Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Management Component Transport Protocol (MCTP) - routing
   4 * implementation.
   5 *
   6 * This is currently based on a simple routing table, with no dst cache. The
   7 * number of routes should stay fairly small, so the lookup cost is small.
   8 *
   9 * Copyright (c) 2021 Code Construct
  10 * Copyright (c) 2021 Google
  11 */
  12
  13#include <linux/idr.h>
  14#include <linux/kconfig.h>
  15#include <linux/mctp.h>
  16#include <linux/netdevice.h>
  17#include <linux/rtnetlink.h>
  18#include <linux/skbuff.h>
  19
  20#include <uapi/linux/if_arp.h>
  21
  22#include <net/mctp.h>
  23#include <net/mctpdevice.h>
  24#include <net/netlink.h>
  25#include <net/sock.h>
  26
  27#include <trace/events/mctp.h>
  28
  29static const unsigned int mctp_message_maxlen = 64 * 1024;
  30static const unsigned long mctp_key_lifetime = 6 * CONFIG_HZ;
  31
  32static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev);
  33
  34/* route output callbacks */
  35static int mctp_route_discard(struct mctp_route *route, struct sk_buff *skb)
  36{
  37	kfree_skb(skb);
  38	return 0;
  39}
  40
  41static struct mctp_sock *mctp_lookup_bind(struct net *net, struct sk_buff *skb)
  42{
  43	struct mctp_skb_cb *cb = mctp_cb(skb);
  44	struct mctp_hdr *mh;
  45	struct sock *sk;
  46	u8 type;
  47
  48	WARN_ON(!rcu_read_lock_held());
  49
  50	/* TODO: look up in skb->cb? */
  51	mh = mctp_hdr(skb);
  52
  53	if (!skb_headlen(skb))
  54		return NULL;
  55
  56	type = (*(u8 *)skb->data) & 0x7f;
  57
  58	sk_for_each_rcu(sk, &net->mctp.binds) {
  59		struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
  60
  61		if (msk->bind_net != MCTP_NET_ANY && msk->bind_net != cb->net)
  62			continue;
  63
  64		if (msk->bind_type != type)
  65			continue;
  66
  67		if (!mctp_address_matches(msk->bind_addr, mh->dest))
  68			continue;
  69
  70		return msk;
  71	}
  72
  73	return NULL;
  74}
  75
  76static bool mctp_key_match(struct mctp_sk_key *key, mctp_eid_t local,
  77			   mctp_eid_t peer, u8 tag)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  78{
 
 
 
  79	if (!mctp_address_matches(key->local_addr, local))
  80		return false;
  81
  82	if (key->peer_addr != peer)
  83		return false;
  84
  85	if (key->tag != tag)
  86		return false;
  87
  88	return true;
  89}
  90
  91/* returns a key (with key->lock held, and refcounted), or NULL if no such
  92 * key exists.
  93 */
  94static struct mctp_sk_key *mctp_lookup_key(struct net *net, struct sk_buff *skb,
  95					   mctp_eid_t peer,
  96					   unsigned long *irqflags)
  97	__acquires(&key->lock)
  98{
  99	struct mctp_sk_key *key, *ret;
 100	unsigned long flags;
 101	struct mctp_hdr *mh;
 102	u8 tag;
 103
 104	mh = mctp_hdr(skb);
 105	tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 106
 107	ret = NULL;
 108	spin_lock_irqsave(&net->mctp.keys_lock, flags);
 109
 110	hlist_for_each_entry(key, &net->mctp.keys, hlist) {
 111		if (!mctp_key_match(key, mh->dest, peer, tag))
 112			continue;
 113
 114		spin_lock(&key->lock);
 115		if (key->valid) {
 116			refcount_inc(&key->refs);
 117			ret = key;
 118			break;
 119		}
 120		spin_unlock(&key->lock);
 121	}
 122
 123	if (ret) {
 124		spin_unlock(&net->mctp.keys_lock);
 125		*irqflags = flags;
 126	} else {
 127		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 128	}
 129
 130	return ret;
 131}
 132
 133static struct mctp_sk_key *mctp_key_alloc(struct mctp_sock *msk,
 
 134					  mctp_eid_t local, mctp_eid_t peer,
 135					  u8 tag, gfp_t gfp)
 136{
 137	struct mctp_sk_key *key;
 138
 139	key = kzalloc(sizeof(*key), gfp);
 140	if (!key)
 141		return NULL;
 142
 
