1// SPDX-License-Identifier: GPL-2.0-only
   2#include <linux/types.h>
   3#include <linux/skbuff.h>
   4#include <linux/socket.h>
   5#include <linux/sysctl.h>
   6#include <linux/net.h>
   7#include <linux/module.h>
   8#include <linux/if_arp.h>
   9#include <linux/ipv6.h>
  10#include <linux/mpls.h>
  11#include <linux/netconf.h>
  12#include <linux/nospec.h>
  13#include <linux/vmalloc.h>
  14#include <linux/percpu.h>
  15#include <net/gso.h>
  16#include <net/ip.h>
  17#include <net/dst.h>
  18#include <net/sock.h>
  19#include <net/arp.h>
  20#include <net/ip_fib.h>
  21#include <net/netevent.h>
  22#include <net/ip_tunnels.h>
  23#include <net/netns/generic.h>
  24#if IS_ENABLED(CONFIG_IPV6)
  25#include <net/ipv6.h>
 
  26#endif
  27#include <net/ipv6_stubs.h>
  28#include <net/rtnh.h>
  29#include "internal.h"
  30
  31/* max memory we will use for mpls_route */
  32#define MAX_MPLS_ROUTE_MEM	4096
  33
  34/* Maximum number of labels to look ahead at when selecting a path of
  35 * a multipath route
  36 */
  37#define MAX_MP_SELECT_LABELS 4
  38
  39#define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
  40
 
  41static int label_limit = (1 << 20) - 1;
  42static int ttl_max = 255;
  43
  44#if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
  45static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
  46{
  47	return sizeof(struct mpls_shim_hdr);
  48}
  49
  50static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
  51	.encap_hlen	= ipgre_mpls_encap_hlen,
  52};
  53
  54static int ipgre_tunnel_encap_add_mpls_ops(void)
  55{
  56	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
  57}
  58
  59static void ipgre_tunnel_encap_del_mpls_ops(void)
  60{
  61	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
  62}
  63#else
  64static int ipgre_tunnel_encap_add_mpls_ops(void)
  65{
  66	return 0;
  67}
  68
  69static void ipgre_tunnel_encap_del_mpls_ops(void)
  70{
  71}
  72#endif
  73
  74static 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);
  77
  78static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
  79{
  80	struct mpls_route *rt = NULL;
  81
  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]);
  86	}
  87	return rt;
  88}
  89
 
 
 
 
 
  90bool mpls_output_possible(const struct net_device *dev)
  91{
  92	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
  93}
  94EXPORT_SYMBOL_GPL(mpls_output_possible);
  95
  96static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
  97{
  98	return (u8 *)nh + rt->rt_via_offset;
 
 
 
  99}
 100
 101static const u8 *mpls_nh_via(const struct mpls_route *rt,
 102			     const struct mpls_nh *nh)
 103{
 104	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
 105}
 106
 107static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
 108{
 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);
 111}
 112
 113unsigned int mpls_dev_mtu(const struct net_device *dev)
 114{
 115	/* The amount of data the layer 2 frame can hold */
 116	return dev->mtu;
 117}
 118EXPORT_SYMBOL_GPL(mpls_dev_mtu);
 119
 120bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
 121{
 122	if (skb->len <= mtu)
 123		return false;
 124
 125	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 126		return false;
 127
 128	return true;
 129}
 130EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
 131
 132void mpls_stats_inc_outucastpkts(struct net_device *dev,
 133				 const struct sk_buff *skb)
 134{
 135	struct mpls_dev *mdev;
 136
 137	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
 138		mdev = mpls_dev_get(dev);
 139		if (mdev)
 140			MPLS_INC_STATS_LEN(mdev, skb->len,
 141					   tx_packets,
 142					   tx_bytes);
 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);
 148
 149		if (in6dev)
 150			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
 151					 IPSTATS_MIB_OUT, skb->len);
 152#endif
 153	}
 154}
 155EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
 156
 157static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
 158{
 159	struct mpls_entry_decoded dec;
 160	unsigned int mpls_hdr_len = 0;
 161	struct mpls_shim_hdr *hdr;
 162	bool eli_seen = false;
 163	int label_index;
 164	u32 hash = 0;
 165
 166	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
 167	     label_index++) {
 168		mpls_hdr_len += sizeof(*hdr);
 169		if (!pskb_may_pull(skb, mpls_hdr_len))
 170			break;
 171
 172		/* Read and decode the current label */
 173		hdr = mpls_hdr(skb) + label_index;
 174		dec = mpls_entry_decode(hdr);
 175
 176		/* RFC6790 - reserved labels MUST NOT be used as keys
 177		 * for the load-balancing function
 178		 */
 179		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
 180			hash = jhash_1word(dec.label, hash);
 181
 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
 186			 * payload
 187			 */
 188			if (eli_seen)
 189				break;
 190		} else if (dec.label == MPLS_LABEL_ENTROPY) {
 191			eli_seen = true;
 192		}
 193
 194		if (!dec.bos)
 195			continue;
 196
 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;
 200
 201			v4hdr = (const struct iphdr *)(hdr + 1);
 202			if (v4hdr->version == 4) {
 203				hash = jhash_3words(ntohl(v4hdr->saddr),
 204						    ntohl(v4hdr->daddr),
 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;
 210
 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);
 215			}
 216		}
 217
 218		break;
 219	}
 220
 221	return hash;
 222}
 223
 224static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
 225{
 226	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
 227}
 228
 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
 232 * access both.
 233 */
 234static const struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
 235						   struct sk_buff *skb)
 236{
 
 237	u32 hash = 0;
 238	int nh_index = 0;
 239	int n = 0;
 240	u8 alive;
 241
 242	/* No need to look further into packet if there's only
 243	 * one path
 244	 */
 245	if (rt->rt_nhn == 1)
 246		return rt->rt_nh;
 247
 248	alive = READ_ONCE(rt->rt_nhn_alive);
 249	if (alive == 0)
 250		return NULL;
 251
 252	hash = mpls_multipath_hash(rt, skb);
 253	nh_index = hash % alive;
 254	if (alive == rt->rt_nhn)
 255		goto out;
 256	for_nexthops(rt) {
 257		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
 258
 259		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 260			continue;
 261		if (n == nh_index)
 262			return nh;
 263		n++;
 264	} endfor_nexthops(rt);
 265
 266out:
 267	return mpls_get_nexthop(rt, nh_index);
 268}
 269
 270static bool mpls_egress(struct net *net, struct mpls_route *rt,
 271			struct sk_buff *skb, struct mpls_entry_decoded dec)
 272{
 273	enum mpls_payload_type payload_type;
 274	bool success = false;
 275
 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.
 280	 *
 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
 284	 * long.
 285	 */
 286	if (!pskb_may_pull(skb, 12))
 287		return false;
 288
 289	payload_type = rt->rt_payload_type;
 290	if (payload_type == MPT_UNSPEC)
 291		payload_type = ip_hdr(skb)->version;
 292
 293	switch (payload_type) {
 294	case MPT_IPV4: {
 295		struct iphdr *hdr4 = ip_hdr(skb);
 296		u8 new_ttl;
 297		skb->protocol = htons(ETH_P_IP);
 298
 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.
 302		 */
 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))
 306			new_ttl = dec.ttl;
 307		else
 308			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
 309
 310		csum_replace2(&hdr4->check,
 311			      htons(hdr4->ttl << 8),
 312			      htons(new_ttl << 8));
 313		hdr4->ttl = new_ttl;
 314		success = true;
 315		break;
 316	}
 317	case MPT_IPV6: {
 318		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
 319		skb->protocol = htons(ETH_P_IPV6);
 320
 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.
 324		 */
 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;
 331		success = true;
 332		break;
 333	}
 334	case MPT_UNSPEC:
 335		/* Should have decided which protocol it is by now */
 336		break;
 337	}
 338
 339	return success;
 340}
 341
 342static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
 343			struct packet_type *pt, struct net_device *orig_dev)
 344{
 345	struct net *net = dev_net(dev);
 346	struct mpls_shim_hdr *hdr;
 347	const struct mpls_nh *nh;
 348	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;
 353	unsigned int hh_len;
 354	unsigned int new_header_size;
 355	unsigned int mtu;
 356	int err;
 357
 358	/* Careful this entire function runs inside of an rcu critical section */
 359
 360	mdev = mpls_dev_get(dev);
 361	if (!mdev)
 362		goto drop;
 363
 364	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
 365			   rx_bytes);
 366
 367	if (!mdev->input_enabled) {
 368		MPLS_INC_STATS(mdev, rx_dropped);
 369		goto drop;
 370	}
 371
 372	if (skb->pkt_type != PACKET_HOST)
 373		goto err;
 374
 375	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 376		goto err;
 377
 378	if (!pskb_may_pull(skb, sizeof(*hdr)))
 379		goto err;
 380
 381	skb_dst_drop(skb);
 382
 383	/* Read and decode the label */
 384	hdr = mpls_hdr(skb);
 385	dec = mpls_entry_decode(hdr);
 386
 387	rt = mpls_route_input_rcu(net, dec.label);
 388	if (!rt) {
 389		MPLS_INC_STATS(mdev, rx_noroute);
 390		goto drop;
 391	}
 392
 393	nh = mpls_select_multipath(rt, skb);
 394	if (!nh)
 395		goto err;
 
 
 
 
 
