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;
624 return ERR_PTR(-EAFNOSUPPORT);
626 memset(&fl6, 0, sizeof(fl6));
627 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
628 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
639 static struct net_device *inet6_fib_lookup_dev(struct net *net,
642 return ERR_PTR(-EAFNOSUPPORT);
646 static struct net_device *find_outdev(struct net *net,
647 struct mpls_route *rt,
648 struct mpls_nh *nh, int oif)
650 struct net_device *dev = NULL;
653 switch (nh->nh_via_table) {
654 case NEIGH_ARP_TABLE:
655 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
658 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
660 case NEIGH_LINK_TABLE:
664 dev = dev_get_by_index(net, oif);
668 return ERR_PTR(-ENODEV);
673 /* The caller is holding rtnl anyways, so release the dev reference */
679 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
680 struct mpls_nh *nh, int oif)
682 struct net_device *dev = NULL;
685 dev = find_outdev(net, rt, nh, oif);
692 /* Ensure this is a supported device */
694 if (!mpls_dev_get(dev))
697 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
698 (dev->addr_len != nh->nh_via_alen))
701 RCU_INIT_POINTER(nh->nh_dev, dev);
703 if (!(dev->flags & IFF_UP)) {
704 nh->nh_flags |= RTNH_F_DEAD;
708 flags = dev_get_flags(dev);
709 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
710 nh->nh_flags |= RTNH_F_LINKDOWN;
719 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
720 u8 via_addr[], struct netlink_ext_ack *extack)
722 struct rtvia *via = nla_data(nla);
726 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
727 NL_SET_ERR_MSG_ATTR(extack, nla,
728 "Invalid attribute length for RTA_VIA");
731 alen = nla_len(nla) -
732 offsetof(struct rtvia, rtvia_addr);
733 if (alen > MAX_VIA_ALEN) {
734 NL_SET_ERR_MSG_ATTR(extack, nla,
735 "Invalid address length for RTA_VIA");
739 /* Validate the address family */
740 switch (via->rtvia_family) {
742 *via_table = NEIGH_LINK_TABLE;
745 *via_table = NEIGH_ARP_TABLE;
750 *via_table = NEIGH_ND_TABLE;
755 /* Unsupported address family */
759 memcpy(via_addr, via->rtvia_addr, alen);
767 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
768 struct mpls_route *rt)
770 struct net *net = cfg->rc_nlinfo.nl_net;
771 struct mpls_nh *nh = rt->rt_nh;
778 nh->nh_labels = cfg->rc_output_labels;
779 for (i = 0; i < nh->nh_labels; i++)
780 nh->nh_label[i] = cfg->rc_output_label[i];
782 nh->nh_via_table = cfg->rc_via_table;
783 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
784 nh->nh_via_alen = cfg->rc_via_alen;
786 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
790 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
799 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
800 struct mpls_nh *nh, int oif, struct nlattr *via,
801 struct nlattr *newdst, u8 max_labels,
802 struct netlink_ext_ack *extack)
810 err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
811 nh->nh_label, extack);
817 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
818 __mpls_nh_via(rt, nh), extack);
822 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
825 err = mpls_nh_assign_dev(net, rt, nh, oif);
835 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
836 u8 cfg_via_alen, u8 *max_via_alen,
845 while (rtnh_ok(rtnh, remaining)) {
846 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
850 attrlen = rtnh_attrlen(rtnh);
851 nla = nla_find(attrs, attrlen, RTA_VIA);
852 if (nla && nla_len(nla) >=
853 offsetof(struct rtvia, rtvia_addr)) {
854 int via_alen = nla_len(nla) -
855 offsetof(struct rtvia, rtvia_addr);
857 if (via_alen <= MAX_VIA_ALEN)
858 *max_via_alen = max_t(u16, *max_via_alen,
862 nla = nla_find(attrs, attrlen, RTA_NEWDST);
864 nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
868 *max_labels = max_t(u8, *max_labels, n_labels);
870 /* number of nexthops is tracked by a u8.
871 * Check for overflow.
