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
31 static int label_limit = (1 << 20) - 1;
33 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
34 struct nlmsghdr *nlh, struct net *net, u32 portid,
35 unsigned int nlm_flags);
37 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
39 struct mpls_route *rt = NULL;
41 if (index < net->mpls.platform_labels) {
42 struct mpls_route __rcu **platform_label =
43 rcu_dereference(net->mpls.platform_label);
44 rt = rcu_dereference(platform_label[index]);
49 static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
51 return rcu_dereference_rtnl(dev->mpls_ptr);
54 bool mpls_output_possible(const struct net_device *dev)
56 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
58 EXPORT_SYMBOL_GPL(mpls_output_possible);
60 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
62 u8 *nh0_via = PTR_ALIGN((u8 *)&rt->rt_nh[rt->rt_nhn], VIA_ALEN_ALIGN);
63 int nh_index = nh - rt->rt_nh;
65 return nh0_via + rt->rt_max_alen * nh_index;
68 static const u8 *mpls_nh_via(const struct mpls_route *rt,
69 const struct mpls_nh *nh)
71 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
74 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
76 /* The size of the layer 2.5 labels to be added for this route */
77 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
80 unsigned int mpls_dev_mtu(const struct net_device *dev)
82 /* The amount of data the layer 2 frame can hold */
85 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
87 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
92 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
97 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
99 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
100 struct sk_buff *skb, bool bos)
102 struct mpls_entry_decoded dec;
103 struct mpls_shim_hdr *hdr;
104 bool eli_seen = false;
109 /* No need to look further into packet if there's only
115 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS && !bos;
117 if (!pskb_may_pull(skb, sizeof(*hdr) * label_index))
120 /* Read and decode the current label */
121 hdr = mpls_hdr(skb) + label_index;
122 dec = mpls_entry_decode(hdr);
124 /* RFC6790 - reserved labels MUST NOT be used as keys
125 * for the load-balancing function
127 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
128 hash = jhash_1word(dec.label, hash);
130 /* The entropy label follows the entropy label
131 * indicator, so this means that the entropy
132 * label was just added to the hash - no need to
133 * go any deeper either in the label stack or in the
138 } else if (dec.label == MPLS_LABEL_ENTROPY) {
143 if (bos && pskb_may_pull(skb, sizeof(*hdr) * label_index +
144 sizeof(struct iphdr))) {
145 const struct iphdr *v4hdr;
147 v4hdr = (const struct iphdr *)(mpls_hdr(skb) +
149 if (v4hdr->version == 4) {
150 hash = jhash_3words(ntohl(v4hdr->saddr),
152 v4hdr->protocol, hash);
153 } else if (v4hdr->version == 6 &&
154 pskb_may_pull(skb, sizeof(*hdr) * label_index +
155 sizeof(struct ipv6hdr))) {
156 const struct ipv6hdr *v6hdr;
158 v6hdr = (const struct ipv6hdr *)(mpls_hdr(skb) +
161 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
162 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
163 hash = jhash_1word(v6hdr->nexthdr, hash);
168 nh_index = hash % rt->rt_nhn;
170 return &rt->rt_nh[nh_index];
173 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
174 struct mpls_entry_decoded dec)
176 enum mpls_payload_type payload_type;
177 bool success = false;
179 /* The IPv4 code below accesses through the IPv4 header
180 * checksum, which is 12 bytes into the packet.
181 * The IPv6 code below accesses through the IPv6 hop limit
182 * which is 8 bytes into the packet.
