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/nospec.h>
12 #include <linux/vmalloc.h>
13 #include <linux/percpu.h>
18 #include <net/ip_fib.h>
19 #include <net/netevent.h>
20 #include <net/ip_tunnels.h>
21 #include <net/netns/generic.h>
22 #if IS_ENABLED(CONFIG_IPV6)
25 #include <net/addrconf.h>
26 #include <net/nexthop.h>
29 /* max memory we will use for mpls_route */
30 #define MAX_MPLS_ROUTE_MEM 4096
32 /* Maximum number of labels to look ahead at when selecting a path of
35 #define MAX_MP_SELECT_LABELS 4
37 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
41 static int label_limit = (1 << 20) - 1;
42 static int ttl_max = 255;
44 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
45 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
47 return sizeof(struct mpls_shim_hdr);
50 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
51 .encap_hlen = ipgre_mpls_encap_hlen,
54 static int ipgre_tunnel_encap_add_mpls_ops(void)
56 return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
59 static void ipgre_tunnel_encap_del_mpls_ops(void)
61 ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
64 static int ipgre_tunnel_encap_add_mpls_ops(void)
69 static void ipgre_tunnel_encap_del_mpls_ops(void)
74 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
75 struct nlmsghdr *nlh, struct net *net, u32 portid,
76 unsigned int nlm_flags);
78 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
80 struct mpls_route *rt = NULL;
82 if (index < net->mpls.platform_labels) {
83 struct mpls_route __rcu **platform_label =
84 rcu_dereference(net->mpls.platform_label);
85 rt = rcu_dereference(platform_label[index]);
90 bool mpls_output_possible(const struct net_device *dev)
92 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
94 EXPORT_SYMBOL_GPL(mpls_output_possible);
96 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
98 return (u8 *)nh + rt->rt_via_offset;
101 static const u8 *mpls_nh_via(const struct mpls_route *rt,
102 const struct mpls_nh *nh)
104 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
107 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
109 /* The size of the layer 2.5 labels to be added for this route */
110 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
113 unsigned int mpls_dev_mtu(const struct net_device *dev)
115 /* The amount of data the layer 2 frame can hold */
118 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
120 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
125 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
130 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
132 void mpls_stats_inc_outucastpkts(struct net_device *dev,
133 const struct sk_buff *skb)
135 struct mpls_dev *mdev;
137 if (skb->protocol == htons(ETH_P_MPLS_UC)) {
138 mdev = mpls_dev_get(dev);
140 MPLS_INC_STATS_LEN(mdev, skb->len,
143 } else if (skb->protocol == htons(ETH_P_IP)) {
144 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
145 #if IS_ENABLED(CONFIG_IPV6)
146 } else if (skb->protocol == htons(ETH_P_IPV6)) {
147 struct inet6_dev *in6dev = __in6_dev_get(dev);
150 IP6_UPD_PO_STATS(dev_net(dev), in6dev,
151 IPSTATS_MIB_OUT, skb->len);
155 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
157 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
159 struct mpls_entry_decoded dec;
160 unsigned int mpls_hdr_len = 0;
161 struct mpls_shim_hdr *hdr;
162 bool eli_seen = false;
166 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
168 mpls_hdr_len += sizeof(*hdr);
169 if (!pskb_may_pull(skb, mpls_hdr_len))
172 /* Read and decode the current label */
173 hdr = mpls_hdr(skb) + label_index;
174 dec = mpls_entry_decode(hdr);
176 /* RFC6790 - reserved labels MUST NOT be used as keys
177 * for the load-balancing function
179 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
180 hash = jhash_1word(dec.label, hash);
182 /* The entropy label follows the entropy label
183 * indicator, so this means that the entropy
184 * label was just added to the hash - no need to
185 * go any deeper either in the label stack or in the
190 } else if (dec.label == MPLS_LABEL_ENTROPY) {
197 /* found bottom label; does skb have room for a header? */
198 if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
199 const struct iphdr *v4hdr;
201 v4hdr = (const struct iphdr *)(hdr + 1);
202 if (v4hdr->version == 4) {
203 hash = jhash_3words(ntohl(v4hdr->saddr),
205 v4hdr->protocol, hash);
206 } else if (v4hdr->version == 6 &&
207 pskb_may_pull(skb, mpls_hdr_len +
208 sizeof(struct ipv6hdr))) {
209 const struct ipv6hdr *v6hdr;
211 v6hdr = (const struct ipv6hdr *)(hdr + 1);
212 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
213 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
214 hash = jhash_1word(v6hdr->nexthdr, hash);
224 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
226 return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
229 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
230 * a next hop (nh->nh_flags) are modified by netdev event handlers.
231 * Since those fields can change at any moment, use READ_ONCE to
234 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
242 /* No need to look further into packet if there's only
248 alive = READ_ONCE(rt->rt_nhn_alive);
252 hash = mpls_multipath_hash(rt, skb);
253 nh_index = hash % alive;
254 if (alive == rt->rt_nhn)
257 unsigned int nh_flags = READ_ONCE(nh->nh_flags);
259 if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
264 } endfor_nexthops(rt);
267 return mpls_get_nexthop(rt, nh_index);
270 static bool mpls_egress(struct net *net, struct mpls_route *rt,
271 struct sk_buff *skb, struct mpls_entry_decoded dec)
273 enum mpls_payload_type payload_type;
274 bool success = false;
276 /* The IPv4 code below accesses through the IPv4 header
277 * checksum, which is 12 bytes into the packet.
278 * The IPv6 code below accesses through the IPv6 hop limit
279 * which is 8 bytes into the packet.
