1 // SPDX-License-Identifier: GPL-2.0-only
3 * GENEVE: Generic Network Virtualization Encapsulation
5 * Copyright (c) 2015 Red Hat, Inc.
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/kernel.h>
11 #include <linux/module.h>
12 #include <linux/etherdevice.h>
13 #include <linux/hash.h>
14 #include <net/ipv6_stubs.h>
15 #include <net/dst_metadata.h>
16 #include <net/gro_cells.h>
17 #include <net/rtnetlink.h>
18 #include <net/geneve.h>
19 #include <net/protocol.h>
21 #define GENEVE_NETDEV_VER "0.6"
23 #define GENEVE_N_VID (1u << 24)
24 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
26 #define VNI_HASH_BITS 10
27 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
29 static bool log_ecn_error = true;
30 module_param(log_ecn_error, bool, 0644);
31 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
34 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
35 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
36 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
38 /* per-network namespace private data for this module */
40 struct list_head geneve_list;
41 struct list_head sock_list;
44 static unsigned int geneve_net_id;
46 struct geneve_dev_node {
47 struct hlist_node hlist;
48 struct geneve_dev *geneve;
51 struct geneve_config {
52 struct ip_tunnel_info info;
54 bool use_udp6_rx_checksums;
56 enum ifla_geneve_df df;
59 /* Pseudo network device */
61 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
62 #if IS_ENABLED(CONFIG_IPV6)
63 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
65 struct net *net; /* netns for packet i/o */
66 struct net_device *dev; /* netdev for geneve tunnel */
67 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
68 #if IS_ENABLED(CONFIG_IPV6)
69 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
71 struct list_head next; /* geneve's per namespace list */
72 struct gro_cells gro_cells;
73 struct geneve_config cfg;
78 struct list_head list;
82 struct hlist_head vni_list[VNI_HASH_SIZE];
85 static inline __u32 geneve_net_vni_hash(u8 vni[3])
89 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
90 return hash_32(vnid, VNI_HASH_BITS);
93 static __be64 vni_to_tunnel_id(const __u8 *vni)
96 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
98 return (__force __be64)(((__force u64)vni[0] << 40) |
99 ((__force u64)vni[1] << 48) |
100 ((__force u64)vni[2] << 56));
104 /* Convert 64 bit tunnel ID to 24 bit VNI. */
105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
108 vni[0] = (__force __u8)(tun_id >> 16);
109 vni[1] = (__force __u8)(tun_id >> 8);
110 vni[2] = (__force __u8)tun_id;
112 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
113 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
114 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
118 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
120 return !memcmp(vni, &tun_id[5], 3);
123 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
125 return gs->sock->sk->sk_family;
128 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
129 __be32 addr, u8 vni[])
131 struct hlist_head *vni_list_head;
132 struct geneve_dev_node *node;
135 /* Find the device for this VNI */
136 hash = geneve_net_vni_hash(vni);
137 vni_list_head = &gs->vni_list[hash];
138 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
139 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
140 addr == node->geneve->cfg.info.key.u.ipv4.dst)
146 #if IS_ENABLED(CONFIG_IPV6)
147 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
148 struct in6_addr addr6, u8 vni[])
150 struct hlist_head *vni_list_head;
151 struct geneve_dev_node *node;
154 /* Find the device for this VNI */
155 hash = geneve_net_vni_hash(vni);
156 vni_list_head = &gs->vni_list[hash];
157 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
158 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
159 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
166 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
168 return (struct genevehdr *)(udp_hdr(skb) + 1);
171 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
174 static u8 zero_vni[3];
177 if (geneve_get_sk_family(gs) == AF_INET) {
181 iph = ip_hdr(skb); /* outer IP header... */
183 if (gs->collect_md) {
187 vni = geneve_hdr(skb)->vni;
191 return geneve_lookup(gs, addr, vni);
192 #if IS_ENABLED(CONFIG_IPV6)
193 } else if (geneve_get_sk_family(gs) == AF_INET6) {
194 static struct in6_addr zero_addr6;
195 struct ipv6hdr *ip6h;
196 struct in6_addr addr6;
198 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
200 if (gs->collect_md) {
204 vni = geneve_hdr(skb)->vni;
208 return geneve6_lookup(gs, addr6, vni);
214 /* geneve receive/decap routine */
215 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
218 struct genevehdr *gnvh = geneve_hdr(skb);
219 struct metadata_dst *tun_dst = NULL;
224 if (ip_tunnel_collect_metadata() || gs->collect_md) {
227 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
228 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
230 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
231 vni_to_tunnel_id(gnvh->vni),
234 geneve->dev->stats.rx_dropped++;
237 /* Update tunnel dst according to Geneve options. */
238 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
239 gnvh->options, gnvh->opt_len * 4,
242 /* Drop packets w/ critical options,
243 * since we don't support any...
245 if (gnvh->critical) {
246 geneve->dev->stats.rx_frame_errors++;
247 geneve->dev->stats.rx_errors++;
252 skb_reset_mac_header(skb);
253 skb->protocol = eth_type_trans(skb, geneve->dev);
254 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
257 skb_dst_set(skb, &tun_dst->dst);
259 /* Ignore packet loops (and multicast echo) */
260 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
261 geneve->dev->stats.rx_errors++;
265 /* Save offset of outer header relative to skb->head,
266 * because we are going to reset the network header to the inner header
267 * and might change skb->head.
