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/netconf.h>
11 #include <linux/vmalloc.h>
12 #include <linux/percpu.h>
17 #include <net/ip_fib.h>
18 #include <net/netevent.h>
19 #include <net/ip_tunnels.h>
20 #include <net/netns/generic.h>
21 #if IS_ENABLED(CONFIG_IPV6)
24 #include <net/addrconf.h>
25 #include <net/nexthop.h>
28 /* max memory we will use for mpls_route */
29 #define MAX_MPLS_ROUTE_MEM 4096
31 /* Maximum number of labels to look ahead at when selecting a path of
34 #define MAX_MP_SELECT_LABELS 4
36 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
40 static int label_limit = (1 << 20) - 1;
41 static int ttl_max = 255;
43 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
44 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
46 return sizeof(struct mpls_shim_hdr);
49 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
50 .encap_hlen = ipgre_mpls_encap_hlen,
53 static int ipgre_tunnel_encap_add_mpls_ops(void)
55 return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
58 static void ipgre_tunnel_encap_del_mpls_ops(void)
60 ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
63 static int ipgre_tunnel_encap_add_mpls_ops(void)
68 static void ipgre_tunnel_encap_del_mpls_ops(void)
73 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
74 struct nlmsghdr *nlh, struct net *net, u32 portid,
75 unsigned int nlm_flags);
77 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
79 struct mpls_route *rt = NULL;
81 if (index < net->mpls.platform_labels) {
82 struct mpls_route __rcu **platform_label =
83 rcu_dereference(net->mpls.platform_label);
84 rt = rcu_dereference(platform_label[index]);
89 bool mpls_output_possible(const struct net_device *dev)
91 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
93 EXPORT_SYMBOL_GPL(mpls_output_possible);
95 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
97 return (u8 *)nh + rt->rt_via_offset;
100 static const u8 *mpls_nh_via(const struct mpls_route *rt,
101 const struct mpls_nh *nh)
103 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
106 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
108 /* The size of the layer 2.5 labels to be added for this route */
109 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
112 unsigned int mpls_dev_mtu(const struct net_device *dev)
114 /* The amount of data the layer 2 frame can hold */
117 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
119 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
124 if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
129 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
131 void mpls_stats_inc_outucastpkts(struct net_device *dev,
132 const struct sk_buff *skb)
134 struct mpls_dev *mdev;
136 if (skb->protocol == htons(ETH_P_MPLS_UC)) {
137 mdev = mpls_dev_get(dev);
139 MPLS_INC_STATS_LEN(mdev, skb->len,
142 } else if (skb->protocol == htons(ETH_P_IP)) {
143 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
144 #if IS_ENABLED(CONFIG_IPV6)
145 } else if (skb->protocol == htons(ETH_P_IPV6)) {
146 struct inet6_dev *in6dev = __in6_dev_get(dev);
149 IP6_UPD_PO_STATS(dev_net(dev), in6dev,
150 IPSTATS_MIB_OUT, skb->len);
154 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
156 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
158 struct mpls_entry_decoded dec;
159 unsigned int mpls_hdr_len = 0;
160 struct mpls_shim_hdr *hdr;
161 bool eli_seen = false;
165 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
167 mpls_hdr_len += sizeof(*hdr);
168 if (!pskb_may_pull(skb, mpls_hdr_len))
171 /* Read and decode the current label */
172 hdr = mpls_hdr(skb) + label_index;
173 dec = mpls_entry_decode(hdr);
175 /* RFC6790 - reserved labels MUST NOT be used as keys
176 * for the load-balancing function
178 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
179 hash = jhash_1word(dec.label, hash);
181 /* The entropy label follows the entropy label
182 * indicator, so this means that the entropy
183 * label was just added to the hash - no need to
184 * go any deeper either in the label stack or in the
189 } else if (dec.label == MPLS_LABEL_ENTROPY) {
196 /* found bottom label; does skb have room for a header? */
197 if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
198 const struct iphdr *v4hdr;
200 v4hdr = (const struct iphdr *)(hdr + 1);
201 if (v4hdr->version == 4) {
202 hash = jhash_3words(ntohl(v4hdr->saddr),
204 v4hdr->protocol, hash);
205 } else if (v4hdr->version == 6 &&
206 pskb_may_pull(skb, mpls_hdr_len +
207 sizeof(struct ipv6hdr))) {
208 const struct ipv6hdr *v6hdr;
210 v6hdr = (const struct ipv6hdr *)(hdr + 1);
211 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
212 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
213 hash = jhash_1word(v6hdr->nexthdr, hash);
223 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
225 return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
228 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
229 * a next hop (nh->nh_flags) are modified by netdev event handlers.
230 * Since those fields can change at any moment, use READ_ONCE to
233 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
241 /* No need to look further into packet if there's only
247 alive = READ_ONCE(rt->rt_nhn_alive);
251 hash = mpls_multipath_hash(rt, skb);
252 nh_index = hash % alive;
253 if (alive == rt->rt_nhn)
256 unsigned int nh_flags = READ_ONCE(nh->nh_flags);
258 if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
263 } endfor_nexthops(rt);
266 return mpls_get_nexthop(rt, nh_index);
269 static bool mpls_egress(struct net *net, struct mpls_route *rt,
270 struct sk_buff *skb, struct mpls_entry_decoded dec)
272 enum mpls_payload_type payload_type;
273 bool success = false;
275 /* The IPv4 code below accesses through the IPv4 header
276 * checksum, which is 12 bytes into the packet.
277 * The IPv6 code below accesses through the IPv6 hop limit
278 * which is 8 bytes into the packet.
280 * For all supported cases there should always be at least 12
281 * bytes of packet data present. The IPv4 header is 20 bytes
282 * without options and the IPv6 header is always 40 bytes
285 if (!pskb_may_pull(skb, 12))
288 payload_type = rt->rt_payload_type;
289 if (payload_type == MPT_UNSPEC)
290 payload_type = ip_hdr(skb)->version;
292 switch (payload_type) {
294 struct iphdr *hdr4 = ip_hdr(skb);
296 skb->protocol = htons(ETH_P_IP);
298 /* If propagating TTL, take the decremented TTL from
299 * the incoming MPLS header, otherwise decrement the
300 * TTL, but only if not 0 to avoid underflow.