877 rtnh = rtnh_next(rtnh, &remaining);
880 /* leftover implies invalid nexthop configuration, discard it */
881 return remaining > 0 ? 0 : nhs;
884 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
885 struct mpls_route *rt, u8 max_labels,
886 struct netlink_ext_ack *extack)
888 struct rtnexthop *rtnh = cfg->rc_mp;
889 struct nlattr *nla_via, *nla_newdst;
890 int remaining = cfg->rc_mp_len;
894 change_nexthops(rt) {
901 if (!rtnh_ok(rtnh, remaining))
904 /* neither weighted multipath nor any flags
907 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
910 attrlen = rtnh_attrlen(rtnh);
912 struct nlattr *attrs = rtnh_attrs(rtnh);
914 nla_via = nla_find(attrs, attrlen, RTA_VIA);
915 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
918 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
919 rtnh->rtnh_ifindex, nla_via, nla_newdst,
924 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
927 rtnh = rtnh_next(rtnh, &remaining);
929 } endfor_nexthops(rt);
939 static bool mpls_label_ok(struct net *net, unsigned int *index,
940 struct netlink_ext_ack *extack)
944 /* Reserved labels may not be set */
945 if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
946 NL_SET_ERR_MSG(extack,
947 "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
951 /* The full 20 bit range may not be supported. */
952 if (is_ok && *index >= net->mpls.platform_labels) {
953 NL_SET_ERR_MSG(extack,
954 "Label >= configured maximum in platform_labels");
958 *index = array_index_nospec(*index, net->mpls.platform_labels);
962 static int mpls_route_add(struct mpls_route_config *cfg,
963 struct netlink_ext_ack *extack)
965 struct mpls_route __rcu **platform_label;
966 struct net *net = cfg->rc_nlinfo.nl_net;
967 struct mpls_route *rt, *old;
974 index = cfg->rc_label;
976 /* If a label was not specified during insert pick one */
977 if ((index == LABEL_NOT_SPECIFIED) &&
978 (cfg->rc_nlflags & NLM_F_CREATE)) {
979 index = find_free_label(net);
982 if (!mpls_label_ok(net, &index, extack))
985 /* Append makes no sense with mpls */
987 if (cfg->rc_nlflags & NLM_F_APPEND) {
988 NL_SET_ERR_MSG(extack, "MPLS does not support route append");
993 platform_label = rtnl_dereference(net->mpls.platform_label);
994 old = rtnl_dereference(platform_label[index]);
995 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
999 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1003 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1008 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1009 cfg->rc_via_alen, &max_via_alen,
1012 max_via_alen = cfg->rc_via_alen;
1013 max_labels = cfg->rc_output_labels;
1018 NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1023 rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1029 rt->rt_protocol = cfg->rc_protocol;
1030 rt->rt_payload_type = cfg->rc_payload_type;
1031 rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1034 err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1036 err = mpls_nh_build_from_cfg(cfg, rt);
1040 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1050 static int mpls_route_del(struct mpls_route_config *cfg,
1051 struct netlink_ext_ack *extack)
1053 struct net *net = cfg->rc_nlinfo.nl_net;
1057 index = cfg->rc_label;
1059 if (!mpls_label_ok(net, &index, extack))
1062 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1069 static void mpls_get_stats(struct mpls_dev *mdev,
1070 struct mpls_link_stats *stats)
1072 struct mpls_pcpu_stats *p;
1075 memset(stats, 0, sizeof(*stats));
1077 for_each_possible_cpu(i) {
1078 struct mpls_link_stats local;
1081 p = per_cpu_ptr(mdev->stats, i);
1083 start = u64_stats_fetch_begin(&p->syncp);
1085 } while (u64_stats_fetch_retry(&p->syncp, start));
1087 stats->rx_packets += local.rx_packets;
1088 stats->rx_bytes += local.rx_bytes;
1089 stats->tx_packets += local.tx_packets;
1090 stats->tx_bytes += local.tx_bytes;
1091 stats->rx_errors += local.rx_errors;
1092 stats->tx_errors += local.tx_errors;
1093 stats->rx_dropped += local.rx_dropped;
1094 stats->tx_dropped += local.tx_dropped;
1095 stats->rx_noroute += local.rx_noroute;
1099 static int mpls_fill_stats_af(struct sk_buff *skb,
1100 const struct net_device *dev)
1102 struct mpls_link_stats *stats;
1103 struct mpls_dev *mdev;
1106 mdev = mpls_dev_get(dev);
1110 nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1111 sizeof(struct mpls_link_stats),
1116 stats = nla_data(nla);
1117 mpls_get_stats(mdev, stats);
1122 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1124 struct mpls_dev *mdev;
1126 mdev = mpls_dev_get(dev);
1130 return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1133 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1134 u32 portid, u32 seq, int event,
1135 unsigned int flags, int type)
1137 struct nlmsghdr *nlh;
1138 struct netconfmsg *ncm;
1141 nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1146 if (type == NETCONFA_ALL)
1149 ncm = nlmsg_data(nlh);
1150 ncm->ncm_family = AF_MPLS;
1152 if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1153 goto nla_put_failure;
1155 if ((all || type == NETCONFA_INPUT) &&
1156 nla_put_s32(skb, NETCONFA_INPUT,
1157 mdev->input_enabled) < 0)
1158 goto nla_put_failure;
1160 nlmsg_end(skb, nlh);
1164 nlmsg_cancel(skb, nlh);
1168 static int mpls_netconf_msgsize_devconf(int type)