281 * For all supported cases there should always be at least 12
282 * bytes of packet data present. The IPv4 header is 20 bytes
283 * without options and the IPv6 header is always 40 bytes
286 if (!pskb_may_pull(skb, 12))
289 payload_type = rt->rt_payload_type;
290 if (payload_type == MPT_UNSPEC)
291 payload_type = ip_hdr(skb)->version;
293 switch (payload_type) {
295 struct iphdr *hdr4 = ip_hdr(skb);
297 skb->protocol = htons(ETH_P_IP);
299 /* If propagating TTL, take the decremented TTL from
300 * the incoming MPLS header, otherwise decrement the
301 * TTL, but only if not 0 to avoid underflow.
303 if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
304 (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
305 net->mpls.ip_ttl_propagate))
308 new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
310 csum_replace2(&hdr4->check,
311 htons(hdr4->ttl << 8),
312 htons(new_ttl << 8));
318 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
319 skb->protocol = htons(ETH_P_IPV6);
321 /* If propagating TTL, take the decremented TTL from
322 * the incoming MPLS header, otherwise decrement the
323 * hop limit, but only if not 0 to avoid underflow.
325 if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
326 (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
327 net->mpls.ip_ttl_propagate))
328 hdr6->hop_limit = dec.ttl;
329 else if (hdr6->hop_limit)
330 hdr6->hop_limit = hdr6->hop_limit - 1;
335 /* Should have decided which protocol it is by now */
342 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
343 struct packet_type *pt, struct net_device *orig_dev)
345 struct net *net = dev_net(dev);
346 struct mpls_shim_hdr *hdr;
347 struct mpls_route *rt;
349 struct mpls_entry_decoded dec;
350 struct net_device *out_dev;
351 struct mpls_dev *out_mdev;
352 struct mpls_dev *mdev;
354 unsigned int new_header_size;
358 /* Careful this entire function runs inside of an rcu critical section */
360 mdev = mpls_dev_get(dev);
364 MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
367 if (!mdev->input_enabled) {
368 MPLS_INC_STATS(mdev, rx_dropped);
372 if (skb->pkt_type != PACKET_HOST)
375 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
378 if (!pskb_may_pull(skb, sizeof(*hdr)))
381 /* Read and decode the label */
383 dec = mpls_entry_decode(hdr);
385 rt = mpls_route_input_rcu(net, dec.label);
387 MPLS_INC_STATS(mdev, rx_noroute);
391 nh = mpls_select_multipath(rt, skb);
396 skb_pull(skb, sizeof(*hdr));
397 skb_reset_network_header(skb);
401 if (skb_warn_if_lro(skb))
404 skb_forward_csum(skb);
406 /* Verify ttl is valid */
411 /* Find the output device */
412 out_dev = rcu_dereference(nh->nh_dev);
413 if (!mpls_output_possible(out_dev))
416 /* Verify the destination can hold the packet */
417 new_header_size = mpls_nh_header_size(nh);
418 mtu = mpls_dev_mtu(out_dev);
419 if (mpls_pkt_too_big(skb, mtu - new_header_size))
422 hh_len = LL_RESERVED_SPACE(out_dev);
423 if (!out_dev->header_ops)
426 /* Ensure there is enough space for the headers in the skb */
427 if (skb_cow(skb, hh_len + new_header_size))
431 skb->protocol = htons(ETH_P_MPLS_UC);
433 if (unlikely(!new_header_size && dec.bos)) {
434 /* Penultimate hop popping */
435 if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
440 skb_push(skb, new_header_size);
441 skb_reset_network_header(skb);
442 /* Push the new labels */
445 for (i = nh->nh_labels - 1; i >= 0; i--) {
446 hdr[i] = mpls_entry_encode(nh->nh_label[i],
452 mpls_stats_inc_outucastpkts(out_dev, skb);
454 /* If via wasn't specified then send out using device address */
455 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
456 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
457 out_dev->dev_addr, skb);
459 err = neigh_xmit(nh->nh_via_table, out_dev,
460 mpls_nh_via(rt, nh), skb);
462 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
467 out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
469 MPLS_INC_STATS(out_mdev, tx_errors);
472 MPLS_INC_STATS(mdev, rx_errors);
478 static struct packet_type mpls_packet_type __read_mostly = {
479 .type = cpu_to_be16(ETH_P_MPLS_UC),
480 .func = mpls_forward,
483 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
484 [RTA_DST] = { .type = NLA_U32 },
485 [RTA_OIF] = { .type = NLA_U32 },
486 [RTA_TTL_PROPAGATE] = { .type = NLA_U8 },
489 struct mpls_route_config {
494 u8 rc_via[MAX_VIA_ALEN];
498 u32 rc_output_label[MAX_NEW_LABELS];
500 enum mpls_payload_type rc_payload_type;
501 struct nl_info rc_nlinfo;
502 struct rtnexthop *rc_mp;
506 /* all nexthops within a route have the same size based on max
507 * number of labels and max via length for a hop
509 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
511 u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
512 struct mpls_route *rt;
515 size = sizeof(*rt) + num_nh * nh_size;
516 if (size > MAX_MPLS_ROUTE_MEM)
517 return ERR_PTR(-EINVAL);
519 rt = kzalloc(size, GFP_KERNEL);
521 return ERR_PTR(-ENOMEM);
524 rt->rt_nhn_alive = num_nh;
525 rt->rt_nh_size = nh_size;
526 rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
531 static void mpls_rt_free(struct mpls_route *rt)
534 kfree_rcu(rt, rt_rcu);
537 static void mpls_notify_route(struct net *net, unsigned index,
538 struct mpls_route *old, struct mpls_route *new,
539 const struct nl_info *info)
541 struct nlmsghdr *nlh = info ? info->nlh : NULL;
542 unsigned portid = info ? info->portid : 0;
543 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
544 struct mpls_route *rt = new ? new : old;
545 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
546 /* Ignore reserved labels for now */
547 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
548 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
551 static void mpls_route_update(struct net *net, unsigned index,
552 struct mpls_route *new,
553 const struct nl_info *info)