302 if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
303 (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
304 net->mpls.ip_ttl_propagate))
307 new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
309 csum_replace2(&hdr4->check,
310 htons(hdr4->ttl << 8),
311 htons(new_ttl << 8));
317 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
318 skb->protocol = htons(ETH_P_IPV6);
320 /* If propagating TTL, take the decremented TTL from
321 * the incoming MPLS header, otherwise decrement the
322 * hop limit, but only if not 0 to avoid underflow.
324 if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
325 (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
326 net->mpls.ip_ttl_propagate))
327 hdr6->hop_limit = dec.ttl;
328 else if (hdr6->hop_limit)
329 hdr6->hop_limit = hdr6->hop_limit - 1;
334 /* Should have decided which protocol it is by now */
341 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
342 struct packet_type *pt, struct net_device *orig_dev)
344 struct net *net = dev_net(dev);
345 struct mpls_shim_hdr *hdr;
346 struct mpls_route *rt;
348 struct mpls_entry_decoded dec;
349 struct net_device *out_dev;
350 struct mpls_dev *out_mdev;
351 struct mpls_dev *mdev;
353 unsigned int new_header_size;
357 /* Careful this entire function runs inside of an rcu critical section */
359 mdev = mpls_dev_get(dev);
363 MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
366 if (!mdev->input_enabled) {
367 MPLS_INC_STATS(mdev, rx_dropped);
371 if (skb->pkt_type != PACKET_HOST)
374 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
377 if (!pskb_may_pull(skb, sizeof(*hdr)))
380 /* Read and decode the label */
382 dec = mpls_entry_decode(hdr);
384 rt = mpls_route_input_rcu(net, dec.label);
386 MPLS_INC_STATS(mdev, rx_noroute);
390 nh = mpls_select_multipath(rt, skb);
395 skb_pull(skb, sizeof(*hdr));
396 skb_reset_network_header(skb);
400 if (skb_warn_if_lro(skb))
403 skb_forward_csum(skb);
405 /* Verify ttl is valid */
410 /* Find the output device */
411 out_dev = rcu_dereference(nh->nh_dev);
412 if (!mpls_output_possible(out_dev))
415 /* Verify the destination can hold the packet */
416 new_header_size = mpls_nh_header_size(nh);
417 mtu = mpls_dev_mtu(out_dev);
418 if (mpls_pkt_too_big(skb, mtu - new_header_size))
421 hh_len = LL_RESERVED_SPACE(out_dev);
422 if (!out_dev->header_ops)
425 /* Ensure there is enough space for the headers in the skb */
426 if (skb_cow(skb, hh_len + new_header_size))
430 skb->protocol = htons(ETH_P_MPLS_UC);
432 if (unlikely(!new_header_size && dec.bos)) {
433 /* Penultimate hop popping */
434 if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
439 skb_push(skb, new_header_size);
440 skb_reset_network_header(skb);
441 /* Push the new labels */
444 for (i = nh->nh_labels - 1; i >= 0; i--) {
445 hdr[i] = mpls_entry_encode(nh->nh_label[i],
451 mpls_stats_inc_outucastpkts(out_dev, skb);
453 /* If via wasn't specified then send out using device address */
454 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
455 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
456 out_dev->dev_addr, skb);
458 err = neigh_xmit(nh->nh_via_table, out_dev,
459 mpls_nh_via(rt, nh), skb);
461 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
466 out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
468 MPLS_INC_STATS(out_mdev, tx_errors);
471 MPLS_INC_STATS(mdev, rx_errors);
477 static struct packet_type mpls_packet_type __read_mostly = {
478 .type = cpu_to_be16(ETH_P_MPLS_UC),
479 .func = mpls_forward,
482 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
483 [RTA_DST] = { .type = NLA_U32 },
484 [RTA_OIF] = { .type = NLA_U32 },
485 [RTA_TTL_PROPAGATE] = { .type = NLA_U8 },
488 struct mpls_route_config {
493 u8 rc_via[MAX_VIA_ALEN];
497 u32 rc_output_label[MAX_NEW_LABELS];
499 enum mpls_payload_type rc_payload_type;
500 struct nl_info rc_nlinfo;
501 struct rtnexthop *rc_mp;
505 /* all nexthops within a route have the same size based on max
506 * number of labels and max via length for a hop
508 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
510 u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
511 struct mpls_route *rt;
514 size = sizeof(*rt) + num_nh * nh_size;
515 if (size > MAX_MPLS_ROUTE_MEM)
516 return ERR_PTR(-EINVAL);
518 rt = kzalloc(size, GFP_KERNEL);
520 return ERR_PTR(-ENOMEM);
523 rt->rt_nhn_alive = num_nh;
524 rt->rt_nh_size = nh_size;
525 rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
530 static void mpls_rt_free(struct mpls_route *rt)
533 kfree_rcu(rt, rt_rcu);
536 static void mpls_notify_route(struct net *net, unsigned index,
537 struct mpls_route *old, struct mpls_route *new,
538 const struct nl_info *info)
540 struct nlmsghdr *nlh = info ? info->nlh : NULL;
541 unsigned portid = info ? info->portid : 0;
542 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
543 struct mpls_route *rt = new ? new : old;
544 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
545 /* Ignore reserved labels for now */
546 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
547 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
550 static void mpls_route_update(struct net *net, unsigned index,
551 struct mpls_route *new,
552 const struct nl_info *info)
554 struct mpls_route __rcu **platform_label;
555 struct mpls_route *rt;
559 platform_label = rtnl_dereference(net->mpls.platform_label);
560 rt = rtnl_dereference(platform_label[index]);
561 rcu_assign_pointer(platform_label[index], new);
563 mpls_notify_route(net, index, rt, new, info);
565 /* If we removed a route free it now */
569 static unsigned find_free_label(struct net *net)
571 struct mpls_route __rcu **platform_label;
572 size_t platform_labels;
575 platform_label = rtnl_dereference(net->mpls.platform_label);
576 platform_labels = net->mpls.platform_labels;
577 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
579 if (!rtnl_dereference(platform_label[index]))
582 return LABEL_NOT_SPECIFIED;
585 #if IS_ENABLED(CONFIG_INET)
586 static struct net_device *inet_fib_lookup_dev(struct net *net,
589 struct net_device *dev;
591 struct in_addr daddr;
593 memcpy(&daddr, addr, sizeof(struct in_addr));
594 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
606 static struct net_device *inet_fib_lookup_dev(struct net *net,
609 return ERR_PTR(-EAFNOSUPPORT);
613 #if IS_ENABLED(CONFIG_IPV6)