 143	key->peer_addr = peer;
 144	key->local_addr = local;
 145	key->tag = tag;
 146	key->sk = &msk->sk;
 147	key->valid = true;
 148	spin_lock_init(&key->lock);
 149	refcount_set(&key->refs, 1);
 150	sock_hold(key->sk);
 151
 152	return key;
 153}
 154
 155void mctp_key_unref(struct mctp_sk_key *key)
 156{
 157	unsigned long flags;
 158
 159	if (!refcount_dec_and_test(&key->refs))
 160		return;
 161
 162	/* even though no refs exist here, the lock allows us to stay
 163	 * consistent with the locking requirement of mctp_dev_release_key
 164	 */
 165	spin_lock_irqsave(&key->lock, flags);
 166	mctp_dev_release_key(key->dev, key);
 167	spin_unlock_irqrestore(&key->lock, flags);
 168
 169	sock_put(key->sk);
 170	kfree(key);
 171}
 172
 173static int mctp_key_add(struct mctp_sk_key *key, struct mctp_sock *msk)
 174{
 175	struct net *net = sock_net(&msk->sk);
 176	struct mctp_sk_key *tmp;
 177	unsigned long flags;
 178	int rc = 0;
 179
 180	spin_lock_irqsave(&net->mctp.keys_lock, flags);
 181
 182	if (sock_flag(&msk->sk, SOCK_DEAD)) {
 183		rc = -EINVAL;
 184		goto out_unlock;
 185	}
 186
 187	hlist_for_each_entry(tmp, &net->mctp.keys, hlist) {
 188		if (mctp_key_match(tmp, key->local_addr, key->peer_addr,
 189				   key->tag)) {
 190			spin_lock(&tmp->lock);
 191			if (tmp->valid)
 192				rc = -EEXIST;
 193			spin_unlock(&tmp->lock);
 194			if (rc)
 195				break;
 196		}
 197	}
 198
 199	if (!rc) {
 200		refcount_inc(&key->refs);
 201		key->expiry = jiffies + mctp_key_lifetime;
 202		timer_reduce(&msk->key_expiry, key->expiry);
 203
 204		hlist_add_head(&key->hlist, &net->mctp.keys);
 205		hlist_add_head(&key->sklist, &msk->keys);
 206	}
 207
 208out_unlock:
 209	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 210
 211	return rc;
 212}
 213
 214/* Helper for mctp_route_input().
 215 * We're done with the key; unlock and unref the key.
 216 * For the usual case of automatic expiry we remove the key from lists.
 217 * In the case that manual allocation is set on a key we release the lock
 218 * and local ref, reset reassembly, but don't remove from lists.
 219 */
 220static void __mctp_key_done_in(struct mctp_sk_key *key, struct net *net,
 221			       unsigned long flags, unsigned long reason)
 222__releases(&key->lock)
 223{
 224	struct sk_buff *skb;
 225
 226	trace_mctp_key_release(key, reason);
 227	skb = key->reasm_head;
 228	key->reasm_head = NULL;
 229
 230	if (!key->manual_alloc) {
 231		key->reasm_dead = true;
 232		key->valid = false;
 233		mctp_dev_release_key(key->dev, key);
 234	}
 235	spin_unlock_irqrestore(&key->lock, flags);
 236
 237	if (!key->manual_alloc) {
 238		spin_lock_irqsave(&net->mctp.keys_lock, flags);
 239		if (!hlist_unhashed(&key->hlist)) {
 240			hlist_del_init(&key->hlist);
 241			hlist_del_init(&key->sklist);
 242			mctp_key_unref(key);
 243		}
 244		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 245	}
 246
 247	/* and one for the local reference */
 248	mctp_key_unref(key);
 249
 250	kfree_skb(skb);
 251}
 252
 253#ifdef CONFIG_MCTP_FLOWS
 254static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key)
 255{
 256	struct mctp_flow *flow;
 257
 258	flow = skb_ext_add(skb, SKB_EXT_MCTP);
 259	if (!flow)
 260		return;
 261
 262	refcount_inc(&key->refs);
 263	flow->key = key;
 264}
 265
 266static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev)
 267{
 268	struct mctp_sk_key *key;
 269	struct mctp_flow *flow;
 270
 271	flow = skb_ext_find(skb, SKB_EXT_MCTP);
 272	if (!flow)
 273		return;
 274
 275	key = flow->key;
 276
 277	if (WARN_ON(key->dev && key->dev != dev))
 278		return;
 279
 280	mctp_dev_set_key(dev, key);
 281}
 282#else
 283static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key) {}
 284static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev) {}
 285#endif
 286
 287static int mctp_frag_queue(struct mctp_sk_key *key, struct sk_buff *skb)
 288{
 289	struct mctp_hdr *hdr = mctp_hdr(skb);
 290	u8 exp_seq, this_seq;
 291
 292	this_seq = (hdr->flags_seq_tag >> MCTP_HDR_SEQ_SHIFT)
 293		& MCTP_HDR_SEQ_MASK;
 294
 295	if (!key->reasm_head) {
 296		key->reasm_head = skb;
 297		key->reasm_tailp = &(skb_shinfo(skb)->frag_list);
 298		key->last_seq = this_seq;
 299		return 0;
 300	}
 301
 302	exp_seq = (key->last_seq + 1) & MCTP_HDR_SEQ_MASK;
 303
 304	if (this_seq != exp_seq)
 305		return -EINVAL;
 306
 307	if (key->reasm_head->len + skb->len > mctp_message_maxlen)
 308		return -EINVAL;
 309
 310	skb->next = NULL;
 311	skb->sk = NULL;
 312	*key->reasm_tailp = skb;
 313	key->reasm_tailp = &skb->next;
 314
 315	key->last_seq = this_seq;
 316
 317	key->reasm_head->data_len += skb->len;
 318	key->reasm_head->len += skb->len;
 319	key->reasm_head->truesize += skb->truesize;
 320
 321	return 0;
 322}
 323
 324static int mctp_route_input(struct mctp_route *route, struct sk_buff *skb)
 325{
 326	struct mctp_sk_key *key, *any_key = NULL;
 327	struct net *net = dev_net(skb->dev);
 328	struct mctp_sock *msk;
 329	struct mctp_hdr *mh;
 
 330	unsigned long f;
 331	u8 tag, flags;
 332	int rc;
 333
 334	msk = NULL;
 335	rc = -EINVAL;
 336
 337	/* we may be receiving a locally-routed packet; drop source sk
 338	 * accounting
 339	 */
 340	skb_orphan(skb);
 341
 342	/* ensure we have enough data for a header and a type */
 343	if (skb->len < sizeof(struct mctp_hdr) + 1)
 344		goto out;
 345
 346	/* grab header, advance data ptr */
 347	mh = mctp_hdr(skb);
 
