1 #include <linux/types.h>
2 #include <linux/skbuff.h>
3 #include <linux/socket.h>
4 #include <linux/sysctl.h>
6 #include <linux/module.h>
7 #include <linux/if_arp.h>
8 #include <linux/ipv6.h>
9 #include <linux/mpls.h>
14 #include <net/ip_fib.h>
15 #include <net/netevent.h>
16 #include <net/netns/generic.h>
19 #define LABEL_NOT_SPECIFIED (1<<20)
20 #define MAX_NEW_LABELS 2
22 /* This maximum ha length copied from the definition of struct neighbour */
23 #define MAX_VIA_ALEN (ALIGN(MAX_ADDR_LEN, sizeof(unsigned long)))
25 struct mpls_route { /* next hop label forwarding entry */
26 struct net_device *rt_dev;
27 struct rcu_head rt_rcu;
28 u32 rt_label[MAX_NEW_LABELS];
29 u8 rt_protocol; /* routing protocol that set this entry */
32 unsigned short rt_via_family;
37 static int label_limit = (1 << 20) - 1;
39 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
40 struct nlmsghdr *nlh, struct net *net, u32 portid,
41 unsigned int nlm_flags);
43 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
45 struct mpls_route *rt = NULL;
47 if (index < net->mpls.platform_labels) {
48 struct mpls_route __rcu **platform_label =
49 rcu_dereference(net->mpls.platform_label);
50 rt = rcu_dereference(platform_label[index]);
55 static bool mpls_output_possible(const struct net_device *dev)
57 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
60 static unsigned int mpls_rt_header_size(const struct mpls_route *rt)
62 /* The size of the layer 2.5 labels to be added for this route */
63 return rt->rt_labels * sizeof(struct mpls_shim_hdr);
66 static unsigned int mpls_dev_mtu(const struct net_device *dev)
68 /* The amount of data the layer 2 frame can hold */
72 static bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
77 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
83 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
84 struct mpls_entry_decoded dec)
86 /* RFC4385 and RFC5586 encode other packets in mpls such that
87 * they don't conflict with the ip version number, making
88 * decoding by examining the ip version correct in everything
89 * except for the strangest cases.
91 * The strange cases if we choose to support them will require
92 * manual configuration.
94 struct iphdr *hdr4 = ip_hdr(skb);
97 if (hdr4->version == 4) {
98 skb->protocol = htons(ETH_P_IP);
99 csum_replace2(&hdr4->check,
100 htons(hdr4->ttl << 8),
101 htons(dec.ttl << 8));
104 else if (hdr4->version == 6) {
105 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
106 skb->protocol = htons(ETH_P_IPV6);
107 hdr6->hop_limit = dec.ttl;
110 /* version 0 and version 1 are used by pseudo wires */
115 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
116 struct packet_type *pt, struct net_device *orig_dev)
118 struct net *net = dev_net(dev);
119 struct mpls_shim_hdr *hdr;
120 struct mpls_route *rt;
121 struct mpls_entry_decoded dec;
122 struct net_device *out_dev;
124 unsigned int new_header_size;
128 /* Careful this entire function runs inside of an rcu critical section */
130 if (skb->pkt_type != PACKET_HOST)
133 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
136 if (!pskb_may_pull(skb, sizeof(*hdr)))
139 /* Read and decode the label */
141 dec = mpls_entry_decode(hdr);
144 skb_pull(skb, sizeof(*hdr));
145 skb_reset_network_header(skb);
149 rt = mpls_route_input_rcu(net, dec.label);
153 /* Find the output device */
154 out_dev = rt->rt_dev;
155 if (!mpls_output_possible(out_dev))
158 if (skb_warn_if_lro(skb))
161 skb_forward_csum(skb);