614 static struct net_device *inet6_fib_lookup_dev(struct net *net,
617 struct net_device *dev;
618 struct dst_entry *dst;
623 return ERR_PTR(-EAFNOSUPPORT);
625 memset(&fl6, 0, sizeof(fl6));
626 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
627 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
638 static struct net_device *inet6_fib_lookup_dev(struct net *net,
641 return ERR_PTR(-EAFNOSUPPORT);
645 static struct net_device *find_outdev(struct net *net,
646 struct mpls_route *rt,
647 struct mpls_nh *nh, int oif)
649 struct net_device *dev = NULL;
652 switch (nh->nh_via_table) {
653 case NEIGH_ARP_TABLE:
654 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
657 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
659 case NEIGH_LINK_TABLE:
663 dev = dev_get_by_index(net, oif);
667 return ERR_PTR(-ENODEV);
672 /* The caller is holding rtnl anyways, so release the dev reference */
678 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
679 struct mpls_nh *nh, int oif)
681 struct net_device *dev = NULL;
684 dev = find_outdev(net, rt, nh, oif);
691 /* Ensure this is a supported device */
693 if (!mpls_dev_get(dev))
696 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
697 (dev->addr_len != nh->nh_via_alen))
700 RCU_INIT_POINTER(nh->nh_dev, dev);
702 if (!(dev->flags & IFF_UP)) {
703 nh->nh_flags |= RTNH_F_DEAD;
707 flags = dev_get_flags(dev);
708 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
709 nh->nh_flags |= RTNH_F_LINKDOWN;
718 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
719 u8 via_addr[], struct netlink_ext_ack *extack)
721 struct rtvia *via = nla_data(nla);
725 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
726 NL_SET_ERR_MSG_ATTR(extack, nla,
727 "Invalid attribute length for RTA_VIA");
730 alen = nla_len(nla) -
731 offsetof(struct rtvia, rtvia_addr);
732 if (alen > MAX_VIA_ALEN) {
733 NL_SET_ERR_MSG_ATTR(extack, nla,
734 "Invalid address length for RTA_VIA");
738 /* Validate the address family */
739 switch (via->rtvia_family) {
741 *via_table = NEIGH_LINK_TABLE;
744 *via_table = NEIGH_ARP_TABLE;
749 *via_table = NEIGH_ND_TABLE;
754 /* Unsupported address family */
758 memcpy(via_addr, via->rtvia_addr, alen);
766 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
767 struct mpls_route *rt)
769 struct net *net = cfg->rc_nlinfo.nl_net;
770 struct mpls_nh *nh = rt->rt_nh;
777 nh->nh_labels = cfg->rc_output_labels;
778 for (i = 0; i < nh->nh_labels; i++)
779 nh->nh_label[i] = cfg->rc_output_label[i];
781 nh->nh_via_table = cfg->rc_via_table;
782 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
783 nh->nh_via_alen = cfg->rc_via_alen;
785 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
789 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
798 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
799 struct mpls_nh *nh, int oif, struct nlattr *via,
800 struct nlattr *newdst, u8 max_labels,
801 struct netlink_ext_ack *extack)
809 err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
810 nh->nh_label, extack);
816 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
817 __mpls_nh_via(rt, nh), extack);
821 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
824 err = mpls_nh_assign_dev(net, rt, nh, oif);
834 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
835 u8 cfg_via_alen, u8 *max_via_alen,
844 while (rtnh_ok(rtnh, remaining)) {
845 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
849 attrlen = rtnh_attrlen(rtnh);
850 nla = nla_find(attrs, attrlen, RTA_VIA);
851 if (nla && nla_len(nla) >=
852 offsetof(struct rtvia, rtvia_addr)) {
853 int via_alen = nla_len(nla) -
854 offsetof(struct rtvia, rtvia_addr);
856 if (via_alen <= MAX_VIA_ALEN)
857 *max_via_alen = max_t(u16, *max_via_alen,
861 nla = nla_find(attrs, attrlen, RTA_NEWDST);
863 nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
867 *max_labels = max_t(u8, *max_labels, n_labels);
869 /* number of nexthops is tracked by a u8.
870 * Check for overflow.
876 rtnh = rtnh_next(rtnh, &remaining);
879 /* leftover implies invalid nexthop configuration, discard it */
880 return remaining > 0 ? 0 : nhs;
883 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
884 struct mpls_route *rt, u8 max_labels,
885 struct netlink_ext_ack *extack)
887 struct rtnexthop *rtnh = cfg->rc_mp;
888 struct nlattr *nla_via, *nla_newdst;
889 int remaining = cfg->rc_mp_len;
893 change_nexthops(rt) {
900 if (!rtnh_ok(rtnh, remaining))
903 /* neither weighted multipath nor any flags
906 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
909 attrlen = rtnh_attrlen(rtnh);
911 struct nlattr *attrs = rtnh_attrs(rtnh);
913 nla_via = nla_find(attrs, attrlen, RTA_VIA);
914 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
917 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
918 rtnh->rtnh_ifindex, nla_via, nla_newdst,
923 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
926 rtnh = rtnh_next(rtnh, &remaining);
928 } endfor_nexthops(rt);
938 static bool mpls_label_ok(struct net *net, unsigned int index,
939 struct netlink_ext_ack *extack)
941 /* Reserved labels may not be set */
942 if (index < MPLS_LABEL_FIRST_UNRESERVED) {
943 NL_SET_ERR_MSG(extack,
944 "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
948 /* The full 20 bit range may not be supported. */
949 if (index >= net->mpls.platform_labels) {
950 NL_SET_ERR_MSG(extack,
951 "Label >= configured maximum in platform_labels");
958 static int mpls_route_add(struct mpls_route_config *cfg,
959 struct netlink_ext_ack *extack)
961 struct mpls_route __rcu **platform_label;
962 struct net *net = cfg->rc_nlinfo.nl_net;
963 struct mpls_route *rt, *old;
970 index = cfg->rc_label;
972 /* If a label was not specified during insert pick one */
973 if ((index == LABEL_NOT_SPECIFIED) &&
974 (cfg->rc_nlflags & NLM_F_CREATE)) {
975 index = find_free_label(net);
978 if (!mpls_label_ok(net, index, extack))
981 /* Append makes no sense with mpls */
983 if (cfg->rc_nlflags & NLM_F_APPEND) {
984 NL_SET_ERR_MSG(extack, "MPLS does not support route append");
989 platform_label = rtnl_dereference(net->mpls.platform_label);
990 old = rtnl_dereference(platform_label[index]);
991 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
995 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
999 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1004 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1005 cfg->rc_via_alen, &max_via_alen,