 348	skb_pull(skb, sizeof(struct mctp_hdr));
 349
 350	if (mh->ver != 1)
 351		goto out;
 352
 353	flags = mh->flags_seq_tag & (MCTP_HDR_FLAG_SOM | MCTP_HDR_FLAG_EOM);
 354	tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 355
 356	rcu_read_lock();
 357
 358	/* lookup socket / reasm context, exactly matching (src,dest,tag).
 359	 * we hold a ref on the key, and key->lock held.
 360	 */
 361	key = mctp_lookup_key(net, skb, mh->src, &f);
 362
 363	if (flags & MCTP_HDR_FLAG_SOM) {
 364		if (key) {
 365			msk = container_of(key->sk, struct mctp_sock, sk);
 366		} else {
 367			/* first response to a broadcast? do a more general
 368			 * key lookup to find the socket, but don't use this
 369			 * key for reassembly - we'll create a more specific
 370			 * one for future packets if required (ie, !EOM).
 
 
 
 371			 */
 372			any_key = mctp_lookup_key(net, skb, MCTP_ADDR_ANY, &f);
 
 373			if (any_key) {
 374				msk = container_of(any_key->sk,
 375						   struct mctp_sock, sk);
 376				spin_unlock_irqrestore(&any_key->lock, f);
 377			}
 378		}
 379
 380		if (!key && !msk && (tag & MCTP_HDR_FLAG_TO))
 381			msk = mctp_lookup_bind(net, skb);
 382
 383		if (!msk) {
 384			rc = -ENOENT;
 385			goto out_unlock;
 386		}
 387
 388		/* single-packet message? deliver to socket, clean up any
 389		 * pending key.
 390		 */
 391		if (flags & MCTP_HDR_FLAG_EOM) {
 392			sock_queue_rcv_skb(&msk->sk, skb);
 393			if (key) {
 394				/* we've hit a pending reassembly; not much we
 395				 * can do but drop it
 396				 */
 397				__mctp_key_done_in(key, net, f,
 398						   MCTP_TRACE_KEY_REPLIED);
 399				key = NULL;
 400			}
 401			rc = 0;
 402			goto out_unlock;
 403		}
 404
 405		/* broadcast response or a bind() - create a key for further
 406		 * packets for this message
 407		 */
 408		if (!key) {
 409			key = mctp_key_alloc(msk, mh->dest, mh->src,
 410					     tag, GFP_ATOMIC);
 411			if (!key) {
 412				rc = -ENOMEM;
 413				goto out_unlock;
 414			}
 415
 416			/* we can queue without the key lock here, as the
 417			 * key isn't observable yet
 418			 */
 419			mctp_frag_queue(key, skb);
 420
 421			/* if the key_add fails, we've raced with another
 422			 * SOM packet with the same src, dest and tag. There's
 423			 * no way to distinguish future packets, so all we
 424			 * can do is drop; we'll free the skb on exit from
 425			 * this function.
 426			 */
 427			rc = mctp_key_add(key, msk);
 428			if (!rc)
 429				trace_mctp_key_acquire(key);
 430
 431			/* we don't need to release key->lock on exit, so
 432			 * clean up here and suppress the unlock via
 433			 * setting to NULL
 434			 */
 435			mctp_key_unref(key);
 436			key = NULL;
 437
 438		} else {
 439			if (key->reasm_head || key->reasm_dead) {
 440				/* duplicate start? drop everything */
 441				__mctp_key_done_in(key, net, f,
 442						   MCTP_TRACE_KEY_INVALIDATED);
 443				rc = -EEXIST;
 444				key = NULL;
 445			} else {
 446				rc = mctp_frag_queue(key, skb);
 447			}
 448		}
 449
 450	} else if (key) {
 451		/* this packet continues a previous message; reassemble
 452		 * using the message-specific key
 453		 */
 454
 455		/* we need to be continuing an existing reassembly... */
 456		if (!key->reasm_head)
 457			rc = -EINVAL;
 458		else
 459			rc = mctp_frag_queue(key, skb);
 460
 461		/* end of message? deliver to socket, and we're done with
 462		 * the reassembly/response key
 463		 */
 464		if (!rc && flags & MCTP_HDR_FLAG_EOM) {
 465			sock_queue_rcv_skb(key->sk, key->reasm_head);
 466			key->reasm_head = NULL;
 467			__mctp_key_done_in(key, net, f, MCTP_TRACE_KEY_REPLIED);
 468			key = NULL;
 469		}
 470
 471	} else {
 472		/* not a start, no matching key */
 473		rc = -ENOENT;
 474	}
 475
 476out_unlock:
 477	rcu_read_unlock();
 478	if (key) {
 479		spin_unlock_irqrestore(&key->lock, f);
 480		mctp_key_unref(key);
 481	}
 482	if (any_key)
 483		mctp_key_unref(any_key);
 484out:
 485	if (rc)
 486		kfree_skb(skb);
 487	return rc;
 488}
 489
 490static unsigned int mctp_route_mtu(struct mctp_route *rt)
 491{
 492	return rt->mtu ?: READ_ONCE(rt->dev->dev->mtu);
 493}
 494
 495static int mctp_route_output(struct mctp_route *route, struct sk_buff *skb)
 496{
 497	struct mctp_skb_cb *cb = mctp_cb(skb);
 498	struct mctp_hdr *hdr = mctp_hdr(skb);
 499	char daddr_buf[MAX_ADDR_LEN];
 500	char *daddr = NULL;
 501	unsigned int mtu;
 502	int rc;
 503
 504	skb->protocol = htons(ETH_P_MCTP);
 505
 506	mtu = READ_ONCE(skb->dev->mtu);
 507	if (skb->len > mtu) {
 508		kfree_skb(skb);
 509		return -EMSGSIZE;
 510	}
 511
 512	if (cb->ifindex) {
 513		/* direct route; use the hwaddr we stashed in sendmsg */
 514		if (cb->halen != skb->dev->addr_len) {
 515			/* sanity check, sendmsg should have already caught this */
 516			kfree_skb(skb);
 517			return -EMSGSIZE;
 518		}
 519		daddr = cb->haddr;
 520	} else {
 521		/* If lookup fails let the device handle daddr==NULL */
 522		if (mctp_neigh_lookup(route->dev, hdr->dest, daddr_buf) == 0)
 523			daddr = daddr_buf;
 524	}
 525
 526	rc = dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
 527			     daddr, skb->dev->dev_addr, skb->len);
 528	if (rc < 0) {
 529		kfree_skb(skb);
 530		return -EHOSTUNREACH;
 531	}
 532
 533	mctp_flow_prepare_output(skb, route->dev);
 534
 535	rc = dev_queue_xmit(skb);
 536	if (rc)
 537		rc = net_xmit_errno(rc);
 538
 539	return rc;
 540}
 541
 542/* route alloc/release */
 543static void mctp_route_release(struct mctp_route *rt)
 544{
 545	if (refcount_dec_and_test(&rt->refs)) {
 546		mctp_dev_put(rt->dev);
 547		kfree_rcu(rt, rcu);
 548	}
 549}
 550
 551/* returns a route with the refcount at 1 */
 552static struct mctp_route *mctp_route_alloc(void)
 553{
 554	struct mctp_route *rt;
 555
 556	rt = kzalloc(sizeof(*rt), GFP_KERNEL);
 557	if (!rt)
 558		return NULL;
 559
 560	INIT_LIST_HEAD(&rt->list);
 561	refcount_set(&rt->refs, 1);
 562	rt->output = mctp_route_discard;
 563
 564	return rt;
 565}
 566
 567unsigned int mctp_default_net(struct net *net)
 568{
 569	return READ_ONCE(net->mctp.default_net);
 570}
 571
 572int mctp_default_net_set(struct net *net, unsigned int index)
 573{
 574	if (index == 0)
 575		return -EINVAL;
 576	WRITE_ONCE(net->mctp.default_net, index);
 577	return 0;
 578}
 579
 580/* tag management */
 581static void mctp_reserve_tag(struct net *net, struct mctp_sk_key *key,
 582			     struct mctp_sock *msk)
 583{
 584	struct netns_mctp *mns = &net->mctp;
 585
 586	lockdep_assert_held(&mns->keys_lock);
 587
 588	key->expiry = jiffies + mctp_key_lifetime;
 589	timer_reduce(&msk->key_expiry, key->expiry);
 590
 591	/* we hold the net->key_lock here, allowing updates to both
 592	 * then net and sk
 593	 */
 594	hlist_add_head_rcu(&key->hlist, &mns->keys);
 595	hlist_add_head_rcu(&key->sklist, &msk->keys);
 596	refcount_inc(&key->refs);
 597}
 598
 599/* Allocate a locally-owned tag value for (saddr, daddr), and reserve
 600 * it for the socket msk
 601 */
 602struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
 603					 mctp_eid_t daddr, mctp_eid_t saddr,
 