1170 int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1171 + nla_total_size(4); /* NETCONFA_IFINDEX */
1174 if (type == NETCONFA_ALL)
1177 if (all || type == NETCONFA_INPUT)
1178 size += nla_total_size(4);
1183 static void mpls_netconf_notify_devconf(struct net *net, int event,
1184 int type, struct mpls_dev *mdev)
1186 struct sk_buff *skb;
1189 skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1193 err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1195 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1196 WARN_ON(err == -EMSGSIZE);
1201 rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1205 rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1208 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1209 [NETCONFA_IFINDEX] = { .len = sizeof(int) },
1212 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1213 struct nlmsghdr *nlh,
1214 struct netlink_ext_ack *extack)
1216 struct net *net = sock_net(in_skb->sk);
1217 struct nlattr *tb[NETCONFA_MAX + 1];
1218 struct netconfmsg *ncm;
1219 struct net_device *dev;
1220 struct mpls_dev *mdev;
1221 struct sk_buff *skb;
1225 err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1226 devconf_mpls_policy, extack);
1231 if (!tb[NETCONFA_IFINDEX])
1234 ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1235 dev = __dev_get_by_index(net, ifindex);
1239 mdev = mpls_dev_get(dev);
1244 skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1248 err = mpls_netconf_fill_devconf(skb, mdev,
1249 NETLINK_CB(in_skb).portid,
1250 nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1253 /* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1254 WARN_ON(err == -EMSGSIZE);
1258 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1263 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1264 struct netlink_callback *cb)
1266 const struct nlmsghdr *nlh = cb->nlh;
1267 struct net *net = sock_net(skb->sk);
1268 struct hlist_head *head;
1269 struct net_device *dev;
1270 struct mpls_dev *mdev;
1274 if (cb->strict_check) {
1275 struct netlink_ext_ack *extack = cb->extack;
1276 struct netconfmsg *ncm;
1278 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1279 NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1283 if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1284 NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1290 s_idx = idx = cb->args[1];
1292 for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1294 head = &net->dev_index_head[h];
1296 cb->seq = net->dev_base_seq;
1297 hlist_for_each_entry_rcu(dev, head, index_hlist) {
1300 mdev = mpls_dev_get(dev);
1303 if (mpls_netconf_fill_devconf(skb, mdev,
1304 NETLINK_CB(cb->skb).portid,
1308 NETCONFA_ALL) < 0) {
1312 nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1325 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
1326 (&((struct mpls_dev *)0)->field)
1328 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1329 void __user *buffer,
1330 size_t *lenp, loff_t *ppos)
1332 int oval = *(int *)ctl->data;
1333 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1336 struct mpls_dev *mdev = ctl->extra1;
1337 int i = (int *)ctl->data - (int *)mdev;
1338 struct net *net = ctl->extra2;
1339 int val = *(int *)ctl->data;
1341 if (i == offsetof(struct mpls_dev, input_enabled) &&
1343 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1344 NETCONFA_INPUT, mdev);
1351 static const struct ctl_table mpls_dev_table[] = {
1353 .procname = "input",
1354 .maxlen = sizeof(int),
1356 .proc_handler = mpls_conf_proc,
1357 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1362 static int mpls_dev_sysctl_register(struct net_device *dev,
1363 struct mpls_dev *mdev)
1365 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1366 struct net *net = dev_net(dev);
1367 struct ctl_table *table;
1370 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1374 /* Table data contains only offsets relative to the base of
1375 * the mdev at this point, so make them absolute.
1377 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1378 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1379 table[i].extra1 = mdev;
1380 table[i].extra2 = net;
1383 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1385 mdev->sysctl = register_net_sysctl(net, path, table);
1389 mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1398 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1399 struct mpls_dev *mdev)
1401 struct net *net = dev_net(dev);
1402 struct ctl_table *table;
1404 table = mdev->sysctl->ctl_table_arg;
1405 unregister_net_sysctl_table(mdev->sysctl);
1408 mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1411 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1413 struct mpls_dev *mdev;
1419 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1421 return ERR_PTR(err);
1423 mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1427 for_each_possible_cpu(i) {
1428 struct mpls_pcpu_stats *mpls_stats;
1430 mpls_stats = per_cpu_ptr(mdev->stats, i);
1431 u64_stats_init(&mpls_stats->syncp);
1436 err = mpls_dev_sysctl_register(dev, mdev);
1440 rcu_assign_pointer(dev->mpls_ptr, mdev);
1445 free_percpu(mdev->stats);
1447 return ERR_PTR(err);
1450 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1452 struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1454 free_percpu(mdev->stats);
1458 static void mpls_ifdown(struct net_device *dev, int event)