1008 max_via_alen = cfg->rc_via_alen;
1009 max_labels = cfg->rc_output_labels;
1014 NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1019 rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1025 rt->rt_protocol = cfg->rc_protocol;
1026 rt->rt_payload_type = cfg->rc_payload_type;
1027 rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1030 err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1032 err = mpls_nh_build_from_cfg(cfg, rt);
1036 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1046 static int mpls_route_del(struct mpls_route_config *cfg,
1047 struct netlink_ext_ack *extack)
1049 struct net *net = cfg->rc_nlinfo.nl_net;
1053 index = cfg->rc_label;
1055 if (!mpls_label_ok(net, index, extack))
1058 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1065 static void mpls_get_stats(struct mpls_dev *mdev,
1066 struct mpls_link_stats *stats)
1068 struct mpls_pcpu_stats *p;
1071 memset(stats, 0, sizeof(*stats));
1073 for_each_possible_cpu(i) {
1074 struct mpls_link_stats local;
1077 p = per_cpu_ptr(mdev->stats, i);
1079 start = u64_stats_fetch_begin(&p->syncp);
1081 } while (u64_stats_fetch_retry(&p->syncp, start));
1083 stats->rx_packets += local.rx_packets;
1084 stats->rx_bytes += local.rx_bytes;
1085 stats->tx_packets += local.tx_packets;
1086 stats->tx_bytes += local.tx_bytes;
1087 stats->rx_errors += local.rx_errors;
1088 stats->tx_errors += local.tx_errors;
1089 stats->rx_dropped += local.rx_dropped;
1090 stats->tx_dropped += local.tx_dropped;
1091 stats->rx_noroute += local.rx_noroute;
1095 static int mpls_fill_stats_af(struct sk_buff *skb,
1096 const struct net_device *dev)
1098 struct mpls_link_stats *stats;
1099 struct mpls_dev *mdev;
1102 mdev = mpls_dev_get(dev);
1106 nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1107 sizeof(struct mpls_link_stats),
1112 stats = nla_data(nla);
1113 mpls_get_stats(mdev, stats);
1118 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1120 struct mpls_dev *mdev;
1122 mdev = mpls_dev_get(dev);
1126 return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1129 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1130 u32 portid, u32 seq, int event,
1131 unsigned int flags, int type)
1133 struct nlmsghdr *nlh;
1134 struct netconfmsg *ncm;
1137 nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1142 if (type == NETCONFA_ALL)
1145 ncm = nlmsg_data(nlh);
1146 ncm->ncm_family = AF_MPLS;
1148 if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1149 goto nla_put_failure;
1151 if ((all || type == NETCONFA_INPUT) &&
1152 nla_put_s32(skb, NETCONFA_INPUT,
1153 mdev->input_enabled) < 0)
1154 goto nla_put_failure;
1156 nlmsg_end(skb, nlh);
1160 nlmsg_cancel(skb, nlh);
1164 static int mpls_netconf_msgsize_devconf(int type)
1166 int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1167 + nla_total_size(4); /* NETCONFA_IFINDEX */
1170 if (type == NETCONFA_ALL)
1173 if (all || type == NETCONFA_INPUT)
1174 size += nla_total_size(4);
1179 static void mpls_netconf_notify_devconf(struct net *net, int event,
1180 int type, struct mpls_dev *mdev)
1182 struct sk_buff *skb;
1185 skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1189 err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1191 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1192 WARN_ON(err == -EMSGSIZE);
1197 rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1201 rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1204 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1205 [NETCONFA_IFINDEX] = { .len = sizeof(int) },
1208 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1209 struct nlmsghdr *nlh,
1210 struct netlink_ext_ack *extack)
1212 struct net *net = sock_net(in_skb->sk);
1213 struct nlattr *tb[NETCONFA_MAX + 1];
1214 struct netconfmsg *ncm;
1215 struct net_device *dev;
1216 struct mpls_dev *mdev;
1217 struct sk_buff *skb;
1221 err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1222 devconf_mpls_policy, NULL);
1227 if (!tb[NETCONFA_IFINDEX])
1230 ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1231 dev = __dev_get_by_index(net, ifindex);
1235 mdev = mpls_dev_get(dev);
1240 skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1244 err = mpls_netconf_fill_devconf(skb, mdev,
1245 NETLINK_CB(in_skb).portid,
1246 nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1249 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1250 WARN_ON(err == -EMSGSIZE);
1254 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1259 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1260 struct netlink_callback *cb)
1262 struct net *net = sock_net(skb->sk);
1263 struct hlist_head *head;
1264 struct net_device *dev;
1265 struct mpls_dev *mdev;
1270 s_idx = idx = cb->args[1];
1272 for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1274 head = &net->dev_index_head[h];
1276 cb->seq = net->dev_base_seq;
1277 hlist_for_each_entry_rcu(dev, head, index_hlist) {
1280 mdev = mpls_dev_get(dev);
1283 if (mpls_netconf_fill_devconf(skb, mdev,
1284 NETLINK_CB(cb->skb).portid,
1288 NETCONFA_ALL) < 0) {
1292 nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1305 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
1306 (&((struct mpls_dev *)0)->field)
1308 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1309 void __user *buffer,
1310 size_t *lenp, loff_t *ppos)
1312 int oval = *(int *)ctl->data;
1313 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1316 struct mpls_dev *mdev = ctl->extra1;
1317 int i = (int *)ctl->data - (int *)mdev;
1318 struct net *net = ctl->extra2;
1319 int val = *(int *)ctl->data;
1321 if (i == offsetof(struct mpls_dev, input_enabled) &&
1323 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1324 NETCONFA_INPUT, mdev);
1331 static const struct ctl_table mpls_dev_table[] = {
1333 .procname = "input",
1334 .maxlen = sizeof(int),
1336 .proc_handler = mpls_conf_proc,
1337 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1342 static int mpls_dev_sysctl_register(struct net_device *dev,
1343 struct mpls_dev *mdev)
1345 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1346 struct net *net = dev_net(dev);
1347 struct ctl_table *table;
1350 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1354 /* Table data contains only offsets relative to the base of
1355 * the mdev at this point, so make them absolute.