184 * For all supported cases there should always be at least 12
185 * bytes of packet data present. The IPv4 header is 20 bytes
186 * without options and the IPv6 header is always 40 bytes
189 if (!pskb_may_pull(skb, 12))
192 payload_type = rt->rt_payload_type;
193 if (payload_type == MPT_UNSPEC)
194 payload_type = ip_hdr(skb)->version;
196 switch (payload_type) {
198 struct iphdr *hdr4 = ip_hdr(skb);
199 skb->protocol = htons(ETH_P_IP);
200 csum_replace2(&hdr4->check,
201 htons(hdr4->ttl << 8),
202 htons(dec.ttl << 8));
208 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
209 skb->protocol = htons(ETH_P_IPV6);
210 hdr6->hop_limit = dec.ttl;
221 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
222 struct packet_type *pt, struct net_device *orig_dev)
224 struct net *net = dev_net(dev);
225 struct mpls_shim_hdr *hdr;
226 struct mpls_route *rt;
228 struct mpls_entry_decoded dec;
229 struct net_device *out_dev;
230 struct mpls_dev *mdev;
232 unsigned int new_header_size;
236 /* Careful this entire function runs inside of an rcu critical section */
238 mdev = mpls_dev_get(dev);
239 if (!mdev || !mdev->input_enabled)
242 if (skb->pkt_type != PACKET_HOST)
245 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
248 if (!pskb_may_pull(skb, sizeof(*hdr)))
251 /* Read and decode the label */
253 dec = mpls_entry_decode(hdr);
256 skb_pull(skb, sizeof(*hdr));
257 skb_reset_network_header(skb);
261 rt = mpls_route_input_rcu(net, dec.label);
265 nh = mpls_select_multipath(rt, skb, dec.bos);
269 /* Find the output device */
270 out_dev = rcu_dereference(nh->nh_dev);
271 if (!mpls_output_possible(out_dev))
274 if (skb_warn_if_lro(skb))
277 skb_forward_csum(skb);
279 /* Verify ttl is valid */
284 /* Verify the destination can hold the packet */
285 new_header_size = mpls_nh_header_size(nh);
286 mtu = mpls_dev_mtu(out_dev);
287 if (mpls_pkt_too_big(skb, mtu - new_header_size))
290 hh_len = LL_RESERVED_SPACE(out_dev);
291 if (!out_dev->header_ops)
294 /* Ensure there is enough space for the headers in the skb */
295 if (skb_cow(skb, hh_len + new_header_size))
299 skb->protocol = htons(ETH_P_MPLS_UC);
301 if (unlikely(!new_header_size && dec.bos)) {
302 /* Penultimate hop popping */
303 if (!mpls_egress(rt, skb, dec))
308 skb_push(skb, new_header_size);
309 skb_reset_network_header(skb);
310 /* Push the new labels */
313 for (i = nh->nh_labels - 1; i >= 0; i--) {
314 hdr[i] = mpls_entry_encode(nh->nh_label[i],
320 err = neigh_xmit(nh->nh_via_table, out_dev, mpls_nh_via(rt, nh), skb);
322 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
331 static struct packet_type mpls_packet_type __read_mostly = {
332 .type = cpu_to_be16(ETH_P_MPLS_UC),
333 .func = mpls_forward,
336 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
337 [RTA_DST] = { .type = NLA_U32 },
338 [RTA_OIF] = { .type = NLA_U32 },
341 struct mpls_route_config {
346 u8 rc_via[MAX_VIA_ALEN];
349 u32 rc_output_label[MAX_NEW_LABELS];
351 enum mpls_payload_type rc_payload_type;
352 struct nl_info rc_nlinfo;
353 struct rtnexthop *rc_mp;
357 static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
359 u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
360 struct mpls_route *rt;
362 rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
364 num_nh * max_alen_aligned,
368 rt->rt_max_alen = max_alen_aligned;
374 static void mpls_rt_free(struct mpls_route *rt)
377 kfree_rcu(rt, rt_rcu);
380 static void mpls_notify_route(struct net *net, unsigned index,
381 struct mpls_route *old, struct mpls_route *new,
382 const struct nl_info *info)
384 struct nlmsghdr *nlh = info ? info->nlh : NULL;
385 unsigned portid = info ? info->portid : 0;
386 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
