1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2013 Nicira, Inc.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/capability.h>
9 #include <linux/module.h>
10 #include <linux/types.h>
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/uaccess.h>
14 #include <linux/skbuff.h>
15 #include <linux/netdevice.h>
17 #include <linux/tcp.h>
18 #include <linux/udp.h>
19 #include <linux/if_arp.h>
20 #include <linux/init.h>
21 #include <linux/in6.h>
22 #include <linux/inetdevice.h>
23 #include <linux/igmp.h>
24 #include <linux/netfilter_ipv4.h>
25 #include <linux/etherdevice.h>
26 #include <linux/if_ether.h>
27 #include <linux/if_vlan.h>
28 #include <linux/rculist.h>
29 #include <linux/err.h>
34 #include <net/protocol.h>
35 #include <net/ip_tunnels.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
41 #include <net/net_namespace.h>
42 #include <net/netns/generic.h>
43 #include <net/rtnetlink.h>
45 #include <net/dst_metadata.h>
47 #if IS_ENABLED(CONFIG_IPV6)
49 #include <net/ip6_fib.h>
50 #include <net/ip6_route.h>
53 static unsigned int ip_tunnel_hash(__be32 key, __be32 remote)
55 return hash_32((__force u32)key ^ (__force u32)remote,
59 static bool ip_tunnel_key_match(const struct ip_tunnel_parm *p,
60 __be16 flags, __be32 key)
62 if (p->i_flags & TUNNEL_KEY) {
63 if (flags & TUNNEL_KEY)
64 return key == p->i_key;
66 /* key expected, none present */
69 return !(flags & TUNNEL_KEY);
72 /* Fallback tunnel: no source, no destination, no key, no options
75 We require exact key match i.e. if a key is present in packet
76 it will match only tunnel with the same key; if it is not present,
77 it will match only keyless tunnel.
79 All keysless packets, if not matched configured keyless tunnels
80 will match fallback tunnel.
81 Given src, dst and key, find appropriate for input tunnel.
83 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
84 int link, __be16 flags,
85 __be32 remote, __be32 local,
88 struct ip_tunnel *t, *cand = NULL;
89 struct hlist_head *head;
90 struct net_device *ndev;
93 hash = ip_tunnel_hash(key, remote);
94 head = &itn->tunnels[hash];
96 hlist_for_each_entry_rcu(t, head, hash_node) {
97 if (local != t->parms.iph.saddr ||
98 remote != t->parms.iph.daddr ||
99 !(t->dev->flags & IFF_UP))
102 if (!ip_tunnel_key_match(&t->parms, flags, key))
105 if (t->parms.link == link)
111 hlist_for_each_entry_rcu(t, head, hash_node) {
112 if (remote != t->parms.iph.daddr ||
113 t->parms.iph.saddr != 0 ||
114 !(t->dev->flags & IFF_UP))
117 if (!ip_tunnel_key_match(&t->parms, flags, key))
120 if (t->parms.link == link)
126 hash = ip_tunnel_hash(key, 0);
127 head = &itn->tunnels[hash];
129 hlist_for_each_entry_rcu(t, head, hash_node) {
130 if ((local != t->parms.iph.saddr || t->parms.iph.daddr != 0) &&
131 (local != t->parms.iph.daddr || !ipv4_is_multicast(local)))
134 if (!(t->dev->flags & IFF_UP))
137 if (!ip_tunnel_key_match(&t->parms, flags, key))
140 if (t->parms.link == link)
146 hlist_for_each_entry_rcu(t, head, hash_node) {
147 if ((!(flags & TUNNEL_NO_KEY) && t->parms.i_key != key) ||
148 t->parms.iph.saddr != 0 ||
149 t->parms.iph.daddr != 0 ||
150 !(t->dev->flags & IFF_UP))
153 if (t->parms.link == link)
162 t = rcu_dereference(itn->collect_md_tun);
163 if (t && t->dev->flags & IFF_UP)
166 ndev = READ_ONCE(itn->fb_tunnel_dev);
167 if (ndev && ndev->flags & IFF_UP)
168 return netdev_priv(ndev);
172 EXPORT_SYMBOL_GPL(ip_tunnel_lookup);
174 static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn,
175 struct ip_tunnel_parm *parms)
179 __be32 i_key = parms->i_key;
181 if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr))
182 remote = parms->iph.daddr;
186 if (!(parms->i_flags & TUNNEL_KEY) && (parms->i_flags & VTI_ISVTI))
189 h = ip_tunnel_hash(i_key, remote);