1357 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1358 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1359 table[i].extra1 = mdev;
1360 table[i].extra2 = net;
1363 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1365 mdev->sysctl = register_net_sysctl(net, path, table);
1369 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1378 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1379 struct mpls_dev *mdev)
1381 struct net *net = dev_net(dev);
1382 struct ctl_table *table;
1384 table = mdev->sysctl->ctl_table_arg;
1385 unregister_net_sysctl_table(mdev->sysctl);
1388 mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1391 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1393 struct mpls_dev *mdev;
1399 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1401 return ERR_PTR(err);
1403 mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1407 for_each_possible_cpu(i) {
1408 struct mpls_pcpu_stats *mpls_stats;
1410 mpls_stats = per_cpu_ptr(mdev->stats, i);
1411 u64_stats_init(&mpls_stats->syncp);
1416 err = mpls_dev_sysctl_register(dev, mdev);
1420 rcu_assign_pointer(dev->mpls_ptr, mdev);
1425 free_percpu(mdev->stats);
1427 return ERR_PTR(err);
1430 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1432 struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1434 free_percpu(mdev->stats);
1438 static void mpls_ifdown(struct net_device *dev, int event)
1440 struct mpls_route __rcu **platform_label;
1441 struct net *net = dev_net(dev);
1445 platform_label = rtnl_dereference(net->mpls.platform_label);
1446 for (index = 0; index < net->mpls.platform_labels; index++) {
1447 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1454 change_nexthops(rt) {
1455 unsigned int nh_flags = nh->nh_flags;
1457 if (rtnl_dereference(nh->nh_dev) != dev)
1462 case NETDEV_UNREGISTER:
1463 nh_flags |= RTNH_F_DEAD;
1466 nh_flags |= RTNH_F_LINKDOWN;
1469 if (event == NETDEV_UNREGISTER)
1470 RCU_INIT_POINTER(nh->nh_dev, NULL);
1472 if (nh->nh_flags != nh_flags)
1473 WRITE_ONCE(nh->nh_flags, nh_flags);
1475 if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1477 if (!rtnl_dereference(nh->nh_dev))
1479 } endfor_nexthops(rt);
1481 WRITE_ONCE(rt->rt_nhn_alive, alive);
1483 /* if there are no more nexthops, delete the route */
1484 if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1485 mpls_route_update(net, index, NULL, NULL);
1489 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1491 struct mpls_route __rcu **platform_label;
1492 struct net *net = dev_net(dev);
1496 platform_label = rtnl_dereference(net->mpls.platform_label);
1497 for (index = 0; index < net->mpls.platform_labels; index++) {
1498 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1504 change_nexthops(rt) {
1505 unsigned int nh_flags = nh->nh_flags;
1506 struct net_device *nh_dev =
1507 rtnl_dereference(nh->nh_dev);
1509 if (!(nh_flags & flags)) {
1517 WRITE_ONCE(nh->nh_flags, nh_flags);
1518 } endfor_nexthops(rt);
1520 WRITE_ONCE(rt->rt_nhn_alive, alive);
1524 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1527 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1528 struct mpls_dev *mdev;
1531 if (event == NETDEV_REGISTER) {
1532 /* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1533 if (dev->type == ARPHRD_ETHER ||
1534 dev->type == ARPHRD_LOOPBACK ||
1535 dev->type == ARPHRD_IPGRE ||
1536 dev->type == ARPHRD_SIT ||
1537 dev->type == ARPHRD_TUNNEL) {
1538 mdev = mpls_add_dev(dev);
1540 return notifier_from_errno(PTR_ERR(mdev));
1545 mdev = mpls_dev_get(dev);
1551 mpls_ifdown(dev, event);
1554 flags = dev_get_flags(dev);
1555 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1556 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1558 mpls_ifup(dev, RTNH_F_DEAD);
1561 flags = dev_get_flags(dev);
1562 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1563 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1565 mpls_ifdown(dev, event);
1567 case NETDEV_UNREGISTER:
1568 mpls_ifdown(dev, event);
1569 mdev = mpls_dev_get(dev);
1571 mpls_dev_sysctl_unregister(dev, mdev);
1572 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1573 call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1576 case NETDEV_CHANGENAME:
1577 mdev = mpls_dev_get(dev);
1581 mpls_dev_sysctl_unregister(dev, mdev);
1582 err = mpls_dev_sysctl_register(dev, mdev);
1584 return notifier_from_errno(err);
1591 static struct notifier_block mpls_dev_notifier = {
1592 .notifier_call = mpls_dev_notify,
1595 static int nla_put_via(struct sk_buff *skb,
1596 u8 table, const void *addr, int alen)
1598 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1599 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1603 int family = AF_UNSPEC;
1605 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1609 if (table <= NEIGH_NR_TABLES)
1610 family = table_to_family[table];
1612 via = nla_data(nla);
1613 via->rtvia_family = family;
1614 memcpy(via->rtvia_addr, addr, alen);
1618 int nla_put_labels(struct sk_buff *skb, int attrtype,
1619 u8 labels, const u32 label[])
1622 struct mpls_shim_hdr *nla_label;
1625 nla = nla_reserve(skb, attrtype, labels*4);
1629 nla_label = nla_data(nla);
1631 for (i = labels - 1; i >= 0; i--) {
1632 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1638 EXPORT_SYMBOL_GPL(nla_put_labels);
1640 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1641 u32 label[], struct netlink_ext_ack *extack)
1643 unsigned len = nla_len(nla);
1644 struct mpls_shim_hdr *nla_label;
1649 /* len needs to be an even multiple of 4 (the label size). Number
1650 * of labels is a u8 so check for overflow.
