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/ip.h>
  16#include <net/dst.h>
  17#include <net/sock.h>
  18#include <net/arp.h>
  19#include <net/ip_fib.h>
  20#include <net/netevent.h>
  21#include <net/ip_tunnels.h>
  22#include <net/netns/generic.h>
  23#if IS_ENABLED(CONFIG_IPV6)
  24#include <net/ipv6.h>
  25#endif
  26#include <net/ipv6_stubs.h>
  27#include <net/rtnh.h>
  28#include "internal.h"
  29
  30/* max memory we will use for mpls_route */
  31#define MAX_MPLS_ROUTE_MEM	4096
  32
  33/* Maximum number of labels to look ahead at when selecting a path of
  34 * a multipath route
  35 */
  36#define MAX_MP_SELECT_LABELS 4
  37
  38#define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
  39
  40static int label_limit = (1 << 20) - 1;
  41static int ttl_max = 255;
  42
  43#if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
  44static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
  45{
  46	return sizeof(struct mpls_shim_hdr);
  47}
  48
  49static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
  50	.encap_hlen	= ipgre_mpls_encap_hlen,
  51};
  52
  53static int ipgre_tunnel_encap_add_mpls_ops(void)
  54{
  55	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
  56}
  57
  58static void ipgre_tunnel_encap_del_mpls_ops(void)
  59{
  60	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
  61}
  62#else
  63static int ipgre_tunnel_encap_add_mpls_ops(void)
  64{
  65	return 0;
  66}
  67
  68static void ipgre_tunnel_encap_del_mpls_ops(void)
  69{
  70}
  71#endif
  72
  73static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
  74		       struct nlmsghdr *nlh, struct net *net, u32 portid,
  75		       unsigned int nlm_flags);
  76
  77static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
  78{
  79	struct mpls_route *rt = NULL;
  80
  81	if (index < net->mpls.platform_labels) {
  82		struct mpls_route __rcu **platform_label =
  83			rcu_dereference(net->mpls.platform_label);
  84		rt = rcu_dereference(platform_label[index]);
  85	}
  86	return rt;
  87}
  88
  89bool mpls_output_possible(const struct net_device *dev)
  90{
  91	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
  92}
  93EXPORT_SYMBOL_GPL(mpls_output_possible);
  94
  95static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
  96{
  97	return (u8 *)nh + rt->rt_via_offset;
  98}
  99
 100static const u8 *mpls_nh_via(const struct mpls_route *rt,
 101			     const struct mpls_nh *nh)
 102{
 103	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
 104}
 105
 106static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
 107{
 108	/* The size of the layer 2.5 labels to be added for this route */
 109	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
 110}
 111
 112unsigned int mpls_dev_mtu(const struct net_device *dev)
 113{
 114	/* The amount of data the layer 2 frame can hold */
 115	return dev->mtu;
 116}
 117EXPORT_SYMBOL_GPL(mpls_dev_mtu);
 118
 119bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
 120{
 121	if (skb->len <= mtu)
 122		return false;
 123
 124	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 125		return false;
 126
 127	return true;
 128}
 129EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
 130
 131void mpls_stats_inc_outucastpkts(struct net_device *dev,
 132				 const struct sk_buff *skb)
 133{
 134	struct mpls_dev *mdev;
 135
 136	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
 137		mdev = mpls_dev_get(dev);
 138		if (mdev)
 139			MPLS_INC_STATS_LEN(mdev, skb->len,
 140					   tx_packets,
 141					   tx_bytes);
 142	} else if (skb->protocol == htons(ETH_P_IP)) {
 143		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
 144#if IS_ENABLED(CONFIG_IPV6)
 145	} else if (skb->protocol == htons(ETH_P_IPV6)) {
 146		struct inet6_dev *in6dev = __in6_dev_get(dev);
 147
 148		if (in6dev)
 149			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
 150					 IPSTATS_MIB_OUT, skb->len);
 151#endif
 152	}
 153}
 154EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
 155
 156static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
 157{
 158	struct mpls_entry_decoded dec;
 159	unsigned int mpls_hdr_len = 0;
 160	struct mpls_shim_hdr *hdr;
 161	bool eli_seen = false;
 162	int label_index;
 163	u32 hash = 0;
 164
 165	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
 166	     label_index++) {
 167		mpls_hdr_len += sizeof(*hdr);
 168		if (!pskb_may_pull(skb, mpls_hdr_len))
 169			break;
 170
 171		/* Read and decode the current label */
 172		hdr = mpls_hdr(skb) + label_index;
 173		dec = mpls_entry_decode(hdr);
 174
 175		/* RFC6790 - reserved labels MUST NOT be used as keys
 176		 * for the load-balancing function
 177		 */
 178		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
 179			hash = jhash_1word(dec.label, hash);
 180
 181			/* The entropy label follows the entropy label
 182			 * indicator, so this means that the entropy
 183			 * label was just added to the hash - no need to
 184			 * go any deeper either in the label stack or in the
 185			 * payload
 186			 */
 187			if (eli_seen)
 188				break;
 189		} else if (dec.label == MPLS_LABEL_ENTROPY) {
 190			eli_seen = true;
 191		}
 192
 193		if (!dec.bos)
 194			continue;
 195
 196		/* found bottom label; does skb have room for a header? */
 197		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
 198			const struct iphdr *v4hdr;
 199
 200			v4hdr = (const struct iphdr *)(hdr + 1);
 201			if (v4hdr->version == 4) {
 202				hash = jhash_3words(ntohl(v4hdr->saddr),
 203						    ntohl(v4hdr->daddr),
 204						    v4hdr->protocol, hash);
 205			} else if (v4hdr->version == 6 &&
 206				   pskb_may_pull(skb, mpls_hdr_len +
 207						 sizeof(struct ipv6hdr))) {
 208				const struct ipv6hdr *v6hdr;
 209
 210				v6hdr = (const struct ipv6hdr *)(hdr + 1);
 211				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
 212				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
 213				hash = jhash_1word(v6hdr->nexthdr, hash);
 214			}
 215		}
 216
 217		break;
 218	}
 219
 220	return hash;
 221}
 222
 223static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
 224{
 225	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
 226}
 227
 228/* number of alive nexthops (rt->rt_nhn_alive) and the flags for
 229 * a next hop (nh->nh_flags) are modified by netdev event handlers.
 230 * Since those fields can change at any moment, use READ_ONCE to
 231 * access both.
 232 */
 233static const struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
 234						   struct sk_buff *skb)
 235{
 236	u32 hash = 0;
 237	int nh_index = 0;
 238	int n = 0;
 239	u8 alive;
 240
 241	/* No need to look further into packet if there's only
 242	 * one path
 243	 */
 244	if (rt->rt_nhn == 1)
 245		return rt->rt_nh;
 246
 247	alive = READ_ONCE(rt->rt_nhn_alive);
 248	if (alive == 0)
 249		return NULL;
 250
 251	hash = mpls_multipath_hash(rt, skb);
 252	nh_index = hash % alive;
 253	if (alive == rt->rt_nhn)
 254		goto out;
 255	for_nexthops(rt) {
 256		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
 257
 258		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 259			continue;
 260		if (n == nh_index)
 261			return nh;
 262		n++;
 263	} endfor_nexthops(rt);
 264
 265out:
 266	return mpls_get_nexthop(rt, nh_index);
 267}
 268
 269static bool mpls_egress(struct net *net, struct mpls_route *rt,
 270			struct sk_buff *skb, struct mpls_entry_decoded dec)
 271{
 272	enum mpls_payload_type payload_type;
 273	bool success = false;
 274
 275	/* The IPv4 code below accesses through the IPv4 header
 276	 * checksum, which is 12 bytes into the packet.
 277	 * The IPv6 code below accesses through the IPv6 hop limit
 278	 * which is 8 bytes into the packet.
 279	 *
 280	 * For all supported cases there should always be at least 12
 281	 * bytes of packet data present.  The IPv4 header is 20 bytes
 282	 * without options and the IPv6 header is always 40 bytes
 283	 * long.
 284	 */
 285	if (!pskb_may_pull(skb, 12))
 286		return false;
 287
 288	payload_type = rt->rt_payload_type;
 289	if (payload_type == MPT_UNSPEC)
 290		payload_type = ip_hdr(skb)->version;
 291
 292	switch (payload_type) {
 293	case MPT_IPV4: {
 294		struct iphdr *hdr4 = ip_hdr(skb);
 295		u8 new_ttl;
 296		skb->protocol = htons(ETH_P_IP);
 297
 298		/* If propagating TTL, take the decremented TTL from
 299		 * the incoming MPLS header, otherwise decrement the
 300		 * TTL, but only if not 0 to avoid underflow.
 301		 */
 302		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
 303		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
 304		     net->mpls.ip_ttl_propagate))
 305			new_ttl = dec.ttl;
 306		else
 307			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
 308
 309		csum_replace2(&hdr4->check,
 310			      htons(hdr4->ttl << 8),
 311			      htons(new_ttl << 8));
 312		hdr4->ttl = new_ttl;
 313		success = true;
 314		break;
 315	}
 316	case MPT_IPV6: {
 317		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
 318		skb->protocol = htons(ETH_P_IPV6);
 319
 320		/* If propagating TTL, take the decremented TTL from
 321		 * the incoming MPLS header, otherwise decrement the
 322		 * hop limit, but only if not 0 to avoid underflow.