555 struct mpls_route __rcu **platform_label;
556 struct mpls_route *rt;
560 platform_label = rtnl_dereference(net->mpls.platform_label);
561 rt = rtnl_dereference(platform_label[index]);
562 rcu_assign_pointer(platform_label[index], new);
564 mpls_notify_route(net, index, rt, new, info);
566 /* If we removed a route free it now */
570 static unsigned find_free_label(struct net *net)
572 struct mpls_route __rcu **platform_label;
573 size_t platform_labels;
576 platform_label = rtnl_dereference(net->mpls.platform_label);
577 platform_labels = net->mpls.platform_labels;
578 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
580 if (!rtnl_dereference(platform_label[index]))
583 return LABEL_NOT_SPECIFIED;
586 #if IS_ENABLED(CONFIG_INET)
587 static struct net_device *inet_fib_lookup_dev(struct net *net,
590 struct net_device *dev;
592 struct in_addr daddr;
594 memcpy(&daddr, addr, sizeof(struct in_addr));
595 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
607 static struct net_device *inet_fib_lookup_dev(struct net *net,
610 return ERR_PTR(-EAFNOSUPPORT);
614 #if IS_ENABLED(CONFIG_IPV6)
615 static struct net_device *inet6_fib_lookup_dev(struct net *net,
618 struct net_device *dev;
619 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 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
629 return ERR_CAST(dst);
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)
943 /* Reserved labels may not be set */
944 if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
945 NL_SET_ERR_MSG(extack,
946 "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
950 /* The full 20 bit range may not be supported. */
951 if (is_ok && *index >= net->mpls.platform_labels) {
952 NL_SET_ERR_MSG(extack,
953 "Label >= configured maximum in platform_labels");
957 *index = array_index_nospec(*index, net->mpls.platform_labels);
961 static int mpls_route_add(struct mpls_route_config *cfg,
962 struct netlink_ext_ack *extack)
964 struct mpls_route __rcu **platform_label;
965 struct net *net = cfg->rc_nlinfo.nl_net;
966 struct mpls_route *rt, *old;
973 index = cfg->rc_label;
975 /* If a label was not specified during insert pick one */
976 if ((index == LABEL_NOT_SPECIFIED) &&
977 (cfg->rc_nlflags & NLM_F_CREATE)) {
978 index = find_free_label(net);
981 if (!mpls_label_ok(net, &index, extack))
984 /* Append makes no sense with mpls */
986 if (cfg->rc_nlflags & NLM_F_APPEND) {
987 NL_SET_ERR_MSG(extack, "MPLS does not support route append");
992 platform_label = rtnl_dereference(net->mpls.platform_label);
993 old = rtnl_dereference(platform_label[index]);
994 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
998 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1002 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1007 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1008 cfg->rc_via_alen, &max_via_alen,
1011 max_via_alen = cfg->rc_via_alen;
1012 max_labels = cfg->rc_output_labels;
1017 NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1022 rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1028 rt->rt_protocol = cfg->rc_protocol;
1029 rt->rt_payload_type = cfg->rc_payload_type;
1030 rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1033 err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1035 err = mpls_nh_build_from_cfg(cfg, rt);
1039 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1049 static int mpls_route_del(struct mpls_route_config *cfg,
1050 struct netlink_ext_ack *extack)
1052 struct net *net = cfg->rc_nlinfo.nl_net;
1056 index = cfg->rc_label;
1058 if (!mpls_label_ok(net, &index, extack))
1061 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1068 static void mpls_get_stats(struct mpls_dev *mdev,
1069 struct mpls_link_stats *stats)
1071 struct mpls_pcpu_stats *p;
1074 memset(stats, 0, sizeof(*stats));
1076 for_each_possible_cpu(i) {
1077 struct mpls_link_stats local;
1080 p = per_cpu_ptr(mdev->stats, i);
1082 start = u64_stats_fetch_begin(&p->syncp);
1084 } while (u64_stats_fetch_retry(&p->syncp, start));
1086 stats->rx_packets += local.rx_packets;
1087 stats->rx_bytes += local.rx_bytes;
1088 stats->tx_packets += local.tx_packets;
1089 stats->tx_bytes += local.tx_bytes;
1090 stats->rx_errors += local.rx_errors;
1091 stats->tx_errors += local.tx_errors;
1092 stats->rx_dropped += local.rx_dropped;
1093 stats->tx_dropped += local.tx_dropped;
1094 stats->rx_noroute += local.rx_noroute;
1098 static int mpls_fill_stats_af(struct sk_buff *skb,
1099 const struct net_device *dev)
1101 struct mpls_link_stats *stats;
1102 struct mpls_dev *mdev;
1105 mdev = mpls_dev_get(dev);
1109 nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1110 sizeof(struct mpls_link_stats),
1115 stats = nla_data(nla);
1116 mpls_get_stats(mdev, stats);
1121 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1123 struct mpls_dev *mdev;
1125 mdev = mpls_dev_get(dev);
1129 return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1132 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1133 u32 portid, u32 seq, int event,
1134 unsigned int flags, int type)
1136 struct nlmsghdr *nlh;
1137 struct netconfmsg *ncm;
1140 nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1145 if (type == NETCONFA_ALL)
1148 ncm = nlmsg_data(nlh);
1149 ncm->ncm_family = AF_MPLS;
1151 if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1152 goto nla_put_failure;
1154 if ((all || type == NETCONFA_INPUT) &&
1155 nla_put_s32(skb, NETCONFA_INPUT,
1156 mdev->input_enabled) < 0)
1157 goto nla_put_failure;
1159 nlmsg_end(skb, nlh);
1163 nlmsg_cancel(skb, nlh);
1167 static int mpls_netconf_msgsize_devconf(int type)
1169 int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1170 + nla_total_size(4); /* NETCONFA_IFINDEX */
1173 if (type == NETCONFA_ALL)
1176 if (all || type == NETCONFA_INPUT)
1177 size += nla_total_size(4);
1182 static void mpls_netconf_notify_devconf(struct net *net, int event,