1652 if (len & 3 || len / 4 > 255) {
1653 NL_SET_ERR_MSG_ATTR(extack, nla,
1654 "Invalid length for labels attribute");
1658 /* Limit the number of new labels allowed */
1660 if (nla_labels > max_labels) {
1661 NL_SET_ERR_MSG(extack, "Too many labels");
1665 /* when label == NULL, caller wants number of labels */
1669 nla_label = nla_data(nla);
1671 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1672 struct mpls_entry_decoded dec;
1673 dec = mpls_entry_decode(nla_label + i);
1675 /* Ensure the bottom of stack flag is properly set
1676 * and ttl and tc are both clear.
1679 NL_SET_ERR_MSG_ATTR(extack, nla,
1680 "TTL in label must be 0");
1685 NL_SET_ERR_MSG_ATTR(extack, nla,
1686 "Traffic class in label must be 0");
1690 if (dec.bos != bos) {
1691 NL_SET_BAD_ATTR(extack, nla);
1693 NL_SET_ERR_MSG(extack,
1694 "BOS bit must be set in first label");
1696 NL_SET_ERR_MSG(extack,
1697 "BOS bit can only be set in first label");
1702 switch (dec.label) {
1703 case MPLS_LABEL_IMPLNULL:
1704 /* RFC3032: This is a label that an LSR may
1705 * assign and distribute, but which never
1706 * actually appears in the encapsulation.
1708 NL_SET_ERR_MSG_ATTR(extack, nla,
1709 "Implicit NULL Label (3) can not be used in encapsulation");
1713 label[i] = dec.label;
1716 *labels = nla_labels;
1719 EXPORT_SYMBOL_GPL(nla_get_labels);
1721 static int rtm_to_route_config(struct sk_buff *skb,
1722 struct nlmsghdr *nlh,
1723 struct mpls_route_config *cfg,
1724 struct netlink_ext_ack *extack)
1727 struct nlattr *tb[RTA_MAX+1];
1731 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy,
1737 rtm = nlmsg_data(nlh);
1739 if (rtm->rtm_family != AF_MPLS) {
1740 NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1743 if (rtm->rtm_dst_len != 20) {
1744 NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1747 if (rtm->rtm_src_len != 0) {
1748 NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1751 if (rtm->rtm_tos != 0) {
1752 NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1755 if (rtm->rtm_table != RT_TABLE_MAIN) {
1756 NL_SET_ERR_MSG(extack,
1757 "MPLS only supports the main route table");
1760 /* Any value is acceptable for rtm_protocol */
1762 /* As mpls uses destination specific addresses
1763 * (or source specific address in the case of multicast)
1764 * all addresses have universal scope.
1766 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1767 NL_SET_ERR_MSG(extack,
1768 "Invalid route scope - MPLS only supports UNIVERSE");
1771 if (rtm->rtm_type != RTN_UNICAST) {
1772 NL_SET_ERR_MSG(extack,
1773 "Invalid route type - MPLS only supports UNICAST");
1776 if (rtm->rtm_flags != 0) {
1777 NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1781 cfg->rc_label = LABEL_NOT_SPECIFIED;
1782 cfg->rc_protocol = rtm->rtm_protocol;
1783 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1784 cfg->rc_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
1785 cfg->rc_nlflags = nlh->nlmsg_flags;
1786 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1787 cfg->rc_nlinfo.nlh = nlh;
1788 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1790 for (index = 0; index <= RTA_MAX; index++) {
1791 struct nlattr *nla = tb[index];
1797 cfg->rc_ifindex = nla_get_u32(nla);
1800 if (nla_get_labels(nla, MAX_NEW_LABELS,
1801 &cfg->rc_output_labels,
1802 cfg->rc_output_label, extack))
1808 if (nla_get_labels(nla, 1, &label_count,
1809 &cfg->rc_label, extack))
1812 if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1813 cfg->rc_label, extack))
1819 if (nla_get_via(nla, &cfg->rc_via_alen,
1820 &cfg->rc_via_table, cfg->rc_via,
1827 cfg->rc_mp = nla_data(nla);
1828 cfg->rc_mp_len = nla_len(nla);
1831 case RTA_TTL_PROPAGATE:
1833 u8 ttl_propagate = nla_get_u8(nla);
1835 if (ttl_propagate > 1) {
1836 NL_SET_ERR_MSG_ATTR(extack, nla,
1837 "RTA_TTL_PROPAGATE can only be 0 or 1");
1840 cfg->rc_ttl_propagate = ttl_propagate ?
1841 MPLS_TTL_PROP_ENABLED :
1842 MPLS_TTL_PROP_DISABLED;
1846 NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1847 /* Unsupported attribute */
1857 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1858 struct netlink_ext_ack *extack)
1860 struct mpls_route_config *cfg;
1863 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1867 err = rtm_to_route_config(skb, nlh, cfg, extack);
1871 err = mpls_route_del(cfg, extack);
1879 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1880 struct netlink_ext_ack *extack)
1882 struct mpls_route_config *cfg;
1885 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1889 err = rtm_to_route_config(skb, nlh, cfg, extack);
1893 err = mpls_route_add(cfg, extack);
1900 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1901 u32 label, struct mpls_route *rt, int flags)
1903 struct net_device *dev;
1904 struct nlmsghdr *nlh;
1907 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1911 rtm = nlmsg_data(nlh);
1912 rtm->rtm_family = AF_MPLS;
1913 rtm->rtm_dst_len = 20;
1914 rtm->rtm_src_len = 0;
1916 rtm->rtm_table = RT_TABLE_MAIN;
1917 rtm->rtm_protocol = rt->rt_protocol;
1918 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1919 rtm->rtm_type = RTN_UNICAST;
1922 if (nla_put_labels(skb, RTA_DST, 1, &label))
1923 goto nla_put_failure;
1925 if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1926 bool ttl_propagate =
1927 rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1929 if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1931 goto nla_put_failure;
1933 if (rt->rt_nhn == 1) {
1934 const struct mpls_nh *nh = rt->rt_nh;
1936 if (nh->nh_labels &&
1937 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1939 goto nla_put_failure;
1940 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1941 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1943 goto nla_put_failure;
1944 dev = rtnl_dereference(nh->nh_dev);
1945 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1946 goto nla_put_failure;
1947 if (nh->nh_flags & RTNH_F_LINKDOWN)
1948 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1949 if (nh->nh_flags & RTNH_F_DEAD)
1950 rtm->rtm_flags |= RTNH_F_DEAD;
1952 struct rtnexthop *rtnh;
1957 mp = nla_nest_start(skb, RTA_MULTIPATH);
1959 goto nla_put_failure;
1962 dev = rtnl_dereference(nh->nh_dev);