387 struct mpls_route *rt = new ? new : old;
388 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
389 /* Ignore reserved labels for now */
390 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
391 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
394 static void mpls_route_update(struct net *net, unsigned index,
395 struct mpls_route *new,
396 const struct nl_info *info)
398 struct mpls_route __rcu **platform_label;
399 struct mpls_route *rt;
403 platform_label = rtnl_dereference(net->mpls.platform_label);
404 rt = rtnl_dereference(platform_label[index]);
405 rcu_assign_pointer(platform_label[index], new);
407 mpls_notify_route(net, index, rt, new, info);
409 /* If we removed a route free it now */
413 static unsigned find_free_label(struct net *net)
415 struct mpls_route __rcu **platform_label;
416 size_t platform_labels;
419 platform_label = rtnl_dereference(net->mpls.platform_label);
420 platform_labels = net->mpls.platform_labels;
421 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
423 if (!rtnl_dereference(platform_label[index]))
426 return LABEL_NOT_SPECIFIED;
429 #if IS_ENABLED(CONFIG_INET)
430 static struct net_device *inet_fib_lookup_dev(struct net *net,
433 struct net_device *dev;
435 struct in_addr daddr;
437 memcpy(&daddr, addr, sizeof(struct in_addr));
438 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
450 static struct net_device *inet_fib_lookup_dev(struct net *net,
453 return ERR_PTR(-EAFNOSUPPORT);
457 #if IS_ENABLED(CONFIG_IPV6)
458 static struct net_device *inet6_fib_lookup_dev(struct net *net,
461 struct net_device *dev;
462 struct dst_entry *dst;
467 return ERR_PTR(-EAFNOSUPPORT);
469 memset(&fl6, 0, sizeof(fl6));
470 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
471 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
482 static struct net_device *inet6_fib_lookup_dev(struct net *net,
485 return ERR_PTR(-EAFNOSUPPORT);
489 static struct net_device *find_outdev(struct net *net,
490 struct mpls_route *rt,
491 struct mpls_nh *nh, int oif)
493 struct net_device *dev = NULL;
496 switch (nh->nh_via_table) {
497 case NEIGH_ARP_TABLE:
498 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
501 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
503 case NEIGH_LINK_TABLE:
507 dev = dev_get_by_index(net, oif);
511 return ERR_PTR(-ENODEV);
513 /* The caller is holding rtnl anyways, so release the dev reference */
519 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
520 struct mpls_nh *nh, int oif)
522 struct net_device *dev = NULL;
525 dev = find_outdev(net, rt, nh, oif);
532 /* Ensure this is a supported device */
534 if (!mpls_dev_get(dev))
537 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
538 (dev->addr_len != nh->nh_via_alen))
541 RCU_INIT_POINTER(nh->nh_dev, dev);
549 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
550 struct mpls_route *rt)
552 struct net *net = cfg->rc_nlinfo.nl_net;
553 struct mpls_nh *nh = rt->rt_nh;
561 /* Ensure only a supported number of labels are present */
562 if (cfg->rc_output_labels > MAX_NEW_LABELS)
565 nh->nh_labels = cfg->rc_output_labels;
566 for (i = 0; i < nh->nh_labels; i++)
567 nh->nh_label[i] = cfg->rc_output_label[i];
569 nh->nh_via_table = cfg->rc_via_table;
570 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
571 nh->nh_via_alen = cfg->rc_via_alen;
573 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
583 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
584 struct mpls_nh *nh, int oif,
585 struct nlattr *via, struct nlattr *newdst)
593 err = nla_get_labels(newdst, MAX_NEW_LABELS,
594 &nh->nh_labels, nh->nh_label);
599 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
600 __mpls_nh_via(rt, nh));
604 err = mpls_nh_assign_dev(net, rt, nh, oif);
