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>
10 #include <linux/vmalloc.h>
15 #include <net/ip_fib.h>
16 #include <net/netevent.h>
17 #include <net/netns/generic.h>
18 #if IS_ENABLED(CONFIG_IPV6)
20 #include <net/addrconf.h>
22 #include <net/nexthop.h>
25 /* Maximum number of labels to look ahead at when selecting a path of
28 #define MAX_MP_SELECT_LABELS 4
30 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
33 static int label_limit = (1 << 20) - 1;
35 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
36 struct nlmsghdr *nlh, struct net *net, u32 portid,
37 unsigned int nlm_flags);
39 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
41 struct mpls_route *rt = NULL;
43 if (index < net->mpls.platform_labels) {
44 struct mpls_route __rcu **platform_label =
45 rcu_dereference(net->mpls.platform_label);
46 rt = rcu_dereference(platform_label[index]);
51 static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
53 return rcu_dereference_rtnl(dev->mpls_ptr);
56 bool mpls_output_possible(const struct net_device *dev)
58 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
60 EXPORT_SYMBOL_GPL(mpls_output_possible);
62 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
64 u8 *nh0_via = PTR_ALIGN((u8 *)&rt->rt_nh[rt->rt_nhn], VIA_ALEN_ALIGN);
65 int nh_index = nh - rt->rt_nh;
67 return nh0_via + rt->rt_max_alen * nh_index;
70 static const u8 *mpls_nh_via(const struct mpls_route *rt,
71 const struct mpls_nh *nh)
73 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
76 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
78 /* The size of the layer 2.5 labels to be added for this route */
79 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
82 unsigned int mpls_dev_mtu(const struct net_device *dev)
84 /* The amount of data the layer 2 frame can hold */
87 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
89 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
94 if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
99 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
101 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
103 struct mpls_entry_decoded dec;
104 unsigned int mpls_hdr_len = 0;
105 struct mpls_shim_hdr *hdr;
106 bool eli_seen = false;
110 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
112 mpls_hdr_len += sizeof(*hdr);
113 if (!pskb_may_pull(skb, mpls_hdr_len))
116 /* Read and decode the current label */
117 hdr = mpls_hdr(skb) + label_index;
118 dec = mpls_entry_decode(hdr);
120 /* RFC6790 - reserved labels MUST NOT be used as keys
121 * for the load-balancing function
123 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
124 hash = jhash_1word(dec.label, hash);
126 /* The entropy label follows the entropy label
127 * indicator, so this means that the entropy
128 * label was just added to the hash - no need to
129 * go any deeper either in the label stack or in the
134 } else if (dec.label == MPLS_LABEL_ENTROPY) {
141 /* found bottom label; does skb have room for a header? */
142 if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
143 const struct iphdr *v4hdr;
145 v4hdr = (const struct iphdr *)(hdr + 1);
146 if (v4hdr->version == 4) {
147 hash = jhash_3words(ntohl(v4hdr->saddr),
149 v4hdr->protocol, hash);
150 } else if (v4hdr->version == 6 &&
151 pskb_may_pull(skb, mpls_hdr_len +
152 sizeof(struct ipv6hdr))) {
153 const struct ipv6hdr *v6hdr;
155 v6hdr = (const struct ipv6hdr *)(hdr + 1);
156 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
157 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
158 hash = jhash_1word(v6hdr->nexthdr, hash);
168 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
171 int alive = ACCESS_ONCE(rt->rt_nhn_alive);
176 /* No need to look further into packet if there's only
185 hash = mpls_multipath_hash(rt, skb);
186 nh_index = hash % alive;
187 if (alive == rt->rt_nhn)
190 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
195 } endfor_nexthops(rt);
198 return &rt->rt_nh[nh_index];
201 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
202 struct mpls_entry_decoded dec)
204 enum mpls_payload_type payload_type;
205 bool success = false;
207 /* The IPv4 code below accesses through the IPv4 header
208 * checksum, which is 12 bytes into the packet.
209 * The IPv6 code below accesses through the IPv6 hop limit
210 * which is 8 bytes into the packet.