1966 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1968 goto nla_put_failure;
1970 rtnh->rtnh_ifindex = dev->ifindex;
1971 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1972 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1975 if (nh->nh_flags & RTNH_F_DEAD) {
1976 rtnh->rtnh_flags |= RTNH_F_DEAD;
1980 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1983 goto nla_put_failure;
1984 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1985 nla_put_via(skb, nh->nh_via_table,
1986 mpls_nh_via(rt, nh),
1988 goto nla_put_failure;
1990 /* length of rtnetlink header + attributes */
1991 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1992 } endfor_nexthops(rt);
1994 if (linkdown == rt->rt_nhn)
1995 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1996 if (dead == rt->rt_nhn)
1997 rtm->rtm_flags |= RTNH_F_DEAD;
1999 nla_nest_end(skb, mp);
2002 nlmsg_end(skb, nlh);
2006 nlmsg_cancel(skb, nlh);
2010 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2012 struct net *net = sock_net(skb->sk);
2013 struct mpls_route __rcu **platform_label;
2014 size_t platform_labels;
2019 index = cb->args[0];
2020 if (index < MPLS_LABEL_FIRST_UNRESERVED)
2021 index = MPLS_LABEL_FIRST_UNRESERVED;
2023 platform_label = rtnl_dereference(net->mpls.platform_label);
2024 platform_labels = net->mpls.platform_labels;
2025 for (; index < platform_labels; index++) {
2026 struct mpls_route *rt;
2027 rt = rtnl_dereference(platform_label[index]);
2031 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2032 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2033 index, rt, NLM_F_MULTI) < 0)
2036 cb->args[0] = index;
2041 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2044 NLMSG_ALIGN(sizeof(struct rtmsg))
2045 + nla_total_size(4) /* RTA_DST */
2046 + nla_total_size(1); /* RTA_TTL_PROPAGATE */
2048 if (rt->rt_nhn == 1) {
2049 struct mpls_nh *nh = rt->rt_nh;
2052 payload += nla_total_size(4); /* RTA_OIF */
2053 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2054 payload += nla_total_size(2 + nh->nh_via_alen);
2055 if (nh->nh_labels) /* RTA_NEWDST */
2056 payload += nla_total_size(nh->nh_labels * 4);
2058 /* each nexthop is packed in an attribute */
2062 if (!rtnl_dereference(nh->nh_dev))
2064 nhsize += nla_total_size(sizeof(struct rtnexthop));
2066 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2067 nhsize += nla_total_size(2 + nh->nh_via_alen);
2069 nhsize += nla_total_size(nh->nh_labels * 4);
2070 } endfor_nexthops(rt);
2071 /* nested attribute */
2072 payload += nla_total_size(nhsize);
2078 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2079 struct nlmsghdr *nlh, struct net *net, u32 portid,
2080 unsigned int nlm_flags)
2082 struct sk_buff *skb;
2083 u32 seq = nlh ? nlh->nlmsg_seq : 0;
2086 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2090 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2092 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2093 WARN_ON(err == -EMSGSIZE);
2097 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2102 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2105 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2106 struct netlink_ext_ack *extack)
2108 struct net *net = sock_net(in_skb->sk);
2109 u32 portid = NETLINK_CB(in_skb).portid;
2110 u32 in_label = LABEL_NOT_SPECIFIED;
2111 struct nlattr *tb[RTA_MAX + 1];
2112 u32 labels[MAX_NEW_LABELS];
2113 struct mpls_shim_hdr *hdr;
2114 unsigned int hdr_size = 0;
2115 struct net_device *dev;
2116 struct mpls_route *rt;
2117 struct rtmsg *rtm, *r;
2118 struct nlmsghdr *nlh;
2119 struct sk_buff *skb;
2124 err = nlmsg_parse(in_nlh, sizeof(*rtm), tb, RTA_MAX,
2125 rtm_mpls_policy, extack);
2129 rtm = nlmsg_data(in_nlh);
2134 if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2135 &in_label, extack)) {
2140 if (!mpls_label_ok(net, in_label, extack)) {
2146 rt = mpls_route_input_rcu(net, in_label);
2152 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2153 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2159 err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2160 RTM_NEWROUTE, in_label, rt, 0);
2162 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2163 WARN_ON(err == -EMSGSIZE);
2167 return rtnl_unicast(skb, net, portid);
2170 if (tb[RTA_NEWDST]) {
2171 if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2172 labels, extack) != 0) {
2177 hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2180 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2186 skb->protocol = htons(ETH_P_MPLS_UC);
2192 if (skb_cow(skb, hdr_size)) {
2197 skb_reserve(skb, hdr_size);
2198 skb_push(skb, hdr_size);
2199 skb_reset_network_header(skb);
2201 /* Push new labels */
2202 hdr = mpls_hdr(skb);
2204 for (i = n_labels - 1; i >= 0; i--) {
2205 hdr[i] = mpls_entry_encode(labels[i],
2211 nh = mpls_select_multipath(rt, skb);
2218 skb_pull(skb, hdr_size);
2219 skb_reset_network_header(skb);
2222 nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2223 RTM_NEWROUTE, sizeof(*r), 0);
2229 r = nlmsg_data(nlh);
2230 r->rtm_family = AF_MPLS;
2231 r->rtm_dst_len = 20;
2233 r->rtm_table = RT_TABLE_MAIN;
2234 r->rtm_type = RTN_UNICAST;
2235 r->rtm_scope = RT_SCOPE_UNIVERSE;
2236 r->rtm_protocol = rt->rt_protocol;
2239 if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2240 goto nla_put_failure;
2242 if (nh->nh_labels &&
2243 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2245 goto nla_put_failure;
2247 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2248 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2250 goto nla_put_failure;
2251 dev = rtnl_dereference(nh->nh_dev);
2252 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2253 goto nla_put_failure;
2255 nlmsg_end(skb, nlh);
2257 err = rtnl_unicast(skb, net, portid);
2262 nlmsg_cancel(skb, nlh);
2269 static int resize_platform_label_table(struct net *net, size_t limit)
2271 size_t size = sizeof(struct mpls_route *) * limit;
2274 struct mpls_route __rcu **labels = NULL, **old;
2275 struct mpls_route *rt0 = NULL, *rt2 = NULL;
2279 labels = kvzalloc(size, GFP_KERNEL);
2284 /* In case the predefined labels need to be populated */
2285 if (limit > MPLS_LABEL_IPV4NULL) {
2286 struct net_device *lo = net->loopback_dev;
