gs6)
1496 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1497 #else
1498 *gs6 = NULL;
1499 #endif
1500 synchronize_net();
1501 }
1502
1503 /* Resumes the geneve device data path for both TX and RX. */
1504 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1505 struct geneve_sock __maybe_unused *gs6)
1506 {
1507 rcu_assign_pointer(geneve->sock4, gs4);
1508 if (gs4)
1509 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1510 #if IS_ENABLED(CONFIG_IPV6)
1511 rcu_assign_pointer(geneve->sock6, gs6);
1512 if (gs6)
1513 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1514 #endif
1515 synchronize_net();
1516 }
1517
1518 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1519 struct nlattr *data[],
1520 struct netlink_ext_ack *extack)
1521 {
1522 struct geneve_dev *geneve = netdev_priv(dev);
1523 struct geneve_sock *gs4, *gs6;
1524 struct ip_tunnel_info info;
1525 bool metadata;
1526 bool use_udp6_rx_checksums;
1527 bool ttl_inherit;
1528 int err;
1529
1530 /* If the geneve device is configured for metadata (or externally
1531 * controlled, for example, OVS), then nothing can be changed.
1532 */
1533 if (geneve->collect_md)
1534 return -EOPNOTSUPP;
1535
1536 /* Start with the existing info. */
1537 memcpy(&info, &geneve->info, sizeof(info));
1538 metadata = geneve->collect_md;
1539 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1540 ttl_inherit = geneve->ttl_inherit;
1541 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1542 &use_udp6_rx_checksums, &ttl_inherit, true);
1543 if (err)
1544 return err;
1545
1546 if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1547 dst_cache_reset(&info.dst_cache);
1548 geneve_link_config(dev, &info, tb);
1549 }
1550
1551 geneve_quiesce(geneve, &gs4, &gs6);
1552 geneve->info = info;
1553 geneve->collect_md = metadata;
1554 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1555 geneve->ttl_inherit = ttl_inherit;
1556 geneve_unquiesce(geneve, gs4, gs6);
1557
1558 return 0;
1559 }
1560
1561 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1562 {
1563 struct geneve_dev *geneve = netdev_priv(dev);
1564
1565 list_del(&geneve->next);
1566 unregister_netdevice_queue(dev, head);
1567 }
1568
1569 static size_t geneve_get_size(const struct net_device *dev)
1570 {
1571 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1572 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1573 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1574 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1575 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1576 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1577 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1578 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1579 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1580 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1581 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1582 0;
1583 }
1584
1585 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1586 {
1587 struct geneve_dev *geneve = netdev_priv(dev);
1588 struct ip_tunnel_info *info = &geneve->info;
1589 bool ttl_inherit = geneve->ttl_inherit;
1590 bool metadata = geneve->collect_md;
1591 __u8 tmp_vni[3];
1592 __u32 vni;
1593
1594 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1595 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1596 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1597 goto nla_put_failure;
1598
1599 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1600 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1601 info->key.u.ipv4.dst))
1602 goto nla_put_failure;
1603 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1604 !!(info->key.tun_flags & TUNNEL_CSUM)))
1605 goto nla_put_failure;
1606
1607 #if IS_ENABLED(CONFIG_IPV6)
1608 } else if (!metadata) {
1609 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1610 &info->key.u.ipv6.dst))
1611 goto nla_put_failure;
1612 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1613 !(info->key.tun_flags & TUNNEL_CSUM)))
1614 goto nla_put_failure;
1615 #endif
1616 }
1617
1618 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1619 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1620 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1621 goto nla_put_failure;
1622
1623 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1624 goto nla_put_failure;
1625
1626 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1627 goto nla_put_failure;
1628
1629 #if IS_ENABLED(CONFIG_IPV6)
1630 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1631 !geneve->use_udp6_rx_checksums))
1632 goto nla_put_failure;
1633 #endif
1634
1635 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1636 goto nla_put_failure;
1637
1638 return 0;
1639
1640 nla_put_failure:
1641 return -EMSGSIZE;
1642 }
1643
1644 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1645 .kind = "geneve",
1646 .maxtype = IFLA_GENEVE_MAX,
1647 .policy = geneve_policy,
1648 .priv_size = sizeof(struct geneve_dev),
1649 .setup = geneve_setup,
1650 .validate = geneve_validate,
1651 .newlink = geneve_newlink,
1652 .changelink = geneve_changelink,
1653 .dellink = geneve_dellink,
1654 .get_size = geneve_get_size,
1655 .fill_info = geneve_fill_info,
1656 };
1657
1658 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1659 u8 name_assign_type, u16 dst_port)
1660 {
1661 struct nlattr *tb[IFLA_MAX + 1];
1662 struct ip_tunnel_info info;
1663 struct net_device *dev;
1664 LIST_HEAD(list_kill);
1665 int err;
1666
1667 memset(tb, 0, sizeof(tb));
1668 dev = rtnl_create_link(net, name, name_assign_type,
1669 &geneve_link_ops, tb);
1670 if (IS_ERR(dev))
1671 return dev;
1672
1673 init_tnl_info(&info, dst_port);
1674 err = geneve_configure(net, dev, NULL, &info, true, true, false);
1675 if (err) {
1676 free_netdev(dev);
1677 return ERR_PTR(err);
1678 }
1679
1680 /* openvswitch users expect packet sizes to be unrestricted,
1681 * so set the largest MTU we can.