 323		 */
 324		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
 325		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
 326		     net->mpls.ip_ttl_propagate))
 327			hdr6->hop_limit = dec.ttl;
 328		else if (hdr6->hop_limit)
 329			hdr6->hop_limit = hdr6->hop_limit - 1;
 330		success = true;
 331		break;
 332	}
 333	case MPT_UNSPEC:
 334		/* Should have decided which protocol it is by now */
 335		break;
 336	}
 337
 338	return success;
 339}
 340
 341static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
 342			struct packet_type *pt, struct net_device *orig_dev)
 343{
 344	struct net *net = dev_net(dev);
 345	struct mpls_shim_hdr *hdr;
 346	const struct mpls_nh *nh;
 347	struct mpls_route *rt;
 348	struct mpls_entry_decoded dec;
 349	struct net_device *out_dev;
 350	struct mpls_dev *out_mdev;
 351	struct mpls_dev *mdev;
 352	unsigned int hh_len;
 353	unsigned int new_header_size;
 354	unsigned int mtu;
 355	int err;
 356
 357	/* Careful this entire function runs inside of an rcu critical section */
 358
 359	mdev = mpls_dev_get(dev);
 360	if (!mdev)
 361		goto drop;
 362
 363	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
 364			   rx_bytes);
 365
 366	if (!mdev->input_enabled) {
 367		MPLS_INC_STATS(mdev, rx_dropped);
 368		goto drop;
 369	}
 370
 371	if (skb->pkt_type != PACKET_HOST)
 372		goto err;
 373
 374	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 375		goto err;
 376
 377	if (!pskb_may_pull(skb, sizeof(*hdr)))
 378		goto err;
 379
 380	skb_dst_drop(skb);
 381
 382	/* Read and decode the label */
 383	hdr = mpls_hdr(skb);
 384	dec = mpls_entry_decode(hdr);
 385
 386	rt = mpls_route_input_rcu(net, dec.label);
 387	if (!rt) {
 388		MPLS_INC_STATS(mdev, rx_noroute);
 389		goto drop;
 390	}
 391
 392	nh = mpls_select_multipath(rt, skb);
 393	if (!nh)
 394		goto err;
 395
 396	/* Pop the label */
 397	skb_pull(skb, sizeof(*hdr));
 398	skb_reset_network_header(skb);
 399
 400	skb_orphan(skb);
 401
 402	if (skb_warn_if_lro(skb))
 403		goto err;
 404
 405	skb_forward_csum(skb);
 406
 407	/* Verify ttl is valid */
 408	if (dec.ttl <= 1)
 409		goto err;
 410
 411	/* Find the output device */
 412	out_dev = nh->nh_dev;
 413	if (!mpls_output_possible(out_dev))
 414		goto tx_err;
 415
 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))
 420		goto tx_err;
 421
 422	hh_len = LL_RESERVED_SPACE(out_dev);
 423	if (!out_dev->header_ops)
 424		hh_len = 0;
 425
 426	/* Ensure there is enough space for the headers in the skb */
 427	if (skb_cow(skb, hh_len + new_header_size))
 428		goto tx_err;
 429
 430	skb->dev = out_dev;
 431	skb->protocol = htons(ETH_P_MPLS_UC);
 432
 433	dec.ttl -= 1;
 434	if (unlikely(!new_header_size && dec.bos)) {
 435		/* Penultimate hop popping */
 436		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
 437			goto err;
 438	} else {
 439		bool bos;
 440		int i;
 441		skb_push(skb, new_header_size);
 442		skb_reset_network_header(skb);
 443		/* Push the new labels */
 444		hdr = mpls_hdr(skb);
 445		bos = dec.bos;
 446		for (i = nh->nh_labels - 1; i >= 0; i--) {
 447			hdr[i] = mpls_entry_encode(nh->nh_label[i],
 448						   dec.ttl, 0, bos);
 449			bos = false;
 450		}
 451	}
 452
 453	mpls_stats_inc_outucastpkts(out_dev, skb);
 454
 455	/* If via wasn't specified then send out using device address */
 456	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
 457		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
 458				 out_dev->dev_addr, skb);
 459	else
 460		err = neigh_xmit(nh->nh_via_table, out_dev,
 461				 mpls_nh_via(rt, nh), skb);
 462	if (err)
 463		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
 464				    __func__, err);
 465	return 0;
 466
 467tx_err:
 468	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
 469	if (out_mdev)
 470		MPLS_INC_STATS(out_mdev, tx_errors);
 471	goto drop;
 472err:
 473	MPLS_INC_STATS(mdev, rx_errors);
 474drop:
 475	kfree_skb(skb);
 476	return NET_RX_DROP;
 477}
 478
 479static struct packet_type mpls_packet_type __read_mostly = {
 480	.type = cpu_to_be16(ETH_P_MPLS_UC),
 481	.func = mpls_forward,
 482};
 483
 484static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
 485	[RTA_DST]		= { .type = NLA_U32 },
 486	[RTA_OIF]		= { .type = NLA_U32 },
 487	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
 488};
 489
 490struct mpls_route_config {
 491	u32			rc_protocol;
 492	u32			rc_ifindex;
 493	u8			rc_via_table;
 494	u8			rc_via_alen;
 495	u8			rc_via[MAX_VIA_ALEN];
 496	u32			rc_label;
 497	u8			rc_ttl_propagate;
 498	u8			rc_output_labels;
 499	u32			rc_output_label[MAX_NEW_LABELS];
 500	u32			rc_nlflags;
 501	enum mpls_payload_type	rc_payload_type;
 502	struct nl_info		rc_nlinfo;
 503	struct rtnexthop	*rc_mp;
 504	int			rc_mp_len;
 505};
 506
 507/* all nexthops within a route have the same size based on max
 508 * number of labels and max via length for a hop
 509 */
 510static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
 511{
 512	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
 513	struct mpls_route *rt;
 514	size_t size;
 515
 516	size = sizeof(*rt) + num_nh * nh_size;
 517	if (size > MAX_MPLS_ROUTE_MEM)
 518		return ERR_PTR(-EINVAL);
 519
 520	rt = kzalloc(size, GFP_KERNEL);
 521	if (!rt)
 522		return ERR_PTR(-ENOMEM);
 523
 524	rt->rt_nhn = num_nh;
 525	rt->rt_nhn_alive = num_nh;
 526	rt->rt_nh_size = nh_size;
 527	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
 528
 529	return rt;
 530}
 531
 532static void mpls_rt_free(struct mpls_route *rt)
 533{
 534	if (rt)
 535		kfree_rcu(rt, rt_rcu);
 536}
 537
 538static void mpls_notify_route(struct net *net, unsigned index,
 539			      struct mpls_route *old, struct mpls_route *new,
 540			      const struct nl_info *info)
 541{
 542	struct nlmsghdr *nlh = info ? info->nlh : NULL;
 543	unsigned portid = info ? info->portid : 0;
 544	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
 545	struct mpls_route *rt = new ? new : old;
 546	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
 547	/* Ignore reserved labels for now */
 548	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
 549		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
 550}
 551
 552static void mpls_route_update(struct net *net, unsigned index,
 553			      struct mpls_route *new,
 554			      const struct nl_info *info)
 555{
 556	struct mpls_route __rcu **platform_label;
 557	struct mpls_route *rt;
 558
 559	ASSERT_RTNL();
 560
 561	platform_label = rtnl_dereference(net->mpls.platform_label);
 562	rt = rtnl_dereference(platform_label[index]);
 563	rcu_assign_pointer(platform_label[index], new);
 564
 565	mpls_notify_route(net, index, rt, new, info);
 566
 567	/* If we removed a route free it now */
 568	mpls_rt_free(rt);
 569}
 570
 571static unsigned find_free_label(struct net *net)
 572{
 573	struct mpls_route __rcu **platform_label;
 574	size_t platform_labels;
 575	unsigned index;
 576
 577	platform_label = rtnl_dereference(net->mpls.platform_label);
 578	platform_labels = net->mpls.platform_labels;
 579	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
 580	     index++) {
 581		if (!rtnl_dereference(platform_label[index]))
 582			return index;
 583	}
 584	return LABEL_NOT_SPECIFIED;
 585}
 586
 587#if IS_ENABLED(CONFIG_INET)
 588static struct net_device *inet_fib_lookup_dev(struct net *net,
 589					      const void *addr)
 590{
 591	struct net_device *dev;
 592	struct rtable *rt;
 593	struct in_addr daddr;
 594
 595	memcpy(&daddr, addr, sizeof(struct in_addr));
 596	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
 597	if (IS_ERR(rt))
 598		return ERR_CAST(rt);
 599
 600	dev = rt->dst.dev;
 601	dev_hold(dev);
 602
 603	ip_rt_put(rt);
 604
 605	return dev;
 606}
 607#else
 608static struct net_device *inet_fib_lookup_dev(struct net *net,
 609					      const void *addr)
 610{
 611	return ERR_PTR(-EAFNOSUPPORT);
 612}
 613#endif
 614
 615#if IS_ENABLED(CONFIG_IPV6)
 616static struct net_device *inet6_fib_lookup_dev(struct net *net,
 617					       const void *addr)
 618{
 619	struct net_device *dev;
 620	struct dst_entry *dst;
 621	struct flowi6 fl6;
 622
 623	if (!ipv6_stub)
 624		return ERR_PTR(-EAFNOSUPPORT);
 625