190 return &itn->tunnels[h];
193 static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t)
195 struct hlist_head *head = ip_bucket(itn, &t->parms);
198 rcu_assign_pointer(itn->collect_md_tun, t);
199 hlist_add_head_rcu(&t->hash_node, head);
202 static void ip_tunnel_del(struct ip_tunnel_net *itn, struct ip_tunnel *t)
205 rcu_assign_pointer(itn->collect_md_tun, NULL);
206 hlist_del_init_rcu(&t->hash_node);
209 static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn,
210 struct ip_tunnel_parm *parms,
213 __be32 remote = parms->iph.daddr;
214 __be32 local = parms->iph.saddr;
215 __be32 key = parms->i_key;
216 __be16 flags = parms->i_flags;
217 int link = parms->link;
218 struct ip_tunnel *t = NULL;
219 struct hlist_head *head = ip_bucket(itn, parms);
221 hlist_for_each_entry_rcu(t, head, hash_node) {
222 if (local == t->parms.iph.saddr &&
223 remote == t->parms.iph.daddr &&
224 link == t->parms.link &&
225 type == t->dev->type &&
226 ip_tunnel_key_match(&t->parms, flags, key))
232 static struct net_device *__ip_tunnel_create(struct net *net,
233 const struct rtnl_link_ops *ops,
234 struct ip_tunnel_parm *parms)
237 struct ip_tunnel *tunnel;
238 struct net_device *dev;
242 if (parms->name[0]) {
243 if (!dev_valid_name(parms->name))
245 strlcpy(name, parms->name, IFNAMSIZ);
247 if (strlen(ops->kind) > (IFNAMSIZ - 3))
249 strcpy(name, ops->kind);
254 dev = alloc_netdev(ops->priv_size, name, NET_NAME_UNKNOWN, ops->setup);
259 dev_net_set(dev, net);
261 dev->rtnl_link_ops = ops;
263 tunnel = netdev_priv(dev);
264 tunnel->parms = *parms;
267 err = register_netdevice(dev);
279 static int ip_tunnel_bind_dev(struct net_device *dev)
281 struct net_device *tdev = NULL;
282 struct ip_tunnel *tunnel = netdev_priv(dev);
283 const struct iphdr *iph;
284 int hlen = LL_MAX_HEADER;
285 int mtu = ETH_DATA_LEN;
286 int t_hlen = tunnel->hlen + sizeof(struct iphdr);
288 iph = &tunnel->parms.iph;
290 /* Guess output device to choose reasonable mtu and needed_headroom */
295 ip_tunnel_init_flow(&fl4, iph->protocol, iph->daddr,
296 iph->saddr, tunnel->parms.o_key,
297 RT_TOS(iph->tos), tunnel->parms.link,
299 rt = ip_route_output_key(tunnel->net, &fl4);
305 if (dev->type != ARPHRD_ETHER)
306 dev->flags |= IFF_POINTOPOINT;
308 dst_cache_reset(&tunnel->dst_cache);
311 if (!tdev && tunnel->parms.link)
312 tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link);
315 hlen = tdev->hard_header_len + tdev->needed_headroom;
316 mtu = min(tdev->mtu, IP_MAX_MTU);
319 dev->needed_headroom = t_hlen + hlen;
320 mtu -= t_hlen + (dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0);
322 if (mtu < IPV4_MIN_MTU)
328 static struct ip_tunnel *ip_tunnel_create(struct net *net,
329 struct ip_tunnel_net *itn,
330 struct ip_tunnel_parm *parms)
332 struct ip_tunnel *nt;
333 struct net_device *dev;
338 dev = __ip_tunnel_create(net, itn->rtnl_link_ops, parms);
340 return ERR_CAST(dev);
342 mtu = ip_tunnel_bind_dev(dev);
343 err = dev_set_mtu(dev, mtu);
345 goto err_dev_set_mtu;
347 nt = netdev_priv(dev);
348 t_hlen = nt->hlen + sizeof(struct iphdr);
349 dev->min_mtu = ETH_MIN_MTU;
350 dev->max_mtu = IP_MAX_MTU - t_hlen;
351 if (dev->type == ARPHRD_ETHER)
352 dev->max_mtu -= dev->hard_header_len;
354 ip_tunnel_add(itn, nt);
358 unregister_netdevice(dev);
362 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb,
363 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst,
366 const struct iphdr *iph = ip_hdr(skb);
369 #ifdef CONFIG_NET_IPGRE_BROADCAST
370 if (ipv4_is_multicast(iph->daddr)) {
371 tunnel->dev->stats.multicast++;
372 skb->pkt_type = PACKET_BROADCAST;
376 if ((!(tpi->flags&TUNNEL_CSUM) && (tunnel->parms.i_flags&TUNNEL_CSUM)) ||
377 ((tpi->flags&TUNNEL_CSUM) && !(tunnel->parms.i_flags&TUNNEL_CSUM))) {