1183 int type, struct mpls_dev *mdev)
1185 struct sk_buff *skb;
1188 skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1192 err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1194 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1195 WARN_ON(err == -EMSGSIZE);
1200 rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1204 rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1207 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1208 [NETCONFA_IFINDEX] = { .len = sizeof(int) },
1211 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1212 struct nlmsghdr *nlh,
1213 struct netlink_ext_ack *extack)
1215 struct net *net = sock_net(in_skb->sk);
1216 struct nlattr *tb[NETCONFA_MAX + 1];
1217 struct netconfmsg *ncm;
1218 struct net_device *dev;
1219 struct mpls_dev *mdev;
1220 struct sk_buff *skb;
1224 err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1225 devconf_mpls_policy, NULL);
1230 if (!tb[NETCONFA_IFINDEX])
1233 ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1234 dev = __dev_get_by_index(net, ifindex);
1238 mdev = mpls_dev_get(dev);
1243 skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1247 err = mpls_netconf_fill_devconf(skb, mdev,
1248 NETLINK_CB(in_skb).portid,
1249 nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1252 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1253 WARN_ON(err == -EMSGSIZE);
1257 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1262 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1263 struct netlink_callback *cb)
1265 struct net *net = sock_net(skb->sk);
1266 struct hlist_head *head;
1267 struct net_device *dev;
1268 struct mpls_dev *mdev;
1273 s_idx = idx = cb->args[1];
1275 for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1277 head = &net->dev_index_head[h];
1279 cb->seq = net->dev_base_seq;
1280 hlist_for_each_entry_rcu(dev, head, index_hlist) {
1283 mdev = mpls_dev_get(dev);
1286 if (mpls_netconf_fill_devconf(skb, mdev,
1287 NETLINK_CB(cb->skb).portid,
1291 NETCONFA_ALL) < 0) {
1295 nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1308 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
1309 (&((struct mpls_dev *)0)->field)
1311 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1312 void __user *buffer,
1313 size_t *lenp, loff_t *ppos)
1315 int oval = *(int *)ctl->data;
1316 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1319 struct mpls_dev *mdev = ctl->extra1;
1320 int i = (int *)ctl->data - (int *)mdev;
1321 struct net *net = ctl->extra2;
1322 int val = *(int *)ctl->data;
1324 if (i == offsetof(struct mpls_dev, input_enabled) &&
1326 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1327 NETCONFA_INPUT, mdev);
1334 static const struct ctl_table mpls_dev_table[] = {
1336 .procname = "input",
1337 .maxlen = sizeof(int),
1339 .proc_handler = mpls_conf_proc,
1340 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1345 static int mpls_dev_sysctl_register(struct net_device *dev,
1346 struct mpls_dev *mdev)
1348 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1349 struct net *net = dev_net(dev);
1350 struct ctl_table *table;
1353 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1357 /* Table data contains only offsets relative to the base of
1358 * the mdev at this point, so make them absolute.
1360 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1361 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1362 table[i].extra1 = mdev;
1363 table[i].extra2 = net;
1366 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1368 mdev->sysctl = register_net_sysctl(net, path, table);
1372 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1381 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1382 struct mpls_dev *mdev)
1384 struct net *net = dev_net(dev);
1385 struct ctl_table *table;
1387 table = mdev->sysctl->ctl_table_arg;
1388 unregister_net_sysctl_table(mdev->sysctl);
1391 mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1394 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1396 struct mpls_dev *mdev;
1402 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1404 return ERR_PTR(err);
1406 mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1410 for_each_possible_cpu(i) {
1411 struct mpls_pcpu_stats *mpls_stats;
1413 mpls_stats = per_cpu_ptr(mdev->stats, i);
1414 u64_stats_init(&mpls_stats->syncp);
1419 err = mpls_dev_sysctl_register(dev, mdev);
1423 rcu_assign_pointer(dev->mpls_ptr, mdev);
1428 free_percpu(mdev->stats);
1430 return ERR_PTR(err);
1433 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1435 struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1437 free_percpu(mdev->stats);
1441 static int mpls_ifdown(struct net_device *dev, int event)
1443 struct mpls_route __rcu **platform_label;
1444 struct net *net = dev_net(dev);
1447 platform_label = rtnl_dereference(net->mpls.platform_label);
1448 for (index = 0; index < net->mpls.platform_labels; index++) {
1449 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1450 bool nh_del = false;
1456 if (event == NETDEV_UNREGISTER) {
1460 struct net_device *nh_dev =
1461 rtnl_dereference(nh->nh_dev);
1463 if (!nh_dev || nh_dev == dev)
1467 } endfor_nexthops(rt);
1469 /* if there are no more nexthops, delete the route */
1470 if (deleted == rt->rt_nhn) {
1471 mpls_route_update(net, index, NULL, NULL);
1476 size_t size = sizeof(*rt) + rt->rt_nhn *
1478 struct mpls_route *orig = rt;
1480 rt = kmalloc(size, GFP_KERNEL);
1483 memcpy(rt, orig, size);
1487 change_nexthops(rt) {
1488 unsigned int nh_flags = nh->nh_flags;
1490 if (rtnl_dereference(nh->nh_dev) != dev)
1495 case NETDEV_UNREGISTER:
1496 nh_flags |= RTNH_F_DEAD;
1499 nh_flags |= RTNH_F_LINKDOWN;
1502 if (event == NETDEV_UNREGISTER)
1503 RCU_INIT_POINTER(nh->nh_dev, NULL);
1505 if (nh->nh_flags != nh_flags)
1506 WRITE_ONCE(nh->nh_flags, nh_flags);