 626	memset(&fl6, 0, sizeof(fl6));
 627	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
 628	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
 629	if (IS_ERR(dst))
 630		return ERR_CAST(dst);
 631
 632	dev = dst->dev;
 633	dev_hold(dev);
 634	dst_release(dst);
 635
 636	return dev;
 637}
 638#else
 639static struct net_device *inet6_fib_lookup_dev(struct net *net,
 640					       const void *addr)
 641{
 642	return ERR_PTR(-EAFNOSUPPORT);
 643}
 644#endif
 645
 646static struct net_device *find_outdev(struct net *net,
 647				      struct mpls_route *rt,
 648				      struct mpls_nh *nh, int oif)
 649{
 650	struct net_device *dev = NULL;
 651
 652	if (!oif) {
 653		switch (nh->nh_via_table) {
 654		case NEIGH_ARP_TABLE:
 655			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
 656			break;
 657		case NEIGH_ND_TABLE:
 658			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
 659			break;
 660		case NEIGH_LINK_TABLE:
 661			break;
 662		}
 663	} else {
 664		dev = dev_get_by_index(net, oif);
 665	}
 666
 667	if (!dev)
 668		return ERR_PTR(-ENODEV);
 669
 670	if (IS_ERR(dev))
 671		return dev;
 672
 673	/* The caller is holding rtnl anyways, so release the dev reference */
 674	dev_put(dev);
 675
 676	return dev;
 677}
 678
 679static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
 680			      struct mpls_nh *nh, int oif)
 681{
 682	struct net_device *dev = NULL;
 683	int err = -ENODEV;
 684
 685	dev = find_outdev(net, rt, nh, oif);
 686	if (IS_ERR(dev)) {
 687		err = PTR_ERR(dev);
 688		dev = NULL;
 689		goto errout;
 690	}
 691
 692	/* Ensure this is a supported device */
 693	err = -EINVAL;
 694	if (!mpls_dev_get(dev))
 695		goto errout;
 696
 697	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
 698	    (dev->addr_len != nh->nh_via_alen))
 699		goto errout;
 700
 701	nh->nh_dev = dev;
 702
 703	if (!(dev->flags & IFF_UP)) {
 704		nh->nh_flags |= RTNH_F_DEAD;
 705	} else {
 706		unsigned int flags;
 707
 708		flags = dev_get_flags(dev);
 709		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
 710			nh->nh_flags |= RTNH_F_LINKDOWN;
 711	}
 712
 713	return 0;
 714
 715errout:
 716	return err;
 717}
 718
 719static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
 720		       u8 via_addr[], struct netlink_ext_ack *extack)
 721{
 722	struct rtvia *via = nla_data(nla);
 723	int err = -EINVAL;
 724	int alen;
 725
 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");
 729		goto errout;
 730	}
 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");
 736		goto errout;
 737	}
 738
 739	/* Validate the address family */
 740	switch (via->rtvia_family) {
 741	case AF_PACKET:
 742		*via_table = NEIGH_LINK_TABLE;
 743		break;
 744	case AF_INET:
 745		*via_table = NEIGH_ARP_TABLE;
 746		if (alen != 4)
 747			goto errout;
 748		break;
 749	case AF_INET6:
 750		*via_table = NEIGH_ND_TABLE;
 751		if (alen != 16)
 752			goto errout;
 753		break;
 754	default:
 755		/* Unsupported address family */
 756		goto errout;
 757	}
 758
 759	memcpy(via_addr, via->rtvia_addr, alen);
 760	*via_alen = alen;
 761	err = 0;
 762
 763errout:
 764	return err;
 765}
 766
 767static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
 768				  struct mpls_route *rt)
 769{
 770	struct net *net = cfg->rc_nlinfo.nl_net;
 771	struct mpls_nh *nh = rt->rt_nh;
 772	int err;
 773	int i;
 774
 775	if (!nh)
 776		return -ENOMEM;
 777
 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];
 781
 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;
 785
 786	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
 787	if (err)
 788		goto errout;
 789
 790	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 791		rt->rt_nhn_alive--;
 792
 793	return 0;
 794
 795errout:
 796	return err;
 797}
 798
 799static 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)
 803{
 804	int err = -ENOMEM;
 805
 806	if (!nh)
 807		goto errout;
 808
 809	if (newdst) {
 810		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
 811				     nh->nh_label, extack);
 812		if (err)
 813			goto errout;
 814	}
 815
 816	if (via) {
 817		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
 818				  __mpls_nh_via(rt, nh), extack);
 819		if (err)
 820			goto errout;
 821	} else {
 822		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
 823	}
 824
 825	err = mpls_nh_assign_dev(net, rt, nh, oif);
 826	if (err)
 827		goto errout;
 828
 829	return 0;
 830
 831errout:
 832	return err;
 833}
 834
 835static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
 836			      u8 cfg_via_alen, u8 *max_via_alen,
 837			      u8 *max_labels)
 838{
 839	int remaining = len;
 840	u8 nhs = 0;
 841
 842	*max_via_alen = 0;
 843	*max_labels = 0;
 844
 845	while (rtnh_ok(rtnh, remaining)) {
 846		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 847		int attrlen;
 848		u8 n_labels = 0;
 849
 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);
 856
 857			if (via_alen <= MAX_VIA_ALEN)
 858				*max_via_alen = max_t(u16, *max_via_alen,
 859						      via_alen);
 860		}
 861
 862		nla = nla_find(attrs, attrlen, RTA_NEWDST);
 863		if (nla &&
 864		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
 865				   NULL, NULL) != 0)
 866			return 0;
 867
 868		*max_labels = max_t(u8, *max_labels, n_labels);
 869
 870		/* number of nexthops is tracked by a u8.
 871		 * Check for overflow.
 872		 */
 873		if (nhs == 255)
 874			return 0;
 875		nhs++;
 876
 877		rtnh = rtnh_next(rtnh, &remaining);
 878	}
 879
 880	/* leftover implies invalid nexthop configuration, discard it */
 881	return remaining > 0 ? 0 : nhs;
 882}
 883
 884static int mpls_nh_build_multi(struct mpls_route_config *cfg,
 885			       struct mpls_route *rt, u8 max_labels,
 886			       struct netlink_ext_ack *extack)
 887{
 888	struct rtnexthop *rtnh = cfg->rc_mp;
 889	struct nlattr *nla_via, *nla_newdst;
 890	int remaining = cfg->rc_mp_len;
 891	int err = 0;
 892	u8 nhs = 0;
 893
 894	change_nexthops(rt) {
 895		int attrlen;
 896
 897		nla_via = NULL;
 898		nla_newdst = NULL;
 899
 900		err = -EINVAL;
 901		if (!rtnh_ok(rtnh, remaining))
 902			goto errout;
 903
 904		/* neither weighted multipath nor any flags
 905		 * are supported
 906		 */
 907		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
 908			goto errout;
 909
 910		attrlen = rtnh_attrlen(rtnh);
 911		if (attrlen > 0) {
 912			struct nlattr *attrs = rtnh_attrs(rtnh);
 913
 914			nla_via = nla_find(attrs, attrlen, RTA_VIA);
 915			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
 916		}
 917
 918		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
 919				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
 920				    max_labels, extack);
 921		if (err)
 922			goto errout;
 923
 924		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 925			rt->rt_nhn_alive--;
 926
 927		rtnh = rtnh_next(rtnh, &remaining);
 928		nhs++;
 929	} endfor_nexthops(rt);
 930
 931	rt->rt_nhn = nhs;
 932
 933	return 0;
 934
 935errout:
 936	return err;
 937}
 938
 939static bool mpls_label_ok(struct net *net, unsigned int *index,
 940			  struct netlink_ext_ack *extack)
 941{
 942	bool is_ok = true;
 943
 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");
 948		is_ok = false;
 949	}
 950
 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");
 955		is_ok = false;
 956	}
 957
 958	*index = array_index_nospec(*index, net->mpls.platform_labels);
 959	return is_ok;
 960}
 961
 962static int mpls_route_add(struct mpls_route_config *cfg,
 963			  struct netlink_ext_ack *extack)
 964{
 965	struct mpls_route __rcu **platform_label;
 966	struct net *net = cfg->rc_nlinfo.nl_net;
 967	struct mpls_route *rt, *old;
 968	int err = -EINVAL;
 969	u8 max_via_alen;
 970	unsigned index;
 971	u8 max_labels;
 972	u8 nhs;
 973
 974	index = cfg->rc_label;
 975
 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);
 980	}
 981
 982	if (!mpls_label_ok(net, &index, extack))
 983		goto errout;
 984
 985	/* Append makes no sense with mpls */
 986	err = -EOPNOTSUPP;
 987	if (cfg->rc_nlflags & NLM_F_APPEND) {
 988		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
 989		goto errout;
 990	}
 991
 992	err = -EEXIST;
 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)