 604					 bool manual, u8 *tagp)
 605{
 606	struct net *net = sock_net(&msk->sk);
 607	struct netns_mctp *mns = &net->mctp;
 608	struct mctp_sk_key *key, *tmp;
 609	unsigned long flags;
 610	u8 tagbits;
 611
 612	/* for NULL destination EIDs, we may get a response from any peer */
 613	if (daddr == MCTP_ADDR_NULL)
 614		daddr = MCTP_ADDR_ANY;
 615
 616	/* be optimistic, alloc now */
 617	key = mctp_key_alloc(msk, saddr, daddr, 0, GFP_KERNEL);
 618	if (!key)
 619		return ERR_PTR(-ENOMEM);
 620
 621	/* 8 possible tag values */
 622	tagbits = 0xff;
 623
 624	spin_lock_irqsave(&mns->keys_lock, flags);
 625
 626	/* Walk through the existing keys, looking for potential conflicting
 627	 * tags. If we find a conflict, clear that bit from tagbits
 628	 */
 629	hlist_for_each_entry(tmp, &mns->keys, hlist) {
 630		/* We can check the lookup fields (*_addr, tag) without the
 631		 * lock held, they don't change over the lifetime of the key.
 632		 */
 633
 
 
 
 
 634		/* if we don't own the tag, it can't conflict */
 635		if (tmp->tag & MCTP_HDR_FLAG_TO)
 636			continue;
 637
 638		if (!(mctp_address_matches(tmp->peer_addr, daddr) &&
 639		      mctp_address_matches(tmp->local_addr, saddr)))
 
 
 
 
 
 
 
 
 640			continue;
 641
 642		spin_lock(&tmp->lock);
 643		/* key must still be valid. If we find a match, clear the
 644		 * potential tag value
 645		 */
 646		if (tmp->valid)
 647			tagbits &= ~(1 << tmp->tag);
 648		spin_unlock(&tmp->lock);
 649
 650		if (!tagbits)
 651			break;
 652	}
 653
 654	if (tagbits) {
 655		key->tag = __ffs(tagbits);
 656		mctp_reserve_tag(net, key, msk);
 657		trace_mctp_key_acquire(key);
 658
 659		key->manual_alloc = manual;
 660		*tagp = key->tag;
 661	}
 662
 663	spin_unlock_irqrestore(&mns->keys_lock, flags);
 664
 665	if (!tagbits) {
 666		mctp_key_unref(key);
 667		return ERR_PTR(-EBUSY);
 668	}
 669
 670	return key;
 671}
 672
 673static struct mctp_sk_key *mctp_lookup_prealloc_tag(struct mctp_sock *msk,
 