269 nh = skb_network_header(skb) - skb->head;
271 skb_reset_network_header(skb);
273 if (!pskb_inet_may_pull(skb)) {
274 DEV_STATS_INC(geneve->dev, rx_length_errors);
275 DEV_STATS_INC(geneve->dev, rx_errors);
279 /* Get the outer header. */
280 oiph = skb->head + nh;
282 if (geneve_get_sk_family(gs) == AF_INET)
283 err = IP_ECN_decapsulate(oiph, skb);
284 #if IS_ENABLED(CONFIG_IPV6)
286 err = IP6_ECN_decapsulate(oiph, skb);
291 if (geneve_get_sk_family(gs) == AF_INET)
292 net_info_ratelimited("non-ECT from %pI4 "
294 &((struct iphdr *)oiph)->saddr,
295 ((struct iphdr *)oiph)->tos);
296 #if IS_ENABLED(CONFIG_IPV6)
298 net_info_ratelimited("non-ECT from %pI6\n",
299 &((struct ipv6hdr *)oiph)->saddr);
303 ++geneve->dev->stats.rx_frame_errors;
304 ++geneve->dev->stats.rx_errors;
310 err = gro_cells_receive(&geneve->gro_cells, skb);
311 if (likely(err == NET_RX_SUCCESS))
312 dev_sw_netstats_rx_add(geneve->dev, len);
316 /* Consume bad packet */
320 /* Setup stats when device is created */
321 static int geneve_init(struct net_device *dev)
323 struct geneve_dev *geneve = netdev_priv(dev);
326 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
330 err = gro_cells_init(&geneve->gro_cells, dev);
332 free_percpu(dev->tstats);
336 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
338 free_percpu(dev->tstats);
339 gro_cells_destroy(&geneve->gro_cells);
345 static void geneve_uninit(struct net_device *dev)
347 struct geneve_dev *geneve = netdev_priv(dev);
349 dst_cache_destroy(&geneve->cfg.info.dst_cache);
350 gro_cells_destroy(&geneve->gro_cells);
351 free_percpu(dev->tstats);
354 /* Callback from net/ipv4/udp.c to receive packets */
355 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
357 struct genevehdr *geneveh;
358 struct geneve_dev *geneve;
359 struct geneve_sock *gs;
362 /* Need UDP and Geneve header to be present */
363 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
366 /* Return packets with reserved bits set */
367 geneveh = geneve_hdr(skb);
368 if (unlikely(geneveh->ver != GENEVE_VER))
371 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
374 gs = rcu_dereference_sk_user_data(sk);
378 geneve = geneve_lookup_skb(gs, skb);
382 opts_len = geneveh->opt_len * 4;
383 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
385 !net_eq(geneve->net, dev_net(geneve->dev)))) {
386 geneve->dev->stats.rx_dropped++;
390 geneve_rx(geneve, gs, skb);
394 /* Consume bad packet */
399 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
400 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
402 struct genevehdr *geneveh;
403 struct geneve_sock *gs;
404 u8 zero_vni[3] = { 0 };
407 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
410 geneveh = geneve_hdr(skb);
411 if (geneveh->ver != GENEVE_VER)
414 if (geneveh->proto_type != htons(ETH_P_TEB))
417 gs = rcu_dereference_sk_user_data(sk);
421 if (geneve_get_sk_family(gs) == AF_INET) {
422 struct iphdr *iph = ip_hdr(skb);
425 if (!gs->collect_md) {
426 vni = geneve_hdr(skb)->vni;
430 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
433 #if IS_ENABLED(CONFIG_IPV6)
434 if (geneve_get_sk_family(gs) == AF_INET6) {
435 struct ipv6hdr *ip6h = ipv6_hdr(skb);
436 struct in6_addr addr6;
438 memset(&addr6, 0, sizeof(struct in6_addr));
440 if (!gs->collect_md) {
441 vni = geneve_hdr(skb)->vni;
445 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
449 return -EPFNOSUPPORT;
452 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
453 __be16 port, bool ipv6_rx_csum)
456 struct udp_port_cfg udp_conf;
459 memset(&udp_conf, 0, sizeof(udp_conf));
462 udp_conf.family = AF_INET6;
463 udp_conf.ipv6_v6only = 1;
464 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
466 udp_conf.family = AF_INET;
467 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
470 udp_conf.local_udp_port = port;
472 /* Open UDP socket */
473 err = udp_sock_create(net, &udp_conf, &sock);
480 static int geneve_hlen(struct genevehdr *gh)
482 return sizeof(*gh) + gh->opt_len * 4;
485 static struct sk_buff *geneve_gro_receive(struct sock *sk,
486 struct list_head *head,
489 struct sk_buff *pp = NULL;
491 struct genevehdr *gh, *gh2;
492 unsigned int hlen, gh_len, off_gnv;
493 const struct packet_offload *ptype;
497 off_gnv = skb_gro_offset(skb);
498 hlen = off_gnv + sizeof(*gh);
499 gh = skb_gro_header_fast(skb, off_gnv);
500 if (skb_gro_header_hard(skb, hlen)) {
501 gh = skb_gro_header_slow(skb, hlen, off_gnv);
506 if (gh->ver != GENEVE_VER || gh->oam)
508 gh_len = geneve_hlen(gh);
510 hlen = off_gnv + gh_len;
511 if (skb_gro_header_hard(skb, hlen)) {
512 gh = skb_gro_header_slow(skb, hlen, off_gnv);
517 list_for_each_entry(p, head, list) {
518 if (!NAPI_GRO_CB(p)->same_flow)
521 gh2 = (struct genevehdr *)(p->data + off_gnv);
522 if (gh->opt_len != gh2->opt_len ||
523 memcmp(gh, gh2, gh_len)) {
524 NAPI_GRO_CB(p)->same_flow = 0;
529 type = gh->proto_type;
532 ptype = gro_find_receive_by_type(type);
536 skb_gro_pull(skb, gh_len);
537 skb_gro_postpull_rcsum(skb, gh, gh_len);
538 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
544 skb_gro_flush_final(skb, pp, flush);
549 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