212 * For all supported cases there should always be at least 12
213 * bytes of packet data present. The IPv4 header is 20 bytes
214 * without options and the IPv6 header is always 40 bytes
217 if (!pskb_may_pull(skb, 12))
220 payload_type = rt->rt_payload_type;
221 if (payload_type == MPT_UNSPEC)
222 payload_type = ip_hdr(skb)->version;
224 switch (payload_type) {
226 struct iphdr *hdr4 = ip_hdr(skb);
227 skb->protocol = htons(ETH_P_IP);
228 csum_replace2(&hdr4->check,
229 htons(hdr4->ttl << 8),
230 htons(dec.ttl << 8));
236 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
237 skb->protocol = htons(ETH_P_IPV6);
238 hdr6->hop_limit = dec.ttl;
249 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
250 struct packet_type *pt, struct net_device *orig_dev)
252 struct net *net = dev_net(dev);
253 struct mpls_shim_hdr *hdr;
254 struct mpls_route *rt;
256 struct mpls_entry_decoded dec;
257 struct net_device *out_dev;
258 struct mpls_dev *mdev;
260 unsigned int new_header_size;
264 /* Careful this entire function runs inside of an rcu critical section */
266 mdev = mpls_dev_get(dev);
267 if (!mdev || !mdev->input_enabled)
270 if (skb->pkt_type != PACKET_HOST)
273 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
276 if (!pskb_may_pull(skb, sizeof(*hdr)))
279 /* Read and decode the label */
281 dec = mpls_entry_decode(hdr);
283 rt = mpls_route_input_rcu(net, dec.label);
287 nh = mpls_select_multipath(rt, skb);
291 /* Find the output device */
292 out_dev = rcu_dereference(nh->nh_dev);
293 if (!mpls_output_possible(out_dev))
297 skb_pull(skb, sizeof(*hdr));
298 skb_reset_network_header(skb);
302 if (skb_warn_if_lro(skb))
305 skb_forward_csum(skb);
307 /* Verify ttl is valid */
312 /* Verify the destination can hold the packet */
313 new_header_size = mpls_nh_header_size(nh);
314 mtu = mpls_dev_mtu(out_dev);
315 if (mpls_pkt_too_big(skb, mtu - new_header_size))
318 hh_len = LL_RESERVED_SPACE(out_dev);
319 if (!out_dev->header_ops)
322 /* Ensure there is enough space for the headers in the skb */
323 if (skb_cow(skb, hh_len + new_header_size))
327 skb->protocol = htons(ETH_P_MPLS_UC);
329 if (unlikely(!new_header_size && dec.bos)) {
330 /* Penultimate hop popping */
331 if (!mpls_egress(rt, skb, dec))
336 skb_push(skb, new_header_size);
337 skb_reset_network_header(skb);
338 /* Push the new labels */
341 for (i = nh->nh_labels - 1; i >= 0; i--) {
342 hdr[i] = mpls_entry_encode(nh->nh_label[i],
348 /* If via wasn't specified then send out using device address */
349 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
350 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
351 out_dev->dev_addr, skb);
353 err = neigh_xmit(nh->nh_via_table, out_dev,
354 mpls_nh_via(rt, nh), skb);
356 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
365 static struct packet_type mpls_packet_type __read_mostly = {
366 .type = cpu_to_be16(ETH_P_MPLS_UC),
367 .func = mpls_forward,
370 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
371 [RTA_DST] = { .type = NLA_U32 },
372 [RTA_OIF] = { .type = NLA_U32 },
375 struct mpls_route_config {
380 u8 rc_via[MAX_VIA_ALEN];
383 u32 rc_output_label[MAX_NEW_LABELS];
385 enum mpls_payload_type rc_payload_type;
386 struct nl_info rc_nlinfo;
387 struct rtnexthop *rc_mp;
391 static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
393 u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
394 struct mpls_route *rt;
396 rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
398 num_nh * max_alen_aligned,
402 rt->rt_nhn_alive = num_nh;
403 rt->rt_max_alen = max_alen_aligned;
409 static void mpls_rt_free(struct mpls_route *rt)
412 kfree_rcu(rt, rt_rcu);
415 static void mpls_notify_route(struct net *net, unsigned index,
416 struct mpls_route *old, struct mpls_route *new,
417 const struct nl_info *info)
419 struct nlmsghdr *nlh = info ? info->nlh : NULL;
420 unsigned portid = info ? info->portid : 0;
421 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