1460 struct mpls_route __rcu **platform_label;
1461 struct net *net = dev_net(dev);
1465 platform_label = rtnl_dereference(net->mpls.platform_label);
1466 for (index = 0; index < net->mpls.platform_labels; index++) {
1467 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1474 change_nexthops(rt) {
1475 unsigned int nh_flags = nh->nh_flags;
1477 if (rtnl_dereference(nh->nh_dev) != dev)
1482 case NETDEV_UNREGISTER:
1483 nh_flags |= RTNH_F_DEAD;
1486 nh_flags |= RTNH_F_LINKDOWN;
1489 if (event == NETDEV_UNREGISTER)
1490 RCU_INIT_POINTER(nh->nh_dev, NULL);
1492 if (nh->nh_flags != nh_flags)
1493 WRITE_ONCE(nh->nh_flags, nh_flags);
1495 if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1497 if (!rtnl_dereference(nh->nh_dev))
1499 } endfor_nexthops(rt);
1501 WRITE_ONCE(rt->rt_nhn_alive, alive);
1503 /* if there are no more nexthops, delete the route */
1504 if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1505 mpls_route_update(net, index, NULL, NULL);
1509 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1511 struct mpls_route __rcu **platform_label;
1512 struct net *net = dev_net(dev);
1516 platform_label = rtnl_dereference(net->mpls.platform_label);
1517 for (index = 0; index < net->mpls.platform_labels; index++) {
1518 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1524 change_nexthops(rt) {
1525 unsigned int nh_flags = nh->nh_flags;
1526 struct net_device *nh_dev =
1527 rtnl_dereference(nh->nh_dev);
1529 if (!(nh_flags & flags)) {
1537 WRITE_ONCE(nh->nh_flags, nh_flags);
1538 } endfor_nexthops(rt);
1540 WRITE_ONCE(rt->rt_nhn_alive, alive);
1544 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1547 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1548 struct mpls_dev *mdev;
1551 if (event == NETDEV_REGISTER) {
1553 /* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1556 if (dev->type == ARPHRD_ETHER ||
1557 dev->type == ARPHRD_LOOPBACK ||
1558 dev->type == ARPHRD_IPGRE ||
1559 dev->type == ARPHRD_IP6GRE ||
1560 dev->type == ARPHRD_SIT ||
1561 dev->type == ARPHRD_TUNNEL) {
1562 mdev = mpls_add_dev(dev);
1564 return notifier_from_errno(PTR_ERR(mdev));
1569 mdev = mpls_dev_get(dev);
1575 mpls_ifdown(dev, event);
1578 flags = dev_get_flags(dev);
1579 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1580 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1582 mpls_ifup(dev, RTNH_F_DEAD);
1585 flags = dev_get_flags(dev);
1586 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1587 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1589 mpls_ifdown(dev, event);
1591 case NETDEV_UNREGISTER:
1592 mpls_ifdown(dev, event);
1593 mdev = mpls_dev_get(dev);
1595 mpls_dev_sysctl_unregister(dev, mdev);
1596 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1597 call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1600 case NETDEV_CHANGENAME:
1601 mdev = mpls_dev_get(dev);
1605 mpls_dev_sysctl_unregister(dev, mdev);
1606 err = mpls_dev_sysctl_register(dev, mdev);
1608 return notifier_from_errno(err);
1615 static struct notifier_block mpls_dev_notifier = {
1616 .notifier_call = mpls_dev_notify,
1619 static int nla_put_via(struct sk_buff *skb,
1620 u8 table, const void *addr, int alen)
1622 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1623 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1627 int family = AF_UNSPEC;
1629 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1633 if (table <= NEIGH_NR_TABLES)
1634 family = table_to_family[table];
1636 via = nla_data(nla);
1637 via->rtvia_family = family;
1638 memcpy(via->rtvia_addr, addr, alen);
1642 int nla_put_labels(struct sk_buff *skb, int attrtype,
1643 u8 labels, const u32 label[])
1646 struct mpls_shim_hdr *nla_label;
1649 nla = nla_reserve(skb, attrtype, labels*4);
1653 nla_label = nla_data(nla);
1655 for (i = labels - 1; i >= 0; i--) {
1656 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1662 EXPORT_SYMBOL_GPL(nla_put_labels);
1664 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1665 u32 label[], struct netlink_ext_ack *extack)
1667 unsigned len = nla_len(nla);
1668 struct mpls_shim_hdr *nla_label;
1673 /* len needs to be an even multiple of 4 (the label size). Number
1674 * of labels is a u8 so check for overflow.
1676 if (len & 3 || len / 4 > 255) {
1677 NL_SET_ERR_MSG_ATTR(extack, nla,
1678 "Invalid length for labels attribute");
1682 /* Limit the number of new labels allowed */
1684 if (nla_labels > max_labels) {
1685 NL_SET_ERR_MSG(extack, "Too many labels");
1689 /* when label == NULL, caller wants number of labels */
1693 nla_label = nla_data(nla);
1695 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1696 struct mpls_entry_decoded dec;
1697 dec = mpls_entry_decode(nla_label + i);
1699 /* Ensure the bottom of stack flag is properly set
1700 * and ttl and tc are both clear.
1703 NL_SET_ERR_MSG_ATTR(extack, nla,
1704 "TTL in label must be 0");
1709 NL_SET_ERR_MSG_ATTR(extack, nla,
1710 "Traffic class in label must be 0");
1714 if (dec.bos != bos) {
1715 NL_SET_BAD_ATTR(extack, nla);
1717 NL_SET_ERR_MSG(extack,
1718 "BOS bit must be set in first label");
1720 NL_SET_ERR_MSG(extack,
1721 "BOS bit can only be set in first label");
1726 switch (dec.label) {
1727 case MPLS_LABEL_IMPLNULL:
1728 /* RFC3032: This is a label that an LSR may
1729 * assign and distribute, but which never
1730 * actually appears in the encapsulation.