552 struct genevehdr *gh;
553 struct packet_offload *ptype;
558 gh = (struct genevehdr *)(skb->data + nhoff);
559 gh_len = geneve_hlen(gh);
560 type = gh->proto_type;
563 ptype = gro_find_complete_by_type(type);
565 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
569 skb_set_inner_mac_header(skb, nhoff + gh_len);
574 /* Create new listen socket if needed */
575 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
576 bool ipv6, bool ipv6_rx_csum)
578 struct geneve_net *gn = net_generic(net, geneve_net_id);
579 struct geneve_sock *gs;
581 struct udp_tunnel_sock_cfg tunnel_cfg;
584 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
586 return ERR_PTR(-ENOMEM);
588 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
591 return ERR_CAST(sock);
596 for (h = 0; h < VNI_HASH_SIZE; ++h)
597 INIT_HLIST_HEAD(&gs->vni_list[h]);
599 /* Initialize the geneve udp offloads structure */
600 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
602 /* Mark socket as an encapsulation socket */
603 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
604 tunnel_cfg.sk_user_data = gs;
605 tunnel_cfg.encap_type = 1;
606 tunnel_cfg.gro_receive = geneve_gro_receive;
607 tunnel_cfg.gro_complete = geneve_gro_complete;
608 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
609 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
610 tunnel_cfg.encap_destroy = NULL;
611 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
612 list_add(&gs->list, &gn->sock_list);
616 static void __geneve_sock_release(struct geneve_sock *gs)
618 if (!gs || --gs->refcnt)
622 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
623 udp_tunnel_sock_release(gs->sock);
627 static void geneve_sock_release(struct geneve_dev *geneve)
629 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
630 #if IS_ENABLED(CONFIG_IPV6)
631 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
633 rcu_assign_pointer(geneve->sock6, NULL);
636 rcu_assign_pointer(geneve->sock4, NULL);
639 __geneve_sock_release(gs4);
640 #if IS_ENABLED(CONFIG_IPV6)
641 __geneve_sock_release(gs6);
645 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
649 struct geneve_sock *gs;
651 list_for_each_entry(gs, &gn->sock_list, list) {
652 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
653 geneve_get_sk_family(gs) == family) {
660 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
662 struct net *net = geneve->net;
663 struct geneve_net *gn = net_generic(net, geneve_net_id);
664 struct geneve_dev_node *node;
665 struct geneve_sock *gs;
669 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
675 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
676 geneve->cfg.use_udp6_rx_checksums);
681 gs->collect_md = geneve->cfg.collect_md;
682 #if IS_ENABLED(CONFIG_IPV6)
684 rcu_assign_pointer(geneve->sock6, gs);
685 node = &geneve->hlist6;
689 rcu_assign_pointer(geneve->sock4, gs);
690 node = &geneve->hlist4;
692 node->geneve = geneve;
694 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
695 hash = geneve_net_vni_hash(vni);
696 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
700 static int geneve_open(struct net_device *dev)
702 struct geneve_dev *geneve = netdev_priv(dev);
703 bool metadata = geneve->cfg.collect_md;
707 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
708 ipv4 = !ipv6 || metadata;
709 #if IS_ENABLED(CONFIG_IPV6)
711 ret = geneve_sock_add(geneve, true);
712 if (ret < 0 && ret != -EAFNOSUPPORT)
717 ret = geneve_sock_add(geneve, false);
719 geneve_sock_release(geneve);
724 static int geneve_stop(struct net_device *dev)
726 struct geneve_dev *geneve = netdev_priv(dev);
728 hlist_del_init_rcu(&geneve->hlist4.hlist);
729 #if IS_ENABLED(CONFIG_IPV6)
730 hlist_del_init_rcu(&geneve->hlist6.hlist);
732 geneve_sock_release(geneve);
736 static void geneve_build_header(struct genevehdr *geneveh,
737 const struct ip_tunnel_info *info)
739 geneveh->ver = GENEVE_VER;
740 geneveh->opt_len = info->options_len / 4;
741 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
742 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
744 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
745 geneveh->proto_type = htons(ETH_P_TEB);
748 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
749 ip_tunnel_info_opts_get(geneveh->options, info);
752 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
753 const struct ip_tunnel_info *info,
754 bool xnet, int ip_hdr_len)
756 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
757 struct genevehdr *gnvh;
761 skb_reset_mac_header(skb);
762 skb_scrub_packet(skb, xnet);
764 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
765 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
766 err = skb_cow_head(skb, min_headroom);
770 err = udp_tunnel_handle_offloads(skb, udp_sum);
774 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
775 geneve_build_header(gnvh, info);
776 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
784 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
785 struct net_device *dev,
786 struct geneve_sock *gs4,
788 const struct ip_tunnel_info *info,
789 __be16 dport, __be16 sport,
792 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