 674						    mctp_eid_t daddr,
 675						    u8 req_tag, u8 *tagp)
 676{
 677	struct net *net = sock_net(&msk->sk);
 678	struct netns_mctp *mns = &net->mctp;
 679	struct mctp_sk_key *key, *tmp;
 680	unsigned long flags;
 681
 682	req_tag &= ~(MCTP_TAG_PREALLOC | MCTP_TAG_OWNER);
 683	key = NULL;
 684
 685	spin_lock_irqsave(&mns->keys_lock, flags);
 686
 687	hlist_for_each_entry(tmp, &mns->keys, hlist) {
 
 
 
 688		if (tmp->tag != req_tag)
 689			continue;
 690
 691		if (!mctp_address_matches(tmp->peer_addr, daddr))
 692			continue;
 693
 694		if (!tmp->manual_alloc)
 695			continue;
 696
 697		spin_lock(&tmp->lock);
 698		if (tmp->valid) {
 699			key = tmp;
 700			refcount_inc(&key->refs);
 701			spin_unlock(&tmp->lock);
 702			break;
 703		}
 704		spin_unlock(&tmp->lock);
 705	}
 706	spin_unlock_irqrestore(&mns->keys_lock, flags);
 707
 708	if (!key)
 709		return ERR_PTR(-ENOENT);
 710
 711	if (tagp)
 712		*tagp = key->tag;
 713
 714	return key;
 715}
 716
 717/* routing lookups */
 718static bool mctp_rt_match_eid(struct mctp_route *rt,
 719			      unsigned int net, mctp_eid_t eid)
 720{
 721	return READ_ONCE(rt->dev->net) == net &&
 722		rt->min <= eid && rt->max >= eid;
 723}
 724
 725/* compares match, used for duplicate prevention */
 726static bool mctp_rt_compare_exact(struct mctp_route *rt1,
 727				  struct mctp_route *rt2)
 728{
 729	ASSERT_RTNL();
 730	return rt1->dev->net == rt2->dev->net &&
 731		rt1->min == rt2->min &&
 732		rt1->max == rt2->max;
 733}
 734
 735struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
 736				     mctp_eid_t daddr)
 737{
 738	struct mctp_route *tmp, *rt = NULL;
 739
 740	rcu_read_lock();
 741
 742	list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 743		/* TODO: add metrics */
 744		if (mctp_rt_match_eid(tmp, dnet, daddr)) {
 745			if (refcount_inc_not_zero(&tmp->refs)) {
 746				rt = tmp;
 747				break;
 748			}
 749		}
 750	}
 751
 752	rcu_read_unlock();
 753
 754	return rt;
 755}
 756
 757static struct mctp_route *mctp_route_lookup_null(struct net *net,
 758						 struct net_device *dev)
 759{
 760	struct mctp_route *tmp, *rt = NULL;
 761
 762	rcu_read_lock();
 763
 764	list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 765		if (tmp->dev->dev == dev && tmp->type == RTN_LOCAL &&
 766		    refcount_inc_not_zero(&tmp->refs)) {
 767			rt = tmp;
 768			break;
 769		}
 770	}
 771
 772	rcu_read_unlock();
 773
 774	return rt;
 775}
 776
 777static int mctp_do_fragment_route(struct mctp_route *rt, struct sk_buff *skb,
 778				  unsigned int mtu, u8 tag)
 779{
 780	const unsigned int hlen = sizeof(struct mctp_hdr);
 781	struct mctp_hdr *hdr, *hdr2;
 782	unsigned int pos, size, headroom;
 783	struct sk_buff *skb2;
 784	int rc;
 785	u8 seq;
 786
 787	hdr = mctp_hdr(skb);
 788	seq = 0;
 789	rc = 0;
 790
 791	if (mtu < hlen + 1) {
 792		kfree_skb(skb);
 793		return -EMSGSIZE;
 794	}
 795
 796	/* keep same headroom as the original skb */
 797	headroom = skb_headroom(skb);
 798
 799	/* we've got the header */
 800	skb_pull(skb, hlen);
 801
 802	for (pos = 0; pos < skb->len;) {
 803		/* size of message payload */
 804		size = min(mtu - hlen, skb->len - pos);
 805
 806		skb2 = alloc_skb(headroom + hlen + size, GFP_KERNEL);
 807		if (!skb2) {
 808			rc = -ENOMEM;
 809			break;
 810		}
 811
 812		/* generic skb copy */
 813		skb2->protocol = skb->protocol;
 814		skb2->priority = skb->priority;
 815		skb2->dev = skb->dev;
 816		memcpy(skb2->cb, skb->cb, sizeof(skb2->cb));
 817
 818		if (skb->sk)
 819			skb_set_owner_w(skb2, skb->sk);
 820
 821		/* establish packet */
 822		skb_reserve(skb2, headroom);
 823		skb_reset_network_header(skb2);
 824		skb_put(skb2, hlen + size);
 825		skb2->transport_header = skb2->network_header + hlen;
 826
 827		/* copy header fields, calculate SOM/EOM flags & seq */
 828		hdr2 = mctp_hdr(skb2);
 829		hdr2->ver = hdr->ver;
 830		hdr2->dest = hdr->dest;
 831		hdr2->src = hdr->src;
 832		hdr2->flags_seq_tag = tag &
 833			(MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 834
 835		if (pos == 0)
 836			hdr2->flags_seq_tag |= MCTP_HDR_FLAG_SOM;
 837
 838		if (pos + size == skb->len)
 839			hdr2->flags_seq_tag |= MCTP_HDR_FLAG_EOM;
 840
 841		hdr2->flags_seq_tag |= seq << MCTP_HDR_SEQ_SHIFT;
 842
 843		/* copy message payload */
 844		skb_copy_bits(skb, pos, skb_transport_header(skb2), size);
 845
 