1732 NL_SET_ERR_MSG_ATTR(extack, nla,
1733 "Implicit NULL Label (3) can not be used in encapsulation");
1737 label[i] = dec.label;
1740 *labels = nla_labels;
1743 EXPORT_SYMBOL_GPL(nla_get_labels);
1745 static int rtm_to_route_config(struct sk_buff *skb,
1746 struct nlmsghdr *nlh,
1747 struct mpls_route_config *cfg,
1748 struct netlink_ext_ack *extack)
1751 struct nlattr *tb[RTA_MAX+1];
1755 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy,
1761 rtm = nlmsg_data(nlh);
1763 if (rtm->rtm_family != AF_MPLS) {
1764 NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1767 if (rtm->rtm_dst_len != 20) {
1768 NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1771 if (rtm->rtm_src_len != 0) {
1772 NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1775 if (rtm->rtm_tos != 0) {
1776 NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1779 if (rtm->rtm_table != RT_TABLE_MAIN) {
1780 NL_SET_ERR_MSG(extack,
1781 "MPLS only supports the main route table");
1784 /* Any value is acceptable for rtm_protocol */
1786 /* As mpls uses destination specific addresses
1787 * (or source specific address in the case of multicast)
1788 * all addresses have universal scope.
1790 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1791 NL_SET_ERR_MSG(extack,
1792 "Invalid route scope - MPLS only supports UNIVERSE");
1795 if (rtm->rtm_type != RTN_UNICAST) {
1796 NL_SET_ERR_MSG(extack,
1797 "Invalid route type - MPLS only supports UNICAST");
1800 if (rtm->rtm_flags != 0) {
1801 NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1805 cfg->rc_label = LABEL_NOT_SPECIFIED;
1806 cfg->rc_protocol = rtm->rtm_protocol;
1807 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1808 cfg->rc_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
1809 cfg->rc_nlflags = nlh->nlmsg_flags;
1810 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1811 cfg->rc_nlinfo.nlh = nlh;
1812 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1814 for (index = 0; index <= RTA_MAX; index++) {
1815 struct nlattr *nla = tb[index];
1821 cfg->rc_ifindex = nla_get_u32(nla);
1824 if (nla_get_labels(nla, MAX_NEW_LABELS,
1825 &cfg->rc_output_labels,
1826 cfg->rc_output_label, extack))
1832 if (nla_get_labels(nla, 1, &label_count,
1833 &cfg->rc_label, extack))
1836 if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1837 &cfg->rc_label, extack))
1843 if (nla_get_via(nla, &cfg->rc_via_alen,
1844 &cfg->rc_via_table, cfg->rc_via,
1851 cfg->rc_mp = nla_data(nla);
1852 cfg->rc_mp_len = nla_len(nla);
1855 case RTA_TTL_PROPAGATE:
1857 u8 ttl_propagate = nla_get_u8(nla);
1859 if (ttl_propagate > 1) {
1860 NL_SET_ERR_MSG_ATTR(extack, nla,
1861 "RTA_TTL_PROPAGATE can only be 0 or 1");
1864 cfg->rc_ttl_propagate = ttl_propagate ?
1865 MPLS_TTL_PROP_ENABLED :
1866 MPLS_TTL_PROP_DISABLED;
1870 NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1871 /* Unsupported attribute */
1881 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1882 struct netlink_ext_ack *extack)
1884 struct mpls_route_config *cfg;
1887 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1891 err = rtm_to_route_config(skb, nlh, cfg, extack);
1895 err = mpls_route_del(cfg, extack);
1903 static int mpls_rtm_newroute(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_add(cfg, extack);
1924 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1925 u32 label, struct mpls_route *rt, int flags)
1927 struct net_device *dev;
1928 struct nlmsghdr *nlh;
1931 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1935 rtm = nlmsg_data(nlh);
1936 rtm->rtm_family = AF_MPLS;
1937 rtm->rtm_dst_len = 20;
1938 rtm->rtm_src_len = 0;
1940 rtm->rtm_table = RT_TABLE_MAIN;
1941 rtm->rtm_protocol = rt->rt_protocol;
1942 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1943 rtm->rtm_type = RTN_UNICAST;
1946 if (nla_put_labels(skb, RTA_DST, 1, &label))
1947 goto nla_put_failure;
1949 if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1950 bool ttl_propagate =
1951 rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1953 if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1955 goto nla_put_failure;
1957 if (rt->rt_nhn == 1) {
1958 const struct mpls_nh *nh = rt->rt_nh;
1960 if (nh->nh_labels &&
1961 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1963 goto nla_put_failure;
1964 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1965 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1967 goto nla_put_failure;
1968 dev = rtnl_dereference(nh->nh_dev);
1969 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1970 goto nla_put_failure;
1971 if (nh->nh_flags & RTNH_F_LINKDOWN)
1972 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1973 if (nh->nh_flags & RTNH_F_DEAD)
1974 rtm->rtm_flags |= RTNH_F_DEAD;
1976 struct rtnexthop *rtnh;
1981 mp = nla_nest_start(skb, RTA_MULTIPATH);
1983 goto nla_put_failure;
1986 dev = rtnl_dereference(nh->nh_dev);