793 struct geneve_dev *geneve = netdev_priv(dev);
794 struct dst_cache *dst_cache;
795 struct rtable *rt = NULL;
799 return ERR_PTR(-EIO);
801 memset(fl4, 0, sizeof(*fl4));
802 fl4->flowi4_mark = skb->mark;
803 fl4->flowi4_proto = IPPROTO_UDP;
804 fl4->daddr = info->key.u.ipv4.dst;
805 fl4->saddr = info->key.u.ipv4.src;
806 fl4->fl4_dport = dport;
807 fl4->fl4_sport = sport;
810 if ((tos == 1) && !geneve->cfg.collect_md) {
811 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
814 fl4->flowi4_tos = RT_TOS(tos);
818 dst_cache = (struct dst_cache *)&info->dst_cache;
820 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
824 rt = ip_route_output_key(geneve->net, fl4);
826 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
827 return ERR_PTR(-ENETUNREACH);
829 if (rt->dst.dev == dev) { /* is this necessary? */
830 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
832 return ERR_PTR(-ELOOP);
835 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
839 #if IS_ENABLED(CONFIG_IPV6)
840 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
841 struct net_device *dev,
842 struct geneve_sock *gs6,
844 const struct ip_tunnel_info *info,
845 __be16 dport, __be16 sport)
847 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
848 struct geneve_dev *geneve = netdev_priv(dev);
849 struct dst_entry *dst = NULL;
850 struct dst_cache *dst_cache;
854 return ERR_PTR(-EIO);
856 memset(fl6, 0, sizeof(*fl6));
857 fl6->flowi6_mark = skb->mark;
858 fl6->flowi6_proto = IPPROTO_UDP;
859 fl6->daddr = info->key.u.ipv6.dst;
860 fl6->saddr = info->key.u.ipv6.src;
861 fl6->fl6_dport = dport;
862 fl6->fl6_sport = sport;
864 prio = info->key.tos;
865 if ((prio == 1) && !geneve->cfg.collect_md) {
866 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
870 fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
871 dst_cache = (struct dst_cache *)&info->dst_cache;
873 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
877 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
880 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
881 return ERR_PTR(-ENETUNREACH);
883 if (dst->dev == dev) { /* is this necessary? */
884 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
886 return ERR_PTR(-ELOOP);
890 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
895 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
896 struct geneve_dev *geneve,
897 const struct ip_tunnel_info *info)
899 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
900 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
901 const struct ip_tunnel_key *key = &info->key;
910 if (!pskb_inet_may_pull(skb))
913 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
914 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
915 geneve->cfg.info.key.tp_dst, sport, &full_tos);
919 err = skb_tunnel_check_pmtu(skb, &rt->dst,
920 GENEVE_IPV4_HLEN + info->options_len,
921 netif_is_any_bridge_port(dev));
923 dst_release(&rt->dst);
926 struct ip_tunnel_info *info;
928 info = skb_tunnel_info(skb);
930 struct ip_tunnel_info *unclone;
932 unclone = skb_tunnel_info_unclone(skb);
933 if (unlikely(!unclone)) {
934 dst_release(&rt->dst);
938 unclone->key.u.ipv4.dst = fl4.saddr;
939 unclone->key.u.ipv4.src = fl4.daddr;
942 if (!pskb_may_pull(skb, ETH_HLEN)) {
943 dst_release(&rt->dst);
947 skb->protocol = eth_type_trans(skb, geneve->dev);
949 dst_release(&rt->dst);
953 if (geneve->cfg.collect_md) {
954 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
957 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
959 tos = ip_tunnel_ecn_encap(full_tos, ip_hdr(skb), skb);
960 if (geneve->cfg.ttl_inherit)
961 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
964 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
966 if (geneve->cfg.df == GENEVE_DF_SET) {
968 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
969 struct ethhdr *eth = eth_hdr(skb);
971 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
973 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
974 struct iphdr *iph = ip_hdr(skb);
976 if (iph->frag_off & htons(IP_DF))
982 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
986 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
987 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
988 !net_eq(geneve->net, dev_net(geneve->dev)),
989 !(info->key.tun_flags & TUNNEL_CSUM));
993 #if IS_ENABLED(CONFIG_IPV6)
994 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
995 struct geneve_dev *geneve,
996 const struct ip_tunnel_info *info)
998 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
999 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1000 const struct ip_tunnel_key *key = &info->key;
1001 struct dst_entry *dst = NULL;
1007 if (!pskb_inet_may_pull(skb))
1010 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1011 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1012 geneve->cfg.info.key.tp_dst, sport);
1014 return PTR_ERR(dst);
1016 err = skb_tunnel_check_pmtu(skb, dst,
1017 GENEVE_IPV6_HLEN + info->options_len,
1018 netif_is_any_bridge_port(dev));
1023 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1026 struct ip_tunnel_info *unclone;