 
 
 846		/* do route */
 847		rc = rt->output(rt, skb2);
 848		if (rc)
 849			break;
 850
 851		seq = (seq + 1) & MCTP_HDR_SEQ_MASK;
 852		pos += size;
 853	}
 854
 855	consume_skb(skb);
 856	return rc;
 857}
 858
 859int mctp_local_output(struct sock *sk, struct mctp_route *rt,
 860		      struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag)
 861{
 862	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
 863	struct mctp_skb_cb *cb = mctp_cb(skb);
 864	struct mctp_route tmp_rt = {0};
 865	struct mctp_sk_key *key;
 866	struct mctp_hdr *hdr;
 867	unsigned long flags;
 
 868	unsigned int mtu;
 869	mctp_eid_t saddr;
 870	bool ext_rt;
 871	int rc;
 872	u8 tag;
 873
 874	rc = -ENODEV;
 875
 876	if (rt) {
 877		ext_rt = false;
 878		if (WARN_ON(!rt->dev))
 879			goto out_release;
 880
 881	} else if (cb->ifindex) {
 882		struct net_device *dev;
 883
 884		ext_rt = true;
 885		rt = &tmp_rt;
 886
 887		rcu_read_lock();
 888		dev = dev_get_by_index_rcu(sock_net(sk), cb->ifindex);
 889		if (!dev) {
 890			rcu_read_unlock();
 891			goto out_free;
 892		}
 893		rt->dev = __mctp_dev_get(dev);
 894		rcu_read_unlock();
 895
 896		if (!rt->dev)
 897			goto out_release;
 898
 899		/* establish temporary route - we set up enough to keep
 900		 * mctp_route_output happy
 901		 */
 902		rt->output = mctp_route_output;
 903		rt->mtu = 0;
 904
 905	} else {
 906		rc = -EINVAL;
 907		goto out_free;
 908	}
 909
 910	spin_lock_irqsave(&rt->dev->addrs_lock, flags);
 911	if (rt->dev->num_addrs == 0) {
 912		rc = -EHOSTUNREACH;
 913	} else {
 914		/* use the outbound interface's first address as our source */
 915		saddr = rt->dev->addrs[0];
 916		rc = 0;
 917	}
 918	spin_unlock_irqrestore(&rt->dev->addrs_lock, flags);
 