1990 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1992 goto nla_put_failure;
1994 rtnh->rtnh_ifindex = dev->ifindex;
1995 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1996 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1999 if (nh->nh_flags & RTNH_F_DEAD) {
2000 rtnh->rtnh_flags |= RTNH_F_DEAD;
2004 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2007 goto nla_put_failure;
2008 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2009 nla_put_via(skb, nh->nh_via_table,
2010 mpls_nh_via(rt, nh),
2012 goto nla_put_failure;
2014 /* length of rtnetlink header + attributes */
2015 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2016 } endfor_nexthops(rt);
2018 if (linkdown == rt->rt_nhn)
2019 rtm->rtm_flags |= RTNH_F_LINKDOWN;
2020 if (dead == rt->rt_nhn)
2021 rtm->rtm_flags |= RTNH_F_DEAD;
2023 nla_nest_end(skb, mp);
2026 nlmsg_end(skb, nlh);
2030 nlmsg_cancel(skb, nlh);
2034 #if IS_ENABLED(CONFIG_INET)
2035 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2036 struct fib_dump_filter *filter,
2037 struct netlink_callback *cb)
2039 return ip_valid_fib_dump_req(net, nlh, filter, cb);
2042 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2043 struct fib_dump_filter *filter,
2044 struct netlink_callback *cb)
2046 struct netlink_ext_ack *extack = cb->extack;
2047 struct nlattr *tb[RTA_MAX + 1];
2051 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2052 NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2056 rtm = nlmsg_data(nlh);
2057 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
2058 rtm->rtm_table || rtm->rtm_scope || rtm->rtm_type ||
2060 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2064 if (rtm->rtm_protocol) {
2065 filter->protocol = rtm->rtm_protocol;
2066 filter->filter_set = 1;
2067 cb->answer_flags = NLM_F_DUMP_FILTERED;
2070 err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2071 rtm_mpls_policy, extack);
2075 for (i = 0; i <= RTA_MAX; ++i) {
2079 ifindex = nla_get_u32(tb[i]);
2080 filter->dev = __dev_get_by_index(net, ifindex);
2083 filter->filter_set = 1;
2085 NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2094 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2095 const struct net_device *dev)
2097 struct net_device *nh_dev;
2099 if (rt->rt_nhn == 1) {
2100 struct mpls_nh *nh = rt->rt_nh;
2102 nh_dev = rtnl_dereference(nh->nh_dev);
2107 nh_dev = rtnl_dereference(nh->nh_dev);
2110 } endfor_nexthops(rt);
2116 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2118 const struct nlmsghdr *nlh = cb->nlh;
2119 struct net *net = sock_net(skb->sk);
2120 struct mpls_route __rcu **platform_label;
2121 struct fib_dump_filter filter = {};
2122 unsigned int flags = NLM_F_MULTI;
2123 size_t platform_labels;
2128 if (cb->strict_check) {
2131 err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2135 /* for MPLS, there is only 1 table with fixed type and flags.
2136 * If either are set in the filter then return nothing.
2138 if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2139 (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2144 index = cb->args[0];
2145 if (index < MPLS_LABEL_FIRST_UNRESERVED)
2146 index = MPLS_LABEL_FIRST_UNRESERVED;
2148 platform_label = rtnl_dereference(net->mpls.platform_label);
2149 platform_labels = net->mpls.platform_labels;
2151 if (filter.filter_set)
2152 flags |= NLM_F_DUMP_FILTERED;
2154 for (; index < platform_labels; index++) {
2155 struct mpls_route *rt;
2157 rt = rtnl_dereference(platform_label[index]);
2161 if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2162 (filter.protocol && rt->rt_protocol != filter.protocol))
2165 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2166 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2167 index, rt, flags) < 0)
2170 cb->args[0] = index;
2175 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2178 NLMSG_ALIGN(sizeof(struct rtmsg))
2179 + nla_total_size(4) /* RTA_DST */
2180 + nla_total_size(1); /* RTA_TTL_PROPAGATE */
2182 if (rt->rt_nhn == 1) {
2183 struct mpls_nh *nh = rt->rt_nh;
2186 payload += nla_total_size(4); /* RTA_OIF */
2187 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2188 payload += nla_total_size(2 + nh->nh_via_alen);
2189 if (nh->nh_labels) /* RTA_NEWDST */
2190 payload += nla_total_size(nh->nh_labels * 4);
2192 /* each nexthop is packed in an attribute */
2196 if (!rtnl_dereference(nh->nh_dev))
2198 nhsize += nla_total_size(sizeof(struct rtnexthop));
2200 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2201 nhsize += nla_total_size(2 + nh->nh_via_alen);
2203 nhsize += nla_total_size(nh->nh_labels * 4);
2204 } endfor_nexthops(rt);
2205 /* nested attribute */
2206 payload += nla_total_size(nhsize);
2212 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2213 struct nlmsghdr *nlh, struct net *net, u32 portid,
2214 unsigned int nlm_flags)
2216 struct sk_buff *skb;
2217 u32 seq = nlh ? nlh->nlmsg_seq : 0;