378 tunnel->dev->stats.rx_crc_errors++;
379 tunnel->dev->stats.rx_errors++;
383 if (tunnel->parms.i_flags&TUNNEL_SEQ) {
384 if (!(tpi->flags&TUNNEL_SEQ) ||
385 (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) {
386 tunnel->dev->stats.rx_fifo_errors++;
387 tunnel->dev->stats.rx_errors++;
390 tunnel->i_seqno = ntohl(tpi->seq) + 1;
393 skb_reset_network_header(skb);
395 err = IP_ECN_decapsulate(iph, skb);
398 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
399 &iph->saddr, iph->tos);
401 ++tunnel->dev->stats.rx_frame_errors;
402 ++tunnel->dev->stats.rx_errors;
407 dev_sw_netstats_rx_add(tunnel->dev, skb->len);
408 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev)));
410 if (tunnel->dev->type == ARPHRD_ETHER) {
411 skb->protocol = eth_type_trans(skb, tunnel->dev);
412 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
414 skb->dev = tunnel->dev;
418 skb_dst_set(skb, (struct dst_entry *)tun_dst);
420 gro_cells_receive(&tunnel->gro_cells, skb);
425 dst_release((struct dst_entry *)tun_dst);
429 EXPORT_SYMBOL_GPL(ip_tunnel_rcv);
431 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops,
434 if (num >= MAX_IPTUN_ENCAP_OPS)
437 return !cmpxchg((const struct ip_tunnel_encap_ops **)
441 EXPORT_SYMBOL(ip_tunnel_encap_add_ops);
443 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops,
448 if (num >= MAX_IPTUN_ENCAP_OPS)
451 ret = (cmpxchg((const struct ip_tunnel_encap_ops **)
453 ops, NULL) == ops) ? 0 : -1;
459 EXPORT_SYMBOL(ip_tunnel_encap_del_ops);
461 int ip_tunnel_encap_setup(struct ip_tunnel *t,
462 struct ip_tunnel_encap *ipencap)
466 memset(&t->encap, 0, sizeof(t->encap));
468 hlen = ip_encap_hlen(ipencap);
472 t->encap.type = ipencap->type;
473 t->encap.sport = ipencap->sport;
474 t->encap.dport = ipencap->dport;
475 t->encap.flags = ipencap->flags;
477 t->encap_hlen = hlen;
478 t->hlen = t->encap_hlen + t->tun_hlen;
482 EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup);
484 static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb,
485 struct rtable *rt, __be16 df,
486 const struct iphdr *inner_iph,
487 int tunnel_hlen, __be32 dst, bool md)
489 struct ip_tunnel *tunnel = netdev_priv(dev);
493 tunnel_hlen = md ? tunnel_hlen : tunnel->hlen;
494 pkt_size = skb->len - tunnel_hlen;
495 pkt_size -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0;
498 mtu = dst_mtu(&rt->dst) - (sizeof(struct iphdr) + tunnel_hlen);
499 mtu -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0;
501 mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
504 if (skb_valid_dst(skb))
505 skb_dst_update_pmtu_no_confirm(skb, mtu);
507 if (skb->protocol == htons(ETH_P_IP)) {
508 if (!skb_is_gso(skb) &&
509 (inner_iph->frag_off & htons(IP_DF)) &&
511 icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
515 #if IS_ENABLED(CONFIG_IPV6)
516 else if (skb->protocol == htons(ETH_P_IPV6)) {
517 struct rt6_info *rt6;
520 rt6 = skb_valid_dst(skb) ? (struct rt6_info *)skb_dst(skb) :
522 daddr = md ? dst : tunnel->parms.iph.daddr;
524 if (rt6 && mtu < dst_mtu(skb_dst(skb)) &&
525 mtu >= IPV6_MIN_MTU) {
526 if ((daddr && !ipv4_is_multicast(daddr)) ||
527 rt6->rt6i_dst.plen == 128) {
528 rt6->rt6i_flags |= RTF_MODIFIED;
529 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
533 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU &&
535 icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
543 void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
544 u8 proto, int tunnel_hlen)
546 struct ip_tunnel *tunnel = netdev_priv(dev);
547 u32 headroom = sizeof(struct iphdr);
548 struct ip_tunnel_info *tun_info;
549 const struct ip_tunnel_key *key;
550 const struct iphdr *inner_iph;
551 struct rtable *rt = NULL;
557 tun_info = skb_tunnel_info(skb);