 996		goto errout;
 997
 998	err = -EEXIST;
 999	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1000		goto errout;
1001
1002	err = -ENOENT;
1003	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1004		goto errout;
1005
1006	err = -EINVAL;
1007	if (cfg->rc_mp) {
1008		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1009					  cfg->rc_via_alen, &max_via_alen,
1010					  &max_labels);
1011	} else {
1012		max_via_alen = cfg->rc_via_alen;
1013		max_labels = cfg->rc_output_labels;
1014		nhs = 1;
1015	}
1016
1017	if (nhs == 0) {
1018		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1019		goto errout;
1020	}
1021
1022	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1023	if (IS_ERR(rt)) {
1024		err = PTR_ERR(rt);
1025		goto errout;
1026	}
1027
1028	rt->rt_protocol = cfg->rc_protocol;
1029	rt->rt_payload_type = cfg->rc_payload_type;
1030	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1031
1032	if (cfg->rc_mp)
1033		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1034	else
1035		err = mpls_nh_build_from_cfg(cfg, rt);
1036	if (err)
1037		goto freert;
1038
1039	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1040
1041	return 0;
1042
1043freert:
1044	mpls_rt_free(rt);
1045errout:
1046	return err;
1047}
1048
1049static int mpls_route_del(struct mpls_route_config *cfg,
1050			  struct netlink_ext_ack *extack)
1051{
1052	struct net *net = cfg->rc_nlinfo.nl_net;
1053	unsigned index;
1054	int err = -EINVAL;
1055
1056	index = cfg->rc_label;
1057
1058	if (!mpls_label_ok(net, &index, extack))
1059		goto errout;
1060
1061	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1062
1063	err = 0;
1064errout:
1065	return err;
1066}
1067
1068static void mpls_get_stats(struct mpls_dev *mdev,
1069			   struct mpls_link_stats *stats)
1070{
1071	struct mpls_pcpu_stats *p;
1072	int i;
1073
1074	memset(stats, 0, sizeof(*stats));
1075
1076	for_each_possible_cpu(i) {
1077		struct mpls_link_stats local;
1078		unsigned int start;
1079
1080		p = per_cpu_ptr(mdev->stats, i);
1081		do {
1082			start = u64_stats_fetch_begin(&p->syncp);
1083			local = p->stats;
1084		} while (u64_stats_fetch_retry(&p->syncp, start));
1085
1086		stats->rx_packets	+= local.rx_packets;
1087		stats->rx_bytes		+= local.rx_bytes;
1088		stats->tx_packets	+= local.tx_packets;
1089		stats->tx_bytes		+= local.tx_bytes;
1090		stats->rx_errors	+= local.rx_errors;
1091		stats->tx_errors	+= local.tx_errors;
1092		stats->rx_dropped	+= local.rx_dropped;
1093		stats->tx_dropped	+= local.tx_dropped;
1094		stats->rx_noroute	+= local.rx_noroute;
1095	}
1096}
1097
1098static int mpls_fill_stats_af(struct sk_buff *skb,
1099			      const struct net_device *dev)
1100{
1101	struct mpls_link_stats *stats;
1102	struct mpls_dev *mdev;
1103	struct nlattr *nla;
1104
1105	mdev = mpls_dev_get(dev);
1106	if (!mdev)
1107		return -ENODATA;
1108
1109	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1110				sizeof(struct mpls_link_stats),
1111				MPLS_STATS_UNSPEC);
1112	if (!nla)
1113		return -EMSGSIZE;
1114
1115	stats = nla_data(nla);
1116	mpls_get_stats(mdev, stats);
1117
1118	return 0;
1119}
1120
1121static size_t mpls_get_stats_af_size(const struct net_device *dev)
1122{
1123	struct mpls_dev *mdev;
1124
1125	mdev = mpls_dev_get(dev);
1126	if (!mdev)
1127		return 0;
1128
1129	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1130}
1131
1132static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1133				     u32 portid, u32 seq, int event,
1134				     unsigned int flags, int type)
1135{
1136	struct nlmsghdr  *nlh;
1137	struct netconfmsg *ncm;
1138	bool all = false;
1139
1140	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1141			flags);
1142	if (!nlh)
1143		return -EMSGSIZE;
1144
1145	if (type == NETCONFA_ALL)
1146		all = true;
1147
1148	ncm = nlmsg_data(nlh);
1149	ncm->ncm_family = AF_MPLS;
1150
1151	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1152		goto nla_put_failure;
1153
1154	if ((all || type == NETCONFA_INPUT) &&
1155	    nla_put_s32(skb, NETCONFA_INPUT,
1156			mdev->input_enabled) < 0)
1157		goto nla_put_failure;
1158
1159	nlmsg_end(skb, nlh);
1160	return 0;
1161
1162nla_put_failure:
1163	nlmsg_cancel(skb, nlh);
1164	return -EMSGSIZE;
1165}
1166
1167static int mpls_netconf_msgsize_devconf(int type)
1168{
1169	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1170			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1171	bool all = false;
1172
1173	if (type == NETCONFA_ALL)
1174		all = true;
1175
1176	if (all || type == NETCONFA_INPUT)
1177		size += nla_total_size(4);
1178
1179	return size;
1180}
1181
1182static void mpls_netconf_notify_devconf(struct net *net, int event,
1183					int type, struct mpls_dev *mdev)
1184{
1185	struct sk_buff *skb;
1186	int err = -ENOBUFS;
1187
1188	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1189	if (!skb)
1190		goto errout;
1191
1192	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1193	if (err < 0) {
1194		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1195		WARN_ON(err == -EMSGSIZE);
1196		kfree_skb(skb);
1197		goto errout;
1198	}
1199
1200	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1201	return;
1202errout:
1203	if (err < 0)
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 hlist_head *head;
1306	struct net_device *dev;
1307	struct mpls_dev *mdev;
1308	int idx, s_idx;
1309	int h, s_h;
1310
1311	if (cb->strict_check) {
1312		struct netlink_ext_ack *extack = cb->extack;
1313		struct netconfmsg *ncm;
1314
1315		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1316			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1317			return -EINVAL;
1318		}
1319
1320		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1321			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1322			return -EINVAL;
1323		}
1324	}
1325
1326	s_h = cb->args[0];
1327	s_idx = idx = cb->args[1];
1328
1329	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1330		idx = 0;
1331		head = &net->dev_index_head[h];
1332		rcu_read_lock();
1333		cb->seq = net->dev_base_seq;
1334		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1335			if (idx < s_idx)
1336				goto cont;
1337			mdev = mpls_dev_get(dev);
1338			if (!mdev)
1339				goto cont;
1340			if (mpls_netconf_fill_devconf(skb, mdev,
1341						      NETLINK_CB(cb->skb).portid,
1342						      nlh->nlmsg_seq,
1343						      RTM_NEWNETCONF,
1344						      NLM_F_MULTI,
1345						      NETCONFA_ALL) < 0) {
1346				rcu_read_unlock();
1347				goto done;
1348			}
1349			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1350cont:
1351			idx++;
1352		}
1353		rcu_read_unlock();
1354	}
1355done:
1356	cb->args[0] = h;
1357	cb->args[1] = idx;
1358
1359	return skb->len;
1360}
1361
1362#define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1363	(&((struct mpls_dev *)0)->field)
1364
1365static int mpls_conf_proc(struct ctl_table *ctl, int write,
1366			  void *buffer, size_t *lenp, loff_t *ppos)
1367{
1368	int oval = *(int *)ctl->data;
1369	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1370
1371	if (write) {
1372		struct mpls_dev *mdev = ctl->extra1;
1373		int i = (int *)ctl->data - (int *)mdev;
1374		struct net *net = ctl->extra2;
1375		int val = *(int *)ctl->data;
1376
1377		if (i == offsetof(struct mpls_dev, input_enabled) &&
1378		    val != oval) {
1379			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1380						    NETCONFA_INPUT, mdev);
1381		}
1382	}
1383
1384	return ret;
1385}
1386
1387static const struct ctl_table mpls_dev_table[] = {
1388	{
1389		.procname	= "input",
1390		.maxlen		= sizeof(int),
1391		.mode		= 0644,
1392		.proc_handler	= mpls_conf_proc,
1393		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1394	},
1395	{ }
1396};
1397
1398static int mpls_dev_sysctl_register(struct net_device *dev,
1399				    struct mpls_dev *mdev)
1400{
1401	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1402	struct net *net = dev_net(dev);
1403	struct ctl_table *table;
1404	int i;
1405
1406	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1407	if (!table)
1408		goto out;
1409
1410	/* Table data contains only offsets relative to the base of
1411	 * the mdev at this point, so make them absolute.