1028 unclone = skb_tunnel_info_unclone(skb);
1029 if (unlikely(!unclone)) {
1034 unclone->key.u.ipv6.dst = fl6.saddr;
1035 unclone->key.u.ipv6.src = fl6.daddr;
1038 if (!pskb_may_pull(skb, ETH_HLEN)) {
1043 skb->protocol = eth_type_trans(skb, geneve->dev);
1049 if (geneve->cfg.collect_md) {
1050 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1053 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1055 if (geneve->cfg.ttl_inherit)
1056 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1059 ttl = ttl ? : ip6_dst_hoplimit(dst);
1061 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1065 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1066 &fl6.saddr, &fl6.daddr, prio, ttl,
1067 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1068 !(info->key.tun_flags & TUNNEL_CSUM));
1073 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1075 struct geneve_dev *geneve = netdev_priv(dev);
1076 struct ip_tunnel_info *info = NULL;
1079 if (geneve->cfg.collect_md) {
1080 info = skb_tunnel_info(skb);
1081 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1082 netdev_dbg(dev, "no tunnel metadata\n");
1084 dev->stats.tx_dropped++;
1085 return NETDEV_TX_OK;
1088 info = &geneve->cfg.info;
1092 #if IS_ENABLED(CONFIG_IPV6)
1093 if (info->mode & IP_TUNNEL_INFO_IPV6)
1094 err = geneve6_xmit_skb(skb, dev, geneve, info);
1097 err = geneve_xmit_skb(skb, dev, geneve, info);
1101 return NETDEV_TX_OK;
1103 if (err != -EMSGSIZE)
1107 dev->stats.collisions++;
1108 else if (err == -ENETUNREACH)
1109 dev->stats.tx_carrier_errors++;
1111 dev->stats.tx_errors++;
1112 return NETDEV_TX_OK;
1115 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1117 if (new_mtu > dev->max_mtu)
1118 new_mtu = dev->max_mtu;
1119 else if (new_mtu < dev->min_mtu)
1120 new_mtu = dev->min_mtu;
1126 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1128 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1129 struct geneve_dev *geneve = netdev_priv(dev);
1132 if (ip_tunnel_info_af(info) == AF_INET) {
1136 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1137 sport = udp_flow_src_port(geneve->net, skb,
1138 1, USHRT_MAX, true);
1140 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1141 geneve->cfg.info.key.tp_dst, sport, NULL);
1146 info->key.u.ipv4.src = fl4.saddr;
1147 #if IS_ENABLED(CONFIG_IPV6)
1148 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1149 struct dst_entry *dst;
1152 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1153 sport = udp_flow_src_port(geneve->net, skb,
1154 1, USHRT_MAX, true);
1156 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1157 geneve->cfg.info.key.tp_dst, sport);
1159 return PTR_ERR(dst);
1162 info->key.u.ipv6.src = fl6.saddr;
1168 info->key.tp_src = sport;
1169 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1173 static const struct net_device_ops geneve_netdev_ops = {
1174 .ndo_init = geneve_init,
1175 .ndo_uninit = geneve_uninit,
1176 .ndo_open = geneve_open,
1177 .ndo_stop = geneve_stop,
1178 .ndo_start_xmit = geneve_xmit,
1179 .ndo_get_stats64 = ip_tunnel_get_stats64,
1180 .ndo_change_mtu = geneve_change_mtu,
1181 .ndo_validate_addr = eth_validate_addr,
1182 .ndo_set_mac_address = eth_mac_addr,
1183 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1186 static void geneve_get_drvinfo(struct net_device *dev,
1187 struct ethtool_drvinfo *drvinfo)
1189 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1190 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1193 static const struct ethtool_ops geneve_ethtool_ops = {
1194 .get_drvinfo = geneve_get_drvinfo,
1195 .get_link = ethtool_op_get_link,
1198 /* Info for udev, that this is a virtual tunnel endpoint */
1199 static struct device_type geneve_type = {
1203 /* Calls the ndo_udp_tunnel_add of the caller in order to
1204 * supply the listening GENEVE udp ports. Callers are expected
1205 * to implement the ndo_udp_tunnel_add.
1207 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1209 struct net *net = dev_net(dev);
1210 struct geneve_net *gn = net_generic(net, geneve_net_id);
1211 struct geneve_sock *gs;
1214 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1216 udp_tunnel_push_rx_port(dev, gs->sock,
1217 UDP_TUNNEL_TYPE_GENEVE);
1219 udp_tunnel_drop_rx_port(dev, gs->sock,
1220 UDP_TUNNEL_TYPE_GENEVE);
1226 /* Initialize the device structure. */
1227 static void geneve_setup(struct net_device *dev)
1231 dev->netdev_ops = &geneve_netdev_ops;
1232 dev->ethtool_ops = &geneve_ethtool_ops;
1233 dev->needs_free_netdev = true;
1235 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1237 dev->features |= NETIF_F_LLTX;
1238 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1239 dev->features |= NETIF_F_RXCSUM;
1240 dev->features |= NETIF_F_GSO_SOFTWARE;
1242 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1243 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1245 /* MTU range: 68 - (something less than 65535) */
1246 dev->min_mtu = ETH_MIN_MTU;
1247 /* The max_mtu calculation does not take account of GENEVE
1248 * options, to avoid excluding potentially valid
1249 * configurations. This will be further reduced by IPvX hdr size.