558 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
559 ip_tunnel_info_af(tun_info) != AF_INET))
561 key = &tun_info->key;
562 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
563 inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
566 if (skb->protocol == htons(ETH_P_IP))
567 tos = inner_iph->tos;
568 else if (skb->protocol == htons(ETH_P_IPV6))
569 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
571 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src,
572 tunnel_id_to_key32(key->tun_id), RT_TOS(tos),
573 0, skb->mark, skb_get_hash(skb));
574 if (tunnel->encap.type != TUNNEL_ENCAP_NONE)
577 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
579 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl4.saddr);
581 rt = ip_route_output_key(tunnel->net, &fl4);
583 dev->stats.tx_carrier_errors++;
587 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
590 if (rt->dst.dev == dev) {
592 dev->stats.collisions++;
596 if (key->tun_flags & TUNNEL_DONT_FRAGMENT)
598 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, tunnel_hlen,
599 key->u.ipv4.dst, true)) {
604 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
607 if (skb->protocol == htons(ETH_P_IP))
608 ttl = inner_iph->ttl;
609 else if (skb->protocol == htons(ETH_P_IPV6))
610 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
612 ttl = ip4_dst_hoplimit(&rt->dst);
615 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len;
616 if (headroom > dev->needed_headroom)
617 dev->needed_headroom = headroom;
619 if (skb_cow_head(skb, dev->needed_headroom)) {
623 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl,
624 df, !net_eq(tunnel->net, dev_net(dev)));
627 dev->stats.tx_errors++;
630 dev->stats.tx_dropped++;
634 EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit);
636 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
637 const struct iphdr *tnl_params, u8 protocol)
639 struct ip_tunnel *tunnel = netdev_priv(dev);
640 struct ip_tunnel_info *tun_info = NULL;
641 const struct iphdr *inner_iph;
642 unsigned int max_headroom; /* The extra header space needed */
643 struct rtable *rt = NULL; /* Route to the other host */
644 bool use_cache = false;
652 inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
653 connected = (tunnel->parms.iph.daddr != 0);
655 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
657 dst = tnl_params->daddr;
662 dev->stats.tx_fifo_errors++;
666 tun_info = skb_tunnel_info(skb);
667 if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX) &&
668 ip_tunnel_info_af(tun_info) == AF_INET &&
669 tun_info->key.u.ipv4.dst) {
670 dst = tun_info->key.u.ipv4.dst;
674 else if (skb->protocol == htons(ETH_P_IP)) {
675 rt = skb_rtable(skb);
676 dst = rt_nexthop(rt, inner_iph->daddr);
678 #if IS_ENABLED(CONFIG_IPV6)
679 else if (skb->protocol == htons(ETH_P_IPV6)) {
680 const struct in6_addr *addr6;
681 struct neighbour *neigh;
682 bool do_tx_error_icmp;
685 neigh = dst_neigh_lookup(skb_dst(skb),
686 &ipv6_hdr(skb)->daddr);
690 addr6 = (const struct in6_addr *)&neigh->primary_key;
691 addr_type = ipv6_addr_type(addr6);
693 if (addr_type == IPV6_ADDR_ANY) {
694 addr6 = &ipv6_hdr(skb)->daddr;
695 addr_type = ipv6_addr_type(addr6);
698 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
699 do_tx_error_icmp = true;
701 do_tx_error_icmp = false;
702 dst = addr6->s6_addr32[3];
704 neigh_release(neigh);
705 if (do_tx_error_icmp)
716 tos = tnl_params->tos;
719 if (skb->protocol == htons(ETH_P_IP)) {
720 tos = inner_iph->tos;
722 } else if (skb->protocol == htons(ETH_P_IPV6)) {
723 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
728 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr,
729 tunnel->parms.o_key, RT_TOS(tos), tunnel->parms.link,
730 tunnel->fwmark, skb_get_hash(skb));