614 static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
615 u8 cfg_via_alen, u8 *max_via_alen)
621 *max_via_alen = cfg_via_alen;
627 while (rtnh_ok(rtnh, remaining)) {
628 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
631 attrlen = rtnh_attrlen(rtnh);
632 nla = nla_find(attrs, attrlen, RTA_VIA);
633 if (nla && nla_len(nla) >=
634 offsetof(struct rtvia, rtvia_addr)) {
635 int via_alen = nla_len(nla) -
636 offsetof(struct rtvia, rtvia_addr);
638 if (via_alen <= MAX_VIA_ALEN)
639 *max_via_alen = max_t(u16, *max_via_alen,
644 rtnh = rtnh_next(rtnh, &remaining);
647 /* leftover implies invalid nexthop configuration, discard it */
648 return remaining > 0 ? 0 : nhs;
651 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
652 struct mpls_route *rt)
654 struct rtnexthop *rtnh = cfg->rc_mp;
655 struct nlattr *nla_via, *nla_newdst;
656 int remaining = cfg->rc_mp_len;
660 change_nexthops(rt) {
667 if (!rtnh_ok(rtnh, remaining))
670 /* neither weighted multipath nor any flags
673 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
676 attrlen = rtnh_attrlen(rtnh);
678 struct nlattr *attrs = rtnh_attrs(rtnh);
680 nla_via = nla_find(attrs, attrlen, RTA_VIA);
681 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
687 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
688 rtnh->rtnh_ifindex, nla_via,
693 rtnh = rtnh_next(rtnh, &remaining);
695 } endfor_nexthops(rt);
705 static int mpls_route_add(struct mpls_route_config *cfg)
707 struct mpls_route __rcu **platform_label;
708 struct net *net = cfg->rc_nlinfo.nl_net;
709 struct mpls_route *rt, *old;
715 index = cfg->rc_label;
717 /* If a label was not specified during insert pick one */
718 if ((index == LABEL_NOT_SPECIFIED) &&
719 (cfg->rc_nlflags & NLM_F_CREATE)) {
720 index = find_free_label(net);
723 /* Reserved labels may not be set */
724 if (index < MPLS_LABEL_FIRST_UNRESERVED)
727 /* The full 20 bit range may not be supported. */
728 if (index >= net->mpls.platform_labels)
731 /* Append makes no sense with mpls */
733 if (cfg->rc_nlflags & NLM_F_APPEND)
737 platform_label = rtnl_dereference(net->mpls.platform_label);
738 old = rtnl_dereference(platform_label[index]);
739 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
743 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
747 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
751 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
752 cfg->rc_via_alen, &max_via_alen);
757 rt = mpls_rt_alloc(nhs, max_via_alen);
761 rt->rt_protocol = cfg->rc_protocol;
762 rt->rt_payload_type = cfg->rc_payload_type;
765 err = mpls_nh_build_multi(cfg, rt);
767 err = mpls_nh_build_from_cfg(cfg, rt);
771 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
781 static int mpls_route_del(struct mpls_route_config *cfg)
783 struct net *net = cfg->rc_nlinfo.nl_net;
787 index = cfg->rc_label;
789 /* Reserved labels may not be removed */
790 if (index < MPLS_LABEL_FIRST_UNRESERVED)
793 /* The full 20 bit range may not be supported */
794 if (index >= net->mpls.platform_labels)
797 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
804 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
805 (&((struct mpls_dev *)0)->field)
807 static const struct ctl_table mpls_dev_table[] = {
810 .maxlen = sizeof(int),
812 .proc_handler = proc_dointvec,
813 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
818 static int mpls_dev_sysctl_register(struct net_device *dev,
819 struct mpls_dev *mdev)
821 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
822 struct ctl_table *table;
825 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
829 /* Table data contains only offsets relative to the base of
830 * the mdev at this point, so make them absolute.