422 struct mpls_route *rt = new ? new : old;
423 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
424 /* Ignore reserved labels for now */
425 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
426 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
429 static void mpls_route_update(struct net *net, unsigned index,
430 struct mpls_route *new,
431 const struct nl_info *info)
433 struct mpls_route __rcu **platform_label;
434 struct mpls_route *rt;
438 platform_label = rtnl_dereference(net->mpls.platform_label);
439 rt = rtnl_dereference(platform_label[index]);
440 rcu_assign_pointer(platform_label[index], new);
442 mpls_notify_route(net, index, rt, new, info);
444 /* If we removed a route free it now */
448 static unsigned find_free_label(struct net *net)
450 struct mpls_route __rcu **platform_label;
451 size_t platform_labels;
454 platform_label = rtnl_dereference(net->mpls.platform_label);
455 platform_labels = net->mpls.platform_labels;
456 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
458 if (!rtnl_dereference(platform_label[index]))
461 return LABEL_NOT_SPECIFIED;
464 #if IS_ENABLED(CONFIG_INET)
465 static struct net_device *inet_fib_lookup_dev(struct net *net,
468 struct net_device *dev;
470 struct in_addr daddr;
472 memcpy(&daddr, addr, sizeof(struct in_addr));
473 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
485 static struct net_device *inet_fib_lookup_dev(struct net *net,
488 return ERR_PTR(-EAFNOSUPPORT);
492 #if IS_ENABLED(CONFIG_IPV6)
493 static struct net_device *inet6_fib_lookup_dev(struct net *net,
496 struct net_device *dev;
497 struct dst_entry *dst;
502 return ERR_PTR(-EAFNOSUPPORT);
504 memset(&fl6, 0, sizeof(fl6));
505 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
506 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
517 static struct net_device *inet6_fib_lookup_dev(struct net *net,
520 return ERR_PTR(-EAFNOSUPPORT);
524 static struct net_device *find_outdev(struct net *net,
525 struct mpls_route *rt,
526 struct mpls_nh *nh, int oif)
528 struct net_device *dev = NULL;
531 switch (nh->nh_via_table) {
532 case NEIGH_ARP_TABLE:
533 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
536 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
538 case NEIGH_LINK_TABLE:
542 dev = dev_get_by_index(net, oif);
546 return ERR_PTR(-ENODEV);
551 /* The caller is holding rtnl anyways, so release the dev reference */
557 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
558 struct mpls_nh *nh, int oif)
560 struct net_device *dev = NULL;
563 dev = find_outdev(net, rt, nh, oif);
570 /* Ensure this is a supported device */
572 if (!mpls_dev_get(dev))
575 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
576 (dev->addr_len != nh->nh_via_alen))
579 RCU_INIT_POINTER(nh->nh_dev, dev);
581 if (!(dev->flags & IFF_UP)) {
582 nh->nh_flags |= RTNH_F_DEAD;
586 flags = dev_get_flags(dev);
587 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
588 nh->nh_flags |= RTNH_F_LINKDOWN;
597 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
598 struct mpls_route *rt)
600 struct net *net = cfg->rc_nlinfo.nl_net;
601 struct mpls_nh *nh = rt->rt_nh;
609 /* Ensure only a supported number of labels are present */
610 if (cfg->rc_output_labels > MAX_NEW_LABELS)
613 nh->nh_labels = cfg->rc_output_labels;
614 for (i = 0; i < nh->nh_labels; i++)
615 nh->nh_label[i] = cfg->rc_output_label[i];
617 nh->nh_via_table = cfg->rc_via_table;
618 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
619 nh->nh_via_alen = cfg->rc_via_alen;
621 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
625 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
634 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
635 struct mpls_nh *nh, int oif, struct nlattr *via,
636 struct nlattr *newdst)
644 err = nla_get_labels(newdst, MAX_NEW_LABELS,
645 &nh->nh_labels, nh->nh_label);
651 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