732 if (ip_tunnel_encap(skb, tunnel, &protocol, &fl4) < 0)
735 if (connected && md) {
736 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
738 rt = dst_cache_get_ip4(&tun_info->dst_cache,
741 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache,
746 rt = ip_route_output_key(tunnel->net, &fl4);
749 dev->stats.tx_carrier_errors++;
753 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
755 else if (!md && connected)
756 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst,
760 if (rt->dst.dev == dev) {
762 dev->stats.collisions++;
766 df = tnl_params->frag_off;
767 if (skb->protocol == htons(ETH_P_IP) && !tunnel->ignore_df)
768 df |= (inner_iph->frag_off & htons(IP_DF));
770 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, 0, 0, false)) {
775 if (tunnel->err_count > 0) {
776 if (time_before(jiffies,
777 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
780 dst_link_failure(skb);
782 tunnel->err_count = 0;
785 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
786 ttl = tnl_params->ttl;
788 if (skb->protocol == htons(ETH_P_IP))
789 ttl = inner_iph->ttl;
790 #if IS_ENABLED(CONFIG_IPV6)
791 else if (skb->protocol == htons(ETH_P_IPV6))
792 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
795 ttl = ip4_dst_hoplimit(&rt->dst);
798 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr)
799 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap);
800 if (max_headroom > dev->needed_headroom)
801 dev->needed_headroom = max_headroom;
803 if (skb_cow_head(skb, dev->needed_headroom)) {
805 dev->stats.tx_dropped++;
810 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl,
811 df, !net_eq(tunnel->net, dev_net(dev)));
814 #if IS_ENABLED(CONFIG_IPV6)
816 dst_link_failure(skb);
819 dev->stats.tx_errors++;
822 EXPORT_SYMBOL_GPL(ip_tunnel_xmit);
824 static void ip_tunnel_update(struct ip_tunnel_net *itn,
826 struct net_device *dev,
827 struct ip_tunnel_parm *p,
831 ip_tunnel_del(itn, t);
832 t->parms.iph.saddr = p->iph.saddr;
833 t->parms.iph.daddr = p->iph.daddr;
834 t->parms.i_key = p->i_key;
835 t->parms.o_key = p->o_key;
836 if (dev->type != ARPHRD_ETHER) {
837 memcpy(dev->dev_addr, &p->iph.saddr, 4);
838 memcpy(dev->broadcast, &p->iph.daddr, 4);
840 ip_tunnel_add(itn, t);
842 t->parms.iph.ttl = p->iph.ttl;
843 t->parms.iph.tos = p->iph.tos;
844 t->parms.iph.frag_off = p->iph.frag_off;
846 if (t->parms.link != p->link || t->fwmark != fwmark) {
849 t->parms.link = p->link;
851 mtu = ip_tunnel_bind_dev(dev);
855 dst_cache_reset(&t->dst_cache);
856 netdev_state_change(dev);
859 int ip_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd)
862 struct ip_tunnel *t = netdev_priv(dev);
863 struct net *net = t->net;
864 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id);
868 if (dev == itn->fb_tunnel_dev) {
869 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
871 t = netdev_priv(dev);
873 memcpy(p, &t->parms, sizeof(*p));
879 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
882 p->iph.frag_off |= htons(IP_DF);
883 if (!(p->i_flags & VTI_ISVTI)) {
884 if (!(p->i_flags & TUNNEL_KEY))
886 if (!(p->o_flags & TUNNEL_KEY))
890 t = ip_tunnel_find(itn, p, itn->type);
892 if (cmd == SIOCADDTUNNEL) {
894 t = ip_tunnel_create(net, itn, p);
895 err = PTR_ERR_OR_ZERO(t);
902 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
909 unsigned int nflags = 0;
911 if (ipv4_is_multicast(p->iph.daddr))
912 nflags = IFF_BROADCAST;
913 else if (p->iph.daddr)
914 nflags = IFF_POINTOPOINT;
916 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
921 t = netdev_priv(dev);
927 ip_tunnel_update(itn, t, dev, p, true, 0);
935 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
938 if (dev == itn->fb_tunnel_dev) {
940 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