1412	 */
1413	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1414		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1415		table[i].extra1 = mdev;
1416		table[i].extra2 = net;
1417	}
1418
1419	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1420
1421	mdev->sysctl = register_net_sysctl(net, path, table);
 
1422	if (!mdev->sysctl)
1423		goto free;
1424
1425	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1426	return 0;
1427
1428free:
1429	kfree(table);
1430out:
1431	mdev->sysctl = NULL;
1432	return -ENOBUFS;
1433}
1434
1435static void mpls_dev_sysctl_unregister(struct net_device *dev,
1436				       struct mpls_dev *mdev)
1437{
1438	struct net *net = dev_net(dev);
1439	struct ctl_table *table;
1440
1441	if (!mdev->sysctl)
1442		return;
1443
1444	table = mdev->sysctl->ctl_table_arg;
1445	unregister_net_sysctl_table(mdev->sysctl);
1446	kfree(table);
1447
1448	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1449}
1450
1451static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1452{
1453	struct mpls_dev *mdev;
1454	int err = -ENOMEM;
1455	int i;
1456
1457	ASSERT_RTNL();
1458
1459	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1460	if (!mdev)
1461		return ERR_PTR(err);
1462
1463	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1464	if (!mdev->stats)
1465		goto free;
1466
1467	for_each_possible_cpu(i) {
1468		struct mpls_pcpu_stats *mpls_stats;
1469
1470		mpls_stats = per_cpu_ptr(mdev->stats, i);
1471		u64_stats_init(&mpls_stats->syncp);
1472	}
1473
1474	mdev->dev = dev;
1475
1476	err = mpls_dev_sysctl_register(dev, mdev);
1477	if (err)
1478		goto free;
1479
1480	rcu_assign_pointer(dev->mpls_ptr, mdev);
1481
1482	return mdev;
1483
1484free:
1485	free_percpu(mdev->stats);
1486	kfree(mdev);
1487	return ERR_PTR(err);
1488}
1489
1490static void mpls_dev_destroy_rcu(struct rcu_head *head)
1491{
1492	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1493
1494	free_percpu(mdev->stats);
1495	kfree(mdev);
1496}
1497
1498static int mpls_ifdown(struct net_device *dev, int event)
1499{
1500	struct mpls_route __rcu **platform_label;
1501	struct net *net = dev_net(dev);
1502	unsigned index;
1503
1504	platform_label = rtnl_dereference(net->mpls.platform_label);
1505	for (index = 0; index < net->mpls.platform_labels; index++) {
1506		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1507		bool nh_del = false;
1508		u8 alive = 0;
1509
1510		if (!rt)
1511			continue;
1512
1513		if (event == NETDEV_UNREGISTER) {
1514			u8 deleted = 0;
1515
1516			for_nexthops(rt) {
1517				if (!nh->nh_dev || nh->nh_dev == dev)
1518					deleted++;
1519				if (nh->nh_dev == dev)
1520					nh_del = true;
1521			} endfor_nexthops(rt);
1522
1523			/* if there are no more nexthops, delete the route */
1524			if (deleted == rt->rt_nhn) {
1525				mpls_route_update(net, index, NULL, NULL);
1526				continue;
1527			}
1528
1529			if (nh_del) {
1530				size_t size = sizeof(*rt) + rt->rt_nhn *
1531					rt->rt_nh_size;
1532				struct mpls_route *orig = rt;
1533
1534				rt = kmemdup(orig, size, GFP_KERNEL);
1535				if (!rt)
1536					return -ENOMEM;
1537			}
1538		}
1539
1540		change_nexthops(rt) {
1541			unsigned int nh_flags = nh->nh_flags;
1542
1543			if (nh->nh_dev != dev)
1544				goto next;
1545
1546			switch (event) {
1547			case NETDEV_DOWN:
1548			case NETDEV_UNREGISTER:
1549				nh_flags |= RTNH_F_DEAD;
1550				fallthrough;
1551			case NETDEV_CHANGE:
1552				nh_flags |= RTNH_F_LINKDOWN;
1553				break;
1554			}
1555			if (event == NETDEV_UNREGISTER)
1556				nh->nh_dev = NULL;
1557
1558			if (nh->nh_flags != nh_flags)
1559				WRITE_ONCE(nh->nh_flags, nh_flags);
1560next:
1561			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1562				alive++;
1563		} endfor_nexthops(rt);
1564
1565		WRITE_ONCE(rt->rt_nhn_alive, alive);
1566
1567		if (nh_del)
1568			mpls_route_update(net, index, rt, NULL);
1569	}
1570
1571	return 0;
1572}
1573
1574static void mpls_ifup(struct net_device *dev, unsigned int flags)
1575{
1576	struct mpls_route __rcu **platform_label;
1577	struct net *net = dev_net(dev);
1578	unsigned index;
1579	u8 alive;
1580
1581	platform_label = rtnl_dereference(net->mpls.platform_label);
1582	for (index = 0; index < net->mpls.platform_labels; index++) {
1583		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1584
1585		if (!rt)
1586			continue;
1587
1588		alive = 0;
1589		change_nexthops(rt) {
1590			unsigned int nh_flags = nh->nh_flags;
1591
1592			if (!(nh_flags & flags)) {
1593				alive++;
1594				continue;
1595			}
1596			if (nh->nh_dev != dev)
1597				continue;
1598			alive++;
1599			nh_flags &= ~flags;
1600			WRITE_ONCE(nh->nh_flags, nh_flags);
1601		} endfor_nexthops(rt);
1602
1603		WRITE_ONCE(rt->rt_nhn_alive, alive);
1604	}
1605}
1606
1607static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1608			   void *ptr)
1609{
1610	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1611	struct mpls_dev *mdev;
1612	unsigned int flags;
1613	int err;
1614
1615	if (event == NETDEV_REGISTER) {
1616		mdev = mpls_add_dev(dev);
1617		if (IS_ERR(mdev))
1618			return notifier_from_errno(PTR_ERR(mdev));
1619
1620		return NOTIFY_OK;
1621	}
1622
1623	mdev = mpls_dev_get(dev);
1624	if (!mdev)
1625		return NOTIFY_OK;
1626
1627	switch (event) {
1628
1629	case NETDEV_DOWN:
1630		err = mpls_ifdown(dev, event);
1631		if (err)
1632			return notifier_from_errno(err);
1633		break;
1634	case NETDEV_UP:
1635		flags = dev_get_flags(dev);
1636		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1637			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1638		else
1639			mpls_ifup(dev, RTNH_F_DEAD);
1640		break;
1641	case NETDEV_CHANGE:
1642		flags = dev_get_flags(dev);
1643		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1644			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1645		} else {
1646			err = mpls_ifdown(dev, event);
1647			if (err)
1648				return notifier_from_errno(err);
1649		}
1650		break;
1651	case NETDEV_UNREGISTER:
1652		err = mpls_ifdown(dev, event);
1653		if (err)
1654			return notifier_from_errno(err);
1655		mdev = mpls_dev_get(dev);
1656		if (mdev) {
1657			mpls_dev_sysctl_unregister(dev, mdev);
1658			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1659			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1660		}
1661		break;
1662	case NETDEV_CHANGENAME:
1663		mdev = mpls_dev_get(dev);
1664		if (mdev) {
1665			mpls_dev_sysctl_unregister(dev, mdev);
1666			err = mpls_dev_sysctl_register(dev, mdev);
1667			if (err)
1668				return notifier_from_errno(err);
1669		}
1670		break;
1671	}
1672	return NOTIFY_OK;
1673}
1674
1675static struct notifier_block mpls_dev_notifier = {
1676	.notifier_call = mpls_dev_notify,
1677};
1678
1679static int nla_put_via(struct sk_buff *skb,
1680		       u8 table, const void *addr, int alen)
1681{
1682	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1683		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1684	};
1685	struct nlattr *nla;
1686	struct rtvia *via;
1687	int family = AF_UNSPEC;
1688
1689	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1690	if (!nla)
1691		return -EMSGSIZE;
1692
1693	if (table <= NEIGH_NR_TABLES)
1694		family = table_to_family[table];
1695
1696	via = nla_data(nla);
1697	via->rtvia_family = family;
1698	memcpy(via->rtvia_addr, addr, alen);
1699	return 0;
1700}
1701
1702int nla_put_labels(struct sk_buff *skb, int attrtype,
1703		   u8 labels, const u32 label[])
1704{
1705	struct nlattr *nla;
1706	struct mpls_shim_hdr *nla_label;
1707	bool bos;
1708	int i;
1709	nla = nla_reserve(skb, attrtype, labels*4);
1710	if (!nla)
1711		return -EMSGSIZE;
1712
1713	nla_label = nla_data(nla);
1714	bos = true;
1715	for (i = labels - 1; i >= 0; i--) {
1716		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1717		bos = false;
1718	}
1719
1720	return 0;
1721}
1722EXPORT_SYMBOL_GPL(nla_put_labels);
1723
1724int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1725		   u32 label[], struct netlink_ext_ack *extack)
1726{
1727	unsigned len = nla_len(nla);
1728	struct mpls_shim_hdr *nla_label;
1729	u8 nla_labels;
1730	bool bos;
1731	int i;
1732
1733	/* len needs to be an even multiple of 4 (the label size). Number
1734	 * of labels is a u8 so check for overflow.
1735	 */
1736	if (len & 3 || len / 4 > 255) {
1737		NL_SET_ERR_MSG_ATTR(extack, nla,
1738				    "Invalid length for labels attribute");
1739		return -EINVAL;
1740	}
1741
1742	/* Limit the number of new labels allowed */
1743	nla_labels = len/4;
1744	if (nla_labels > max_labels) {
1745		NL_SET_ERR_MSG(extack, "Too many labels");
1746		return -EINVAL;
1747	}
1748
1749	/* when label == NULL, caller wants number of labels */
1750	if (!label)
1751		goto out;
1752
1753	nla_label = nla_data(nla);
1754	bos = true;
1755	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1756		struct mpls_entry_decoded dec;
1757		dec = mpls_entry_decode(nla_label + i);
1758
1759		/* Ensure the bottom of stack flag is properly set
1760		 * and ttl and tc are both clear.
1761		 */
1762		if (dec.ttl) {
1763			NL_SET_ERR_MSG_ATTR(extack, nla,
1764					    "TTL in label must be 0");
1765			return -EINVAL;
1766		}
1767
1768		if (dec.tc) {
1769			NL_SET_ERR_MSG_ATTR(extack, nla,
1770					    "Traffic class in label must be 0");
1771			return -EINVAL;
1772		}
1773
1774		if (dec.bos != bos) {
1775			NL_SET_BAD_ATTR(extack, nla);
1776			if (bos) {
1777				NL_SET_ERR_MSG(extack,
1778					       "BOS bit must be set in first label");
1779			} else {
1780				NL_SET_ERR_MSG(extack,
1781					       "BOS bit can only be set in first label");
1782			}
1783			return -EINVAL;
1784		}
1785
1786		switch (dec.label) {
1787		case MPLS_LABEL_IMPLNULL:
1788			/* RFC3032: This is a label that an LSR may
1789			 * assign and distribute, but which never
1790			 * actually appears in the encapsulation.