1508 if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1510 } endfor_nexthops(rt);
1512 WRITE_ONCE(rt->rt_nhn_alive, alive);
1515 mpls_route_update(net, index, rt, NULL);
1521 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1523 struct mpls_route __rcu **platform_label;
1524 struct net *net = dev_net(dev);
1528 platform_label = rtnl_dereference(net->mpls.platform_label);
1529 for (index = 0; index < net->mpls.platform_labels; index++) {
1530 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1536 change_nexthops(rt) {
1537 unsigned int nh_flags = nh->nh_flags;
1538 struct net_device *nh_dev =
1539 rtnl_dereference(nh->nh_dev);
1541 if (!(nh_flags & flags)) {
1549 WRITE_ONCE(nh->nh_flags, nh_flags);
1550 } endfor_nexthops(rt);
1552 WRITE_ONCE(rt->rt_nhn_alive, alive);
1556 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1559 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1560 struct mpls_dev *mdev;
1563 if (event == NETDEV_REGISTER) {
1565 /* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1568 if (dev->type == ARPHRD_ETHER ||
1569 dev->type == ARPHRD_LOOPBACK ||
1570 dev->type == ARPHRD_IPGRE ||
1571 dev->type == ARPHRD_IP6GRE ||
1572 dev->type == ARPHRD_SIT ||
1573 dev->type == ARPHRD_TUNNEL) {
1574 mdev = mpls_add_dev(dev);
1576 return notifier_from_errno(PTR_ERR(mdev));
1581 mdev = mpls_dev_get(dev);
1589 err = mpls_ifdown(dev, event);
1591 return notifier_from_errno(err);
1594 flags = dev_get_flags(dev);
1595 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1596 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1598 mpls_ifup(dev, RTNH_F_DEAD);
1601 flags = dev_get_flags(dev);
1602 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1603 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1605 err = mpls_ifdown(dev, event);
1607 return notifier_from_errno(err);
1610 case NETDEV_UNREGISTER:
1611 err = mpls_ifdown(dev, event);
1613 return notifier_from_errno(err);
1614 mdev = mpls_dev_get(dev);
1616 mpls_dev_sysctl_unregister(dev, mdev);
1617 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1618 call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1621 case NETDEV_CHANGENAME:
1622 mdev = mpls_dev_get(dev);
1624 mpls_dev_sysctl_unregister(dev, mdev);
1625 err = mpls_dev_sysctl_register(dev, mdev);
1627 return notifier_from_errno(err);
1634 static struct notifier_block mpls_dev_notifier = {
1635 .notifier_call = mpls_dev_notify,
1638 static int nla_put_via(struct sk_buff *skb,
1639 u8 table, const void *addr, int alen)
1641 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1642 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1646 int family = AF_UNSPEC;
1648 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1652 if (table <= NEIGH_NR_TABLES)
1653 family = table_to_family[table];
1655 via = nla_data(nla);
1656 via->rtvia_family = family;
1657 memcpy(via->rtvia_addr, addr, alen);
1661 int nla_put_labels(struct sk_buff *skb, int attrtype,
1662 u8 labels, const u32 label[])
1665 struct mpls_shim_hdr *nla_label;
1668 nla = nla_reserve(skb, attrtype, labels*4);
1672 nla_label = nla_data(nla);
1674 for (i = labels - 1; i >= 0; i--) {
1675 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1681 EXPORT_SYMBOL_GPL(nla_put_labels);
1683 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1684 u32 label[], struct netlink_ext_ack *extack)
1686 unsigned len = nla_len(nla);
1687 struct mpls_shim_hdr *nla_label;
1692 /* len needs to be an even multiple of 4 (the label size). Number
1693 * of labels is a u8 so check for overflow.
1695 if (len & 3 || len / 4 > 255) {
1696 NL_SET_ERR_MSG_ATTR(extack, nla,
1697 "Invalid length for labels attribute");
1701 /* Limit the number of new labels allowed */
1703 if (nla_labels > max_labels) {
1704 NL_SET_ERR_MSG(extack, "Too many labels");
1708 /* when label == NULL, caller wants number of labels */
1712 nla_label = nla_data(nla);
1714 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1715 struct mpls_entry_decoded dec;
1716 dec = mpls_entry_decode(nla_label + i);
1718 /* Ensure the bottom of stack flag is properly set
1719 * and ttl and tc are both clear.
1722 NL_SET_ERR_MSG_ATTR(extack, nla,
1723 "TTL in label must be 0");
1728 NL_SET_ERR_MSG_ATTR(extack, nla,
1729 "Traffic class in label must be 0");
1733 if (dec.bos != bos) {
1734 NL_SET_BAD_ATTR(extack, nla);
1736 NL_SET_ERR_MSG(extack,
1737 "BOS bit must be set in first label");
1739 NL_SET_ERR_MSG(extack,
1740 "BOS bit can only be set in first label");
1745 switch (dec.label) {
1746 case MPLS_LABEL_IMPLNULL:
1747 /* RFC3032: This is a label that an LSR may
1748 * assign and distribute, but which never
1749 * actually appears in the encapsulation.
1751 NL_SET_ERR_MSG_ATTR(extack, nla,
1752 "Implicit NULL Label (3) can not be used in encapsulation");
1756 label[i] = dec.label;
1759 *labels = nla_labels;
1762 EXPORT_SYMBOL_GPL(nla_get_labels);
1764 static int rtm_to_route_config(struct sk_buff *skb,
1765 struct nlmsghdr *nlh,
1766 struct mpls_route_config *cfg,
1767 struct netlink_ext_ack *extack)
1770 struct nlattr *tb[RTA_MAX+1];
1774 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy,
1780 rtm = nlmsg_data(nlh);
1782 if (rtm->rtm_family != AF_MPLS) {
1783 NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1786 if (rtm->rtm_dst_len != 20) {
1787 NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1790 if (rtm->rtm_src_len != 0) {
1791 NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1794 if (rtm->rtm_tos != 0) {
1795 NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1798 if (rtm->rtm_table != RT_TABLE_MAIN) {
1799 NL_SET_ERR_MSG(extack,
1800 "MPLS only supports the main route table");
1803 /* Any value is acceptable for rtm_protocol */
1805 /* As mpls uses destination specific addresses
1806 * (or source specific address in the case of multicast)
1807 * all addresses have universal scope.