 396
 397	/* Pop the label */
 398	skb_pull(skb, sizeof(*hdr));
 399	skb_reset_network_header(skb);
 400
 401	skb_orphan(skb);
 402
 403	if (skb_warn_if_lro(skb))
 404		goto err;
 405
 406	skb_forward_csum(skb);
 407
 408	/* Verify ttl is valid */
 409	if (dec.ttl <= 1)
 410		goto err;
 411
 412	/* Find the output device */
 413	out_dev = nh->nh_dev;
 414	if (!mpls_output_possible(out_dev))
 415		goto tx_err;
 416
 417	/* Verify the destination can hold the packet */
 418	new_header_size = mpls_nh_header_size(nh);
 419	mtu = mpls_dev_mtu(out_dev);
 420	if (mpls_pkt_too_big(skb, mtu - new_header_size))
 421		goto tx_err;
 422
 423	hh_len = LL_RESERVED_SPACE(out_dev);
 424	if (!out_dev->header_ops)
 425		hh_len = 0;
 426
 427	/* Ensure there is enough space for the headers in the skb */
 428	if (skb_cow(skb, hh_len + new_header_size))
 429		goto tx_err;
 430
 431	skb->dev = out_dev;
 432	skb->protocol = htons(ETH_P_MPLS_UC);
 433
 434	dec.ttl -= 1;
 435	if (unlikely(!new_header_size && dec.bos)) {
 436		/* Penultimate hop popping */
 437		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
 438			goto err;
 439	} else {
 440		bool bos;
 441		int i;
 442		skb_push(skb, new_header_size);
 443		skb_reset_network_header(skb);
 444		/* Push the new labels */
 445		hdr = mpls_hdr(skb);
 446		bos = dec.bos;
 447		for (i = nh->nh_labels - 1; i >= 0; i--) {
 448			hdr[i] = mpls_entry_encode(nh->nh_label[i],
 449						   dec.ttl, 0, bos);
 450			bos = false;
 451		}
 452	}
 453
 454	mpls_stats_inc_outucastpkts(out_dev, skb);
 455
 456	/* If via wasn't specified then send out using device address */
 457	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
 458		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
 459				 out_dev->dev_addr, skb);
 460	else
 461		err = neigh_xmit(nh->nh_via_table, out_dev,
 462				 mpls_nh_via(rt, nh), skb);
 463	if (err)
 464		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
 465				    __func__, err);
 466	return 0;
 467
 468tx_err:
 469	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
 470	if (out_mdev)
 471		MPLS_INC_STATS(out_mdev, tx_errors);
 472	goto drop;
 473err:
 474	MPLS_INC_STATS(mdev, rx_errors);
 475drop:
 476	kfree_skb(skb);
 477	return NET_RX_DROP;
 478}
 479
 480static struct packet_type mpls_packet_type __read_mostly = {
 481	.type = cpu_to_be16(ETH_P_MPLS_UC),
 482	.func = mpls_forward,
 483};
 484
 485static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
 486	[RTA_DST]		= { .type = NLA_U32 },
 487	[RTA_OIF]		= { .type = NLA_U32 },
 488	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
 489};
 490
 491struct mpls_route_config {
 492	u32			rc_protocol;
 493	u32			rc_ifindex;
 494	u8			rc_via_table;
 495	u8			rc_via_alen;
 496	u8			rc_via[MAX_VIA_ALEN];
 497	u32			rc_label;
 498	u8			rc_ttl_propagate;
 499	u8			rc_output_labels;
 500	u32			rc_output_label[MAX_NEW_LABELS];
 501	u32			rc_nlflags;
 502	enum mpls_payload_type	rc_payload_type;
 503	struct nl_info		rc_nlinfo;
 504	struct rtnexthop	*rc_mp;
 505	int			rc_mp_len;
 506};
 507
 508/* all nexthops within a route have the same size based on max
 509 * number of labels and max via length for a hop
 510 */
 511static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
 512{
 513	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
 514	struct mpls_route *rt;
 515	size_t size;
 516
 517	size = sizeof(*rt) + num_nh * nh_size;
 518	if (size > MAX_MPLS_ROUTE_MEM)
 519		return ERR_PTR(-EINVAL);
 520
 521	rt = kzalloc(size, GFP_KERNEL);
 522	if (!rt)
 523		return ERR_PTR(-ENOMEM);
 524
 525	rt->rt_nhn = num_nh;
 526	rt->rt_nhn_alive = num_nh;
 527	rt->rt_nh_size = nh_size;
 528	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
 
 529
 530	return rt;
 531}
 532
 533static void mpls_rt_free(struct mpls_route *rt)
 534{
 535	if (rt)
 536		kfree_rcu(rt, rt_rcu);
 537}
 538
 539static void mpls_notify_route(struct net *net, unsigned index,
 540			      struct mpls_route *old, struct mpls_route *new,
 541			      const struct nl_info *info)
 542{
 543	struct nlmsghdr *nlh = info ? info->nlh : NULL;
 544	unsigned portid = info ? info->portid : 0;
 545	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
 546	struct mpls_route *rt = new ? new : old;
 547	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
 548	/* Ignore reserved labels for now */
 549	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
 550		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
 551}
 552
 553static void mpls_route_update(struct net *net, unsigned index,
 554			      struct mpls_route *new,
 555			      const struct nl_info *info)
 556{
 557	struct mpls_route __rcu **platform_label;
 558	struct mpls_route *rt;
 559
 560	ASSERT_RTNL();
 561
 562	platform_label = rtnl_dereference(net->mpls.platform_label);
 563	rt = rtnl_dereference(platform_label[index]);
 564	rcu_assign_pointer(platform_label[index], new);
 565
 566	mpls_notify_route(net, index, rt, new, info);
 567
 568	/* If we removed a route free it now */
 569	mpls_rt_free(rt);
 570}
 571
 572static unsigned find_free_label(struct net *net)
 573{
 574	struct mpls_route __rcu **platform_label;
 575	size_t platform_labels;
 576	unsigned index;
 577
 578	platform_label = rtnl_dereference(net->mpls.platform_label);
 579	platform_labels = net->mpls.platform_labels;
 580	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
 581	     index++) {
 582		if (!rtnl_dereference(platform_label[index]))
 583			return index;
 584	}
 585	return LABEL_NOT_SPECIFIED;
 586}
 587
 588#if IS_ENABLED(CONFIG_INET)
 589static struct net_device *inet_fib_lookup_dev(struct net *net,
 590					      const void *addr)
 591{
 592	struct net_device *dev;
 593	struct rtable *rt;
 594	struct in_addr daddr;
 595
 596	memcpy(&daddr, addr, sizeof(struct in_addr));
 597	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0, RT_SCOPE_UNIVERSE);
 598	if (IS_ERR(rt))
 599		return ERR_CAST(rt);
 600
 601	dev = rt->dst.dev;
 602	dev_hold(dev);
 603
 604	ip_rt_put(rt);
 605
 606	return dev;
 607}
 608#else
 609static struct net_device *inet_fib_lookup_dev(struct net *net,
 610					      const void *addr)
 611{
 612	return ERR_PTR(-EAFNOSUPPORT);
 613}
 614#endif
 615
 616#if IS_ENABLED(CONFIG_IPV6)
 617static struct net_device *inet6_fib_lookup_dev(struct net *net,
 618					       const void *addr)
 619{
 620	struct net_device *dev;
 621	struct dst_entry *dst;
 622	struct flowi6 fl6;
 
 623
 624	if (!ipv6_stub)
 625		return ERR_PTR(-EAFNOSUPPORT);
 626
 627	memset(&fl6, 0, sizeof(fl6));
 628	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
 629	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
 630	if (IS_ERR(dst))
 631		return ERR_CAST(dst);
 632
 633	dev = dst->dev;
 634	dev_hold(dev);
 635	dst_release(dst);
 636
 637	return dev;
 638}
 639#else
 640static struct net_device *inet6_fib_lookup_dev(struct net *net,
 641					       const void *addr)
 642{
 643	return ERR_PTR(-EAFNOSUPPORT);
 644}
 645#endif
 646
 647static struct net_device *find_outdev(struct net *net,
 648				      struct mpls_route *rt,
 649				      struct mpls_nh *nh, int oif)
 650{
 651	struct net_device *dev = NULL;
 652
 653	if (!oif) {
 654		switch (nh->nh_via_table) {
 655		case NEIGH_ARP_TABLE:
 656			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
 657			break;
 658		case NEIGH_ND_TABLE:
 659			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
 660			break;
 661		case NEIGH_LINK_TABLE:
 662			break;
 663		}
 664	} else {
 665		dev = dev_get_by_index(net, oif);
 666	}
 667
 668	if (!dev)
 669		return ERR_PTR(-ENODEV);
 670
 671	if (IS_ERR(dev))
 672		return dev;
 673
 674	/* The caller is holding rtnl anyways, so release the dev reference */
 675	dev_put(dev);
 676
 677	return dev;
 678}
 679
 680static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
 681			      struct mpls_nh *nh, int oif)
 682{
 683	struct net_device *dev = NULL;
 684	int err = -ENODEV;
 685
 686	dev = find_outdev(net, rt, nh, oif);
 687	if (IS_ERR(dev)) {
 688		err = PTR_ERR(dev);
 689		dev = NULL;
 690		goto errout;
 691	}
 692
 693	/* Ensure this is a supported device */
 694	err = -EINVAL;
 695	if (!mpls_dev_get(dev))
 696		goto errout;
 697
 698	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
 699	    (dev->addr_len != nh->nh_via_alen))
 700		goto errout;
 701
 702	nh->nh_dev = dev;
 703
 704	if (!(dev->flags & IFF_UP)) {
 705		nh->nh_flags |= RTNH_F_DEAD;
 706	} else {
 707		unsigned int flags;
 708
 709		flags = dev_get_flags(dev);
 710		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
 711			nh->nh_flags |= RTNH_F_LINKDOWN;
 712	}
 713
 714	return 0;
 715
 716errout:
 717	return err;
 718}
 719
 720static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
 721		       u8 via_addr[], struct netlink_ext_ack *extack)
 722{
 723	struct rtvia *via = nla_data(nla);
 724	int err = -EINVAL;
 725	int alen;
 726
 727	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 728		NL_SET_ERR_MSG_ATTR(extack, nla,
 729				    "Invalid attribute length for RTA_VIA");
 730		goto errout;
 731	}
 732	alen = nla_len(nla) -
 733			offsetof(struct rtvia, rtvia_addr);
 734	if (alen > MAX_VIA_ALEN) {
 735		NL_SET_ERR_MSG_ATTR(extack, nla,
 736				    "Invalid address length for RTA_VIA");
 737		goto errout;
 738	}
 739
 740	/* Validate the address family */
 741	switch (via->rtvia_family) {
 742	case AF_PACKET:
 743		*via_table = NEIGH_LINK_TABLE;
 744		break;
 745	case AF_INET:
 746		*via_table = NEIGH_ARP_TABLE;
 747		if (alen != 4)
 748			goto errout;
 749		break;
 750	case AF_INET6:
 751		*via_table = NEIGH_ND_TABLE;
 752		if (alen != 16)
 753			goto errout;
 754		break;
 755	default:
 756		/* Unsupported address family */
 757		goto errout;
 758	}
 759
 760	memcpy(via_addr, via->rtvia_addr, alen);
 761	*via_alen = alen;
 762	err = 0;
 763
 764errout:
 765	return err;
 766}
 767
 768static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
 769				  struct mpls_route *rt)
 770{
 771	struct net *net = cfg->rc_nlinfo.nl_net;
 772	struct mpls_nh *nh = rt->rt_nh;
 773	int err;
 774	int i;
 775
 776	if (!nh)
 777		return -ENOMEM;
 778
 