832 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
833 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
835 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
837 mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
849 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
851 struct ctl_table *table;
853 table = mdev->sysctl->ctl_table_arg;
854 unregister_net_sysctl_table(mdev->sysctl);
858 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
860 struct mpls_dev *mdev;
865 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
869 err = mpls_dev_sysctl_register(dev, mdev);
873 rcu_assign_pointer(dev->mpls_ptr, mdev);
882 static void mpls_ifdown(struct net_device *dev)
884 struct mpls_route __rcu **platform_label;
885 struct net *net = dev_net(dev);
886 struct mpls_dev *mdev;
889 platform_label = rtnl_dereference(net->mpls.platform_label);
890 for (index = 0; index < net->mpls.platform_labels; index++) {
891 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
895 if (rtnl_dereference(nh->nh_dev) != dev)
898 } endfor_nexthops(rt);
901 mdev = mpls_dev_get(dev);
905 mpls_dev_sysctl_unregister(mdev);
907 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
909 kfree_rcu(mdev, rcu);
912 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
915 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
916 struct mpls_dev *mdev;
919 case NETDEV_REGISTER:
920 /* For now just support ethernet devices */
921 if ((dev->type == ARPHRD_ETHER) ||
922 (dev->type == ARPHRD_LOOPBACK)) {
923 mdev = mpls_add_dev(dev);
925 return notifier_from_errno(PTR_ERR(mdev));
929 case NETDEV_UNREGISTER:
932 case NETDEV_CHANGENAME:
933 mdev = mpls_dev_get(dev);
937 mpls_dev_sysctl_unregister(mdev);
938 err = mpls_dev_sysctl_register(dev, mdev);
940 return notifier_from_errno(err);
947 static struct notifier_block mpls_dev_notifier = {
948 .notifier_call = mpls_dev_notify,
951 static int nla_put_via(struct sk_buff *skb,
952 u8 table, const void *addr, int alen)
954 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
955 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
959 int family = AF_UNSPEC;
961 nla = nla_reserve(skb, RTA_VIA, alen + 2);
965 if (table <= NEIGH_NR_TABLES)
966 family = table_to_family[table];
969 via->rtvia_family = family;
970 memcpy(via->rtvia_addr, addr, alen);
974 int nla_put_labels(struct sk_buff *skb, int attrtype,
975 u8 labels, const u32 label[])
978 struct mpls_shim_hdr *nla_label;
981 nla = nla_reserve(skb, attrtype, labels*4);
985 nla_label = nla_data(nla);
987 for (i = labels - 1; i >= 0; i--) {
988 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
994 EXPORT_SYMBOL_GPL(nla_put_labels);
996 int nla_get_labels(const struct nlattr *nla,
997 u32 max_labels, u8 *labels, u32 label[])
999 unsigned len = nla_len(nla);
1000 unsigned nla_labels;
1001 struct mpls_shim_hdr *nla_label;
1005 /* len needs to be an even multiple of 4 (the label size) */
1009 /* Limit the number of new labels allowed */
1011 if (nla_labels > max_labels)
1014 nla_label = nla_data(nla);
1016 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1017 struct mpls_entry_decoded dec;
1018 dec = mpls_entry_decode(nla_label + i);
1020 /* Ensure the bottom of stack flag is properly set
1021 * and ttl and tc are both clear.
1023 if ((dec.bos != bos) || dec.ttl || dec.tc)
1026 switch (dec.label) {
1027 case MPLS_LABEL_IMPLNULL:
1028 /* RFC3032: This is a label that an LSR may
1029 * assign and distribute, but which never
1030 * actually appears in the encapsulation.