2220 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2224 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2226 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2227 WARN_ON(err == -EMSGSIZE);
2231 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2236 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2239 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2240 struct netlink_ext_ack *extack)
2242 struct net *net = sock_net(in_skb->sk);
2243 u32 portid = NETLINK_CB(in_skb).portid;
2244 u32 in_label = LABEL_NOT_SPECIFIED;
2245 struct nlattr *tb[RTA_MAX + 1];
2246 u32 labels[MAX_NEW_LABELS];
2247 struct mpls_shim_hdr *hdr;
2248 unsigned int hdr_size = 0;
2249 struct net_device *dev;
2250 struct mpls_route *rt;
2251 struct rtmsg *rtm, *r;
2252 struct nlmsghdr *nlh;
2253 struct sk_buff *skb;
2258 err = nlmsg_parse(in_nlh, sizeof(*rtm), tb, RTA_MAX,
2259 rtm_mpls_policy, extack);
2263 rtm = nlmsg_data(in_nlh);
2268 if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2269 &in_label, extack)) {
2274 if (!mpls_label_ok(net, &in_label, extack)) {
2280 rt = mpls_route_input_rcu(net, in_label);
2286 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2287 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2293 err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2294 RTM_NEWROUTE, in_label, rt, 0);
2296 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2297 WARN_ON(err == -EMSGSIZE);
2301 return rtnl_unicast(skb, net, portid);
2304 if (tb[RTA_NEWDST]) {
2305 if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2306 labels, extack) != 0) {
2311 hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2314 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2320 skb->protocol = htons(ETH_P_MPLS_UC);
2326 if (skb_cow(skb, hdr_size)) {
2331 skb_reserve(skb, hdr_size);
2332 skb_push(skb, hdr_size);
2333 skb_reset_network_header(skb);
2335 /* Push new labels */
2336 hdr = mpls_hdr(skb);
2338 for (i = n_labels - 1; i >= 0; i--) {
2339 hdr[i] = mpls_entry_encode(labels[i],
2345 nh = mpls_select_multipath(rt, skb);
2352 skb_pull(skb, hdr_size);
2353 skb_reset_network_header(skb);
2356 nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2357 RTM_NEWROUTE, sizeof(*r), 0);
2363 r = nlmsg_data(nlh);
2364 r->rtm_family = AF_MPLS;
2365 r->rtm_dst_len = 20;
2367 r->rtm_table = RT_TABLE_MAIN;
2368 r->rtm_type = RTN_UNICAST;
2369 r->rtm_scope = RT_SCOPE_UNIVERSE;
2370 r->rtm_protocol = rt->rt_protocol;
2373 if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2374 goto nla_put_failure;
2376 if (nh->nh_labels &&
2377 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2379 goto nla_put_failure;
2381 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2382 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2384 goto nla_put_failure;
2385 dev = rtnl_dereference(nh->nh_dev);
2386 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2387 goto nla_put_failure;
2389 nlmsg_end(skb, nlh);
2391 err = rtnl_unicast(skb, net, portid);
2396 nlmsg_cancel(skb, nlh);
2403 static int resize_platform_label_table(struct net *net, size_t limit)
2405 size_t size = sizeof(struct mpls_route *) * limit;
2408 struct mpls_route __rcu **labels = NULL, **old;
2409 struct mpls_route *rt0 = NULL, *rt2 = NULL;
2413 labels = kvzalloc(size, GFP_KERNEL);
2418 /* In case the predefined labels need to be populated */
2419 if (limit > MPLS_LABEL_IPV4NULL) {
2420 struct net_device *lo = net->loopback_dev;
2421 rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2424 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2425 rt0->rt_protocol = RTPROT_KERNEL;
2426 rt0->rt_payload_type = MPT_IPV4;
2427 rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2428 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2429 rt0->rt_nh->nh_via_alen = lo->addr_len;
2430 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2433 if (limit > MPLS_LABEL_IPV6NULL) {
2434 struct net_device *lo = net->loopback_dev;
2435 rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2438 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2439 rt2->rt_protocol = RTPROT_KERNEL;
2440 rt2->rt_payload_type = MPT_IPV6;
2441 rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2442 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2443 rt2->rt_nh->nh_via_alen = lo->addr_len;
2444 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2449 /* Remember the original table */
2450 old = rtnl_dereference(net->mpls.platform_label);
2451 old_limit = net->mpls.platform_labels;
2453 /* Free any labels beyond the new table */
2454 for (index = limit; index < old_limit; index++)
2455 mpls_route_update(net, index, NULL, NULL);
2457 /* Copy over the old labels */
2459 if (old_limit < limit)
2460 cp_size = old_limit * sizeof(struct mpls_route *);
2462 memcpy(labels, old, cp_size);
2464 /* If needed set the predefined labels */