 
 
 
 
 779	nh->nh_labels = cfg->rc_output_labels;
 780	for (i = 0; i < nh->nh_labels; i++)
 781		nh->nh_label[i] = cfg->rc_output_label[i];
 782
 783	nh->nh_via_table = cfg->rc_via_table;
 784	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
 785	nh->nh_via_alen = cfg->rc_via_alen;
 786
 787	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
 788	if (err)
 789		goto errout;
 790
 791	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 792		rt->rt_nhn_alive--;
 793
 794	return 0;
 795
 796errout:
 797	return err;
 798}
 799
 800static int mpls_nh_build(struct net *net, struct mpls_route *rt,
 801			 struct mpls_nh *nh, int oif, struct nlattr *via,
 802			 struct nlattr *newdst, u8 max_labels,
 803			 struct netlink_ext_ack *extack)
 804{
 805	int err = -ENOMEM;
 806
 807	if (!nh)
 808		goto errout;
 809
 810	if (newdst) {
 811		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
 812				     nh->nh_label, extack);
 813		if (err)
 814			goto errout;
 815	}
 816
 817	if (via) {
 818		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
 819				  __mpls_nh_via(rt, nh), extack);
 820		if (err)
 821			goto errout;
 822	} else {
 823		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
 824	}
 825
 826	err = mpls_nh_assign_dev(net, rt, nh, oif);
 827	if (err)
 828		goto errout;
 829
 830	return 0;
 831
 832errout:
 833	return err;
 834}
 835
 836static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
 837			      u8 cfg_via_alen, u8 *max_via_alen,
 838			      u8 *max_labels)
 839{
 
 840	int remaining = len;
 841	u8 nhs = 0;
 
 
 
 
 842
 843	*max_via_alen = 0;
 844	*max_labels = 0;
 845
 846	while (rtnh_ok(rtnh, remaining)) {
 847		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 848		int attrlen;
 849		u8 n_labels = 0;
 850
 851		attrlen = rtnh_attrlen(rtnh);
 852		nla = nla_find(attrs, attrlen, RTA_VIA);
 853		if (nla && nla_len(nla) >=
 854		    offsetof(struct rtvia, rtvia_addr)) {
 855			int via_alen = nla_len(nla) -
 856				offsetof(struct rtvia, rtvia_addr);
 857
 858			if (via_alen <= MAX_VIA_ALEN)
 859				*max_via_alen = max_t(u16, *max_via_alen,
 860						      via_alen);
 861		}
 862
 863		nla = nla_find(attrs, attrlen, RTA_NEWDST);
 864		if (nla &&
 865		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
 866				   NULL, NULL) != 0)
 867			return 0;
 868
 869		*max_labels = max_t(u8, *max_labels, n_labels);
 870
 871		/* number of nexthops is tracked by a u8.
 872		 * Check for overflow.
 873		 */
 874		if (nhs == 255)
 875			return 0;
 876		nhs++;
 877
 878		rtnh = rtnh_next(rtnh, &remaining);
 879	}
 880
 881	/* leftover implies invalid nexthop configuration, discard it */
 882	return remaining > 0 ? 0 : nhs;
 883}
 884
 885static int mpls_nh_build_multi(struct mpls_route_config *cfg,
 886			       struct mpls_route *rt, u8 max_labels,
 887			       struct netlink_ext_ack *extack)
 888{
 889	struct rtnexthop *rtnh = cfg->rc_mp;
 890	struct nlattr *nla_via, *nla_newdst;
 891	int remaining = cfg->rc_mp_len;
 
 892	int err = 0;
 893	u8 nhs = 0;
 894
 895	change_nexthops(rt) {
 896		int attrlen;
 897
 898		nla_via = NULL;
 899		nla_newdst = NULL;
 900
 901		err = -EINVAL;
 902		if (!rtnh_ok(rtnh, remaining))
 903			goto errout;
 904
 905		/* neither weighted multipath nor any flags
 906		 * are supported
 907		 */
 908		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
 909			goto errout;
 910
 911		attrlen = rtnh_attrlen(rtnh);
 912		if (attrlen > 0) {
 913			struct nlattr *attrs = rtnh_attrs(rtnh);
 914
 915			nla_via = nla_find(attrs, attrlen, RTA_VIA);
 916			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
 917		}
 918
 919		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
 920				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
 921				    max_labels, extack);
 922		if (err)
 923			goto errout;
 924
 925		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 926			rt->rt_nhn_alive--;
 927
 928		rtnh = rtnh_next(rtnh, &remaining);
 929		nhs++;
 930	} endfor_nexthops(rt);
 931
 932	rt->rt_nhn = nhs;
 933
 934	return 0;
 935
 936errout:
 937	return err;
 938}
 939
 940static bool mpls_label_ok(struct net *net, unsigned int *index,
 941			  struct netlink_ext_ack *extack)
 942{
 943	bool is_ok = true;
 944
 945	/* Reserved labels may not be set */
 946	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
 947		NL_SET_ERR_MSG(extack,
 948			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
 949		is_ok = false;
 950	}
 951
 952	/* The full 20 bit range may not be supported. */
 953	if (is_ok && *index >= net->mpls.platform_labels) {
 954		NL_SET_ERR_MSG(extack,
 955			       "Label >= configured maximum in platform_labels");
 956		is_ok = false;
 957	}
 958
 959	*index = array_index_nospec(*index, net->mpls.platform_labels);
 960	return is_ok;
 961}
 962
 963static int mpls_route_add(struct mpls_route_config *cfg,
 964			  struct netlink_ext_ack *extack)
 965{
 966	struct mpls_route __rcu **platform_label;
 967	struct net *net = cfg->rc_nlinfo.nl_net;
 968	struct mpls_route *rt, *old;
 969	int err = -EINVAL;
 970	u8 max_via_alen;
 971	unsigned index;
 972	u8 max_labels;
 973	u8 nhs;
 974
 975	index = cfg->rc_label;
 976
 977	/* If a label was not specified during insert pick one */
 978	if ((index == LABEL_NOT_SPECIFIED) &&
 979	    (cfg->rc_nlflags & NLM_F_CREATE)) {
 980		index = find_free_label(net);
 981	}
 982
 983	if (!mpls_label_ok(net, &index, extack))
 
 
 
 
 