652 __mpls_nh_via(rt, nh));
656 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
659 err = mpls_nh_assign_dev(net, rt, nh, oif);
669 static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
670 u8 cfg_via_alen, u8 *max_via_alen)
676 *max_via_alen = cfg_via_alen;
682 while (rtnh_ok(rtnh, remaining)) {
683 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
686 attrlen = rtnh_attrlen(rtnh);
687 nla = nla_find(attrs, attrlen, RTA_VIA);
688 if (nla && nla_len(nla) >=
689 offsetof(struct rtvia, rtvia_addr)) {
690 int via_alen = nla_len(nla) -
691 offsetof(struct rtvia, rtvia_addr);
693 if (via_alen <= MAX_VIA_ALEN)
694 *max_via_alen = max_t(u16, *max_via_alen,
699 rtnh = rtnh_next(rtnh, &remaining);
702 /* leftover implies invalid nexthop configuration, discard it */
703 return remaining > 0 ? 0 : nhs;
706 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
707 struct mpls_route *rt)
709 struct rtnexthop *rtnh = cfg->rc_mp;
710 struct nlattr *nla_via, *nla_newdst;
711 int remaining = cfg->rc_mp_len;
715 change_nexthops(rt) {
722 if (!rtnh_ok(rtnh, remaining))
725 /* neither weighted multipath nor any flags
728 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
731 attrlen = rtnh_attrlen(rtnh);
733 struct nlattr *attrs = rtnh_attrs(rtnh);
735 nla_via = nla_find(attrs, attrlen, RTA_VIA);
736 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
739 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
740 rtnh->rtnh_ifindex, nla_via, nla_newdst);
744 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
747 rtnh = rtnh_next(rtnh, &remaining);
749 } endfor_nexthops(rt);
759 static int mpls_route_add(struct mpls_route_config *cfg)
761 struct mpls_route __rcu **platform_label;
762 struct net *net = cfg->rc_nlinfo.nl_net;
763 struct mpls_route *rt, *old;
769 index = cfg->rc_label;
771 /* If a label was not specified during insert pick one */
772 if ((index == LABEL_NOT_SPECIFIED) &&
773 (cfg->rc_nlflags & NLM_F_CREATE)) {
774 index = find_free_label(net);
777 /* Reserved labels may not be set */
778 if (index < MPLS_LABEL_FIRST_UNRESERVED)
781 /* The full 20 bit range may not be supported. */
782 if (index >= net->mpls.platform_labels)
785 /* Append makes no sense with mpls */
787 if (cfg->rc_nlflags & NLM_F_APPEND)
791 platform_label = rtnl_dereference(net->mpls.platform_label);
792 old = rtnl_dereference(platform_label[index]);
793 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
797 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
801 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
805 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
806 cfg->rc_via_alen, &max_via_alen);
811 rt = mpls_rt_alloc(nhs, max_via_alen);
815 rt->rt_protocol = cfg->rc_protocol;
816 rt->rt_payload_type = cfg->rc_payload_type;
819 err = mpls_nh_build_multi(cfg, rt);
821 err = mpls_nh_build_from_cfg(cfg, rt);
825 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
835 static int mpls_route_del(struct mpls_route_config *cfg)
837 struct net *net = cfg->rc_nlinfo.nl_net;
841 index = cfg->rc_label;
843 /* Reserved labels may not be removed */
844 if (index < MPLS_LABEL_FIRST_UNRESERVED)
847 /* The full 20 bit range may not be supported */
848 if (index >= net->mpls.platform_labels)
851 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
858 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
859 (&((struct mpls_dev *)0)->field)
861 static const struct ctl_table mpls_dev_table[] = {
864 .maxlen = sizeof(int),
866 .proc_handler = proc_dointvec,
867 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
872 static int mpls_dev_sysctl_register(struct net_device *dev,
873 struct mpls_dev *mdev)
875 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
876 struct ctl_table *table;
879 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
883 /* Table data contains only offsets relative to the base of
884 * the mdev at this point, so make them absolute.