1035 label[i] = dec.label;
1037 *labels = nla_labels;
1040 EXPORT_SYMBOL_GPL(nla_get_labels);
1042 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1043 u8 *via_table, u8 via_addr[])
1045 struct rtvia *via = nla_data(nla);
1049 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1051 alen = nla_len(nla) -
1052 offsetof(struct rtvia, rtvia_addr);
1053 if (alen > MAX_VIA_ALEN)
1056 /* Validate the address family */
1057 switch (via->rtvia_family) {
1059 *via_table = NEIGH_LINK_TABLE;
1062 *via_table = NEIGH_ARP_TABLE;
1067 *via_table = NEIGH_ND_TABLE;
1072 /* Unsupported address family */
1076 memcpy(via_addr, via->rtvia_addr, alen);
1084 static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
1085 struct mpls_route_config *cfg)
1088 struct nlattr *tb[RTA_MAX+1];
1092 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1097 rtm = nlmsg_data(nlh);
1098 memset(cfg, 0, sizeof(*cfg));
1100 if (rtm->rtm_family != AF_MPLS)
1102 if (rtm->rtm_dst_len != 20)
1104 if (rtm->rtm_src_len != 0)
1106 if (rtm->rtm_tos != 0)
1108 if (rtm->rtm_table != RT_TABLE_MAIN)
1110 /* Any value is acceptable for rtm_protocol */
1112 /* As mpls uses destination specific addresses
1113 * (or source specific address in the case of multicast)
1114 * all addresses have universal scope.
1116 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1118 if (rtm->rtm_type != RTN_UNICAST)
1120 if (rtm->rtm_flags != 0)
1123 cfg->rc_label = LABEL_NOT_SPECIFIED;
1124 cfg->rc_protocol = rtm->rtm_protocol;
1125 cfg->rc_nlflags = nlh->nlmsg_flags;
1126 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1127 cfg->rc_nlinfo.nlh = nlh;
1128 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1130 for (index = 0; index <= RTA_MAX; index++) {
1131 struct nlattr *nla = tb[index];
1137 cfg->rc_ifindex = nla_get_u32(nla);
1140 if (nla_get_labels(nla, MAX_NEW_LABELS,
1141 &cfg->rc_output_labels,
1142 cfg->rc_output_label))
1148 if (nla_get_labels(nla, 1, &label_count,
1152 /* Reserved labels may not be set */
1153 if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1160 if (nla_get_via(nla, &cfg->rc_via_alen,
1161 &cfg->rc_via_table, cfg->rc_via))
1167 cfg->rc_mp = nla_data(nla);
1168 cfg->rc_mp_len = nla_len(nla);
1172 /* Unsupported attribute */
1182 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1184 struct mpls_route_config cfg;
1187 err = rtm_to_route_config(skb, nlh, &cfg);
1191 return mpls_route_del(&cfg);
1195 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1197 struct mpls_route_config cfg;
1200 err = rtm_to_route_config(skb, nlh, &cfg);
1204 return mpls_route_add(&cfg);
1207 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1208 u32 label, struct mpls_route *rt, int flags)
1210 struct net_device *dev;
1211 struct nlmsghdr *nlh;
1214 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1218 rtm = nlmsg_data(nlh);
1219 rtm->rtm_family = AF_MPLS;
1220 rtm->rtm_dst_len = 20;
1221 rtm->rtm_src_len = 0;
1223 rtm->rtm_table = RT_TABLE_MAIN;
1224 rtm->rtm_protocol = rt->rt_protocol;
1225 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1226 rtm->rtm_type = RTN_UNICAST;
1229 if (nla_put_labels(skb, RTA_DST, 1, &label))
1230 goto nla_put_failure;
1231 if (rt->rt_nhn == 1) {
1232 const struct mpls_nh *nh = rt->rt_nh;
1234 if (nh->nh_labels &&
1235 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1237 goto nla_put_failure;
1238 if (nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1240 goto nla_put_failure;
1241 dev = rtnl_dereference(nh->nh_dev);
1242 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1243 goto nla_put_failure;
1245 struct rtnexthop *rtnh;
1248 mp = nla_nest_start(skb, RTA_MULTIPATH);
1250 goto nla_put_failure;
1253 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1255 goto nla_put_failure;
1257 dev = rtnl_dereference(nh->nh_dev);