 984		goto errout;
 985
 986	/* Append makes no sense with mpls */
 987	err = -EOPNOTSUPP;
 988	if (cfg->rc_nlflags & NLM_F_APPEND) {
 989		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
 990		goto errout;
 991	}
 992
 993	err = -EEXIST;
 994	platform_label = rtnl_dereference(net->mpls.platform_label);
 995	old = rtnl_dereference(platform_label[index]);
 996	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
 997		goto errout;
 998
 999	err = -EEXIST;
1000	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1001		goto errout;
1002
1003	err = -ENOENT;
1004	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1005		goto errout;
1006
1007	err = -EINVAL;
1008	if (cfg->rc_mp) {
1009		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1010					  cfg->rc_via_alen, &max_via_alen,
1011					  &max_labels);
1012	} else {
1013		max_via_alen = cfg->rc_via_alen;
1014		max_labels = cfg->rc_output_labels;
1015		nhs = 1;
1016	}
1017
1018	if (nhs == 0) {
1019		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1020		goto errout;
1021	}
1022
1023	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1024	if (IS_ERR(rt)) {
1025		err = PTR_ERR(rt);
1026		goto errout;
1027	}
1028
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;
1032
1033	if (cfg->rc_mp)
1034		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1035	else
1036		err = mpls_nh_build_from_cfg(cfg, rt);
1037	if (err)
1038		goto freert;
1039
1040	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1041
1042	return 0;
1043
1044freert:
1045	mpls_rt_free(rt);
1046errout:
1047	return err;
1048}
1049
1050static int mpls_route_del(struct mpls_route_config *cfg,
1051			  struct netlink_ext_ack *extack)
1052{
1053	struct net *net = cfg->rc_nlinfo.nl_net;
1054	unsigned index;
1055	int err = -EINVAL;
1056
1057	index = cfg->rc_label;
1058
1059	if (!mpls_label_ok(net, &index, extack))
1060		goto errout;
1061
1062	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1063
1064	err = 0;
1065errout:
1066	return err;
1067}
1068
1069static void mpls_get_stats(struct mpls_dev *mdev,
1070			   struct mpls_link_stats *stats)
1071{
1072	struct mpls_pcpu_stats *p;
1073	int i;
1074
1075	memset(stats, 0, sizeof(*stats));
1076
1077	for_each_possible_cpu(i) {
1078		struct mpls_link_stats local;
1079		unsigned int start;
1080
1081		p = per_cpu_ptr(mdev->stats, i);
1082		do {
1083			start = u64_stats_fetch_begin(&p->syncp);
1084			local = p->stats;
1085		} while (u64_stats_fetch_retry(&p->syncp, start));
1086
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;
1096	}
1097}
1098
1099static int mpls_fill_stats_af(struct sk_buff *skb,
1100			      const struct net_device *dev)
1101{
1102	struct mpls_link_stats *stats;
1103	struct mpls_dev *mdev;
1104	struct nlattr *nla;
1105
1106	mdev = mpls_dev_get(dev);
1107	if (!mdev)
1108		return -ENODATA;
1109
1110	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1111				sizeof(struct mpls_link_stats),
1112				MPLS_STATS_UNSPEC);
1113	if (!nla)
1114		return -EMSGSIZE;
1115
1116	stats = nla_data(nla);
1117	mpls_get_stats(mdev, stats);
1118
1119	return 0;
1120}
1121
1122static size_t mpls_get_stats_af_size(const struct net_device *dev)
1123{
1124	struct mpls_dev *mdev;
1125
1126	mdev = mpls_dev_get(dev);
1127	if (!mdev)
1128		return 0;
1129
1130	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1131}
1132
1133static 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)
1136{
1137	struct nlmsghdr  *nlh;
1138	struct netconfmsg *ncm;
1139	bool all = false;
1140
1141	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1142			flags);
1143	if (!nlh)
1144		return -EMSGSIZE;
1145
1146	if (type == NETCONFA_ALL)
1147		all = true;
1148
1149	ncm = nlmsg_data(nlh);
1150	ncm->ncm_family = AF_MPLS;
1151
1152	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1153		goto nla_put_failure;
1154
1155	if ((all || type == NETCONFA_INPUT) &&
1156	    nla_put_s32(skb, NETCONFA_INPUT,
1157			READ_ONCE(mdev->input_enabled)) < 0)
1158		goto nla_put_failure;
1159
1160	nlmsg_end(skb, nlh);
1161	return 0;
1162
1163nla_put_failure:
1164	nlmsg_cancel(skb, nlh);
1165	return -EMSGSIZE;
1166}
1167
1168static int mpls_netconf_msgsize_devconf(int type)
1169{
1170	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1171			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1172	bool all = false;
1173
1174	if (type == NETCONFA_ALL)
1175		all = true;
1176
1177	if (all || type == NETCONFA_INPUT)
1178		size += nla_total_size(4);
1179
1180	return size;
1181}
1182
1183static void mpls_netconf_notify_devconf(struct net *net, int event,
1184					int type, struct mpls_dev *mdev)
1185{
1186	struct sk_buff *skb;
1187	int err = -ENOBUFS;
1188
1189	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1190	if (!skb)
1191		goto errout;
1192
1193	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1194	if (err < 0) {
1195		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1196		WARN_ON(err == -EMSGSIZE);
1197		kfree_skb(skb);
1198		goto errout;
1199	}
1200
1201	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1202	return;
1203errout:
1204	rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1205}
1206
1207static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1208	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1209};
1210
1211static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1212				      const struct nlmsghdr *nlh,
1213				      struct nlattr **tb,
1214				      struct netlink_ext_ack *extack)
1215{
1216	int i, err;
1217
1218	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1219		NL_SET_ERR_MSG_MOD(extack,
1220				   "Invalid header for netconf get request");
1221		return -EINVAL;
1222	}
1223
1224	if (!netlink_strict_get_check(skb))
1225		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1226					      tb, NETCONFA_MAX,
1227					      devconf_mpls_policy, extack);
1228
1229	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1230					    tb, NETCONFA_MAX,
1231					    devconf_mpls_policy, extack);
1232	if (err)
1233		return err;
1234
1235	for (i = 0; i <= NETCONFA_MAX; i++) {
1236		if (!tb[i])
1237			continue;
1238
1239		switch (i) {
1240		case NETCONFA_IFINDEX:
1241			break;
1242		default:
1243			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1244			return -EINVAL;
1245		}
1246	}
1247
1248	return 0;
1249}
1250
1251static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1252				    struct nlmsghdr *nlh,
1253				    struct netlink_ext_ack *extack)
1254{
1255	struct net *net = sock_net(in_skb->sk);
1256	struct nlattr *tb[NETCONFA_MAX + 1];
1257	struct net_device *dev;
1258	struct mpls_dev *mdev;
1259	struct sk_buff *skb;
1260	int ifindex;
1261	int err;
1262
1263	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1264	if (err < 0)
1265		goto errout;
1266
1267	err = -EINVAL;
1268	if (!tb[NETCONFA_IFINDEX])
1269		goto errout;
1270
1271	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1272	dev = __dev_get_by_index(net, ifindex);
1273	if (!dev)
1274		goto errout;
1275
1276	mdev = mpls_dev_get(dev);
1277	if (!mdev)
1278		goto errout;
1279
1280	err = -ENOBUFS;
1281	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1282	if (!skb)
1283		goto errout;
1284
1285	err = mpls_netconf_fill_devconf(skb, mdev,
1286					NETLINK_CB(in_skb).portid,
1287					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1288					NETCONFA_ALL);
1289	if (err < 0) {
1290		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1291		WARN_ON(err == -EMSGSIZE);
1292		kfree_skb(skb);
1293		goto errout;
1294	}
1295	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1296errout:
1297	return err;
1298}
1299
1300static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1301				     struct netlink_callback *cb)
1302{
1303	const struct nlmsghdr *nlh = cb->nlh;
1304	struct net *net = sock_net(skb->sk);
1305	struct {
1306		unsigned long ifindex;
1307	} *ctx = (void *)cb->ctx;
1308	struct net_device *dev;
1309	struct mpls_dev *mdev;
1310	int err = 0;
1311
1312	if (cb->strict_check) {
1313		struct netlink_ext_ack *extack = cb->extack;
1314		struct netconfmsg *ncm;
1315
1316		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1317			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1318			return -EINVAL;
1319		}
1320
1321		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1322			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1323			return -EINVAL;
1324		}
1325	}
1326
1327	rcu_read_lock();
1328	for_each_netdev_dump(net, dev, ctx->ifindex) {
1329		mdev = mpls_dev_get(dev);
1330		if (!mdev)
1331			continue;
1332		err = mpls_netconf_fill_devconf(skb, mdev,
1333						NETLINK_CB(cb->skb).portid,
1334						nlh->nlmsg_seq,
1335						RTM_NEWNETCONF,
1336						NLM_F_MULTI,
1337						NETCONFA_ALL);
1338		if (err < 0)
1339			break;
1340	}
1341	rcu_read_unlock();
1342
1343	return err;
1344}
1345
1346#define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1347	(&((struct mpls_dev *)0)->field)
1348
1349static int mpls_conf_proc(const struct ctl_table *ctl, int write,
1350			  void *buffer, size_t *lenp, loff_t *ppos)
1351{
1352	int oval = *(int *)ctl->data;
1353	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1354
1355	if (write) {
1356		struct mpls_dev *mdev = ctl->extra1;
1357		int i = (int *)ctl->data - (int *)mdev;
1358		struct net *net = ctl->extra2;
1359		int val = *(int *)ctl->data;
1360
1361		if (i == offsetof(struct mpls_dev, input_enabled) &&
1362		    val != oval) {
1363			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1364						    NETCONFA_INPUT, mdev);
1365		}
1366	}
1367
1368	return ret;
1369}
1370
1371static const struct ctl_table mpls_dev_table[] = {
1372	{
1373		.procname	= "input",
1374		.maxlen		= sizeof(int),
1375		.mode		= 0644,
1376		.proc_handler	= mpls_conf_proc,
1377		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1378	},
 
1379};
1380
1381static int mpls_dev_sysctl_register(struct net_device *dev,
1382				    struct mpls_dev *mdev)
1383{
1384	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1385	size_t table_size = ARRAY_SIZE(mpls_dev_table);
1386	struct net *net = dev_net(dev);
1387	struct ctl_table *table;
1388	int i;
1389
1390	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1391	if (!table)
1392		goto out;
1393
1394	/* Table data contains only offsets relative to the base of
1395	 * the mdev at this point, so make them absolute.
1396	 */
1397	for (i = 0; i < table_size; i++) {
1398		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1399		table[i].extra1 = mdev;
1400		table[i].extra2 = net;
1401	}
1402
1403	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1404
1405	mdev->sysctl = register_net_sysctl_sz(net, path, table, table_size);
1406	if (!mdev->sysctl)
1407		goto free;
1408
1409	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1410	return 0;
1411
1412free:
1413	kfree(table);
1414out:
1415	mdev->sysctl = NULL;
1416	return -ENOBUFS;
1417}
1418
1419static void mpls_dev_sysctl_unregister(struct net_device *dev,
1420				       struct mpls_dev *mdev)
1421{
1422	struct net *net = dev_net(dev);
1423	const struct ctl_table *table;
1424
1425	if (!mdev->sysctl)
1426		return;
1427
1428	table = mdev->sysctl->ctl_table_arg;
1429	unregister_net_sysctl_table(mdev->sysctl);
1430	kfree(table);
1431
1432	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1433}
1434
1435static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1436{
1437	struct mpls_dev *mdev;
1438	int err = -ENOMEM;
1439	int i;
1440
1441	ASSERT_RTNL();
1442
1443	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1444	if (!mdev)
1445		return ERR_PTR(err);
1446
1447	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1448	if (!mdev->stats)
1449		goto free;
1450
1451	for_each_possible_cpu(i) {
1452		struct mpls_pcpu_stats *mpls_stats;
1453
1454		mpls_stats = per_cpu_ptr(mdev->stats, i);
1455		u64_stats_init(&mpls_stats->syncp);
1456	}
1457
1458	mdev->dev = dev;
1459
1460	err = mpls_dev_sysctl_register(dev, mdev);
1461	if (err)
1462		goto free;
1463
1464	rcu_assign_pointer(dev->mpls_ptr, mdev);
1465
1466	return mdev;
1467
1468free:
1469	free_percpu(mdev->stats);
1470	kfree(mdev);
1471	return ERR_PTR(err);
1472}
1473
1474static void mpls_dev_destroy_rcu(struct rcu_head *head)
1475{
1476	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1477
1478	free_percpu(mdev->stats);
1479	kfree(mdev);
1480}
1481
1482static int mpls_ifdown(struct net_device *dev, int event)
1483{
1484	struct mpls_route __rcu **platform_label;
1485	struct net *net = dev_net(dev);
1486	unsigned index;
1487
1488	platform_label = rtnl_dereference(net->mpls.platform_label);
1489	for (index = 0; index < net->mpls.platform_labels; index++) {
1490		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1491		bool nh_del = false;
1492		u8 alive = 0;
1493
1494		if (!rt)
1495			continue;
1496
1497		if (event == NETDEV_UNREGISTER) {
1498			u8 deleted = 0;
1499
1500			for_nexthops(rt) {
1501				if (!nh->nh_dev || nh->nh_dev == dev)
1502					deleted++;
1503				if (nh->nh_dev == dev)
1504					nh_del = true;
1505			} endfor_nexthops(rt);
1506
1507			/* if there are no more nexthops, delete the route */
1508			if (deleted == rt->rt_nhn) {
1509				mpls_route_update(net, index, NULL, NULL);
1510				continue;
1511			}
1512
1513			if (nh_del) {
1514				size_t size = sizeof(*rt) + rt->rt_nhn *
1515					rt->rt_nh_size;
1516				struct mpls_route *orig = rt;
1517
1518				rt = kmemdup(orig, size, GFP_KERNEL);
1519				if (!rt)
1520					return -ENOMEM;
1521			}
1522		}
1523
1524		change_nexthops(rt) {
1525			unsigned int nh_flags = nh->nh_flags;
1526
1527			if (nh->nh_dev != dev)
1528				goto next;
1529
1530			switch (event) {
1531			case NETDEV_DOWN:
1532			case NETDEV_UNREGISTER:
1533				nh_flags |= RTNH_F_DEAD;
1534				fallthrough;
1535			case NETDEV_CHANGE:
1536				nh_flags |= RTNH_F_LINKDOWN;
 