2465 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2466 (limit > MPLS_LABEL_IPV6NULL)) {
2467 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2471 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2472 (limit > MPLS_LABEL_IPV4NULL)) {
2473 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2477 /* Update the global pointers */
2478 net->mpls.platform_labels = limit;
2479 rcu_assign_pointer(net->mpls.platform_label, labels);
2500 static int mpls_platform_labels(struct ctl_table *table, int write,
2501 void __user *buffer, size_t *lenp, loff_t *ppos)
2503 struct net *net = table->data;
2504 int platform_labels = net->mpls.platform_labels;
2506 struct ctl_table tmp = {
2507 .procname = table->procname,
2508 .data = &platform_labels,
2509 .maxlen = sizeof(int),
2510 .mode = table->mode,
2512 .extra2 = &label_limit,
2515 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2517 if (write && ret == 0)
2518 ret = resize_platform_label_table(net, platform_labels);
2523 #define MPLS_NS_SYSCTL_OFFSET(field) \
2524 (&((struct net *)0)->field)
2526 static const struct ctl_table mpls_table[] = {
2528 .procname = "platform_labels",
2530 .maxlen = sizeof(int),
2532 .proc_handler = mpls_platform_labels,
2535 .procname = "ip_ttl_propagate",
2536 .data = MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2537 .maxlen = sizeof(int),
2539 .proc_handler = proc_dointvec_minmax,
2544 .procname = "default_ttl",
2545 .data = MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2546 .maxlen = sizeof(int),
2548 .proc_handler = proc_dointvec_minmax,
2555 static int mpls_net_init(struct net *net)
2557 struct ctl_table *table;
2560 net->mpls.platform_labels = 0;
2561 net->mpls.platform_label = NULL;
2562 net->mpls.ip_ttl_propagate = 1;
2563 net->mpls.default_ttl = 255;
2565 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2569 /* Table data contains only offsets relative to the base of
2570 * the mdev at this point, so make them absolute.
2572 for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2573 table[i].data = (char *)net + (uintptr_t)table[i].data;
2575 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2576 if (net->mpls.ctl == NULL) {
2584 static void mpls_net_exit(struct net *net)
2586 struct mpls_route __rcu **platform_label;
2587 size_t platform_labels;
2588 struct ctl_table *table;
2591 table = net->mpls.ctl->ctl_table_arg;
2592 unregister_net_sysctl_table(net->mpls.ctl);
2595 /* An rcu grace period has passed since there was a device in
2596 * the network namespace (and thus the last in flight packet)
2597 * left this network namespace. This is because
2598 * unregister_netdevice_many and netdev_run_todo has completed
2599 * for each network device that was in this network namespace.
2601 * As such no additional rcu synchronization is necessary when
2602 * freeing the platform_label table.
2605 platform_label = rtnl_dereference(net->mpls.platform_label);
2606 platform_labels = net->mpls.platform_labels;
2607 for (index = 0; index < platform_labels; index++) {
2608 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2609 RCU_INIT_POINTER(platform_label[index], NULL);
2610 mpls_notify_route(net, index, rt, NULL, NULL);
2615 kvfree(platform_label);
2618 static struct pernet_operations mpls_net_ops = {
2619 .init = mpls_net_init,
2620 .exit = mpls_net_exit,
2623 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2625 .fill_stats_af = mpls_fill_stats_af,
2626 .get_stats_af_size = mpls_get_stats_af_size,
2629 static int __init mpls_init(void)
2633 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2635 err = register_pernet_subsys(&mpls_net_ops);
2639 err = register_netdevice_notifier(&mpls_dev_notifier);
2641 goto out_unregister_pernet;
2643 dev_add_pack(&mpls_packet_type);
2645 rtnl_af_register(&mpls_af_ops);
2647 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2648 mpls_rtm_newroute, NULL, 0);
2649 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2650 mpls_rtm_delroute, NULL, 0);
2651 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2652 mpls_getroute, mpls_dump_routes, 0);
2653 rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2654 mpls_netconf_get_devconf,
2655 mpls_netconf_dump_devconf, 0);
2656 err = ipgre_tunnel_encap_add_mpls_ops();
2658 pr_err("Can't add mpls over gre tunnel ops\n");
2664 out_unregister_pernet:
2665 unregister_pernet_subsys(&mpls_net_ops);
2668 module_init(mpls_init);
2670 static void __exit mpls_exit(void)
2672 rtnl_unregister_all(PF_MPLS);
2673 rtnl_af_unregister(&mpls_af_ops);
2674 dev_remove_pack(&mpls_packet_type);
2675 unregister_netdevice_notifier(&mpls_dev_notifier);
2676 unregister_pernet_subsys(&mpls_net_ops);
2677 ipgre_tunnel_encap_del_mpls_ops();
2679 module_exit(mpls_exit);
2681 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2682 MODULE_LICENSE("GPL v2");
2683 MODULE_ALIAS_NETPROTO(PF_MPLS);