886 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
887 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
889 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
891 mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
903 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
905 struct ctl_table *table;
907 table = mdev->sysctl->ctl_table_arg;
908 unregister_net_sysctl_table(mdev->sysctl);
912 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
914 struct mpls_dev *mdev;
919 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
923 err = mpls_dev_sysctl_register(dev, mdev);
927 rcu_assign_pointer(dev->mpls_ptr, mdev);
936 static void mpls_ifdown(struct net_device *dev, int event)
938 struct mpls_route __rcu **platform_label;
939 struct net *net = dev_net(dev);
942 platform_label = rtnl_dereference(net->mpls.platform_label);
943 for (index = 0; index < net->mpls.platform_labels; index++) {
944 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
949 change_nexthops(rt) {
950 if (rtnl_dereference(nh->nh_dev) != dev)
954 case NETDEV_UNREGISTER:
955 nh->nh_flags |= RTNH_F_DEAD;
958 nh->nh_flags |= RTNH_F_LINKDOWN;
959 ACCESS_ONCE(rt->rt_nhn_alive) = rt->rt_nhn_alive - 1;
962 if (event == NETDEV_UNREGISTER)
963 RCU_INIT_POINTER(nh->nh_dev, NULL);
964 } endfor_nexthops(rt);
968 static void mpls_ifup(struct net_device *dev, unsigned int nh_flags)
970 struct mpls_route __rcu **platform_label;
971 struct net *net = dev_net(dev);
975 platform_label = rtnl_dereference(net->mpls.platform_label);
976 for (index = 0; index < net->mpls.platform_labels; index++) {
977 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
983 change_nexthops(rt) {
984 struct net_device *nh_dev =
985 rtnl_dereference(nh->nh_dev);
987 if (!(nh->nh_flags & nh_flags)) {
994 nh->nh_flags &= ~nh_flags;
995 } endfor_nexthops(rt);
997 ACCESS_ONCE(rt->rt_nhn_alive) = alive;
1001 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1004 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1005 struct mpls_dev *mdev;
1008 if (event == NETDEV_REGISTER) {
1009 /* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1010 if (dev->type == ARPHRD_ETHER ||
1011 dev->type == ARPHRD_LOOPBACK ||
1012 dev->type == ARPHRD_IPGRE ||
1013 dev->type == ARPHRD_SIT ||
1014 dev->type == ARPHRD_TUNNEL) {
1015 mdev = mpls_add_dev(dev);
1017 return notifier_from_errno(PTR_ERR(mdev));
1022 mdev = mpls_dev_get(dev);
1028 mpls_ifdown(dev, event);
1031 flags = dev_get_flags(dev);
1032 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1033 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1035 mpls_ifup(dev, RTNH_F_DEAD);
1038 flags = dev_get_flags(dev);
1039 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1040 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1042 mpls_ifdown(dev, event);
1044 case NETDEV_UNREGISTER:
1045 mpls_ifdown(dev, event);
1046 mdev = mpls_dev_get(dev);
1048 mpls_dev_sysctl_unregister(mdev);
1049 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1050 kfree_rcu(mdev, rcu);
1053 case NETDEV_CHANGENAME:
1054 mdev = mpls_dev_get(dev);
1058 mpls_dev_sysctl_unregister(mdev);
1059 err = mpls_dev_sysctl_register(dev, mdev);
1061 return notifier_from_errno(err);
1068 static struct notifier_block mpls_dev_notifier = {
1069 .notifier_call = mpls_dev_notify,
1072 static int nla_put_via(struct sk_buff *skb,
1073 u8 table, const void *addr, int alen)
1075 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1076 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1080 int family = AF_UNSPEC;
1082 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1086 if (table <= NEIGH_NR_TABLES)
1087 family = table_to_family[table];
1089 via = nla_data(nla);
1090 via->rtvia_family = family;
1091 memcpy(via->rtvia_addr, addr, alen);
1095 int nla_put_labels(struct sk_buff *skb, int attrtype,
1096 u8 labels, const u32 label[])
1099 struct mpls_shim_hdr *nla_label;
1102 nla = nla_reserve(skb, attrtype, labels*4);
1106 nla_label = nla_data(nla);
1108 for (i = labels - 1; i >= 0; i--) {
1109 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1115 EXPORT_SYMBOL_GPL(nla_put_labels);
1117 int nla_get_labels(const struct nlattr *nla,
1118 u32 max_labels, u8 *labels, u32 label[])
1120 unsigned len = nla_len(nla);
1121 unsigned nla_labels;
1122 struct mpls_shim_hdr *nla_label;
1126 /* len needs to be an even multiple of 4 (the label size) */
1130 /* Limit the number of new labels allowed */
1132 if (nla_labels > max_labels)
1135 nla_label = nla_data(nla);
1137 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1138 struct mpls_entry_decoded dec;
1139 dec = mpls_entry_decode(nla_label + i);
1141 /* Ensure the bottom of stack flag is properly set
1142 * and ttl and tc are both clear.