163 /* Verify ttl is valid */
168 /* Verify the destination can hold the packet */
169 new_header_size = mpls_rt_header_size(rt);
170 mtu = mpls_dev_mtu(out_dev);
171 if (mpls_pkt_too_big(skb, mtu - new_header_size))
174 hh_len = LL_RESERVED_SPACE(out_dev);
175 if (!out_dev->header_ops)
178 /* Ensure there is enough space for the headers in the skb */
179 if (skb_cow(skb, hh_len + new_header_size))
183 skb->protocol = htons(ETH_P_MPLS_UC);
185 if (unlikely(!new_header_size && dec.bos)) {
186 /* Penultimate hop popping */
187 if (!mpls_egress(rt, skb, dec))
192 skb_push(skb, new_header_size);
193 skb_reset_network_header(skb);
194 /* Push the new labels */
197 for (i = rt->rt_labels - 1; i >= 0; i--) {
198 hdr[i] = mpls_entry_encode(rt->rt_label[i], dec.ttl, 0, bos);
203 err = neigh_xmit(rt->rt_via_family, out_dev, rt->rt_via, skb);
205 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
214 static struct packet_type mpls_packet_type __read_mostly = {
215 .type = cpu_to_be16(ETH_P_MPLS_UC),
216 .func = mpls_forward,
219 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
220 [RTA_DST] = { .type = NLA_U32 },
221 [RTA_OIF] = { .type = NLA_U32 },
224 struct mpls_route_config {
229 u8 rc_via[MAX_VIA_ALEN];
231 u32 rc_output_labels;
232 u32 rc_output_label[MAX_NEW_LABELS];
234 struct nl_info rc_nlinfo;
237 static struct mpls_route *mpls_rt_alloc(size_t alen)
239 struct mpls_route *rt;
241 rt = kzalloc(GFP_KERNEL, sizeof(*rt) + alen);
243 rt->rt_via_alen = alen;
247 static void mpls_rt_free(struct mpls_route *rt)
250 kfree_rcu(rt, rt_rcu);
253 static void mpls_notify_route(struct net *net, unsigned index,
254 struct mpls_route *old, struct mpls_route *new,
255 const struct nl_info *info)
257 struct nlmsghdr *nlh = info ? info->nlh : NULL;
258 unsigned portid = info ? info->portid : 0;
259 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
260 struct mpls_route *rt = new ? new : old;
261 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
262 /* Ignore reserved labels for now */
263 if (rt && (index >= 16))
264 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
267 static void mpls_route_update(struct net *net, unsigned index,
268 struct net_device *dev, struct mpls_route *new,
269 const struct nl_info *info)
271 struct mpls_route *rt, *old = NULL;
275 rt = net->mpls.platform_label[index];
276 if (!dev || (rt && (rt->rt_dev == dev))) {
277 rcu_assign_pointer(net->mpls.platform_label[index], new);
281 mpls_notify_route(net, index, old, new, info);
283 /* If we removed a route free it now */
287 static unsigned find_free_label(struct net *net)
290 for (index = 16; index < net->mpls.platform_labels; index++) {
291 if (!net->mpls.platform_label[index])
294 return LABEL_NOT_SPECIFIED;
297 static int mpls_route_add(struct mpls_route_config *cfg)
299 struct net *net = cfg->rc_nlinfo.nl_net;
300 struct net_device *dev = NULL;
301 struct mpls_route *rt, *old;
306 index = cfg->rc_label;
308 /* If a label was not specified during insert pick one */
309 if ((index == LABEL_NOT_SPECIFIED) &&
310 (cfg->rc_nlflags & NLM_F_CREATE)) {
311 index = find_free_label(net);
314 /* The first 16 labels are reserved, and may not be set */
318 /* The full 20 bit range may not be supported. */
319 if (index >= net->mpls.platform_labels)
322 /* Ensure only a supported number of labels are present */
323 if (cfg->rc_output_labels > MAX_NEW_LABELS)
327 dev = dev_get_by_index(net, cfg->rc_ifindex);
331 /* For now just support ethernet devices */