 919
 920	if (rc)
 921		goto out_release;
 922
 923	if (req_tag & MCTP_TAG_OWNER) {
 924		if (req_tag & MCTP_TAG_PREALLOC)
 925			key = mctp_lookup_prealloc_tag(msk, daddr,
 926						       req_tag, &tag);
 927		else
 928			key = mctp_alloc_local_tag(msk, daddr, saddr,
 929						   false, &tag);
 930
 931		if (IS_ERR(key)) {
 932			rc = PTR_ERR(key);
 933			goto out_release;
 934		}
 935		mctp_skb_set_flow(skb, key);
 936		/* done with the key in this scope */
 937		mctp_key_unref(key);
 938		tag |= MCTP_HDR_FLAG_TO;
 939	} else {
 940		key = NULL;
 941		tag = req_tag & MCTP_TAG_MASK;
 942	}
 943
 944	skb->protocol = htons(ETH_P_MCTP);
 945	skb->priority = 0;
 946	skb_reset_transport_header(skb);
 947	skb_push(skb, sizeof(struct mctp_hdr));
 948	skb_reset_network_header(skb);
 949	skb->dev = rt->dev->dev;
 950
 951	/* cb->net will have been set on initial ingress */
 952	cb->src = saddr;
 953
 954	/* set up common header fields */
 955	hdr = mctp_hdr(skb);
 956	hdr->ver = 1;
 957	hdr->dest = daddr;
 958	hdr->src = saddr;
 959
 960	mtu = mctp_route_mtu(rt);
 961
 962	if (skb->len + sizeof(struct mctp_hdr) <= mtu) {
 963		hdr->flags_seq_tag = MCTP_HDR_FLAG_SOM |
 964			MCTP_HDR_FLAG_EOM | tag;
 965		rc = rt->output(rt, skb);
 966	} else {
 967		rc = mctp_do_fragment_route(rt, skb, mtu, tag);
 968	}
 969
 970	/* route output functions consume the skb, even on error */
 971	skb = NULL;
 972
 973out_release:
 974	if (!ext_rt)
 975		mctp_route_release(rt);
 976
 977	mctp_dev_put(tmp_rt.dev);
 978
 979out_free:
 980	kfree_skb(skb);
 981	return rc;
 982}
 983
 984/* route management */
 985static int mctp_route_add(struct mctp_dev *mdev, mctp_eid_t daddr_start,
 986			  unsigned int daddr_extent, unsigned int mtu,
 987			  unsigned char type)
 988{
 989	int (*rtfn)(struct mctp_route *rt, struct sk_buff *skb);
 990	struct net *net = dev_net(mdev->dev);
 991	struct mctp_route *rt, *ert;
 992
 993	if (!mctp_address_unicast(daddr_start))
 994		return -EINVAL;
 995
 996	if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
 997		return -EINVAL;
 998
 999	switch (type) {
1000	case RTN_LOCAL:
1001		rtfn = mctp_route_input;
1002		break;
1003	case RTN_UNICAST:
1004		rtfn = mctp_route_output;
1005		break;
1006	default:
1007		return -EINVAL;
1008	}
1009
1010	rt = mctp_route_alloc();
1011	if (!rt)
1012		return -ENOMEM;
1013
1014	rt->min = daddr_start;
1015	rt->max = daddr_start + daddr_extent;
1016	rt->mtu = mtu;
1017	rt->dev = mdev;
1018	mctp_dev_hold(rt->dev);
1019	rt->type = type;
1020	rt->output = rtfn;
1021
1022	ASSERT_RTNL();
1023	/* Prevent duplicate identical routes. */
1024	list_for_each_entry(ert, &net->mctp.routes, list) {
1025		if (mctp_rt_compare_exact(rt, ert)) {
1026			mctp_route_release(rt);
1027			return -EEXIST;
1028		}
1029	}
1030
1031	list_add_rcu(&rt->list, &net->mctp.routes);
1032
1033	return 0;
1034}
1035
1036static int mctp_route_remove(struct mctp_dev *mdev, mctp_eid_t daddr_start,
1037			     unsigned int daddr_extent, unsigned char type)
1038{
1039	struct net *net = dev_net(mdev->dev);
1040	struct mctp_route *rt, *tmp;
1041	mctp_eid_t daddr_end;
1042	bool dropped;
1043
1044	if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
1045		return -EINVAL;
1046
1047	daddr_end = daddr_start + daddr_extent;
1048	dropped = false;
1049
1050	ASSERT_RTNL();
1051
1052	list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1053		if (rt->dev == mdev &&
1054		    rt->min == daddr_start && rt->max == daddr_end &&
1055		    rt->type == type) {
1056			list_del_rcu(&rt->list);
1057			/* TODO: immediate RTM_DELROUTE */
1058			mctp_route_release(rt);
1059			dropped = true;
1060		}
1061	}
1062
1063	return dropped ? 0 : -ENOENT;
1064}
1065
1066int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr)
1067{
1068	return mctp_route_add(mdev, addr, 0, 0, RTN_LOCAL);
1069}
1070
1071int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr)
1072{
1073	return mctp_route_remove(mdev, addr, 0, RTN_LOCAL);
1074}
1075
1076/* removes all entries for a given device */
1077void mctp_route_remove_dev(struct mctp_dev *mdev)
1078{
1079	struct net *net = dev_net(mdev->dev);
1080	struct mctp_route *rt, *tmp;
1081
1082	ASSERT_RTNL();
1083	list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1084		if (rt->dev == mdev) {
1085			list_del_rcu(&rt->list);
1086			/* TODO: immediate RTM_DELROUTE */
1087			mctp_route_release(rt);
1088		}
1089	}
1090}
1091
1092/* Incoming packet-handling */
1093
1094static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,
1095				struct packet_type *pt,
1096				struct net_device *orig_dev)
1097{
1098	struct net *net = dev_net(dev);
1099	struct mctp_dev *mdev;
1100	struct mctp_skb_cb *cb;
1101	struct mctp_route *rt;
1102	struct mctp_hdr *mh;
1103
1104	rcu_read_lock();
1105	mdev = __mctp_dev_get(dev);
1106	rcu_read_unlock();
1107	if (!mdev) {
1108		/* basic non-data sanity checks */
1109		goto err_drop;
1110	}
1111
1112	if (!pskb_may_pull(skb, sizeof(struct mctp_hdr)))
1113		goto err_drop;
1114
1115	skb_reset_transport_header(skb);
1116	skb_reset_network_header(skb);
1117
1118	/* We have enough for a header; decode and route */