1791			 */
1792			NL_SET_ERR_MSG_ATTR(extack, nla,
1793					    "Implicit NULL Label (3) can not be used in encapsulation");
1794			return -EINVAL;
1795		}
1796
1797		label[i] = dec.label;
1798	}
1799out:
1800	*labels = nla_labels;
1801	return 0;
1802}
1803EXPORT_SYMBOL_GPL(nla_get_labels);
1804
1805static int rtm_to_route_config(struct sk_buff *skb,
1806			       struct nlmsghdr *nlh,
1807			       struct mpls_route_config *cfg,
1808			       struct netlink_ext_ack *extack)
1809{
1810	struct rtmsg *rtm;
1811	struct nlattr *tb[RTA_MAX+1];
1812	int index;
1813	int err;
1814
1815	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1816				     rtm_mpls_policy, extack);
1817	if (err < 0)
1818		goto errout;
1819
1820	err = -EINVAL;
1821	rtm = nlmsg_data(nlh);
1822
1823	if (rtm->rtm_family != AF_MPLS) {
1824		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1825		goto errout;
1826	}
1827	if (rtm->rtm_dst_len != 20) {
1828		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1829		goto errout;
1830	}
1831	if (rtm->rtm_src_len != 0) {
1832		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1833		goto errout;
1834	}
1835	if (rtm->rtm_tos != 0) {
1836		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1837		goto errout;
1838	}
1839	if (rtm->rtm_table != RT_TABLE_MAIN) {
1840		NL_SET_ERR_MSG(extack,
1841			       "MPLS only supports the main route table");
1842		goto errout;
1843	}
1844	/* Any value is acceptable for rtm_protocol */
1845
1846	/* As mpls uses destination specific addresses
1847	 * (or source specific address in the case of multicast)
1848	 * all addresses have universal scope.
1849	 */
1850	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1851		NL_SET_ERR_MSG(extack,
1852			       "Invalid route scope  - MPLS only supports UNIVERSE");
1853		goto errout;
1854	}
1855	if (rtm->rtm_type != RTN_UNICAST) {
1856		NL_SET_ERR_MSG(extack,
1857			       "Invalid route type - MPLS only supports UNICAST");
1858		goto errout;
1859	}
1860	if (rtm->rtm_flags != 0) {
1861		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1862		goto errout;
1863	}
1864
1865	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1866	cfg->rc_protocol	= rtm->rtm_protocol;
1867	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1868	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1869	cfg->rc_nlflags		= nlh->nlmsg_flags;
1870	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1871	cfg->rc_nlinfo.nlh	= nlh;
1872	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1873
1874	for (index = 0; index <= RTA_MAX; index++) {
1875		struct nlattr *nla = tb[index];
1876		if (!nla)
1877			continue;
1878
1879		switch (index) {
1880		case RTA_OIF:
1881			cfg->rc_ifindex = nla_get_u32(nla);
1882			break;
1883		case RTA_NEWDST:
1884			if (nla_get_labels(nla, MAX_NEW_LABELS,
1885					   &cfg->rc_output_labels,
1886					   cfg->rc_output_label, extack))
1887				goto errout;
1888			break;
1889		case RTA_DST:
1890		{
1891			u8 label_count;
1892			if (nla_get_labels(nla, 1, &label_count,
1893					   &cfg->rc_label, extack))
1894				goto errout;
1895
1896			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1897					   &cfg->rc_label, extack))
1898				goto errout;
1899			break;
1900		}
1901		case RTA_GATEWAY:
1902			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1903			goto errout;
1904		case RTA_VIA:
1905		{
1906			if (nla_get_via(nla, &cfg->rc_via_alen,
1907					&cfg->rc_via_table, cfg->rc_via,
1908					extack))
1909				goto errout;
1910			break;
1911		}
1912		case RTA_MULTIPATH:
1913		{
1914			cfg->rc_mp = nla_data(nla);
1915			cfg->rc_mp_len = nla_len(nla);
1916			break;
1917		}
1918		case RTA_TTL_PROPAGATE:
1919		{
1920			u8 ttl_propagate = nla_get_u8(nla);
1921
1922			if (ttl_propagate > 1) {
1923				NL_SET_ERR_MSG_ATTR(extack, nla,
1924						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1925				goto errout;
1926			}
1927			cfg->rc_ttl_propagate = ttl_propagate ?
1928				MPLS_TTL_PROP_ENABLED :
1929				MPLS_TTL_PROP_DISABLED;
1930			break;
1931		}
1932		default:
1933			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1934			/* Unsupported attribute */
1935			goto errout;
1936		}
1937	}
1938
1939	err = 0;
1940errout:
1941	return err;
1942}
1943
1944static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1945			     struct netlink_ext_ack *extack)
1946{
1947	struct mpls_route_config *cfg;
1948	int err;
1949
1950	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1951	if (!cfg)
1952		return -ENOMEM;
1953
1954	err = rtm_to_route_config(skb, nlh, cfg, extack);
1955	if (err < 0)
1956		goto out;
1957
1958	err = mpls_route_del(cfg, extack);
1959out:
1960	kfree(cfg);
1961
1962	return err;
1963}
1964
1965
1966static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1967			     struct netlink_ext_ack *extack)
1968{
1969	struct mpls_route_config *cfg;
1970	int err;
1971
1972	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1973	if (!cfg)
1974		return -ENOMEM;
1975
1976	err = rtm_to_route_config(skb, nlh, cfg, extack);
1977	if (err < 0)
1978		goto out;
1979
1980	err = mpls_route_add(cfg, extack);
1981out:
1982	kfree(cfg);
1983
1984	return err;
1985}
1986
1987static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1988			   u32 label, struct mpls_route *rt, int flags)
1989{
1990	struct net_device *dev;
1991	struct nlmsghdr *nlh;
1992	struct rtmsg *rtm;
1993
1994	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1995	if (nlh == NULL)
1996		return -EMSGSIZE;
1997
1998	rtm = nlmsg_data(nlh);
1999	rtm->rtm_family = AF_MPLS;
2000	rtm->rtm_dst_len = 20;
2001	rtm->rtm_src_len = 0;
2002	rtm->rtm_tos = 0;
2003	rtm->rtm_table = RT_TABLE_MAIN;
2004	rtm->rtm_protocol = rt->rt_protocol;
2005	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
2006	rtm->rtm_type = RTN_UNICAST;
2007	rtm->rtm_flags = 0;
2008
2009	if (nla_put_labels(skb, RTA_DST, 1, &label))
2010		goto nla_put_failure;
2011
2012	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
2013		bool ttl_propagate =
2014			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2015
2016		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2017			       ttl_propagate))
2018			goto nla_put_failure;
2019	}
2020	if (rt->rt_nhn == 1) {
2021		const struct mpls_nh *nh = rt->rt_nh;
2022
2023		if (nh->nh_labels &&
2024		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2025				   nh->nh_label))
2026			goto nla_put_failure;
2027		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2028		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2029				nh->nh_via_alen))
2030			goto nla_put_failure;
2031		dev = nh->nh_dev;
2032		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2033			goto nla_put_failure;
2034		if (nh->nh_flags & RTNH_F_LINKDOWN)
2035			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2036		if (nh->nh_flags & RTNH_F_DEAD)
2037			rtm->rtm_flags |= RTNH_F_DEAD;
2038	} else {
2039		struct rtnexthop *rtnh;
2040		struct nlattr *mp;
2041		u8 linkdown = 0;
2042		u8 dead = 0;
2043
2044		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2045		if (!mp)
2046			goto nla_put_failure;
2047
2048		for_nexthops(rt) {
2049			dev = nh->nh_dev;
2050			if (!dev)
2051				continue;
2052
2053			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2054			if (!rtnh)
2055				goto nla_put_failure;
2056
2057			rtnh->rtnh_ifindex = dev->ifindex;
2058			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2059				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2060				linkdown++;
2061			}
2062			if (nh->nh_flags & RTNH_F_DEAD) {
2063				rtnh->rtnh_flags |= RTNH_F_DEAD;
2064				dead++;
2065			}
2066
2067			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2068							    nh->nh_labels,
2069							    nh->nh_label))
2070				goto nla_put_failure;
2071			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2072			    nla_put_via(skb, nh->nh_via_table,
2073					mpls_nh_via(rt, nh),
2074					nh->nh_via_alen))
2075				goto nla_put_failure;
2076
2077			/* length of rtnetlink header + attributes */
2078			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2079		} endfor_nexthops(rt);
2080
2081		if (linkdown == rt->rt_nhn)
2082			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2083		if (dead == rt->rt_nhn)
2084			rtm->rtm_flags |= RTNH_F_DEAD;
2085
2086		nla_nest_end(skb, mp);
2087	}
2088
2089	nlmsg_end(skb, nlh);
2090	return 0;
2091
2092nla_put_failure:
2093	nlmsg_cancel(skb, nlh);
2094	return -EMSGSIZE;
2095}
2096
2097#if IS_ENABLED(CONFIG_INET)
2098static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2099				   struct fib_dump_filter *filter,
2100				   struct netlink_callback *cb)
2101{
2102	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2103}
2104#else
2105static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2106				   struct fib_dump_filter *filter,
2107				   struct netlink_callback *cb)
2108{
2109	struct netlink_ext_ack *extack = cb->extack;
2110	struct nlattr *tb[RTA_MAX + 1];
2111	struct rtmsg *rtm;
2112	int err, i;
2113
2114	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2115		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2116		return -EINVAL;
2117	}
2118
2119	rtm = nlmsg_data(nlh);
2120	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2121	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2122	    rtm->rtm_flags) {
2123		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2124		return -EINVAL;
2125	}
2126
2127	if (rtm->rtm_protocol) {
2128		filter->protocol = rtm->rtm_protocol;
2129		filter->filter_set = 1;
2130		cb->answer_flags = NLM_F_DUMP_FILTERED;
2131	}
2132
2133	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2134					    rtm_mpls_policy, extack);
2135	if (err < 0)
2136		return err;
2137
2138	for (i = 0; i <= RTA_MAX; ++i) {
2139		int ifindex;
2140
2141		if (i == RTA_OIF) {
2142			ifindex = nla_get_u32(tb[i]);
2143			filter->dev = __dev_get_by_index(net, ifindex);
2144			if (!filter->dev)
2145				return -ENODEV;
2146			filter->filter_set = 1;
2147		} else if (tb[i]) {
2148			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2149			return -EINVAL;
2150		}
2151	}
2152
2153	return 0;
2154}
2155#endif
2156
2157static bool mpls_rt_uses_dev(struct mpls_route *rt,
2158			     const struct net_device *dev)
2159{
2160	if (rt->rt_nhn == 1) {
2161		struct mpls_nh *nh = rt->rt_nh;
2162
2163		if (nh->nh_dev == dev)
2164			return true;
2165	} else {
2166		for_nexthops(rt) {
2167			if (nh->nh_dev == dev)
2168				return true;
2169		} endfor_nexthops(rt);
2170	}
2171
2172	return false;
2173}
2174
2175static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2176{
2177	const struct nlmsghdr *nlh = cb->nlh;
2178	struct net *net = sock_net(skb->sk);
2179	struct mpls_route __rcu **platform_label;
2180	struct fib_dump_filter filter = {};
2181	unsigned int flags = NLM_F_MULTI;
2182	size_t platform_labels;
2183	unsigned int index;
2184
2185	ASSERT_RTNL();
2186
2187	if (cb->strict_check) {
2188		int err;
2189
2190		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2191		if (err < 0)
2192			return err;
2193
2194		/* for MPLS, there is only 1 table with fixed type and flags.