2287 rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2290 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2291 rt0->rt_protocol = RTPROT_KERNEL;
2292 rt0->rt_payload_type = MPT_IPV4;
2293 rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2294 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2295 rt0->rt_nh->nh_via_alen = lo->addr_len;
2296 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2299 if (limit > MPLS_LABEL_IPV6NULL) {
2300 struct net_device *lo = net->loopback_dev;
2301 rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2304 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2305 rt2->rt_protocol = RTPROT_KERNEL;
2306 rt2->rt_payload_type = MPT_IPV6;
2307 rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2308 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2309 rt2->rt_nh->nh_via_alen = lo->addr_len;
2310 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2315 /* Remember the original table */
2316 old = rtnl_dereference(net->mpls.platform_label);
2317 old_limit = net->mpls.platform_labels;
2319 /* Free any labels beyond the new table */
2320 for (index = limit; index < old_limit; index++)
2321 mpls_route_update(net, index, NULL, NULL);
2323 /* Copy over the old labels */
2325 if (old_limit < limit)
2326 cp_size = old_limit * sizeof(struct mpls_route *);
2328 memcpy(labels, old, cp_size);
2330 /* If needed set the predefined labels */
2331 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2332 (limit > MPLS_LABEL_IPV6NULL)) {
2333 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2337 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2338 (limit > MPLS_LABEL_IPV4NULL)) {
2339 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2343 /* Update the global pointers */
2344 net->mpls.platform_labels = limit;
2345 rcu_assign_pointer(net->mpls.platform_label, labels);
2366 static int mpls_platform_labels(struct ctl_table *table, int write,
2367 void __user *buffer, size_t *lenp, loff_t *ppos)
2369 struct net *net = table->data;
2370 int platform_labels = net->mpls.platform_labels;
2372 struct ctl_table tmp = {
2373 .procname = table->procname,
2374 .data = &platform_labels,
2375 .maxlen = sizeof(int),
2376 .mode = table->mode,
2378 .extra2 = &label_limit,
2381 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2383 if (write && ret == 0)
2384 ret = resize_platform_label_table(net, platform_labels);
2389 #define MPLS_NS_SYSCTL_OFFSET(field) \
2390 (&((struct net *)0)->field)
2392 static const struct ctl_table mpls_table[] = {
2394 .procname = "platform_labels",
2396 .maxlen = sizeof(int),
2398 .proc_handler = mpls_platform_labels,
2401 .procname = "ip_ttl_propagate",
2402 .data = MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2403 .maxlen = sizeof(int),
2405 .proc_handler = proc_dointvec_minmax,
2410 .procname = "default_ttl",
2411 .data = MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2412 .maxlen = sizeof(int),
2414 .proc_handler = proc_dointvec_minmax,
2421 static int mpls_net_init(struct net *net)
2423 struct ctl_table *table;
2426 net->mpls.platform_labels = 0;
2427 net->mpls.platform_label = NULL;
2428 net->mpls.ip_ttl_propagate = 1;
2429 net->mpls.default_ttl = 255;
2431 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2435 /* Table data contains only offsets relative to the base of
2436 * the mdev at this point, so make them absolute.
2438 for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2439 table[i].data = (char *)net + (uintptr_t)table[i].data;
2441 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2442 if (net->mpls.ctl == NULL) {
2450 static void mpls_net_exit(struct net *net)
2452 struct mpls_route __rcu **platform_label;
2453 size_t platform_labels;
2454 struct ctl_table *table;
2457 table = net->mpls.ctl->ctl_table_arg;
2458 unregister_net_sysctl_table(net->mpls.ctl);
2461 /* An rcu grace period has passed since there was a device in
2462 * the network namespace (and thus the last in flight packet)
2463 * left this network namespace. This is because
2464 * unregister_netdevice_many and netdev_run_todo has completed
2465 * for each network device that was in this network namespace.
2467 * As such no additional rcu synchronization is necessary when
2468 * freeing the platform_label table.
2471 platform_label = rtnl_dereference(net->mpls.platform_label);
2472 platform_labels = net->mpls.platform_labels;
2473 for (index = 0; index < platform_labels; index++) {
2474 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2475 RCU_INIT_POINTER(platform_label[index], NULL);
2476 mpls_notify_route(net, index, rt, NULL, NULL);
2481 kvfree(platform_label);
2484 static struct pernet_operations mpls_net_ops = {
2485 .init = mpls_net_init,
2486 .exit = mpls_net_exit,
2489 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2491 .fill_stats_af = mpls_fill_stats_af,
2492 .get_stats_af_size = mpls_get_stats_af_size,
2495 static int __init mpls_init(void)
2499 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2501 err = register_pernet_subsys(&mpls_net_ops);
2505 err = register_netdevice_notifier(&mpls_dev_notifier);
2507 goto out_unregister_pernet;
2509 dev_add_pack(&mpls_packet_type);
2511 rtnl_af_register(&mpls_af_ops);
2513 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, 0);
2514 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, 0);
2515 rtnl_register(PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes,
2517 rtnl_register(PF_MPLS, RTM_GETNETCONF, mpls_netconf_get_devconf,
2518 mpls_netconf_dump_devconf, 0);
2519 err = ipgre_tunnel_encap_add_mpls_ops();
2521 pr_err("Can't add mpls over gre tunnel ops\n");
2527 out_unregister_pernet:
2528 unregister_pernet_subsys(&mpls_net_ops);
2531 module_init(mpls_init);
2533 static void __exit mpls_exit(void)
2535 rtnl_unregister_all(PF_MPLS);
2536 rtnl_af_unregister(&mpls_af_ops);
2537 dev_remove_pack(&mpls_packet_type);
2538 unregister_netdevice_notifier(&mpls_dev_notifier);
2539 unregister_pernet_subsys(&mpls_net_ops);
2540 ipgre_tunnel_encap_del_mpls_ops();
2542 module_exit(mpls_exit);
2544 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2545 MODULE_LICENSE("GPL v2");
2546 MODULE_ALIAS_NETPROTO(PF_MPLS);