944 if (t == netdev_priv(itn->fb_tunnel_dev))
948 unregister_netdevice(dev);
959 EXPORT_SYMBOL_GPL(ip_tunnel_ctl);
961 int ip_tunnel_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
963 struct ip_tunnel_parm p;
966 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
968 err = dev->netdev_ops->ndo_tunnel_ctl(dev, &p, cmd);
969 if (!err && copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
973 EXPORT_SYMBOL_GPL(ip_tunnel_ioctl);
975 int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict)
977 struct ip_tunnel *tunnel = netdev_priv(dev);
978 int t_hlen = tunnel->hlen + sizeof(struct iphdr);
979 int max_mtu = IP_MAX_MTU - t_hlen;
981 if (dev->type == ARPHRD_ETHER)
982 max_mtu -= dev->hard_header_len;
984 if (new_mtu < ETH_MIN_MTU)
987 if (new_mtu > max_mtu) {
997 EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu);
999 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1001 return __ip_tunnel_change_mtu(dev, new_mtu, true);
1003 EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu);
1005 static void ip_tunnel_dev_free(struct net_device *dev)
1007 struct ip_tunnel *tunnel = netdev_priv(dev);
1009 gro_cells_destroy(&tunnel->gro_cells);
1010 dst_cache_destroy(&tunnel->dst_cache);
1011 free_percpu(dev->tstats);
1014 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head)
1016 struct ip_tunnel *tunnel = netdev_priv(dev);
1017 struct ip_tunnel_net *itn;
1019 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id);
1021 if (itn->fb_tunnel_dev != dev) {
1022 ip_tunnel_del(itn, netdev_priv(dev));
1023 unregister_netdevice_queue(dev, head);
1026 EXPORT_SYMBOL_GPL(ip_tunnel_dellink);
1028 struct net *ip_tunnel_get_link_net(const struct net_device *dev)
1030 struct ip_tunnel *tunnel = netdev_priv(dev);
1034 EXPORT_SYMBOL(ip_tunnel_get_link_net);
1036 int ip_tunnel_get_iflink(const struct net_device *dev)
1038 struct ip_tunnel *tunnel = netdev_priv(dev);
1040 return tunnel->parms.link;
1042 EXPORT_SYMBOL(ip_tunnel_get_iflink);
1044 int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id,
1045 struct rtnl_link_ops *ops, char *devname)
1047 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id);
1048 struct ip_tunnel_parm parms;
1051 itn->rtnl_link_ops = ops;
1052 for (i = 0; i < IP_TNL_HASH_SIZE; i++)
1053 INIT_HLIST_HEAD(&itn->tunnels[i]);
1055 if (!ops || !net_has_fallback_tunnels(net)) {
1056 struct ip_tunnel_net *it_init_net;
1058 it_init_net = net_generic(&init_net, ip_tnl_net_id);
1059 itn->type = it_init_net->type;
1060 itn->fb_tunnel_dev = NULL;
1064 memset(&parms, 0, sizeof(parms));
1066 strlcpy(parms.name, devname, IFNAMSIZ);
1069 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms);
1070 /* FB netdevice is special: we have one, and only one per netns.
1071 * Allowing to move it to another netns is clearly unsafe.
1073 if (!IS_ERR(itn->fb_tunnel_dev)) {
1074 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1075 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev);
1076 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev));
1077 itn->type = itn->fb_tunnel_dev->type;
1081 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev);
1083 EXPORT_SYMBOL_GPL(ip_tunnel_init_net);
1085 static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn,
1086 struct list_head *head,
1087 struct rtnl_link_ops *ops)
1089 struct net_device *dev, *aux;
1092 for_each_netdev_safe(net, dev, aux)
1093 if (dev->rtnl_link_ops == ops)
1094 unregister_netdevice_queue(dev, head);
1096 for (h = 0; h < IP_TNL_HASH_SIZE; h++) {
1097 struct ip_tunnel *t;
1098 struct hlist_node *n;
1099 struct hlist_head *thead = &itn->tunnels[h];
1101 hlist_for_each_entry_safe(t, n, thead, hash_node)
1102 /* If dev is in the same netns, it has already
1103 * been added to the list by the previous loop.