333 if ((dev->type != ARPHRD_ETHER) && (dev->type != ARPHRD_LOOPBACK))
337 if ((cfg->rc_via_family == AF_PACKET) &&
338 (dev->addr_len != cfg->rc_via_alen))
341 /* Append makes no sense with mpls */
343 if (cfg->rc_nlflags & NLM_F_APPEND)
347 old = net->mpls.platform_label[index];
348 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
352 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
356 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
360 rt = mpls_rt_alloc(cfg->rc_via_alen);
364 rt->rt_labels = cfg->rc_output_labels;
365 for (i = 0; i < rt->rt_labels; i++)
366 rt->rt_label[i] = cfg->rc_output_label[i];
367 rt->rt_protocol = cfg->rc_protocol;
369 rt->rt_via_family = cfg->rc_via_family;
370 memcpy(rt->rt_via, cfg->rc_via, cfg->rc_via_alen);
372 mpls_route_update(net, index, NULL, rt, &cfg->rc_nlinfo);
383 static int mpls_route_del(struct mpls_route_config *cfg)
385 struct net *net = cfg->rc_nlinfo.nl_net;
389 index = cfg->rc_label;
391 /* The first 16 labels are reserved, and may not be removed */
395 /* The full 20 bit range may not be supported */
396 if (index >= net->mpls.platform_labels)
399 mpls_route_update(net, index, NULL, NULL, &cfg->rc_nlinfo);
406 static void mpls_ifdown(struct net_device *dev)
408 struct net *net = dev_net(dev);
411 for (index = 0; index < net->mpls.platform_labels; index++) {
412 struct mpls_route *rt = net->mpls.platform_label[index];
415 if (rt->rt_dev != dev)
421 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
424 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
427 case NETDEV_UNREGISTER:
434 static struct notifier_block mpls_dev_notifier = {
435 .notifier_call = mpls_dev_notify,
438 static int nla_put_via(struct sk_buff *skb,
439 u16 family, const void *addr, int alen)
444 nla = nla_reserve(skb, RTA_VIA, alen + 2);
449 via->rtvia_family = family;
450 memcpy(via->rtvia_addr, addr, alen);
454 int nla_put_labels(struct sk_buff *skb, int attrtype,
455 u8 labels, const u32 label[])
458 struct mpls_shim_hdr *nla_label;
461 nla = nla_reserve(skb, attrtype, labels*4);
465 nla_label = nla_data(nla);
467 for (i = labels - 1; i >= 0; i--) {
468 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
475 int nla_get_labels(const struct nlattr *nla,
476 u32 max_labels, u32 *labels, u32 label[])
478 unsigned len = nla_len(nla);
480 struct mpls_shim_hdr *nla_label;
484 /* len needs to be an even multiple of 4 (the label size) */
488 /* Limit the number of new labels allowed */
490 if (nla_labels > max_labels)
493 nla_label = nla_data(nla);
495 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
496 struct mpls_entry_decoded dec;
497 dec = mpls_entry_decode(nla_label + i);
499 /* Ensure the bottom of stack flag is properly set
500 * and ttl and tc are both clear.
502 if ((dec.bos != bos) || dec.ttl || dec.tc)
505 label[i] = dec.label;
507 *labels = nla_labels;
511 static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
512 struct mpls_route_config *cfg)
515 struct nlattr *tb[RTA_MAX+1];
519 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
524 rtm = nlmsg_data(nlh);
525 memset(cfg, 0, sizeof(*cfg));
527 if (rtm->rtm_family != AF_MPLS)
529 if (rtm->rtm_dst_len != 20)
531 if (rtm->rtm_src_len != 0)
533 if (rtm->rtm_tos != 0)
535 if (rtm->rtm_table != RT_TABLE_MAIN)
537 /* Any value is acceptable for rtm_protocol */
539 /* As mpls uses destination specific addresses
540 * (or source specific address in the case of multicast)
541 * all addresses have universal scope.