1809 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1810 NL_SET_ERR_MSG(extack,
1811 "Invalid route scope - MPLS only supports UNIVERSE");
1814 if (rtm->rtm_type != RTN_UNICAST) {
1815 NL_SET_ERR_MSG(extack,
1816 "Invalid route type - MPLS only supports UNICAST");
1819 if (rtm->rtm_flags != 0) {
1820 NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1824 cfg->rc_label = LABEL_NOT_SPECIFIED;
1825 cfg->rc_protocol = rtm->rtm_protocol;
1826 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1827 cfg->rc_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
1828 cfg->rc_nlflags = nlh->nlmsg_flags;
1829 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1830 cfg->rc_nlinfo.nlh = nlh;
1831 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1833 for (index = 0; index <= RTA_MAX; index++) {
1834 struct nlattr *nla = tb[index];
1840 cfg->rc_ifindex = nla_get_u32(nla);
1843 if (nla_get_labels(nla, MAX_NEW_LABELS,
1844 &cfg->rc_output_labels,
1845 cfg->rc_output_label, extack))
1851 if (nla_get_labels(nla, 1, &label_count,
1852 &cfg->rc_label, extack))
1855 if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1856 &cfg->rc_label, extack))
1861 NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1865 if (nla_get_via(nla, &cfg->rc_via_alen,
1866 &cfg->rc_via_table, cfg->rc_via,
1873 cfg->rc_mp = nla_data(nla);
1874 cfg->rc_mp_len = nla_len(nla);
1877 case RTA_TTL_PROPAGATE:
1879 u8 ttl_propagate = nla_get_u8(nla);
1881 if (ttl_propagate > 1) {
1882 NL_SET_ERR_MSG_ATTR(extack, nla,
1883 "RTA_TTL_PROPAGATE can only be 0 or 1");
1886 cfg->rc_ttl_propagate = ttl_propagate ?
1887 MPLS_TTL_PROP_ENABLED :
1888 MPLS_TTL_PROP_DISABLED;
1892 NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1893 /* Unsupported attribute */
1903 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1904 struct netlink_ext_ack *extack)
1906 struct mpls_route_config *cfg;
1909 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1913 err = rtm_to_route_config(skb, nlh, cfg, extack);
1917 err = mpls_route_del(cfg, extack);
1925 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1926 struct netlink_ext_ack *extack)
1928 struct mpls_route_config *cfg;
1931 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1935 err = rtm_to_route_config(skb, nlh, cfg, extack);
1939 err = mpls_route_add(cfg, extack);
1946 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1947 u32 label, struct mpls_route *rt, int flags)
1949 struct net_device *dev;
1950 struct nlmsghdr *nlh;
1953 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1957 rtm = nlmsg_data(nlh);
1958 rtm->rtm_family = AF_MPLS;
1959 rtm->rtm_dst_len = 20;
1960 rtm->rtm_src_len = 0;
1962 rtm->rtm_table = RT_TABLE_MAIN;
1963 rtm->rtm_protocol = rt->rt_protocol;
1964 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1965 rtm->rtm_type = RTN_UNICAST;
1968 if (nla_put_labels(skb, RTA_DST, 1, &label))
1969 goto nla_put_failure;
1971 if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1972 bool ttl_propagate =
1973 rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1975 if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1977 goto nla_put_failure;
1979 if (rt->rt_nhn == 1) {
1980 const struct mpls_nh *nh = rt->rt_nh;
1982 if (nh->nh_labels &&
1983 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1985 goto nla_put_failure;
1986 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1987 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1989 goto nla_put_failure;
1990 dev = rtnl_dereference(nh->nh_dev);
1991 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1992 goto nla_put_failure;
1993 if (nh->nh_flags & RTNH_F_LINKDOWN)
1994 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1995 if (nh->nh_flags & RTNH_F_DEAD)
1996 rtm->rtm_flags |= RTNH_F_DEAD;
1998 struct rtnexthop *rtnh;
2003 mp = nla_nest_start(skb, RTA_MULTIPATH);
2005 goto nla_put_failure;
2008 dev = rtnl_dereference(nh->nh_dev);
2012 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2014 goto nla_put_failure;
2016 rtnh->rtnh_ifindex = dev->ifindex;
2017 if (nh->nh_flags & RTNH_F_LINKDOWN) {
2018 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2021 if (nh->nh_flags & RTNH_F_DEAD) {
2022 rtnh->rtnh_flags |= RTNH_F_DEAD;
2026 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2029 goto nla_put_failure;
2030 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2031 nla_put_via(skb, nh->nh_via_table,
2032 mpls_nh_via(rt, nh),
2034 goto nla_put_failure;
2036 /* length of rtnetlink header + attributes */
2037 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2038 } endfor_nexthops(rt);
2040 if (linkdown == rt->rt_nhn)
2041 rtm->rtm_flags |= RTNH_F_LINKDOWN;
2042 if (dead == rt->rt_nhn)
2043 rtm->rtm_flags |= RTNH_F_DEAD;
2045 nla_nest_end(skb, mp);
2048 nlmsg_end(skb, nlh);
2052 nlmsg_cancel(skb, nlh);
2056 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2058 struct net *net = sock_net(skb->sk);
2059 struct mpls_route __rcu **platform_label;
2060 size_t platform_labels;
2065 index = cb->args[0];
2066 if (index < MPLS_LABEL_FIRST_UNRESERVED)
2067 index = MPLS_LABEL_FIRST_UNRESERVED;
2069 platform_label = rtnl_dereference(net->mpls.platform_label);
2070 platform_labels = net->mpls.platform_labels;
2071 for (; index < platform_labels; index++) {
2072 struct mpls_route *rt;
2073 rt = rtnl_dereference(platform_label[index]);
2077 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2078 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2079 index, rt, NLM_F_MULTI) < 0)