1144 if ((dec.bos != bos) || dec.ttl || dec.tc)
1147 switch (dec.label) {
1148 case MPLS_LABEL_IMPLNULL:
1149 /* RFC3032: This is a label that an LSR may
1150 * assign and distribute, but which never
1151 * actually appears in the encapsulation.
1156 label[i] = dec.label;
1158 *labels = nla_labels;
1161 EXPORT_SYMBOL_GPL(nla_get_labels);
1163 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1164 u8 *via_table, u8 via_addr[])
1166 struct rtvia *via = nla_data(nla);
1170 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1172 alen = nla_len(nla) -
1173 offsetof(struct rtvia, rtvia_addr);
1174 if (alen > MAX_VIA_ALEN)
1177 /* Validate the address family */
1178 switch (via->rtvia_family) {
1180 *via_table = NEIGH_LINK_TABLE;
1183 *via_table = NEIGH_ARP_TABLE;
1188 *via_table = NEIGH_ND_TABLE;
1193 /* Unsupported address family */
1197 memcpy(via_addr, via->rtvia_addr, alen);
1205 static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
1206 struct mpls_route_config *cfg)
1209 struct nlattr *tb[RTA_MAX+1];
1213 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1218 rtm = nlmsg_data(nlh);
1219 memset(cfg, 0, sizeof(*cfg));
1221 if (rtm->rtm_family != AF_MPLS)
1223 if (rtm->rtm_dst_len != 20)
1225 if (rtm->rtm_src_len != 0)
1227 if (rtm->rtm_tos != 0)
1229 if (rtm->rtm_table != RT_TABLE_MAIN)
1231 /* Any value is acceptable for rtm_protocol */
1233 /* As mpls uses destination specific addresses
1234 * (or source specific address in the case of multicast)
1235 * all addresses have universal scope.
1237 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1239 if (rtm->rtm_type != RTN_UNICAST)
1241 if (rtm->rtm_flags != 0)
1244 cfg->rc_label = LABEL_NOT_SPECIFIED;
1245 cfg->rc_protocol = rtm->rtm_protocol;
1246 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1247 cfg->rc_nlflags = nlh->nlmsg_flags;
1248 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1249 cfg->rc_nlinfo.nlh = nlh;
1250 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1252 for (index = 0; index <= RTA_MAX; index++) {
1253 struct nlattr *nla = tb[index];
1259 cfg->rc_ifindex = nla_get_u32(nla);
1262 if (nla_get_labels(nla, MAX_NEW_LABELS,
1263 &cfg->rc_output_labels,
1264 cfg->rc_output_label))
1270 if (nla_get_labels(nla, 1, &label_count,
1274 /* Reserved labels may not be set */
1275 if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1282 if (nla_get_via(nla, &cfg->rc_via_alen,
1283 &cfg->rc_via_table, cfg->rc_via))
1289 cfg->rc_mp = nla_data(nla);
1290 cfg->rc_mp_len = nla_len(nla);
1294 /* Unsupported attribute */
1304 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1306 struct mpls_route_config cfg;
1309 err = rtm_to_route_config(skb, nlh, &cfg);
1313 return mpls_route_del(&cfg);
1317 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1319 struct mpls_route_config cfg;
1322 err = rtm_to_route_config(skb, nlh, &cfg);
1326 return mpls_route_add(&cfg);
1329 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1330 u32 label, struct mpls_route *rt, int flags)
1332 struct net_device *dev;
1333 struct nlmsghdr *nlh;
1336 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1340 rtm = nlmsg_data(nlh);
1341 rtm->rtm_family = AF_MPLS;
1342 rtm->rtm_dst_len = 20;
1343 rtm->rtm_src_len = 0;
1345 rtm->rtm_table = RT_TABLE_MAIN;
1346 rtm->rtm_protocol = rt->rt_protocol;
1347 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1348 rtm->rtm_type = RTN_UNICAST;
1351 if (nla_put_labels(skb, RTA_DST, 1, &label))
1352 goto nla_put_failure;
1353 if (rt->rt_nhn == 1) {
1354 const struct mpls_nh *nh = rt->rt_nh;
1356 if (nh->nh_labels &&
1357 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1359 goto nla_put_failure;
1360 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1361 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1363 goto nla_put_failure;
1364 dev = rtnl_dereference(nh->nh_dev);
1365 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1366 goto nla_put_failure;
1367 if (nh->nh_flags & RTNH_F_LINKDOWN)
1368 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1369 if (nh->nh_flags & RTNH_F_DEAD)
1370 rtm->rtm_flags |= RTNH_F_DEAD;
1372 struct rtnexthop *rtnh;