 
1537				break;
1538			}
1539			if (event == NETDEV_UNREGISTER)
1540				nh->nh_dev = NULL;
1541
1542			if (nh->nh_flags != nh_flags)
1543				WRITE_ONCE(nh->nh_flags, nh_flags);
1544next:
1545			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1546				alive++;
1547		} endfor_nexthops(rt);
1548
1549		WRITE_ONCE(rt->rt_nhn_alive, alive);
1550
1551		if (nh_del)
1552			mpls_route_update(net, index, rt, NULL);
1553	}
1554
1555	return 0;
1556}
1557
1558static void mpls_ifup(struct net_device *dev, unsigned int flags)
1559{
1560	struct mpls_route __rcu **platform_label;
1561	struct net *net = dev_net(dev);
1562	unsigned index;
1563	u8 alive;
1564
1565	platform_label = rtnl_dereference(net->mpls.platform_label);
1566	for (index = 0; index < net->mpls.platform_labels; index++) {
1567		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1568
1569		if (!rt)
1570			continue;
1571
1572		alive = 0;
1573		change_nexthops(rt) {
1574			unsigned int nh_flags = nh->nh_flags;
 
1575
1576			if (!(nh_flags & flags)) {
1577				alive++;
1578				continue;
1579			}
1580			if (nh->nh_dev != dev)
1581				continue;
1582			alive++;
1583			nh_flags &= ~flags;
1584			WRITE_ONCE(nh->nh_flags, nh_flags);
1585		} endfor_nexthops(rt);
1586
1587		WRITE_ONCE(rt->rt_nhn_alive, alive);
1588	}
1589}
1590
1591static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1592			   void *ptr)
1593{
1594	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1595	struct mpls_dev *mdev;
1596	unsigned int flags;
1597	int err;
1598
1599	if (event == NETDEV_REGISTER) {
1600		mdev = mpls_add_dev(dev);
1601		if (IS_ERR(mdev))
1602			return notifier_from_errno(PTR_ERR(mdev));
1603
 
 
 
 
 
 
1604		return NOTIFY_OK;
1605	}
1606
1607	mdev = mpls_dev_get(dev);
1608	if (!mdev)
1609		return NOTIFY_OK;
1610
1611	switch (event) {
1612
1613	case NETDEV_DOWN:
1614		err = mpls_ifdown(dev, event);
1615		if (err)
1616			return notifier_from_errno(err);
1617		break;
1618	case NETDEV_UP:
1619		flags = dev_get_flags(dev);
1620		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1621			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1622		else
1623			mpls_ifup(dev, RTNH_F_DEAD);
1624		break;
1625	case NETDEV_CHANGE:
1626		flags = dev_get_flags(dev);
1627		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1628			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1629		} else {
1630			err = mpls_ifdown(dev, event);
1631			if (err)
1632				return notifier_from_errno(err);
1633		}
1634		break;
1635	case NETDEV_UNREGISTER:
1636		err = mpls_ifdown(dev, event);
1637		if (err)
1638			return notifier_from_errno(err);
1639		mdev = mpls_dev_get(dev);
1640		if (mdev) {
1641			mpls_dev_sysctl_unregister(dev, mdev);
1642			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1643			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1644		}
1645		break;
1646	case NETDEV_CHANGENAME:
1647		mdev = mpls_dev_get(dev);
1648		if (mdev) {
1649			mpls_dev_sysctl_unregister(dev, mdev);
 
 
1650			err = mpls_dev_sysctl_register(dev, mdev);
1651			if (err)
1652				return notifier_from_errno(err);
1653		}
1654		break;
1655	}
1656	return NOTIFY_OK;
1657}
1658
1659static struct notifier_block mpls_dev_notifier = {
1660	.notifier_call = mpls_dev_notify,
1661};
1662
1663static int nla_put_via(struct sk_buff *skb,
1664		       u8 table, const void *addr, int alen)
1665{
1666	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1667		AF_INET, AF_INET6, AF_PACKET,
1668	};
1669	struct nlattr *nla;
1670	struct rtvia *via;
1671	int family = AF_UNSPEC;
1672
1673	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1674	if (!nla)
1675		return -EMSGSIZE;
1676
1677	if (table <= NEIGH_NR_TABLES)
1678		family = table_to_family[table];
1679
1680	via = nla_data(nla);
1681	via->rtvia_family = family;
1682	memcpy(via->rtvia_addr, addr, alen);
1683	return 0;
1684}
1685
1686int nla_put_labels(struct sk_buff *skb, int attrtype,
1687		   u8 labels, const u32 label[])
1688{
1689	struct nlattr *nla;
1690	struct mpls_shim_hdr *nla_label;
1691	bool bos;
1692	int i;
1693	nla = nla_reserve(skb, attrtype, labels*4);
1694	if (!nla)
1695		return -EMSGSIZE;
1696
1697	nla_label = nla_data(nla);
1698	bos = true;
1699	for (i = labels - 1; i >= 0; i--) {
1700		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1701		bos = false;
1702	}
1703
1704	return 0;
1705}
1706EXPORT_SYMBOL_GPL(nla_put_labels);
1707
1708int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1709		   u32 label[], struct netlink_ext_ack *extack)
1710{
1711	unsigned len = nla_len(nla);
 
1712	struct mpls_shim_hdr *nla_label;
1713	u8 nla_labels;
1714	bool bos;
1715	int i;
1716
1717	/* len needs to be an even multiple of 4 (the label size). Number
1718	 * of labels is a u8 so check for overflow.
1719	 */
1720	if (len & 3 || len / 4 > 255) {
1721		NL_SET_ERR_MSG_ATTR(extack, nla,
1722				    "Invalid length for labels attribute");
1723		return -EINVAL;
1724	}
1725
1726	/* Limit the number of new labels allowed */
1727	nla_labels = len/4;
1728	if (nla_labels > max_labels) {
1729		NL_SET_ERR_MSG(extack, "Too many labels");
1730		return -EINVAL;
1731	}
1732
1733	/* when label == NULL, caller wants number of labels */
1734	if (!label)
1735		goto out;
1736
1737	nla_label = nla_data(nla);
1738	bos = true;
1739	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1740		struct mpls_entry_decoded dec;
1741		dec = mpls_entry_decode(nla_label + i);
1742
1743		/* Ensure the bottom of stack flag is properly set
1744		 * and ttl and tc are both clear.
1745		 */
1746		if (dec.ttl) {
1747			NL_SET_ERR_MSG_ATTR(extack, nla,
1748					    "TTL in label must be 0");
1749			return -EINVAL;
1750		}
1751
1752		if (dec.tc) {
1753			NL_SET_ERR_MSG_ATTR(extack, nla,
1754					    "Traffic class in label must be 0");
1755			return -EINVAL;
1756		}
1757
1758		if (dec.bos != bos) {
1759			NL_SET_BAD_ATTR(extack, nla);
1760			if (bos) {
1761				NL_SET_ERR_MSG(extack,
1762					       "BOS bit must be set in first label");
1763			} else {
1764				NL_SET_ERR_MSG(extack,
1765					       "BOS bit can only be set in first label");
1766			}
1767			return -EINVAL;
1768		}
1769
1770		switch (dec.label) {
1771		case MPLS_LABEL_IMPLNULL:
1772			/* RFC3032: This is a label that an LSR may
1773			 * assign and distribute, but which never
1774			 * actually appears in the encapsulation.
1775			 */
1776			NL_SET_ERR_MSG_ATTR(extack, nla,
1777					    "Implicit NULL Label (3) can not be used in encapsulation");
1778			return -EINVAL;
1779		}
1780
1781		label[i] = dec.label;
1782	}
1783out:
1784	*labels = nla_labels;
1785	return 0;
1786}
1787EXPORT_SYMBOL_GPL(nla_get_labels);
1788
1789static int rtm_to_route_config(struct sk_buff *skb,
1790			       struct nlmsghdr *nlh,
1791			       struct mpls_route_config *cfg,
1792			       struct netlink_ext_ack *extack)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1793{
1794	struct rtmsg *rtm;
1795	struct nlattr *tb[RTA_MAX+1];
1796	int index;
1797	int err;
1798
1799	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1800				     rtm_mpls_policy, extack);
1801	if (err < 0)
1802		goto errout;
1803
1804	err = -EINVAL;
1805	rtm = nlmsg_data(nlh);
 