1251 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1253 netif_keep_dst(dev);
1254 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1255 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1256 eth_hw_addr_random(dev);
1259 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1260 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1261 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1262 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1263 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1264 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1265 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1266 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1267 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1268 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1269 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1270 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1271 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1272 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1275 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1276 struct netlink_ext_ack *extack)
1278 if (tb[IFLA_ADDRESS]) {
1279 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1280 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1281 "Provided link layer address is not Ethernet");
1285 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1286 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1287 "Provided Ethernet address is not unicast");
1288 return -EADDRNOTAVAIL;
1293 NL_SET_ERR_MSG(extack,
1294 "Not enough attributes provided to perform the operation");
1298 if (data[IFLA_GENEVE_ID]) {
1299 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1301 if (vni >= GENEVE_N_VID) {
1302 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1303 "Geneve ID must be lower than 16777216");
1308 if (data[IFLA_GENEVE_DF]) {
1309 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1311 if (df < 0 || df > GENEVE_DF_MAX) {
1312 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1313 "Invalid DF attribute");
1321 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1322 const struct ip_tunnel_info *info,
1323 bool *tun_on_same_port,
1324 bool *tun_collect_md)
1326 struct geneve_dev *geneve, *t = NULL;
1328 *tun_on_same_port = false;
1329 *tun_collect_md = false;
1330 list_for_each_entry(geneve, &gn->geneve_list, next) {
1331 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1332 *tun_collect_md = geneve->cfg.collect_md;
1333 *tun_on_same_port = true;
1335 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1336 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1337 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1343 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1345 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1346 info->key.ttl || info->key.label || info->key.tp_src ||
1347 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1350 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1351 struct ip_tunnel_info *b)
1353 if (ip_tunnel_info_af(a) == AF_INET)
1354 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1356 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1359 static int geneve_configure(struct net *net, struct net_device *dev,
1360 struct netlink_ext_ack *extack,
1361 const struct geneve_config *cfg)
1363 struct geneve_net *gn = net_generic(net, geneve_net_id);
1364 struct geneve_dev *t, *geneve = netdev_priv(dev);
1365 const struct ip_tunnel_info *info = &cfg->info;
1366 bool tun_collect_md, tun_on_same_port;
1369 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1370 NL_SET_ERR_MSG(extack,
1371 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1378 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1382 /* make enough headroom for basic scenario */
1383 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1384 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1385 encap_len += sizeof(struct iphdr);
1386 dev->max_mtu -= sizeof(struct iphdr);
1388 encap_len += sizeof(struct ipv6hdr);
1389 dev->max_mtu -= sizeof(struct ipv6hdr);
1391 dev->needed_headroom = encap_len + ETH_HLEN;
1393 if (cfg->collect_md) {
1394 if (tun_on_same_port) {
1395 NL_SET_ERR_MSG(extack,
1396 "There can be only one externally controlled device on a destination port");
1400 if (tun_collect_md) {
1401 NL_SET_ERR_MSG(extack,
1402 "There already exists an externally controlled device on this destination port");
1407 dst_cache_reset(&geneve->cfg.info.dst_cache);
1408 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1410 err = register_netdevice(dev);
1414 list_add(&geneve->next, &gn->geneve_list);
1418 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1420 memset(info, 0, sizeof(*info));
1421 info->key.tp_dst = htons(dst_port);
1424 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1425 struct netlink_ext_ack *extack,
1426 struct geneve_config *cfg, bool changelink)
1428 struct ip_tunnel_info *info = &cfg->info;
1431 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1432 NL_SET_ERR_MSG(extack,
1433 "Cannot specify both IPv4 and IPv6 Remote addresses");
1437 if (data[IFLA_GENEVE_REMOTE]) {
1438 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1439 attrtype = IFLA_GENEVE_REMOTE;
1443 info->key.u.ipv4.dst =
1444 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1446 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1447 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1448 "Remote IPv4 address cannot be Multicast");
1453 if (data[IFLA_GENEVE_REMOTE6]) {
1454 #if IS_ENABLED(CONFIG_IPV6)
1455 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1456 attrtype = IFLA_GENEVE_REMOTE6;
1460 info->mode = IP_TUNNEL_INFO_IPV6;
1461 info->key.u.ipv6.dst =
1462 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1464 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1465 IPV6_ADDR_LINKLOCAL) {
1466 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1467 "Remote IPv6 address cannot be link-local");