1259 rtnh->rtnh_ifindex = dev->ifindex;
1260 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1263 goto nla_put_failure;
1264 if (nla_put_via(skb, nh->nh_via_table,
1265 mpls_nh_via(rt, nh),
1267 goto nla_put_failure;
1269 /* length of rtnetlink header + attributes */
1270 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1271 } endfor_nexthops(rt);
1273 nla_nest_end(skb, mp);
1276 nlmsg_end(skb, nlh);
1280 nlmsg_cancel(skb, nlh);
1284 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1286 struct net *net = sock_net(skb->sk);
1287 struct mpls_route __rcu **platform_label;
1288 size_t platform_labels;
1293 index = cb->args[0];
1294 if (index < MPLS_LABEL_FIRST_UNRESERVED)
1295 index = MPLS_LABEL_FIRST_UNRESERVED;
1297 platform_label = rtnl_dereference(net->mpls.platform_label);
1298 platform_labels = net->mpls.platform_labels;
1299 for (; index < platform_labels; index++) {
1300 struct mpls_route *rt;
1301 rt = rtnl_dereference(platform_label[index]);
1305 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1306 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1307 index, rt, NLM_F_MULTI) < 0)
1310 cb->args[0] = index;
1315 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1318 NLMSG_ALIGN(sizeof(struct rtmsg))
1319 + nla_total_size(4); /* RTA_DST */
1321 if (rt->rt_nhn == 1) {
1322 struct mpls_nh *nh = rt->rt_nh;
1325 payload += nla_total_size(4); /* RTA_OIF */
1326 payload += nla_total_size(2 + nh->nh_via_alen); /* RTA_VIA */
1327 if (nh->nh_labels) /* RTA_NEWDST */
1328 payload += nla_total_size(nh->nh_labels * 4);
1330 /* each nexthop is packed in an attribute */
1334 nhsize += nla_total_size(sizeof(struct rtnexthop));
1335 nhsize += nla_total_size(2 + nh->nh_via_alen);
1337 nhsize += nla_total_size(nh->nh_labels * 4);
1338 } endfor_nexthops(rt);
1339 /* nested attribute */
1340 payload += nla_total_size(nhsize);
1346 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1347 struct nlmsghdr *nlh, struct net *net, u32 portid,
1348 unsigned int nlm_flags)
1350 struct sk_buff *skb;
1351 u32 seq = nlh ? nlh->nlmsg_seq : 0;
1354 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1358 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1360 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1361 WARN_ON(err == -EMSGSIZE);
1365 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1370 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1373 static int resize_platform_label_table(struct net *net, size_t limit)
1375 size_t size = sizeof(struct mpls_route *) * limit;
1378 struct mpls_route __rcu **labels = NULL, **old;
1379 struct mpls_route *rt0 = NULL, *rt2 = NULL;
1383 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1385 labels = vzalloc(size);
1391 /* In case the predefined labels need to be populated */
1392 if (limit > MPLS_LABEL_IPV4NULL) {
1393 struct net_device *lo = net->loopback_dev;
1394 rt0 = mpls_rt_alloc(1, lo->addr_len);
1397 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1398 rt0->rt_protocol = RTPROT_KERNEL;
1399 rt0->rt_payload_type = MPT_IPV4;
1400 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1401 rt0->rt_nh->nh_via_alen = lo->addr_len;
1402 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1405 if (limit > MPLS_LABEL_IPV6NULL) {
1406 struct net_device *lo = net->loopback_dev;
1407 rt2 = mpls_rt_alloc(1, lo->addr_len);
1410 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1411 rt2->rt_protocol = RTPROT_KERNEL;
1412 rt2->rt_payload_type = MPT_IPV6;
1413 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1414 rt2->rt_nh->nh_via_alen = lo->addr_len;
1415 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1420 /* Remember the original table */
1421 old = rtnl_dereference(net->mpls.platform_label);
1422 old_limit = net->mpls.platform_labels;
1424 /* Free any labels beyond the new table */