1806
1807	if (rtm->rtm_family != AF_MPLS) {
1808		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1809		goto errout;
1810	}
1811	if (rtm->rtm_dst_len != 20) {
1812		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1813		goto errout;
1814	}
1815	if (rtm->rtm_src_len != 0) {
1816		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1817		goto errout;
1818	}
1819	if (rtm->rtm_tos != 0) {
1820		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1821		goto errout;
1822	}
1823	if (rtm->rtm_table != RT_TABLE_MAIN) {
1824		NL_SET_ERR_MSG(extack,
1825			       "MPLS only supports the main route table");
1826		goto errout;
1827	}
1828	/* Any value is acceptable for rtm_protocol */
1829
1830	/* As mpls uses destination specific addresses
1831	 * (or source specific address in the case of multicast)
1832	 * all addresses have universal scope.
1833	 */
1834	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1835		NL_SET_ERR_MSG(extack,
1836			       "Invalid route scope  - MPLS only supports UNIVERSE");
1837		goto errout;
1838	}
1839	if (rtm->rtm_type != RTN_UNICAST) {
1840		NL_SET_ERR_MSG(extack,
1841			       "Invalid route type - MPLS only supports UNICAST");
1842		goto errout;
1843	}
1844	if (rtm->rtm_flags != 0) {
1845		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1846		goto errout;
1847	}
1848
1849	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1850	cfg->rc_protocol	= rtm->rtm_protocol;
1851	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1852	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1853	cfg->rc_nlflags		= nlh->nlmsg_flags;
1854	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1855	cfg->rc_nlinfo.nlh	= nlh;
1856	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1857
1858	for (index = 0; index <= RTA_MAX; index++) {
1859		struct nlattr *nla = tb[index];
1860		if (!nla)
1861			continue;
1862
1863		switch (index) {
1864		case RTA_OIF:
1865			cfg->rc_ifindex = nla_get_u32(nla);
1866			break;
1867		case RTA_NEWDST:
1868			if (nla_get_labels(nla, MAX_NEW_LABELS,
1869					   &cfg->rc_output_labels,
1870					   cfg->rc_output_label, extack))
1871				goto errout;
1872			break;
1873		case RTA_DST:
1874		{
1875			u8 label_count;
1876			if (nla_get_labels(nla, 1, &label_count,
1877					   &cfg->rc_label, extack))
1878				goto errout;
1879
1880			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1881					   &cfg->rc_label, extack))
1882				goto errout;
 
1883			break;
1884		}
1885		case RTA_GATEWAY:
1886			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1887			goto errout;
1888		case RTA_VIA:
1889		{
1890			if (nla_get_via(nla, &cfg->rc_via_alen,
1891					&cfg->rc_via_table, cfg->rc_via,
1892					extack))
1893				goto errout;
1894			break;
1895		}
1896		case RTA_MULTIPATH:
1897		{
1898			cfg->rc_mp = nla_data(nla);
1899			cfg->rc_mp_len = nla_len(nla);
1900			break;
1901		}
1902		case RTA_TTL_PROPAGATE:
1903		{
1904			u8 ttl_propagate = nla_get_u8(nla);
1905
1906			if (ttl_propagate > 1) {
1907				NL_SET_ERR_MSG_ATTR(extack, nla,
1908						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1909				goto errout;
1910			}
1911			cfg->rc_ttl_propagate = ttl_propagate ?
1912				MPLS_TTL_PROP_ENABLED :
1913				MPLS_TTL_PROP_DISABLED;
1914			break;
1915		}
1916		default:
1917			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1918			/* Unsupported attribute */
1919			goto errout;
1920		}
1921	}
1922
1923	err = 0;
1924errout:
1925	return err;
1926}
1927
1928static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1929			     struct netlink_ext_ack *extack)
1930{
1931	struct mpls_route_config *cfg;
1932	int err;
1933
1934	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1935	if (!cfg)
1936		return -ENOMEM;
1937
1938	err = rtm_to_route_config(skb, nlh, cfg, extack);
1939	if (err < 0)
1940		goto out;
1941
1942	err = mpls_route_del(cfg, extack);
1943out:
1944	kfree(cfg);
1945
1946	return err;
1947}
1948
1949
1950static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1951			     struct netlink_ext_ack *extack)
1952{
1953	struct mpls_route_config *cfg;
1954	int err;
1955
1956	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1957	if (!cfg)
1958		return -ENOMEM;
1959
1960	err = rtm_to_route_config(skb, nlh, cfg, extack);
1961	if (err < 0)
1962		goto out;
1963
1964	err = mpls_route_add(cfg, extack);
1965out:
1966	kfree(cfg);
1967
1968	return err;
1969}
1970
1971static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1972			   u32 label, struct mpls_route *rt, int flags)
1973{
1974	struct net_device *dev;
1975	struct nlmsghdr *nlh;
1976	struct rtmsg *rtm;
1977
1978	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1979	if (nlh == NULL)
1980		return -EMSGSIZE;
1981
1982	rtm = nlmsg_data(nlh);
1983	rtm->rtm_family = AF_MPLS;
1984	rtm->rtm_dst_len = 20;
1985	rtm->rtm_src_len = 0;
1986	rtm->rtm_tos = 0;
1987	rtm->rtm_table = RT_TABLE_MAIN;
1988	rtm->rtm_protocol = rt->rt_protocol;
1989	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1990	rtm->rtm_type = RTN_UNICAST;
1991	rtm->rtm_flags = 0;
1992
1993	if (nla_put_labels(skb, RTA_DST, 1, &label))
1994		goto nla_put_failure;
1995
1996	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1997		bool ttl_propagate =
1998			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1999
2000		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2001			       ttl_propagate))
2002			goto nla_put_failure;
2003	}
2004	if (rt->rt_nhn == 1) {
2005		const struct mpls_nh *nh = rt->rt_nh;
2006
2007		if (nh->nh_labels &&
2008		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2009				   nh->nh_label))
2010			goto nla_put_failure;
2011		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2012		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2013				nh->nh_via_alen))
2014			goto nla_put_failure;
2015		dev = nh->nh_dev;
2016		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2017			goto nla_put_failure;
2018		if (nh->nh_flags & RTNH_F_LINKDOWN)
2019			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2020		if (nh->nh_flags & RTNH_F_DEAD)
2021			rtm->rtm_flags |= RTNH_F_DEAD;
2022	} else {
2023		struct rtnexthop *rtnh;
2024		struct nlattr *mp;
2025		u8 linkdown = 0;
2026		u8 dead = 0;
2027
2028		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2029		if (!mp)
2030			goto nla_put_failure;
2031
2032		for_nexthops(rt) {
2033			dev = nh->nh_dev;
2034			if (!dev)
2035				continue;
2036
2037			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2038			if (!rtnh)
2039				goto nla_put_failure;
2040
2041			rtnh->rtnh_ifindex = dev->ifindex;
 