2082 cb->args[0] = index;
2087 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2090 NLMSG_ALIGN(sizeof(struct rtmsg))
2091 + nla_total_size(4) /* RTA_DST */
2092 + nla_total_size(1); /* RTA_TTL_PROPAGATE */
2094 if (rt->rt_nhn == 1) {
2095 struct mpls_nh *nh = rt->rt_nh;
2098 payload += nla_total_size(4); /* RTA_OIF */
2099 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2100 payload += nla_total_size(2 + nh->nh_via_alen);
2101 if (nh->nh_labels) /* RTA_NEWDST */
2102 payload += nla_total_size(nh->nh_labels * 4);
2104 /* each nexthop is packed in an attribute */
2108 if (!rtnl_dereference(nh->nh_dev))
2110 nhsize += nla_total_size(sizeof(struct rtnexthop));
2112 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2113 nhsize += nla_total_size(2 + nh->nh_via_alen);
2115 nhsize += nla_total_size(nh->nh_labels * 4);
2116 } endfor_nexthops(rt);
2117 /* nested attribute */
2118 payload += nla_total_size(nhsize);
2124 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2125 struct nlmsghdr *nlh, struct net *net, u32 portid,
2126 unsigned int nlm_flags)
2128 struct sk_buff *skb;
2129 u32 seq = nlh ? nlh->nlmsg_seq : 0;
2132 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2136 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2138 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2139 WARN_ON(err == -EMSGSIZE);
2143 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2148 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2151 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2152 struct netlink_ext_ack *extack)
2154 struct net *net = sock_net(in_skb->sk);
2155 u32 portid = NETLINK_CB(in_skb).portid;
2156 u32 in_label = LABEL_NOT_SPECIFIED;
2157 struct nlattr *tb[RTA_MAX + 1];
2158 u32 labels[MAX_NEW_LABELS];
2159 struct mpls_shim_hdr *hdr;
2160 unsigned int hdr_size = 0;
2161 struct net_device *dev;
2162 struct mpls_route *rt;
2163 struct rtmsg *rtm, *r;
2164 struct nlmsghdr *nlh;
2165 struct sk_buff *skb;
2170 err = nlmsg_parse(in_nlh, sizeof(*rtm), tb, RTA_MAX,
2171 rtm_mpls_policy, extack);
2175 rtm = nlmsg_data(in_nlh);
2180 if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2181 &in_label, extack)) {
2186 if (!mpls_label_ok(net, &in_label, extack)) {
2192 rt = mpls_route_input_rcu(net, in_label);
2198 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2199 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2205 err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2206 RTM_NEWROUTE, in_label, rt, 0);
2208 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2209 WARN_ON(err == -EMSGSIZE);
2213 return rtnl_unicast(skb, net, portid);
2216 if (tb[RTA_NEWDST]) {
2217 if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2218 labels, extack) != 0) {
2223 hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2226 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2232 skb->protocol = htons(ETH_P_MPLS_UC);
2238 if (skb_cow(skb, hdr_size)) {
2243 skb_reserve(skb, hdr_size);
2244 skb_push(skb, hdr_size);
2245 skb_reset_network_header(skb);
2247 /* Push new labels */
2248 hdr = mpls_hdr(skb);
2250 for (i = n_labels - 1; i >= 0; i--) {
2251 hdr[i] = mpls_entry_encode(labels[i],
2257 nh = mpls_select_multipath(rt, skb);
2264 skb_pull(skb, hdr_size);
2265 skb_reset_network_header(skb);
2268 nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2269 RTM_NEWROUTE, sizeof(*r), 0);
2275 r = nlmsg_data(nlh);
2276 r->rtm_family = AF_MPLS;
2277 r->rtm_dst_len = 20;
2279 r->rtm_table = RT_TABLE_MAIN;
2280 r->rtm_type = RTN_UNICAST;
2281 r->rtm_scope = RT_SCOPE_UNIVERSE;
2282 r->rtm_protocol = rt->rt_protocol;
2285 if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2286 goto nla_put_failure;
2288 if (nh->nh_labels &&
2289 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2291 goto nla_put_failure;
2293 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2294 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2296 goto nla_put_failure;
2297 dev = rtnl_dereference(nh->nh_dev);
2298 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2299 goto nla_put_failure;
2301 nlmsg_end(skb, nlh);
2303 err = rtnl_unicast(skb, net, portid);
2308 nlmsg_cancel(skb, nlh);
2315 static int resize_platform_label_table(struct net *net, size_t limit)
2317 size_t size = sizeof(struct mpls_route *) * limit;
2320 struct mpls_route __rcu **labels = NULL, **old;
2321 struct mpls_route *rt0 = NULL, *rt2 = NULL;
2325 labels = kvzalloc(size, GFP_KERNEL);
2330 /* In case the predefined labels need to be populated */
2331 if (limit > MPLS_LABEL_IPV4NULL) {
2332 struct net_device *lo = net->loopback_dev;
2333 rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2336 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2337 rt0->rt_protocol = RTPROT_KERNEL;
2338 rt0->rt_payload_type = MPT_IPV4;
2339 rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2340 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2341 rt0->rt_nh->nh_via_alen = lo->addr_len;
2342 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2345 if (limit > MPLS_LABEL_IPV6NULL) {
2346 struct net_device *lo = net->loopback_dev;
2347 rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2350 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2351 rt2->rt_protocol = RTPROT_KERNEL;
2352 rt2->rt_payload_type = MPT_IPV6;
2353 rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2354 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2355 rt2->rt_nh->nh_via_alen = lo->addr_len;