1425 for (index = limit; index < old_limit; index++)
1426 mpls_route_update(net, index, NULL, NULL);
1428 /* Copy over the old labels */
1430 if (old_limit < limit)
1431 cp_size = old_limit * sizeof(struct mpls_route *);
1433 memcpy(labels, old, cp_size);
1435 /* If needed set the predefined labels */
1436 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1437 (limit > MPLS_LABEL_IPV6NULL)) {
1438 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1442 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1443 (limit > MPLS_LABEL_IPV4NULL)) {
1444 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1448 /* Update the global pointers */
1449 net->mpls.platform_labels = limit;
1450 rcu_assign_pointer(net->mpls.platform_label, labels);
1471 static int mpls_platform_labels(struct ctl_table *table, int write,
1472 void __user *buffer, size_t *lenp, loff_t *ppos)
1474 struct net *net = table->data;
1475 int platform_labels = net->mpls.platform_labels;
1477 struct ctl_table tmp = {
1478 .procname = table->procname,
1479 .data = &platform_labels,
1480 .maxlen = sizeof(int),
1481 .mode = table->mode,
1483 .extra2 = &label_limit,
1486 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1488 if (write && ret == 0)
1489 ret = resize_platform_label_table(net, platform_labels);
1494 static const struct ctl_table mpls_table[] = {
1496 .procname = "platform_labels",
1498 .maxlen = sizeof(int),
1500 .proc_handler = mpls_platform_labels,
1505 static int mpls_net_init(struct net *net)
1507 struct ctl_table *table;
1509 net->mpls.platform_labels = 0;
1510 net->mpls.platform_label = NULL;
1512 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1516 table[0].data = net;
1517 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1518 if (net->mpls.ctl == NULL) {
1526 static void mpls_net_exit(struct net *net)
1528 struct mpls_route __rcu **platform_label;
1529 size_t platform_labels;
1530 struct ctl_table *table;
1533 table = net->mpls.ctl->ctl_table_arg;
1534 unregister_net_sysctl_table(net->mpls.ctl);
1537 /* An rcu grace period has passed since there was a device in
1538 * the network namespace (and thus the last in flight packet)
1539 * left this network namespace. This is because
1540 * unregister_netdevice_many and netdev_run_todo has completed
1541 * for each network device that was in this network namespace.
1543 * As such no additional rcu synchronization is necessary when
1544 * freeing the platform_label table.
1547 platform_label = rtnl_dereference(net->mpls.platform_label);
1548 platform_labels = net->mpls.platform_labels;
1549 for (index = 0; index < platform_labels; index++) {
1550 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1551 RCU_INIT_POINTER(platform_label[index], NULL);
1556 kvfree(platform_label);
1559 static struct pernet_operations mpls_net_ops = {
1560 .init = mpls_net_init,
1561 .exit = mpls_net_exit,
1564 static int __init mpls_init(void)
1568 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1570 err = register_pernet_subsys(&mpls_net_ops);
1574 err = register_netdevice_notifier(&mpls_dev_notifier);
1576 goto out_unregister_pernet;
1578 dev_add_pack(&mpls_packet_type);
1580 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1581 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1582 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1587 out_unregister_pernet:
1588 unregister_pernet_subsys(&mpls_net_ops);
1591 module_init(mpls_init);
1593 static void __exit mpls_exit(void)
1595 rtnl_unregister_all(PF_MPLS);
1596 dev_remove_pack(&mpls_packet_type);
1597 unregister_netdevice_notifier(&mpls_dev_notifier);
1598 unregister_pernet_subsys(&mpls_net_ops);
1600 module_exit(mpls_exit);
1602 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1603 MODULE_LICENSE("GPL v2");
1604 MODULE_ALIAS_NETPROTO(PF_MPLS);