 
2042			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2043				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2044				linkdown++;
2045			}
2046			if (nh->nh_flags & RTNH_F_DEAD) {
2047				rtnh->rtnh_flags |= RTNH_F_DEAD;
2048				dead++;
2049			}
2050
2051			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2052							    nh->nh_labels,
2053							    nh->nh_label))
2054				goto nla_put_failure;
2055			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2056			    nla_put_via(skb, nh->nh_via_table,
2057					mpls_nh_via(rt, nh),
2058					nh->nh_via_alen))
2059				goto nla_put_failure;
2060
2061			/* length of rtnetlink header + attributes */
2062			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2063		} endfor_nexthops(rt);
2064
2065		if (linkdown == rt->rt_nhn)
2066			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2067		if (dead == rt->rt_nhn)
2068			rtm->rtm_flags |= RTNH_F_DEAD;
2069
2070		nla_nest_end(skb, mp);
2071	}
2072
2073	nlmsg_end(skb, nlh);
2074	return 0;
2075
2076nla_put_failure:
2077	nlmsg_cancel(skb, nlh);
2078	return -EMSGSIZE;
2079}
2080
2081#if IS_ENABLED(CONFIG_INET)
2082static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2083				   struct fib_dump_filter *filter,
2084				   struct netlink_callback *cb)
2085{
2086	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2087}
2088#else
2089static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2090				   struct fib_dump_filter *filter,
2091				   struct netlink_callback *cb)
2092{
2093	struct netlink_ext_ack *extack = cb->extack;
2094	struct nlattr *tb[RTA_MAX + 1];
2095	struct rtmsg *rtm;
2096	int err, i;
2097
2098	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2099		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2100		return -EINVAL;
2101	}
2102
2103	rtm = nlmsg_data(nlh);
2104	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2105	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2106	    rtm->rtm_flags) {
2107		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2108		return -EINVAL;
2109	}
2110
2111	if (rtm->rtm_protocol) {
2112		filter->protocol = rtm->rtm_protocol;
2113		filter->filter_set = 1;
2114		cb->answer_flags = NLM_F_DUMP_FILTERED;
2115	}
2116
2117	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2118					    rtm_mpls_policy, extack);
2119	if (err < 0)
2120		return err;
2121
2122	for (i = 0; i <= RTA_MAX; ++i) {
2123		int ifindex;
2124
2125		if (i == RTA_OIF) {
2126			ifindex = nla_get_u32(tb[i]);
2127			filter->dev = __dev_get_by_index(net, ifindex);
2128			if (!filter->dev)
2129				return -ENODEV;
2130			filter->filter_set = 1;
2131		} else if (tb[i]) {
2132			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2133			return -EINVAL;
2134		}
2135	}
2136
2137	return 0;
2138}
2139#endif
2140
2141static bool mpls_rt_uses_dev(struct mpls_route *rt,
2142			     const struct net_device *dev)
2143{
2144	if (rt->rt_nhn == 1) {
2145		struct mpls_nh *nh = rt->rt_nh;
2146
2147		if (nh->nh_dev == dev)
2148			return true;
2149	} else {
2150		for_nexthops(rt) {
2151			if (nh->nh_dev == dev)
2152				return true;
2153		} endfor_nexthops(rt);
2154	}
2155
2156	return false;
2157}
2158
2159static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2160{
2161	const struct nlmsghdr *nlh = cb->nlh;
2162	struct net *net = sock_net(skb->sk);
2163	struct mpls_route __rcu **platform_label;
2164	struct fib_dump_filter filter = {
2165		.rtnl_held = true,
2166	};
2167	unsigned int flags = NLM_F_MULTI;
2168	size_t platform_labels;
2169	unsigned int index;
2170
2171	ASSERT_RTNL();
2172
2173	if (cb->strict_check) {
2174		int err;
2175
2176		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2177		if (err < 0)
2178			return err;
2179
2180		/* for MPLS, there is only 1 table with fixed type and flags.
2181		 * If either are set in the filter then return nothing.
2182		 */
2183		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2184		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2185		     filter.flags)
2186			return skb->len;
2187	}
2188
2189	index = cb->args[0];
2190	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2191		index = MPLS_LABEL_FIRST_UNRESERVED;
2192
2193	platform_label = rtnl_dereference(net->mpls.platform_label);
2194	platform_labels = net->mpls.platform_labels;
2195
2196	if (filter.filter_set)
2197		flags |= NLM_F_DUMP_FILTERED;
2198
2199	for (; index < platform_labels; index++) {
2200		struct mpls_route *rt;
2201
2202		rt = rtnl_dereference(platform_label[index]);
2203		if (!rt)
2204			continue;
2205
2206		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2207		    (filter.protocol && rt->rt_protocol != filter.protocol))
2208			continue;
2209
2210		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2211				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2212				    index, rt, flags) < 0)
2213			break;
2214	}
2215	cb->args[0] = index;
2216
2217	return skb->len;
2218}
2219
2220static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2221{
2222	size_t payload =
2223		NLMSG_ALIGN(sizeof(struct rtmsg))
2224		+ nla_total_size(4)			/* RTA_DST */
2225		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2226
2227	if (rt->rt_nhn == 1) {
2228		struct mpls_nh *nh = rt->rt_nh;
2229
2230		if (nh->nh_dev)
2231			payload += nla_total_size(4); /* RTA_OIF */
2232		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2233			payload += nla_total_size(2 + nh->nh_via_alen);
2234		if (nh->nh_labels) /* RTA_NEWDST */
2235			payload += nla_total_size(nh->nh_labels * 4);
2236	} else {
2237		/* each nexthop is packed in an attribute */
2238		size_t nhsize = 0;
2239
2240		for_nexthops(rt) {
2241			if (!nh->nh_dev)
2242				continue;
2243			nhsize += nla_total_size(sizeof(struct rtnexthop));
2244			/* RTA_VIA */
2245			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2246				nhsize += nla_total_size(2 + nh->nh_via_alen);
2247			if (nh->nh_labels)
2248				nhsize += nla_total_size(nh->nh_labels * 4);
2249		} endfor_nexthops(rt);
2250		/* nested attribute */
2251		payload += nla_total_size(nhsize);
2252	}
2253
2254	return payload;
2255}
2256
2257static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2258		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2259		       unsigned int nlm_flags)
2260{
2261	struct sk_buff *skb;
2262	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2263	int err = -ENOBUFS;
2264
2265	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2266	if (skb == NULL)
2267		goto errout;
2268
2269	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2270	if (err < 0) {
2271		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2272		WARN_ON(err == -EMSGSIZE);
2273		kfree_skb(skb);
2274		goto errout;
2275	}
2276	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2277
2278	return;
2279errout:
2280	rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2281}
2282
2283static int mpls_valid_getroute_req(struct sk_buff *skb,
2284				   const struct nlmsghdr *nlh,
2285				   struct nlattr **tb,
2286				   struct netlink_ext_ack *extack)
2287{
2288	struct rtmsg *rtm;
2289	int i, err;
2290
2291	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2292		NL_SET_ERR_MSG_MOD(extack,
2293				   "Invalid header for get route request");
2294		return -EINVAL;
2295	}
2296
2297	if (!netlink_strict_get_check(skb))
2298		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2299					      rtm_mpls_policy, extack);
2300
2301	rtm = nlmsg_data(nlh);
2302	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2303	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2304	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2305		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2306		return -EINVAL;
2307	}
2308	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2309		NL_SET_ERR_MSG_MOD(extack,
2310				   "Invalid flags for get route request");
2311		return -EINVAL;
2312	}
2313
2314	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2315					    rtm_mpls_policy, extack);
2316	if (err)
2317		return err;
2318
2319	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2320		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2321		return -EINVAL;
2322	}
2323
2324	for (i = 0; i <= RTA_MAX; i++) {
2325		if (!tb[i])
2326			continue;
2327
2328		switch (i) {
2329		case RTA_DST:
2330		case RTA_NEWDST:
2331			break;
2332		default:
2333			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2334			return -EINVAL;
2335		}
2336	}
2337
2338	return 0;
2339}
2340
2341static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2342			 struct netlink_ext_ack *extack)
2343{
2344	struct net *net = sock_net(in_skb->sk);
2345	u32 portid = NETLINK_CB(in_skb).portid;
2346	u32 in_label = LABEL_NOT_SPECIFIED;
2347	struct nlattr *tb[RTA_MAX + 1];
2348	u32 labels[MAX_NEW_LABELS];
2349	struct mpls_shim_hdr *hdr;
2350	unsigned int hdr_size = 0;
2351	const struct mpls_nh *nh;
2352	struct net_device *dev;
2353	struct mpls_route *rt;
2354	struct rtmsg *rtm, *r;
2355	struct nlmsghdr *nlh;
2356	struct sk_buff *skb;
2357	u8 n_labels;
2358	int err;
2359
2360	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2361	if (err < 0)
2362		goto errout;
2363
2364	rtm = nlmsg_data(in_nlh);
2365
2366	if (tb[RTA_DST]) {
2367		u8 label_count;
2368
2369		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2370				   &in_label, extack)) {
2371			err = -EINVAL;
2372			goto errout;
2373		}
2374
2375		if (!mpls_label_ok(net, &in_label, extack)) {
2376			err = -EINVAL;
2377			goto errout;
2378		}
2379	}
2380
2381	rt = mpls_route_input_rcu(net, in_label);
2382	if (!rt) {
2383		err = -ENETUNREACH;
2384		goto errout;
2385	}
2386
2387	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2388		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2389		if (!skb) {
2390			err = -ENOBUFS;
2391			goto errout;
2392		}
2393
2394		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2395				      RTM_NEWROUTE, in_label, rt, 0);
2396		if (err < 0) {
2397			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2398			WARN_ON(err == -EMSGSIZE);
2399			goto errout_free;
2400		}
2401
2402		return rtnl_unicast(skb, net, portid);
2403	}
2404
2405	if (tb[RTA_NEWDST]) {
2406		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2407				   labels, extack) != 0) {
2408			err = -EINVAL;
2409			goto errout;
2410		}
2411
2412		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2413	}
2414
2415	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2416	if (!skb) {
2417		err = -ENOBUFS;
2418		goto errout;
2419	}
2420
2421	skb->protocol = htons(ETH_P_MPLS_UC);
2422
2423	if (hdr_size) {
2424		bool bos;
2425		int i;
2426
2427		if (skb_cow(skb, hdr_size)) {
2428			err = -ENOBUFS;
2429			goto errout_free;
2430		}
2431
2432		skb_reserve(skb, hdr_size);
2433		skb_push(skb, hdr_size);
2434		skb_reset_network_header(skb);
2435
2436		/* Push new labels */
2437		hdr = mpls_hdr(skb);
2438		bos = true;
2439		for (i = n_labels - 1; i >= 0; i--) {
2440			hdr[i] = mpls_entry_encode(labels[i],
2441						   1, 0, bos);
2442			bos = false;
2443		}
2444	}
2445
2446	nh = mpls_select_multipath(rt, skb);
2447	if (!nh) {
2448		err = -ENETUNREACH;
2449		goto errout_free;
2450	}
2451
2452	if (hdr_size) {
2453		skb_pull(skb, hdr_size);
2454		skb_reset_network_header(skb);
2455	}
2456
2457	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2458			RTM_NEWROUTE, sizeof(*r), 0);
2459	if (!nlh) {
2460		err = -EMSGSIZE;
2461		goto errout_free;
2462	}
2463
2464	r = nlmsg_data(nlh);
2465	r->rtm_family	 = AF_MPLS;
2466	r->rtm_dst_len	= 20;
2467	r->rtm_src_len	= 0;
2468	r->rtm_table	= RT_TABLE_MAIN;
2469	r->rtm_type	= RTN_UNICAST;
2470	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2471	r->rtm_protocol = rt->rt_protocol;
2472	r->rtm_flags	= 0;
2473
2474	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2475		goto nla_put_failure;
2476
2477	if (nh->nh_labels &&
2478	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2479			   nh->nh_label))
2480		goto nla_put_failure;
2481
2482	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2483	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2484			nh->nh_via_alen))
2485		goto nla_put_failure;
2486	dev = nh->nh_dev;
2487	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2488		goto nla_put_failure;
2489
2490	nlmsg_end(skb, nlh);
2491
2492	err = rtnl_unicast(skb, net, portid);
2493errout:
2494	return err;
2495
2496nla_put_failure:
2497	nlmsg_cancel(skb, nlh);
2498	err = -EMSGSIZE;
2499errout_free:
2500	kfree_skb(skb);
2501	return err;
2502}
2503
2504static int resize_platform_label_table(struct net *net, size_t limit)
2505{
2506	size_t size = sizeof(struct mpls_route *) * limit;
2507	size_t old_limit;
2508	size_t cp_size;
2509	struct mpls_route __rcu **labels = NULL, **old;
2510	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2511	unsigned index;
2512
2513	if (size) {
2514		labels = kvzalloc(size, GFP_KERNEL);
 