543 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
545 if (rtm->rtm_type != RTN_UNICAST)
547 if (rtm->rtm_flags != 0)
550 cfg->rc_label = LABEL_NOT_SPECIFIED;
551 cfg->rc_protocol = rtm->rtm_protocol;
552 cfg->rc_nlflags = nlh->nlmsg_flags;
553 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
554 cfg->rc_nlinfo.nlh = nlh;
555 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
557 for (index = 0; index <= RTA_MAX; index++) {
558 struct nlattr *nla = tb[index];
564 cfg->rc_ifindex = nla_get_u32(nla);
567 if (nla_get_labels(nla, MAX_NEW_LABELS,
568 &cfg->rc_output_labels,
569 cfg->rc_output_label))
575 if (nla_get_labels(nla, 1, &label_count,
579 /* The first 16 labels are reserved, and may not be set */
580 if (cfg->rc_label < 16)
587 struct rtvia *via = nla_data(nla);
588 cfg->rc_via_family = via->rtvia_family;
589 cfg->rc_via_alen = nla_len(nla) - 2;
590 if (cfg->rc_via_alen > MAX_VIA_ALEN)
593 /* Validate the address family */
594 switch(cfg->rc_via_family) {
598 if (cfg->rc_via_alen != 4)
602 if (cfg->rc_via_alen != 16)
606 /* Unsupported address family */
610 memcpy(cfg->rc_via, via->rtvia_addr, cfg->rc_via_alen);
614 /* Unsupported attribute */
624 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
626 struct mpls_route_config cfg;
629 err = rtm_to_route_config(skb, nlh, &cfg);
633 return mpls_route_del(&cfg);
637 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
639 struct mpls_route_config cfg;
642 err = rtm_to_route_config(skb, nlh, &cfg);
646 return mpls_route_add(&cfg);
649 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
650 u32 label, struct mpls_route *rt, int flags)
652 struct nlmsghdr *nlh;
655 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
659 rtm = nlmsg_data(nlh);
660 rtm->rtm_family = AF_MPLS;
661 rtm->rtm_dst_len = 20;
662 rtm->rtm_src_len = 0;
664 rtm->rtm_table = RT_TABLE_MAIN;
665 rtm->rtm_protocol = rt->rt_protocol;
666 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
667 rtm->rtm_type = RTN_UNICAST;
671 nla_put_labels(skb, RTA_NEWDST, rt->rt_labels, rt->rt_label))
672 goto nla_put_failure;
673 if (nla_put_via(skb, rt->rt_via_family, rt->rt_via, rt->rt_via_alen))
674 goto nla_put_failure;
675 if (rt->rt_dev && nla_put_u32(skb, RTA_OIF, rt->rt_dev->ifindex))
676 goto nla_put_failure;
677 if (nla_put_labels(skb, RTA_DST, 1, &label))
678 goto nla_put_failure;
684 nlmsg_cancel(skb, nlh);
688 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
690 struct net *net = sock_net(skb->sk);
699 for (; index < net->mpls.platform_labels; index++) {
700 struct mpls_route *rt;
701 rt = net->mpls.platform_label[index];
705 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
706 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
707 index, rt, NLM_F_MULTI) < 0)
715 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
718 NLMSG_ALIGN(sizeof(struct rtmsg))
719 + nla_total_size(2 + rt->rt_via_alen) /* RTA_VIA */
720 + nla_total_size(4); /* RTA_DST */
721 if (rt->rt_labels) /* RTA_NEWDST */
722 payload += nla_total_size(rt->rt_labels * 4);
723 if (rt->rt_dev) /* RTA_OIF */
724 payload += nla_total_size(4);
728 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
729 struct nlmsghdr *nlh, struct net *net, u32 portid,
730 unsigned int nlm_flags)
733 u32 seq = nlh ? nlh->nlmsg_seq : 0;
736 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
740 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
742 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
743 WARN_ON(err == -EMSGSIZE);
747 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
752 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
755 static int resize_platform_label_table(struct net *net, size_t limit)
757 size_t size = sizeof(struct mpls_route *) * limit;
760 struct mpls_route __rcu **labels = NULL, **old;
761 struct mpls_route *rt0 = NULL, *rt2 = NULL;
765 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
767 labels = vzalloc(size);
773 /* In case the predefined labels need to be populated */
774 if (limit > LABEL_IPV4_EXPLICIT_NULL) {
775 struct net_device *lo = net->loopback_dev;