1470 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1471 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1472 "Remote IPv6 address cannot be Multicast");
1475 info->key.tun_flags |= TUNNEL_CSUM;
1476 cfg->use_udp6_rx_checksums = true;
1478 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1479 "IPv6 support not enabled in the kernel");
1480 return -EPFNOSUPPORT;
1484 if (data[IFLA_GENEVE_ID]) {
1489 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1490 tvni[0] = (vni & 0x00ff0000) >> 16;
1491 tvni[1] = (vni & 0x0000ff00) >> 8;
1492 tvni[2] = vni & 0x000000ff;
1494 tunid = vni_to_tunnel_id(tvni);
1495 if (changelink && (tunid != info->key.tun_id)) {
1496 attrtype = IFLA_GENEVE_ID;
1499 info->key.tun_id = tunid;
1502 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1503 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1504 cfg->ttl_inherit = true;
1506 cfg->ttl_inherit = false;
1507 } else if (data[IFLA_GENEVE_TTL]) {
1508 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1509 cfg->ttl_inherit = false;
1512 if (data[IFLA_GENEVE_TOS])
1513 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1515 if (data[IFLA_GENEVE_DF])
1516 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1518 if (data[IFLA_GENEVE_LABEL]) {
1519 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1520 IPV6_FLOWLABEL_MASK;
1521 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1522 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1523 "Label attribute only applies for IPv6 Geneve devices");
1528 if (data[IFLA_GENEVE_PORT]) {
1530 attrtype = IFLA_GENEVE_PORT;
1533 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1536 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1538 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1541 cfg->collect_md = true;
1544 if (data[IFLA_GENEVE_UDP_CSUM]) {
1546 attrtype = IFLA_GENEVE_UDP_CSUM;
1549 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1550 info->key.tun_flags |= TUNNEL_CSUM;
1553 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1554 #if IS_ENABLED(CONFIG_IPV6)
1556 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1559 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1560 info->key.tun_flags &= ~TUNNEL_CSUM;
1562 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1563 "IPv6 support not enabled in the kernel");
1564 return -EPFNOSUPPORT;
1568 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1569 #if IS_ENABLED(CONFIG_IPV6)
1571 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1574 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1575 cfg->use_udp6_rx_checksums = false;
1577 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1578 "IPv6 support not enabled in the kernel");
1579 return -EPFNOSUPPORT;
1585 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1586 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1590 static void geneve_link_config(struct net_device *dev,
1591 struct ip_tunnel_info *info, struct nlattr *tb[])
1593 struct geneve_dev *geneve = netdev_priv(dev);
1597 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1601 switch (ip_tunnel_info_af(info)) {
1603 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1604 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1606 if (!IS_ERR(rt) && rt->dst.dev) {
1607 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1612 #if IS_ENABLED(CONFIG_IPV6)
1614 struct rt6_info *rt;
1616 if (!__in6_dev_get(dev))
1619 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1622 if (rt && rt->dst.dev)
1623 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1633 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1636 static int geneve_newlink(struct net *net, struct net_device *dev,
1637 struct nlattr *tb[], struct nlattr *data[],
1638 struct netlink_ext_ack *extack)
1640 struct geneve_config cfg = {
1641 .df = GENEVE_DF_UNSET,
1642 .use_udp6_rx_checksums = false,
1643 .ttl_inherit = false,
1644 .collect_md = false,
1648 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1649 err = geneve_nl2info(tb, data, extack, &cfg, false);
1653 err = geneve_configure(net, dev, extack, &cfg);
1657 geneve_link_config(dev, &cfg.info, tb);
1662 /* Quiesces the geneve device data path for both TX and RX.
1664 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1665 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1666 * to complete for the existing set of in-flight packets to be transmitted,
1667 * then we would have quiesced the transmit data path. All the future packets
1668 * will get dropped until we unquiesce the data path.
1670 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1671 * if we set that to NULL under RCU and wait for synchronize_net() to
1672 * complete, then we would have quiesced the receive data path.
1674 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1675 struct geneve_sock **gs6)
1677 *gs4 = rtnl_dereference(geneve->sock4);
1678 rcu_assign_pointer(geneve->sock4, NULL);
1680 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1681 #if IS_ENABLED(CONFIG_IPV6)
1682 *gs6 = rtnl_dereference(geneve->sock6);
1683 rcu_assign_pointer(geneve->sock6, NULL);
1685 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1692 /* Resumes the geneve device data path for both TX and RX. */
1693 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1694 struct geneve_sock __maybe_unused *gs6)
1696 rcu_assign_pointer(geneve->sock4, gs4);
1698 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1699 #if IS_ENABLED(CONFIG_IPV6)
1700 rcu_assign_pointer(geneve->sock6, gs6);
1702 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1707 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1708 struct nlattr *data[],
1709 struct netlink_ext_ack *extack)
1711 struct geneve_dev *geneve = netdev_priv(dev);
1712 struct geneve_sock *gs4, *gs6;
1713 struct geneve_config cfg;
1716 /* If the geneve device is configured for metadata (or externally
1717 * controlled, for example, OVS), then nothing can be changed.