1377 mp = nla_nest_start(skb, RTA_MULTIPATH);
1379 goto nla_put_failure;
1382 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1384 goto nla_put_failure;
1386 dev = rtnl_dereference(nh->nh_dev);
1388 rtnh->rtnh_ifindex = dev->ifindex;
1389 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1390 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1393 if (nh->nh_flags & RTNH_F_DEAD) {
1394 rtnh->rtnh_flags |= RTNH_F_DEAD;
1398 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1401 goto nla_put_failure;
1402 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1403 nla_put_via(skb, nh->nh_via_table,
1404 mpls_nh_via(rt, nh),
1406 goto nla_put_failure;
1408 /* length of rtnetlink header + attributes */
1409 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1410 } endfor_nexthops(rt);
1412 if (linkdown == rt->rt_nhn)
1413 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1414 if (dead == rt->rt_nhn)
1415 rtm->rtm_flags |= RTNH_F_DEAD;
1417 nla_nest_end(skb, mp);
1420 nlmsg_end(skb, nlh);
1424 nlmsg_cancel(skb, nlh);
1428 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1430 struct net *net = sock_net(skb->sk);
1431 struct mpls_route __rcu **platform_label;
1432 size_t platform_labels;
1437 index = cb->args[0];
1438 if (index < MPLS_LABEL_FIRST_UNRESERVED)
1439 index = MPLS_LABEL_FIRST_UNRESERVED;
1441 platform_label = rtnl_dereference(net->mpls.platform_label);
1442 platform_labels = net->mpls.platform_labels;
1443 for (; index < platform_labels; index++) {
1444 struct mpls_route *rt;
1445 rt = rtnl_dereference(platform_label[index]);
1449 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1450 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1451 index, rt, NLM_F_MULTI) < 0)
1454 cb->args[0] = index;
1459 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1462 NLMSG_ALIGN(sizeof(struct rtmsg))
1463 + nla_total_size(4); /* RTA_DST */
1465 if (rt->rt_nhn == 1) {
1466 struct mpls_nh *nh = rt->rt_nh;
1469 payload += nla_total_size(4); /* RTA_OIF */
1470 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1471 payload += nla_total_size(2 + nh->nh_via_alen);
1472 if (nh->nh_labels) /* RTA_NEWDST */
1473 payload += nla_total_size(nh->nh_labels * 4);
1475 /* each nexthop is packed in an attribute */
1479 nhsize += nla_total_size(sizeof(struct rtnexthop));
1481 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1482 nhsize += nla_total_size(2 + nh->nh_via_alen);
1484 nhsize += nla_total_size(nh->nh_labels * 4);
1485 } endfor_nexthops(rt);
1486 /* nested attribute */
1487 payload += nla_total_size(nhsize);
1493 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1494 struct nlmsghdr *nlh, struct net *net, u32 portid,
1495 unsigned int nlm_flags)
1497 struct sk_buff *skb;
1498 u32 seq = nlh ? nlh->nlmsg_seq : 0;
1501 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1505 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1507 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1508 WARN_ON(err == -EMSGSIZE);
1512 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1517 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1520 static int resize_platform_label_table(struct net *net, size_t limit)
1522 size_t size = sizeof(struct mpls_route *) * limit;
1525 struct mpls_route __rcu **labels = NULL, **old;
1526 struct mpls_route *rt0 = NULL, *rt2 = NULL;
1530 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1532 labels = vzalloc(size);
1538 /* In case the predefined labels need to be populated */
1539 if (limit > MPLS_LABEL_IPV4NULL) {
1540 struct net_device *lo = net->loopback_dev;
1541 rt0 = mpls_rt_alloc(1, lo->addr_len);
1544 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1545 rt0->rt_protocol = RTPROT_KERNEL;
1546 rt0->rt_payload_type = MPT_IPV4;
1547 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1548 rt0->rt_nh->nh_via_alen = lo->addr_len;
1549 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1552 if (limit > MPLS_LABEL_IPV6NULL) {
1553 struct net_device *lo = net->loopback_dev;
1554 rt2 = mpls_rt_alloc(1, lo->addr_len);
1557 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1558 rt2->rt_protocol = RTPROT_KERNEL;
1559 rt2->rt_payload_type = MPT_IPV6;