1682 */
1683 err = geneve_change_mtu(dev, IP_MAX_MTU);
1684 if (err)
1685 goto err;
1686
1687 err = rtnl_configure_link(dev, NULL);
1688 if (err < 0)
1689 goto err;
1690
1691 return dev;
1692 err:
1693 geneve_dellink(dev, &list_kill);
1694 unregister_netdevice_many(&list_kill);
1695 return ERR_PTR(err);
1696 }
1697 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1698
1699 static int geneve_netdevice_event(struct notifier_block *unused,
1700 unsigned long event, void *ptr)
1701 {
1702 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1703
1704 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1705 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1706 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1707 } else if (event == NETDEV_UNREGISTER) {
1708 geneve_offload_rx_ports(dev, false);
1709 } else if (event == NETDEV_REGISTER) {
1710 geneve_offload_rx_ports(dev, true);
1711 }
1712
1713 return NOTIFY_DONE;
1714 }
1715
1716 static struct notifier_block geneve_notifier_block __read_mostly = {
1717 .notifier_call = geneve_netdevice_event,
1718 };
1719
1720 static __net_init int geneve_init_net(struct net *net)
1721 {
1722 struct geneve_net *gn = net_generic(net, geneve_net_id);
1723
1724 INIT_LIST_HEAD(&gn->geneve_list);
1725 INIT_LIST_HEAD(&gn->sock_list);
1726 return 0;
1727 }
1728
1729 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1730 {
1731 struct geneve_net *gn = net_generic(net, geneve_net_id);
1732 struct geneve_dev *geneve, *next;
1733 struct net_device *dev, *aux;
1734
1735 /* gather any geneve devices that were moved into this ns */
1736 for_each_netdev_safe(net, dev, aux)
1737 if (dev->rtnl_link_ops == &geneve_link_ops)
1738 unregister_netdevice_queue(dev, head);
1739
1740 /* now gather any other geneve devices that were created in this ns */
1741 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1742 /* If geneve->dev is in the same netns, it was already added
1743 * to the list by the previous loop.
1744 */
1745 if (!net_eq(dev_net(geneve->dev), net))
1746 unregister_netdevice_queue(geneve->dev, head);
1747 }
1748
1749 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1750 }
1751
1752 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1753 {
1754 struct net *net;
1755 LIST_HEAD(list);
1756
1757 rtnl_lock();
1758 list_for_each_entry(net, net_list, exit_list)
1759 geneve_destroy_tunnels(net, &list);
1760
1761 /* unregister the devices gathered above */
1762 unregister_netdevice_many(&list);
1763 rtnl_unlock();
1764 }
1765
1766 static struct pernet_operations geneve_net_ops = {
1767 .init = geneve_init_net,
1768 .exit_batch = geneve_exit_batch_net,
1769 .id = &geneve_net_id,
1770 .size = sizeof(struct geneve_net),
1771 };
1772
1773 static int __init geneve_init_module(void)
1774 {
1775 int rc;
1776
1777 rc = register_pernet_subsys(&geneve_net_ops);
1778 if (rc)
1779 goto out1;
1780
1781 rc = register_netdevice_notifier(&geneve_notifier_block);
1782 if (rc)
1783 goto out2;
1784
1785 rc = rtnl_link_register(&geneve_link_ops);
1786 if (rc)
1787 goto out3;
1788
1789 return 0;
1790 out3:
1791 unregister_netdevice_notifier(&geneve_notifier_block);
1792 out2:
1793 unregister_pernet_subsys(&geneve_net_ops);
1794 out1:
1795 return rc;
1796 }
1797 late_initcall(geneve_init_module);
1798
1799 static void __exit geneve_cleanup_module(void)
1800 {
1801 rtnl_link_unregister(&geneve_link_ops);
1802 unregister_netdevice_notifier(&geneve_notifier_block);
1803 unregister_pernet_subsys(&geneve_net_ops);
1804 }
1805 module_exit(geneve_cleanup_module);
1806
1807 MODULE_LICENSE("GPL");
1808 MODULE_VERSION(GENEVE_NETDEV_VER);
1809 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1810 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1811 MODULE_ALIAS_RTNL_LINK("geneve");
Linux with added FPC-III support