1719 if (geneve->cfg.collect_md)
1722 /* Start with the existing info. */
1723 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1724 err = geneve_nl2info(tb, data, extack, &cfg, true);
1728 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1729 dst_cache_reset(&cfg.info.dst_cache);
1730 geneve_link_config(dev, &cfg.info, tb);
1733 geneve_quiesce(geneve, &gs4, &gs6);
1734 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1735 geneve_unquiesce(geneve, gs4, gs6);
1740 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1742 struct geneve_dev *geneve = netdev_priv(dev);
1744 list_del(&geneve->next);
1745 unregister_netdevice_queue(dev, head);
1748 static size_t geneve_get_size(const struct net_device *dev)
1750 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1751 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1752 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1753 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1754 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1755 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1756 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1757 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1758 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1759 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1760 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1761 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1765 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1767 struct geneve_dev *geneve = netdev_priv(dev);
1768 struct ip_tunnel_info *info = &geneve->cfg.info;
1769 bool ttl_inherit = geneve->cfg.ttl_inherit;
1770 bool metadata = geneve->cfg.collect_md;
1774 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1775 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1776 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1777 goto nla_put_failure;
1779 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1780 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1781 info->key.u.ipv4.dst))
1782 goto nla_put_failure;
1783 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1784 !!(info->key.tun_flags & TUNNEL_CSUM)))
1785 goto nla_put_failure;
1787 #if IS_ENABLED(CONFIG_IPV6)
1788 } else if (!metadata) {
1789 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1790 &info->key.u.ipv6.dst))
1791 goto nla_put_failure;
1792 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1793 !(info->key.tun_flags & TUNNEL_CSUM)))
1794 goto nla_put_failure;
1798 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1799 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1800 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1801 goto nla_put_failure;
1803 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1804 goto nla_put_failure;
1806 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1807 goto nla_put_failure;
1809 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1810 goto nla_put_failure;
1812 #if IS_ENABLED(CONFIG_IPV6)
1813 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1814 !geneve->cfg.use_udp6_rx_checksums))
1815 goto nla_put_failure;
1818 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1819 goto nla_put_failure;
1827 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1829 .maxtype = IFLA_GENEVE_MAX,
1830 .policy = geneve_policy,
1831 .priv_size = sizeof(struct geneve_dev),
1832 .setup = geneve_setup,
1833 .validate = geneve_validate,
1834 .newlink = geneve_newlink,
1835 .changelink = geneve_changelink,
1836 .dellink = geneve_dellink,
1837 .get_size = geneve_get_size,
1838 .fill_info = geneve_fill_info,
1841 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1842 u8 name_assign_type, u16 dst_port)
1844 struct nlattr *tb[IFLA_MAX + 1];
1845 struct net_device *dev;
1846 LIST_HEAD(list_kill);
1848 struct geneve_config cfg = {
1849 .df = GENEVE_DF_UNSET,
1850 .use_udp6_rx_checksums = true,
1851 .ttl_inherit = false,
1855 memset(tb, 0, sizeof(tb));
1856 dev = rtnl_create_link(net, name, name_assign_type,
1857 &geneve_link_ops, tb, NULL);
1861 init_tnl_info(&cfg.info, dst_port);
1862 err = geneve_configure(net, dev, NULL, &cfg);
1865 return ERR_PTR(err);
1868 /* openvswitch users expect packet sizes to be unrestricted,
1869 * so set the largest MTU we can.
1871 err = geneve_change_mtu(dev, IP_MAX_MTU);
1875 err = rtnl_configure_link(dev, NULL);
1881 geneve_dellink(dev, &list_kill);
1882 unregister_netdevice_many(&list_kill);
1883 return ERR_PTR(err);
1885 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1887 static int geneve_netdevice_event(struct notifier_block *unused,
1888 unsigned long event, void *ptr)
1890 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1892 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1893 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1894 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1895 } else if (event == NETDEV_UNREGISTER) {
1896 if (!dev->udp_tunnel_nic_info)
1897 geneve_offload_rx_ports(dev, false);
1898 } else if (event == NETDEV_REGISTER) {
1899 if (!dev->udp_tunnel_nic_info)
1900 geneve_offload_rx_ports(dev, true);
1906 static struct notifier_block geneve_notifier_block __read_mostly = {
1907 .notifier_call = geneve_netdevice_event,
1910 static __net_init int geneve_init_net(struct net *net)
1912 struct geneve_net *gn = net_generic(net, geneve_net_id);
1914 INIT_LIST_HEAD(&gn->geneve_list);
1915 INIT_LIST_HEAD(&gn->sock_list);
1919 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1921 struct geneve_net *gn = net_generic(net, geneve_net_id);
1922 struct geneve_dev *geneve, *next;
1923 struct net_device *dev, *aux;
1925 /* gather any geneve devices that were moved into this ns */
1926 for_each_netdev_safe(net, dev, aux)
1927 if (dev->rtnl_link_ops == &geneve_link_ops)
1928 unregister_netdevice_queue(dev, head);
1930 /* now gather any other geneve devices that were created in this ns */
1931 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1932 /* If geneve->dev is in the same netns, it was already added
1933 * to the list by the previous loop.
1935 if (!net_eq(dev_net(geneve->dev), net))
1936 unregister_netdevice_queue(geneve->dev, head);
1940 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1946 list_for_each_entry(net, net_list, exit_list)
1947 geneve_destroy_tunnels(net, &list);
1949 /* unregister the devices gathered above */
1950 unregister_netdevice_many(&list);
1953 list_for_each_entry(net, net_list, exit_list) {
1954 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1956 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1960 static struct pernet_operations geneve_net_ops = {
1961 .init = geneve_init_net,
1962 .exit_batch = geneve_exit_batch_net,
1963 .id = &geneve_net_id,
1964 .size = sizeof(struct geneve_net),
1967 static int __init geneve_init_module(void)
1971 rc = register_pernet_subsys(&geneve_net_ops);
1975 rc = register_netdevice_notifier(&geneve_notifier_block);
1979 rc = rtnl_link_register(&geneve_link_ops);
1985 unregister_netdevice_notifier(&geneve_notifier_block);
1987 unregister_pernet_subsys(&geneve_net_ops);
1991 late_initcall(geneve_init_module);
1993 static void __exit geneve_cleanup_module(void)
1995 rtnl_link_unregister(&geneve_link_ops);
1996 unregister_netdevice_notifier(&geneve_notifier_block);
1997 unregister_pernet_subsys(&geneve_net_ops);
1999 module_exit(geneve_cleanup_module);
2001 MODULE_LICENSE("GPL");
2002 MODULE_VERSION(GENEVE_NETDEV_VER);
2003 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2004 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2005 MODULE_ALIAS_RTNL_LINK("geneve");