1560 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1561 rt2->rt_nh->nh_via_alen = lo->addr_len;
1562 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1567 /* Remember the original table */
1568 old = rtnl_dereference(net->mpls.platform_label);
1569 old_limit = net->mpls.platform_labels;
1571 /* Free any labels beyond the new table */
1572 for (index = limit; index < old_limit; index++)
1573 mpls_route_update(net, index, NULL, NULL);
1575 /* Copy over the old labels */
1577 if (old_limit < limit)
1578 cp_size = old_limit * sizeof(struct mpls_route *);
1580 memcpy(labels, old, cp_size);
1582 /* If needed set the predefined labels */
1583 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1584 (limit > MPLS_LABEL_IPV6NULL)) {
1585 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1589 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1590 (limit > MPLS_LABEL_IPV4NULL)) {
1591 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1595 /* Update the global pointers */
1596 net->mpls.platform_labels = limit;
1597 rcu_assign_pointer(net->mpls.platform_label, labels);
1618 static int mpls_platform_labels(struct ctl_table *table, int write,
1619 void __user *buffer, size_t *lenp, loff_t *ppos)
1621 struct net *net = table->data;
1622 int platform_labels = net->mpls.platform_labels;
1624 struct ctl_table tmp = {
1625 .procname = table->procname,
1626 .data = &platform_labels,
1627 .maxlen = sizeof(int),
1628 .mode = table->mode,
1630 .extra2 = &label_limit,
1633 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1635 if (write && ret == 0)
1636 ret = resize_platform_label_table(net, platform_labels);
1641 static const struct ctl_table mpls_table[] = {
1643 .procname = "platform_labels",
1645 .maxlen = sizeof(int),
1647 .proc_handler = mpls_platform_labels,
1652 static int mpls_net_init(struct net *net)
1654 struct ctl_table *table;
1656 net->mpls.platform_labels = 0;
1657 net->mpls.platform_label = NULL;
1659 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1663 table[0].data = net;
1664 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1665 if (net->mpls.ctl == NULL) {
1673 static void mpls_net_exit(struct net *net)
1675 struct mpls_route __rcu **platform_label;
1676 size_t platform_labels;
1677 struct ctl_table *table;
1680 table = net->mpls.ctl->ctl_table_arg;
1681 unregister_net_sysctl_table(net->mpls.ctl);
1684 /* An rcu grace period has passed since there was a device in
1685 * the network namespace (and thus the last in flight packet)
1686 * left this network namespace. This is because
1687 * unregister_netdevice_many and netdev_run_todo has completed
1688 * for each network device that was in this network namespace.
1690 * As such no additional rcu synchronization is necessary when
1691 * freeing the platform_label table.
1694 platform_label = rtnl_dereference(net->mpls.platform_label);
1695 platform_labels = net->mpls.platform_labels;
1696 for (index = 0; index < platform_labels; index++) {
1697 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1698 RCU_INIT_POINTER(platform_label[index], NULL);
1703 kvfree(platform_label);
1706 static struct pernet_operations mpls_net_ops = {
1707 .init = mpls_net_init,
1708 .exit = mpls_net_exit,
1711 static int __init mpls_init(void)
1715 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1717 err = register_pernet_subsys(&mpls_net_ops);
1721 err = register_netdevice_notifier(&mpls_dev_notifier);
1723 goto out_unregister_pernet;
1725 dev_add_pack(&mpls_packet_type);
1727 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1728 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1729 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1734 out_unregister_pernet:
1735 unregister_pernet_subsys(&mpls_net_ops);
1738 module_init(mpls_init);
1740 static void __exit mpls_exit(void)
1742 rtnl_unregister_all(PF_MPLS);
1743 dev_remove_pack(&mpls_packet_type);
1744 unregister_netdevice_notifier(&mpls_dev_notifier);
1745 unregister_pernet_subsys(&mpls_net_ops);
1747 module_exit(mpls_exit);
1749 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1750 MODULE_LICENSE("GPL v2");
1751 MODULE_ALIAS_NETPROTO(PF_MPLS);