776 rt0 = mpls_rt_alloc(lo->addr_len);
780 rt0->rt_protocol = RTPROT_KERNEL;
781 rt0->rt_via_family = AF_PACKET;
782 memcpy(rt0->rt_via, lo->dev_addr, lo->addr_len);
784 if (limit > LABEL_IPV6_EXPLICIT_NULL) {
785 struct net_device *lo = net->loopback_dev;
786 rt2 = mpls_rt_alloc(lo->addr_len);
790 rt2->rt_protocol = RTPROT_KERNEL;
791 rt2->rt_via_family = AF_PACKET;
792 memcpy(rt2->rt_via, lo->dev_addr, lo->addr_len);
796 /* Remember the original table */
797 old = net->mpls.platform_label;
798 old_limit = net->mpls.platform_labels;
800 /* Free any labels beyond the new table */
801 for (index = limit; index < old_limit; index++)
802 mpls_route_update(net, index, NULL, NULL, NULL);
804 /* Copy over the old labels */
806 if (old_limit < limit)
807 cp_size = old_limit * sizeof(struct mpls_route *);
809 memcpy(labels, old, cp_size);
811 /* If needed set the predefined labels */
812 if ((old_limit <= LABEL_IPV6_EXPLICIT_NULL) &&
813 (limit > LABEL_IPV6_EXPLICIT_NULL)) {
814 labels[LABEL_IPV6_EXPLICIT_NULL] = rt2;
818 if ((old_limit <= LABEL_IPV4_EXPLICIT_NULL) &&
819 (limit > LABEL_IPV4_EXPLICIT_NULL)) {
820 labels[LABEL_IPV4_EXPLICIT_NULL] = rt0;
824 /* Update the global pointers */
825 net->mpls.platform_labels = limit;
826 net->mpls.platform_label = labels;
847 static int mpls_platform_labels(struct ctl_table *table, int write,
848 void __user *buffer, size_t *lenp, loff_t *ppos)
850 struct net *net = table->data;
851 int platform_labels = net->mpls.platform_labels;
853 struct ctl_table tmp = {
854 .procname = table->procname,
855 .data = &platform_labels,
856 .maxlen = sizeof(int),
859 .extra2 = &label_limit,
862 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
864 if (write && ret == 0)
865 ret = resize_platform_label_table(net, platform_labels);
870 static struct ctl_table mpls_table[] = {
872 .procname = "platform_labels",
874 .maxlen = sizeof(int),
876 .proc_handler = mpls_platform_labels,
881 static int mpls_net_init(struct net *net)
883 struct ctl_table *table;
885 net->mpls.platform_labels = 0;
886 net->mpls.platform_label = NULL;
888 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
893 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
894 if (net->mpls.ctl == NULL)
900 static void mpls_net_exit(struct net *net)
902 struct ctl_table *table;
905 table = net->mpls.ctl->ctl_table_arg;
906 unregister_net_sysctl_table(net->mpls.ctl);
909 /* An rcu grace period haselapsed since there was a device in
910 * the network namespace (and thus the last in fqlight packet)
911 * left this network namespace. This is because
912 * unregister_netdevice_many and netdev_run_todo has completed
913 * for each network device that was in this network namespace.
915 * As such no additional rcu synchronization is necessary when
916 * freeing the platform_label table.
919 for (index = 0; index < net->mpls.platform_labels; index++) {
920 struct mpls_route *rt = net->mpls.platform_label[index];
921 rcu_assign_pointer(net->mpls.platform_label[index], NULL);
926 kvfree(net->mpls.platform_label);
929 static struct pernet_operations mpls_net_ops = {
930 .init = mpls_net_init,
931 .exit = mpls_net_exit,
934 static int __init mpls_init(void)
938 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
940 err = register_pernet_subsys(&mpls_net_ops);
944 err = register_netdevice_notifier(&mpls_dev_notifier);
946 goto out_unregister_pernet;
948 dev_add_pack(&mpls_packet_type);
950 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
951 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
952 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
957 out_unregister_pernet:
958 unregister_pernet_subsys(&mpls_net_ops);
961 module_init(mpls_init);
963 static void __exit mpls_exit(void)
965 rtnl_unregister_all(PF_MPLS);
966 dev_remove_pack(&mpls_packet_type);
967 unregister_netdevice_notifier(&mpls_dev_notifier);
968 unregister_pernet_subsys(&mpls_net_ops);
970 module_exit(mpls_exit);
972 MODULE_DESCRIPTION("MultiProtocol Label Switching");
973 MODULE_LICENSE("GPL v2");
974 MODULE_ALIAS_NETPROTO(PF_MPLS);