 
 
2515		if (!labels)
2516			goto nolabels;
2517	}
2518
2519	/* In case the predefined labels need to be populated */
2520	if (limit > MPLS_LABEL_IPV4NULL) {
2521		struct net_device *lo = net->loopback_dev;
2522		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2523		if (IS_ERR(rt0))
2524			goto nort0;
2525		rt0->rt_nh->nh_dev = lo;
2526		rt0->rt_protocol = RTPROT_KERNEL;
2527		rt0->rt_payload_type = MPT_IPV4;
2528		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2529		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2530		rt0->rt_nh->nh_via_alen = lo->addr_len;
2531		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2532		       lo->addr_len);
2533	}
2534	if (limit > MPLS_LABEL_IPV6NULL) {
2535		struct net_device *lo = net->loopback_dev;
2536		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2537		if (IS_ERR(rt2))
2538			goto nort2;
2539		rt2->rt_nh->nh_dev = lo;
2540		rt2->rt_protocol = RTPROT_KERNEL;
2541		rt2->rt_payload_type = MPT_IPV6;
2542		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2543		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2544		rt2->rt_nh->nh_via_alen = lo->addr_len;
2545		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2546		       lo->addr_len);
2547	}
2548
2549	rtnl_lock();
2550	/* Remember the original table */
2551	old = rtnl_dereference(net->mpls.platform_label);
2552	old_limit = net->mpls.platform_labels;
2553
2554	/* Free any labels beyond the new table */
2555	for (index = limit; index < old_limit; index++)
2556		mpls_route_update(net, index, NULL, NULL);
2557
2558	/* Copy over the old labels */
2559	cp_size = size;
2560	if (old_limit < limit)
2561		cp_size = old_limit * sizeof(struct mpls_route *);
2562
2563	memcpy(labels, old, cp_size);
2564
2565	/* If needed set the predefined labels */
2566	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2567	    (limit > MPLS_LABEL_IPV6NULL)) {
2568		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2569		rt2 = NULL;
2570	}
2571
2572	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2573	    (limit > MPLS_LABEL_IPV4NULL)) {
2574		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2575		rt0 = NULL;
2576	}
2577
2578	/* Update the global pointers */
2579	net->mpls.platform_labels = limit;
2580	rcu_assign_pointer(net->mpls.platform_label, labels);
2581
2582	rtnl_unlock();
2583
2584	mpls_rt_free(rt2);
2585	mpls_rt_free(rt0);
2586
2587	if (old) {
2588		synchronize_rcu();
2589		kvfree(old);
2590	}
2591	return 0;
2592
2593nort2:
2594	mpls_rt_free(rt0);
2595nort0:
2596	kvfree(labels);
2597nolabels:
2598	return -ENOMEM;
2599}
2600
2601static int mpls_platform_labels(const struct ctl_table *table, int write,
2602				void *buffer, size_t *lenp, loff_t *ppos)
2603{
2604	struct net *net = table->data;
2605	int platform_labels = net->mpls.platform_labels;
2606	int ret;
2607	struct ctl_table tmp = {
2608		.procname	= table->procname,
2609		.data		= &platform_labels,
2610		.maxlen		= sizeof(int),
2611		.mode		= table->mode,
2612		.extra1		= SYSCTL_ZERO,
2613		.extra2		= &label_limit,
2614	};
2615
2616	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2617
2618	if (write && ret == 0)
2619		ret = resize_platform_label_table(net, platform_labels);
2620
2621	return ret;
2622}
2623
2624#define MPLS_NS_SYSCTL_OFFSET(field)		\
2625	(&((struct net *)0)->field)
2626
2627static const struct ctl_table mpls_table[] = {
2628	{
2629		.procname	= "platform_labels",
2630		.data		= NULL,
2631		.maxlen		= sizeof(int),
2632		.mode		= 0644,
2633		.proc_handler	= mpls_platform_labels,
2634	},
2635	{
2636		.procname	= "ip_ttl_propagate",
2637		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2638		.maxlen		= sizeof(int),
2639		.mode		= 0644,
2640		.proc_handler	= proc_dointvec_minmax,
2641		.extra1		= SYSCTL_ZERO,
2642		.extra2		= SYSCTL_ONE,
2643	},
2644	{
2645		.procname	= "default_ttl",
2646		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2647		.maxlen		= sizeof(int),
2648		.mode		= 0644,
2649		.proc_handler	= proc_dointvec_minmax,
2650		.extra1		= SYSCTL_ONE,
2651		.extra2		= &ttl_max,
2652	},
2653};
2654
2655static int mpls_net_init(struct net *net)
2656{
2657	size_t table_size = ARRAY_SIZE(mpls_table);
2658	struct ctl_table *table;
2659	int i;
2660
2661	net->mpls.platform_labels = 0;
2662	net->mpls.platform_label = NULL;
2663	net->mpls.ip_ttl_propagate = 1;
2664	net->mpls.default_ttl = 255;
2665
2666	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2667	if (table == NULL)
2668		return -ENOMEM;
2669
2670	/* Table data contains only offsets relative to the base of
2671	 * the mdev at this point, so make them absolute.
2672	 */
2673	for (i = 0; i < table_size; i++)
2674		table[i].data = (char *)net + (uintptr_t)table[i].data;
2675
2676	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2677					       table_size);
2678	if (net->mpls.ctl == NULL) {
2679		kfree(table);
2680		return -ENOMEM;
2681	}
2682
2683	return 0;
2684}
2685
2686static void mpls_net_exit(struct net *net)
2687{
2688	struct mpls_route __rcu **platform_label;
2689	size_t platform_labels;
2690	const struct ctl_table *table;
2691	unsigned int index;
2692
2693	table = net->mpls.ctl->ctl_table_arg;
2694	unregister_net_sysctl_table(net->mpls.ctl);
2695	kfree(table);
2696
2697	/* An rcu grace period has passed since there was a device in
2698	 * the network namespace (and thus the last in flight packet)
2699	 * left this network namespace.  This is because
2700	 * unregister_netdevice_many and netdev_run_todo has completed
2701	 * for each network device that was in this network namespace.
2702	 *
2703	 * As such no additional rcu synchronization is necessary when
2704	 * freeing the platform_label table.
2705	 */
2706	rtnl_lock();
2707	platform_label = rtnl_dereference(net->mpls.platform_label);
2708	platform_labels = net->mpls.platform_labels;
2709	for (index = 0; index < platform_labels; index++) {
2710		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2711		RCU_INIT_POINTER(platform_label[index], NULL);
2712		mpls_notify_route(net, index, rt, NULL, NULL);
2713		mpls_rt_free(rt);
2714	}
2715	rtnl_unlock();
2716
2717	kvfree(platform_label);
2718}
2719
2720static struct pernet_operations mpls_net_ops = {
2721	.init = mpls_net_init,
2722	.exit = mpls_net_exit,
2723};
2724
2725static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2726	.family		   = AF_MPLS,
2727	.fill_stats_af	   = mpls_fill_stats_af,
2728	.get_stats_af_size = mpls_get_stats_af_size,
2729};
2730
2731static const struct rtnl_msg_handler mpls_rtnl_msg_handlers[] __initdata_or_module = {
2732	{THIS_MODULE, PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, 0},
2733	{THIS_MODULE, PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, 0},
2734	{THIS_MODULE, PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes, 0},
2735	{THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2736	 mpls_netconf_get_devconf, mpls_netconf_dump_devconf,
2737	 RTNL_FLAG_DUMP_UNLOCKED},
2738};
2739
2740static int __init mpls_init(void)
2741{
2742	int err;
2743
2744	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2745
2746	err = register_pernet_subsys(&mpls_net_ops);
2747	if (err)
2748		goto out;
2749
2750	err = register_netdevice_notifier(&mpls_dev_notifier);
2751	if (err)
2752		goto out_unregister_pernet;
2753
2754	dev_add_pack(&mpls_packet_type);
2755
2756	err = rtnl_af_register(&mpls_af_ops);
2757	if (err)
2758		goto out_unregister_dev_type;
2759
2760	err = rtnl_register_many(mpls_rtnl_msg_handlers);
2761	if (err)
2762		goto out_unregister_rtnl_af;
2763
2764	err = ipgre_tunnel_encap_add_mpls_ops();
2765	if (err) {
2766		pr_err("Can't add mpls over gre tunnel ops\n");
2767		goto out_unregister_rtnl;
2768	}
2769
2770	err = 0;
2771out:
2772	return err;
2773
2774out_unregister_rtnl:
2775	rtnl_unregister_many(mpls_rtnl_msg_handlers);
2776out_unregister_rtnl_af:
2777	rtnl_af_unregister(&mpls_af_ops);
2778out_unregister_dev_type:
2779	dev_remove_pack(&mpls_packet_type);
2780out_unregister_pernet:
2781	unregister_pernet_subsys(&mpls_net_ops);
2782	goto out;
2783}
2784module_init(mpls_init);
2785
2786static void __exit mpls_exit(void)
2787{
2788	rtnl_unregister_all(PF_MPLS);
2789	rtnl_af_unregister(&mpls_af_ops);
2790	dev_remove_pack(&mpls_packet_type);
2791	unregister_netdevice_notifier(&mpls_dev_notifier);
2792	unregister_pernet_subsys(&mpls_net_ops);
2793	ipgre_tunnel_encap_del_mpls_ops();
2794}
2795module_exit(mpls_exit);
2796
2797MODULE_DESCRIPTION("MultiProtocol Label Switching");
2798MODULE_LICENSE("GPL v2");
2799MODULE_ALIAS_NETPROTO(PF_MPLS);