2195		 * If either are set in the filter then return nothing.
2196		 */
2197		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2198		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2199		     filter.flags)
2200			return skb->len;
2201	}
2202
2203	index = cb->args[0];
2204	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2205		index = MPLS_LABEL_FIRST_UNRESERVED;
2206
2207	platform_label = rtnl_dereference(net->mpls.platform_label);
2208	platform_labels = net->mpls.platform_labels;
2209
2210	if (filter.filter_set)
2211		flags |= NLM_F_DUMP_FILTERED;
2212
2213	for (; index < platform_labels; index++) {
2214		struct mpls_route *rt;
2215
2216		rt = rtnl_dereference(platform_label[index]);
2217		if (!rt)
2218			continue;
2219
2220		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2221		    (filter.protocol && rt->rt_protocol != filter.protocol))
2222			continue;
2223
2224		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2225				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2226				    index, rt, flags) < 0)
2227			break;
2228	}
2229	cb->args[0] = index;
2230
2231	return skb->len;
2232}
2233
2234static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2235{
2236	size_t payload =
2237		NLMSG_ALIGN(sizeof(struct rtmsg))
2238		+ nla_total_size(4)			/* RTA_DST */
2239		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2240
2241	if (rt->rt_nhn == 1) {
2242		struct mpls_nh *nh = rt->rt_nh;
2243
2244		if (nh->nh_dev)
2245			payload += nla_total_size(4); /* RTA_OIF */
2246		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2247			payload += nla_total_size(2 + nh->nh_via_alen);
2248		if (nh->nh_labels) /* RTA_NEWDST */
2249			payload += nla_total_size(nh->nh_labels * 4);
2250	} else {
2251		/* each nexthop is packed in an attribute */
2252		size_t nhsize = 0;
2253
2254		for_nexthops(rt) {
2255			if (!nh->nh_dev)
2256				continue;
2257			nhsize += nla_total_size(sizeof(struct rtnexthop));
2258			/* RTA_VIA */
2259			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2260				nhsize += nla_total_size(2 + nh->nh_via_alen);
2261			if (nh->nh_labels)
2262				nhsize += nla_total_size(nh->nh_labels * 4);
2263		} endfor_nexthops(rt);
2264		/* nested attribute */
2265		payload += nla_total_size(nhsize);
2266	}
2267
2268	return payload;
2269}
2270
2271static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2272		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2273		       unsigned int nlm_flags)
2274{
2275	struct sk_buff *skb;
2276	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2277	int err = -ENOBUFS;
2278
2279	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2280	if (skb == NULL)
2281		goto errout;
2282
2283	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2284	if (err < 0) {
2285		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2286		WARN_ON(err == -EMSGSIZE);
2287		kfree_skb(skb);
2288		goto errout;
2289	}
2290	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2291
2292	return;
2293errout:
2294	if (err < 0)
2295		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2296}
2297
2298static int mpls_valid_getroute_req(struct sk_buff *skb,
2299				   const struct nlmsghdr *nlh,
2300				   struct nlattr **tb,
2301				   struct netlink_ext_ack *extack)
2302{
2303	struct rtmsg *rtm;
2304	int i, err;
2305
2306	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2307		NL_SET_ERR_MSG_MOD(extack,
2308				   "Invalid header for get route request");
2309		return -EINVAL;
2310	}
2311
2312	if (!netlink_strict_get_check(skb))
2313		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2314					      rtm_mpls_policy, extack);
2315
2316	rtm = nlmsg_data(nlh);
2317	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2318	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2319	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2320		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2321		return -EINVAL;
2322	}
2323	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2324		NL_SET_ERR_MSG_MOD(extack,
2325				   "Invalid flags for get route request");
2326		return -EINVAL;
2327	}
2328
2329	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2330					    rtm_mpls_policy, extack);
2331	if (err)
2332		return err;
2333
2334	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2335		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2336		return -EINVAL;
2337	}
2338
2339	for (i = 0; i <= RTA_MAX; i++) {
2340		if (!tb[i])
2341			continue;
2342
2343		switch (i) {
2344		case RTA_DST:
2345		case RTA_NEWDST:
2346			break;
2347		default:
2348			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2349			return -EINVAL;
2350		}
2351	}
2352
2353	return 0;
2354}
2355
2356static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2357			 struct netlink_ext_ack *extack)
2358{
2359	struct net *net = sock_net(in_skb->sk);
2360	u32 portid = NETLINK_CB(in_skb).portid;
2361	u32 in_label = LABEL_NOT_SPECIFIED;
2362	struct nlattr *tb[RTA_MAX + 1];
2363	u32 labels[MAX_NEW_LABELS];
2364	struct mpls_shim_hdr *hdr;
2365	unsigned int hdr_size = 0;
2366	const struct mpls_nh *nh;
2367	struct net_device *dev;
2368	struct mpls_route *rt;
2369	struct rtmsg *rtm, *r;
2370	struct nlmsghdr *nlh;
2371	struct sk_buff *skb;
2372	u8 n_labels;
2373	int err;
2374
2375	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2376	if (err < 0)
2377		goto errout;
2378
2379	rtm = nlmsg_data(in_nlh);
2380
2381	if (tb[RTA_DST]) {
2382		u8 label_count;
2383
2384		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2385				   &in_label, extack)) {
2386			err = -EINVAL;
2387			goto errout;
2388		}
2389
2390		if (!mpls_label_ok(net, &in_label, extack)) {
2391			err = -EINVAL;
2392			goto errout;
2393		}
2394	}
2395
2396	rt = mpls_route_input_rcu(net, in_label);
2397	if (!rt) {
2398		err = -ENETUNREACH;
2399		goto errout;
2400	}
2401
2402	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2403		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2404		if (!skb) {
2405			err = -ENOBUFS;
2406			goto errout;
2407		}
2408
2409		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2410				      RTM_NEWROUTE, in_label, rt, 0);
2411		if (err < 0) {
2412			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2413			WARN_ON(err == -EMSGSIZE);
2414			goto errout_free;
2415		}
2416
2417		return rtnl_unicast(skb, net, portid);
2418	}
2419
2420	if (tb[RTA_NEWDST]) {
2421		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2422				   labels, extack) != 0) {
2423			err = -EINVAL;
2424			goto errout;
2425		}
2426
2427		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2428	}
2429
2430	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2431	if (!skb) {
2432		err = -ENOBUFS;
2433		goto errout;
2434	}
2435
2436	skb->protocol = htons(ETH_P_MPLS_UC);
2437
2438	if (hdr_size) {
2439		bool bos;
2440		int i;
2441
2442		if (skb_cow(skb, hdr_size)) {
2443			err = -ENOBUFS;
2444			goto errout_free;
2445		}
2446
2447		skb_reserve(skb, hdr_size);
2448		skb_push(skb, hdr_size);
2449		skb_reset_network_header(skb);
2450
2451		/* Push new labels */
2452		hdr = mpls_hdr(skb);
2453		bos = true;
2454		for (i = n_labels - 1; i >= 0; i--) {
2455			hdr[i] = mpls_entry_encode(labels[i],
2456						   1, 0, bos);
2457			bos = false;
2458		}
2459	}
2460
2461	nh = mpls_select_multipath(rt, skb);
2462	if (!nh) {
2463		err = -ENETUNREACH;
2464		goto errout_free;
2465	}
2466
2467	if (hdr_size) {
2468		skb_pull(skb, hdr_size);
2469		skb_reset_network_header(skb);
2470	}
2471
2472	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2473			RTM_NEWROUTE, sizeof(*r), 0);
2474	if (!nlh) {
2475		err = -EMSGSIZE;
2476		goto errout_free;
2477	}
2478
2479	r = nlmsg_data(nlh);
2480	r->rtm_family	 = AF_MPLS;
2481	r->rtm_dst_len	= 20;
2482	r->rtm_src_len	= 0;
2483	r->rtm_table	= RT_TABLE_MAIN;
2484	r->rtm_type	= RTN_UNICAST;
2485	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2486	r->rtm_protocol = rt->rt_protocol;
2487	r->rtm_flags	= 0;
2488
2489	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2490		goto nla_put_failure;
2491
2492	if (nh->nh_labels &&
2493	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2494			   nh->nh_label))
2495		goto nla_put_failure;
2496
2497	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2498	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2499			nh->nh_via_alen))