1105 if (!net_eq(dev_net(t->dev), net))
1106 unregister_netdevice_queue(t->dev, head);
1110 void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id,
1111 struct rtnl_link_ops *ops)
1113 struct ip_tunnel_net *itn;
1118 list_for_each_entry(net, net_list, exit_list) {
1119 itn = net_generic(net, id);
1120 ip_tunnel_destroy(net, itn, &list, ops);
1122 unregister_netdevice_many(&list);
1125 EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets);
1127 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
1128 struct ip_tunnel_parm *p, __u32 fwmark)
1130 struct ip_tunnel *nt;
1131 struct net *net = dev_net(dev);
1132 struct ip_tunnel_net *itn;
1136 nt = netdev_priv(dev);
1137 itn = net_generic(net, nt->ip_tnl_net_id);
1139 if (nt->collect_md) {
1140 if (rtnl_dereference(itn->collect_md_tun))
1143 if (ip_tunnel_find(itn, p, dev->type))
1149 nt->fwmark = fwmark;
1150 err = register_netdevice(dev);
1152 goto err_register_netdevice;
1154 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1155 eth_hw_addr_random(dev);
1157 mtu = ip_tunnel_bind_dev(dev);
1159 unsigned int max = IP_MAX_MTU - (nt->hlen + sizeof(struct iphdr));
1161 if (dev->type == ARPHRD_ETHER)
1162 max -= dev->hard_header_len;
1164 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU, max);
1167 err = dev_set_mtu(dev, mtu);
1169 goto err_dev_set_mtu;
1171 ip_tunnel_add(itn, nt);
1175 unregister_netdevice(dev);
1176 err_register_netdevice:
1179 EXPORT_SYMBOL_GPL(ip_tunnel_newlink);
1181 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[],
1182 struct ip_tunnel_parm *p, __u32 fwmark)
1184 struct ip_tunnel *t;
1185 struct ip_tunnel *tunnel = netdev_priv(dev);
1186 struct net *net = tunnel->net;
1187 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id);
1189 if (dev == itn->fb_tunnel_dev)
1192 t = ip_tunnel_find(itn, p, dev->type);
1200 if (dev->type != ARPHRD_ETHER) {
1201 unsigned int nflags = 0;
1203 if (ipv4_is_multicast(p->iph.daddr))
1204 nflags = IFF_BROADCAST;
1205 else if (p->iph.daddr)
1206 nflags = IFF_POINTOPOINT;
1208 if ((dev->flags ^ nflags) &
1209 (IFF_POINTOPOINT | IFF_BROADCAST))
1214 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark);
1217 EXPORT_SYMBOL_GPL(ip_tunnel_changelink);
1219 int ip_tunnel_init(struct net_device *dev)
1221 struct ip_tunnel *tunnel = netdev_priv(dev);
1222 struct iphdr *iph = &tunnel->parms.iph;
1225 dev->needs_free_netdev = true;
1226 dev->priv_destructor = ip_tunnel_dev_free;
1227 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1231 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL);
1233 free_percpu(dev->tstats);
1237 err = gro_cells_init(&tunnel->gro_cells, dev);
1239 dst_cache_destroy(&tunnel->dst_cache);
1240 free_percpu(dev->tstats);
1245 tunnel->net = dev_net(dev);
1246 strcpy(tunnel->parms.name, dev->name);
1250 if (tunnel->collect_md)
1251 netif_keep_dst(dev);
1254 EXPORT_SYMBOL_GPL(ip_tunnel_init);
1256 void ip_tunnel_uninit(struct net_device *dev)
1258 struct ip_tunnel *tunnel = netdev_priv(dev);
1259 struct net *net = tunnel->net;
1260 struct ip_tunnel_net *itn;
1262 itn = net_generic(net, tunnel->ip_tnl_net_id);
1263 ip_tunnel_del(itn, netdev_priv(dev));
1264 if (itn->fb_tunnel_dev == dev)
1265 WRITE_ONCE(itn->fb_tunnel_dev, NULL);
1267 dst_cache_reset(&tunnel->dst_cache);
1269 EXPORT_SYMBOL_GPL(ip_tunnel_uninit);
1271 /* Do least required initialization, rest of init is done in tunnel_init call */
1272 void ip_tunnel_setup(struct net_device *dev, unsigned int net_id)
1274 struct ip_tunnel *tunnel = netdev_priv(dev);
1275 tunnel->ip_tnl_net_id = net_id;
1277 EXPORT_SYMBOL_GPL(ip_tunnel_setup);
1279 MODULE_LICENSE("GPL");