2356 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2361 /* Remember the original table */
2362 old = rtnl_dereference(net->mpls.platform_label);
2363 old_limit = net->mpls.platform_labels;
2365 /* Free any labels beyond the new table */
2366 for (index = limit; index < old_limit; index++)
2367 mpls_route_update(net, index, NULL, NULL);
2369 /* Copy over the old labels */
2371 if (old_limit < limit)
2372 cp_size = old_limit * sizeof(struct mpls_route *);
2374 memcpy(labels, old, cp_size);
2376 /* If needed set the predefined labels */
2377 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2378 (limit > MPLS_LABEL_IPV6NULL)) {
2379 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2383 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2384 (limit > MPLS_LABEL_IPV4NULL)) {
2385 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2389 /* Update the global pointers */
2390 net->mpls.platform_labels = limit;
2391 rcu_assign_pointer(net->mpls.platform_label, labels);
2412 static int mpls_platform_labels(struct ctl_table *table, int write,
2413 void __user *buffer, size_t *lenp, loff_t *ppos)
2415 struct net *net = table->data;
2416 int platform_labels = net->mpls.platform_labels;
2418 struct ctl_table tmp = {
2419 .procname = table->procname,
2420 .data = &platform_labels,
2421 .maxlen = sizeof(int),
2422 .mode = table->mode,
2424 .extra2 = &label_limit,
2427 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2429 if (write && ret == 0)
2430 ret = resize_platform_label_table(net, platform_labels);
2435 #define MPLS_NS_SYSCTL_OFFSET(field) \
2436 (&((struct net *)0)->field)
2438 static const struct ctl_table mpls_table[] = {
2440 .procname = "platform_labels",
2442 .maxlen = sizeof(int),
2444 .proc_handler = mpls_platform_labels,
2447 .procname = "ip_ttl_propagate",
2448 .data = MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2449 .maxlen = sizeof(int),
2451 .proc_handler = proc_dointvec_minmax,
2456 .procname = "default_ttl",
2457 .data = MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2458 .maxlen = sizeof(int),
2460 .proc_handler = proc_dointvec_minmax,
2467 static int mpls_net_init(struct net *net)
2469 struct ctl_table *table;
2472 net->mpls.platform_labels = 0;
2473 net->mpls.platform_label = NULL;
2474 net->mpls.ip_ttl_propagate = 1;
2475 net->mpls.default_ttl = 255;
2477 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2481 /* Table data contains only offsets relative to the base of
2482 * the mdev at this point, so make them absolute.
2484 for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2485 table[i].data = (char *)net + (uintptr_t)table[i].data;
2487 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2488 if (net->mpls.ctl == NULL) {
2496 static void mpls_net_exit(struct net *net)
2498 struct mpls_route __rcu **platform_label;
2499 size_t platform_labels;
2500 struct ctl_table *table;
2503 table = net->mpls.ctl->ctl_table_arg;
2504 unregister_net_sysctl_table(net->mpls.ctl);
2507 /* An rcu grace period has passed since there was a device in
2508 * the network namespace (and thus the last in flight packet)
2509 * left this network namespace. This is because
2510 * unregister_netdevice_many and netdev_run_todo has completed
2511 * for each network device that was in this network namespace.
2513 * As such no additional rcu synchronization is necessary when
2514 * freeing the platform_label table.
2517 platform_label = rtnl_dereference(net->mpls.platform_label);
2518 platform_labels = net->mpls.platform_labels;
2519 for (index = 0; index < platform_labels; index++) {
2520 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2521 RCU_INIT_POINTER(platform_label[index], NULL);
2522 mpls_notify_route(net, index, rt, NULL, NULL);
2527 kvfree(platform_label);
2530 static struct pernet_operations mpls_net_ops = {
2531 .init = mpls_net_init,
2532 .exit = mpls_net_exit,
2535 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2537 .fill_stats_af = mpls_fill_stats_af,
2538 .get_stats_af_size = mpls_get_stats_af_size,
2541 static int __init mpls_init(void)
2545 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2547 err = register_pernet_subsys(&mpls_net_ops);
2551 err = register_netdevice_notifier(&mpls_dev_notifier);
2553 goto out_unregister_pernet;
2555 dev_add_pack(&mpls_packet_type);
2557 rtnl_af_register(&mpls_af_ops);
2559 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2560 mpls_rtm_newroute, NULL, 0);
2561 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2562 mpls_rtm_delroute, NULL, 0);
2563 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2564 mpls_getroute, mpls_dump_routes, 0);
2565 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2566 mpls_netconf_get_devconf,
2567 mpls_netconf_dump_devconf, 0);
2568 err = ipgre_tunnel_encap_add_mpls_ops();
2570 pr_err("Can't add mpls over gre tunnel ops\n");
2576 out_unregister_pernet:
2577 unregister_pernet_subsys(&mpls_net_ops);
2580 module_init(mpls_init);
2582 static void __exit mpls_exit(void)
2584 rtnl_unregister_all(PF_MPLS);
2585 rtnl_af_unregister(&mpls_af_ops);
2586 dev_remove_pack(&mpls_packet_type);
2587 unregister_netdevice_notifier(&mpls_dev_notifier);
2588 unregister_pernet_subsys(&mpls_net_ops);
2589 ipgre_tunnel_encap_del_mpls_ops();
2591 module_exit(mpls_exit);
2593 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2594 MODULE_LICENSE("GPL v2");
2595 MODULE_ALIAS_NETPROTO(PF_MPLS);