2500		goto nla_put_failure;
2501	dev = nh->nh_dev;
2502	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2503		goto nla_put_failure;
2504
2505	nlmsg_end(skb, nlh);
2506
2507	err = rtnl_unicast(skb, net, portid);
2508errout:
2509	return err;
2510
2511nla_put_failure:
2512	nlmsg_cancel(skb, nlh);
2513	err = -EMSGSIZE;
2514errout_free:
2515	kfree_skb(skb);
2516	return err;
2517}
2518
2519static int resize_platform_label_table(struct net *net, size_t limit)
2520{
2521	size_t size = sizeof(struct mpls_route *) * limit;
2522	size_t old_limit;
2523	size_t cp_size;
2524	struct mpls_route __rcu **labels = NULL, **old;
2525	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2526	unsigned index;
2527
2528	if (size) {
2529		labels = kvzalloc(size, GFP_KERNEL);
2530		if (!labels)
2531			goto nolabels;
2532	}
2533
2534	/* In case the predefined labels need to be populated */
2535	if (limit > MPLS_LABEL_IPV4NULL) {
2536		struct net_device *lo = net->loopback_dev;
2537		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2538		if (IS_ERR(rt0))
2539			goto nort0;
2540		rt0->rt_nh->nh_dev = lo;
2541		rt0->rt_protocol = RTPROT_KERNEL;
2542		rt0->rt_payload_type = MPT_IPV4;
2543		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2544		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2545		rt0->rt_nh->nh_via_alen = lo->addr_len;
2546		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2547		       lo->addr_len);
2548	}
2549	if (limit > MPLS_LABEL_IPV6NULL) {
2550		struct net_device *lo = net->loopback_dev;
2551		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2552		if (IS_ERR(rt2))
2553			goto nort2;
2554		rt2->rt_nh->nh_dev = lo;
2555		rt2->rt_protocol = RTPROT_KERNEL;
2556		rt2->rt_payload_type = MPT_IPV6;
2557		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2558		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2559		rt2->rt_nh->nh_via_alen = lo->addr_len;
2560		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2561		       lo->addr_len);
2562	}
2563
2564	rtnl_lock();
2565	/* Remember the original table */
2566	old = rtnl_dereference(net->mpls.platform_label);
2567	old_limit = net->mpls.platform_labels;
2568
2569	/* Free any labels beyond the new table */
2570	for (index = limit; index < old_limit; index++)
2571		mpls_route_update(net, index, NULL, NULL);
2572
2573	/* Copy over the old labels */
2574	cp_size = size;
2575	if (old_limit < limit)
2576		cp_size = old_limit * sizeof(struct mpls_route *);
2577
2578	memcpy(labels, old, cp_size);
2579
2580	/* If needed set the predefined labels */
2581	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2582	    (limit > MPLS_LABEL_IPV6NULL)) {
2583		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2584		rt2 = NULL;
2585	}
2586
2587	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2588	    (limit > MPLS_LABEL_IPV4NULL)) {
2589		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2590		rt0 = NULL;
2591	}
2592
2593	/* Update the global pointers */
2594	net->mpls.platform_labels = limit;
2595	rcu_assign_pointer(net->mpls.platform_label, labels);
2596
2597	rtnl_unlock();
2598
2599	mpls_rt_free(rt2);
2600	mpls_rt_free(rt0);
2601
2602	if (old) {
2603		synchronize_rcu();
2604		kvfree(old);
2605	}
2606	return 0;
2607
2608nort2:
2609	mpls_rt_free(rt0);
2610nort0:
2611	kvfree(labels);
2612nolabels:
2613	return -ENOMEM;
2614}
2615
2616static int mpls_platform_labels(struct ctl_table *table, int write,
2617				void *buffer, size_t *lenp, loff_t *ppos)
2618{
2619	struct net *net = table->data;
2620	int platform_labels = net->mpls.platform_labels;
2621	int ret;
2622	struct ctl_table tmp = {
2623		.procname	= table->procname,
2624		.data		= &platform_labels,
2625		.maxlen		= sizeof(int),
2626		.mode		= table->mode,
2627		.extra1		= SYSCTL_ZERO,
2628		.extra2		= &label_limit,
2629	};
2630
2631	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2632
2633	if (write && ret == 0)
2634		ret = resize_platform_label_table(net, platform_labels);
2635
2636	return ret;
2637}
2638
2639#define MPLS_NS_SYSCTL_OFFSET(field)		\
2640	(&((struct net *)0)->field)
2641
2642static const struct ctl_table mpls_table[] = {
2643	{
2644		.procname	= "platform_labels",
2645		.data		= NULL,
2646		.maxlen		= sizeof(int),
2647		.mode		= 0644,
2648		.proc_handler	= mpls_platform_labels,
2649	},
2650	{
2651		.procname	= "ip_ttl_propagate",
2652		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2653		.maxlen		= sizeof(int),
2654		.mode		= 0644,
2655		.proc_handler	= proc_dointvec_minmax,
2656		.extra1		= SYSCTL_ZERO,
2657		.extra2		= SYSCTL_ONE,
2658	},
2659	{
2660		.procname	= "default_ttl",
2661		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2662		.maxlen		= sizeof(int),
2663		.mode		= 0644,
2664		.proc_handler	= proc_dointvec_minmax,
2665		.extra1		= SYSCTL_ONE,
2666		.extra2		= &ttl_max,
2667	},
2668	{ }
2669};
2670
2671static int mpls_net_init(struct net *net)
2672{
2673	struct ctl_table *table;
2674	int i;
2675
2676	net->mpls.platform_labels = 0;
2677	net->mpls.platform_label = NULL;
2678	net->mpls.ip_ttl_propagate = 1;
2679	net->mpls.default_ttl = 255;
2680
2681	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2682	if (table == NULL)
2683		return -ENOMEM;
2684
2685	/* Table data contains only offsets relative to the base of
2686	 * the mdev at this point, so make them absolute.
2687	 */
2688	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2689		table[i].data = (char *)net + (uintptr_t)table[i].data;
2690
2691	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
 
2692	if (net->mpls.ctl == NULL) {
2693		kfree(table);
2694		return -ENOMEM;
2695	}
2696
2697	return 0;
2698}
2699
2700static void mpls_net_exit(struct net *net)
2701{
2702	struct mpls_route __rcu **platform_label;
2703	size_t platform_labels;
2704	struct ctl_table *table;
2705	unsigned int index;
2706
2707	table = net->mpls.ctl->ctl_table_arg;
2708	unregister_net_sysctl_table(net->mpls.ctl);
2709	kfree(table);
2710
2711	/* An rcu grace period has passed since there was a device in
2712	 * the network namespace (and thus the last in flight packet)
2713	 * left this network namespace.  This is because
2714	 * unregister_netdevice_many and netdev_run_todo has completed
2715	 * for each network device that was in this network namespace.
2716	 *
2717	 * As such no additional rcu synchronization is necessary when
2718	 * freeing the platform_label table.
2719	 */
2720	rtnl_lock();
2721	platform_label = rtnl_dereference(net->mpls.platform_label);
2722	platform_labels = net->mpls.platform_labels;
2723	for (index = 0; index < platform_labels; index++) {
2724		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2725		RCU_INIT_POINTER(platform_label[index], NULL);
2726		mpls_notify_route(net, index, rt, NULL, NULL);
2727		mpls_rt_free(rt);
2728	}
2729	rtnl_unlock();
2730
2731	kvfree(platform_label);
2732}
2733
2734static struct pernet_operations mpls_net_ops = {
2735	.init = mpls_net_init,
2736	.exit = mpls_net_exit,
2737};
2738
2739static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2740	.family		   = AF_MPLS,
2741	.fill_stats_af	   = mpls_fill_stats_af,
2742	.get_stats_af_size = mpls_get_stats_af_size,
2743};
2744
2745static int __init mpls_init(void)
2746{
2747	int err;
2748
2749	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2750
2751	err = register_pernet_subsys(&mpls_net_ops);
2752	if (err)
2753		goto out;
2754
2755	err = register_netdevice_notifier(&mpls_dev_notifier);
2756	if (err)
2757		goto out_unregister_pernet;
2758
2759	dev_add_pack(&mpls_packet_type);
2760
2761	rtnl_af_register(&mpls_af_ops);
2762
2763	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2764			     mpls_rtm_newroute, NULL, 0);
2765	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2766			     mpls_rtm_delroute, NULL, 0);
2767	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2768			     mpls_getroute, mpls_dump_routes, 0);
2769	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2770			     mpls_netconf_get_devconf,
2771			     mpls_netconf_dump_devconf, 0);
2772	err = ipgre_tunnel_encap_add_mpls_ops();
2773	if (err)
2774		pr_err("Can't add mpls over gre tunnel ops\n");
2775
2776	err = 0;
2777out:
2778	return err;
2779
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);