1119	mh = mctp_hdr(skb);
1120	if (mh->ver < MCTP_VER_MIN || mh->ver > MCTP_VER_MAX)
1121		goto err_drop;
1122
1123	/* source must be valid unicast or null; drop reserved ranges and
1124	 * broadcast
1125	 */
1126	if (!(mctp_address_unicast(mh->src) || mctp_address_null(mh->src)))
1127		goto err_drop;
1128
1129	/* dest address: as above, but allow broadcast */
1130	if (!(mctp_address_unicast(mh->dest) || mctp_address_null(mh->dest) ||
1131	      mctp_address_broadcast(mh->dest)))
1132		goto err_drop;
1133
1134	/* MCTP drivers must populate halen/haddr */
1135	if (dev->type == ARPHRD_MCTP) {
1136		cb = mctp_cb(skb);
1137	} else {
1138		cb = __mctp_cb(skb);
1139		cb->halen = 0;
1140	}
1141	cb->net = READ_ONCE(mdev->net);
1142	cb->ifindex = dev->ifindex;
1143
1144	rt = mctp_route_lookup(net, cb->net, mh->dest);
1145
1146	/* NULL EID, but addressed to our physical address */
1147	if (!rt && mh->dest == MCTP_ADDR_NULL && skb->pkt_type == PACKET_HOST)
1148		rt = mctp_route_lookup_null(net, dev);
1149
1150	if (!rt)
1151		goto err_drop;
1152
1153	rt->output(rt, skb);
1154	mctp_route_release(rt);
1155	mctp_dev_put(mdev);
1156
1157	return NET_RX_SUCCESS;
1158
1159err_drop:
1160	kfree_skb(skb);
1161	mctp_dev_put(mdev);
1162	return NET_RX_DROP;
1163}
1164
1165static struct packet_type mctp_packet_type = {
1166	.type = cpu_to_be16(ETH_P_MCTP),
1167	.func = mctp_pkttype_receive,
1168};
1169
1170/* netlink interface */
1171
1172static const struct nla_policy rta_mctp_policy[RTA_MAX + 1] = {
1173	[RTA_DST]		= { .type = NLA_U8 },
1174	[RTA_METRICS]		= { .type = NLA_NESTED },
1175	[RTA_OIF]		= { .type = NLA_U32 },
1176};
1177
1178/* Common part for RTM_NEWROUTE and RTM_DELROUTE parsing.
1179 * tb must hold RTA_MAX+1 elements.
1180 */
1181static int mctp_route_nlparse(struct sk_buff *skb, struct nlmsghdr *nlh,
1182			      struct netlink_ext_ack *extack,
1183			      struct nlattr **tb, struct rtmsg **rtm,
1184			      struct mctp_dev **mdev, mctp_eid_t *daddr_start)
1185{
1186	struct net *net = sock_net(skb->sk);
1187	struct net_device *dev;
1188	unsigned int ifindex;
1189	int rc;
1190
1191	rc = nlmsg_parse(nlh, sizeof(struct rtmsg), tb, RTA_MAX,
1192			 rta_mctp_policy, extack);
1193	if (rc < 0) {
1194		NL_SET_ERR_MSG(extack, "incorrect format");
1195		return rc;
1196	}
1197
1198	if (!tb[RTA_DST]) {
1199		NL_SET_ERR_MSG(extack, "dst EID missing");
1200		return -EINVAL;
1201	}
1202	*daddr_start = nla_get_u8(tb[RTA_DST]);
1203
1204	if (!tb[RTA_OIF]) {
1205		NL_SET_ERR_MSG(extack, "ifindex missing");
1206		return -EINVAL;
1207	}
1208	ifindex = nla_get_u32(tb[RTA_OIF]);
1209
1210	*rtm = nlmsg_data(nlh);
1211	if ((*rtm)->rtm_family != AF_MCTP) {
1212		NL_SET_ERR_MSG(extack, "route family must be AF_MCTP");
1213		return -EINVAL;
1214	}
1215
1216	dev = __dev_get_by_index(net, ifindex);
1217	if (!dev) {
1218		NL_SET_ERR_MSG(extack, "bad ifindex");
1219		return -ENODEV;
1220	}
1221	*mdev = mctp_dev_get_rtnl(dev);
1222	if (!*mdev)
1223		return -ENODEV;
1224
1225	if (dev->flags & IFF_LOOPBACK) {
1226		NL_SET_ERR_MSG(extack, "no routes to loopback");
1227		return -EINVAL;
1228	}
1229
1230	return 0;
1231}
1232
1233static const struct nla_policy rta_metrics_policy[RTAX_MAX + 1] = {
1234	[RTAX_MTU]		= { .type = NLA_U32 },
1235};
1236
1237static int mctp_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1238			 struct netlink_ext_ack *extack)
1239{
1240	struct nlattr *tb[RTA_MAX + 1];
1241	struct nlattr *tbx[RTAX_MAX + 1];
1242	mctp_eid_t daddr_start;
1243	struct mctp_dev *mdev;
1244	struct rtmsg *rtm;
1245	unsigned int mtu;
1246	int rc;
1247
1248	rc = mctp_route_nlparse(skb, nlh, extack, tb,
1249				&rtm, &mdev, &daddr_start);
1250	if (rc < 0)
1251		return rc;
1252
1253	if (rtm->rtm_type != RTN_UNICAST) {
1254		NL_SET_ERR_MSG(extack, "rtm_type must be RTN_UNICAST");
1255		return -EINVAL;
1256	}
1257
1258	mtu = 0;
1259	if (tb[RTA_METRICS]) {
1260		rc = nla_parse_nested(tbx, RTAX_MAX, tb[RTA_METRICS],
1261				      rta_metrics_policy, NULL);
1262		if (rc < 0)
1263			return rc;
1264		if (tbx[RTAX_MTU])
1265			mtu = nla_get_u32(tbx[RTAX_MTU]);
1266	}
1267
1268	rc = mctp_route_add(mdev, daddr_start, rtm->rtm_dst_len, mtu,
1269			    rtm->rtm_type);
1270	return rc;
1271}
1272
1273static int mctp_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1274			 struct netlink_ext_ack *extack)
1275{
1276	struct nlattr *tb[RTA_MAX + 1];
1277	mctp_eid_t daddr_start;
1278	struct mctp_dev *mdev;
1279	struct rtmsg *rtm;
1280	int rc;
1281
1282	rc = mctp_route_nlparse(skb, nlh, extack, tb,
1283				&rtm, &mdev, &daddr_start);
1284	if (rc < 0)
1285		return rc;
1286
1287	/* we only have unicast routes */
1288	if (rtm->rtm_type != RTN_UNICAST)
1289		return -EINVAL;
1290
1291	rc = mctp_route_remove(mdev, daddr_start, rtm->rtm_dst_len, RTN_UNICAST);
1292	return rc;
1293}
1294
1295static int mctp_fill_rtinfo(struct sk_buff *skb, struct mctp_route *rt,
1296			    u32 portid, u32 seq, int event, unsigned int flags)
1297{
1298	struct nlmsghdr *nlh;
1299	struct rtmsg *hdr;
1300	void *metrics;
1301
1302	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*hdr), flags);
1303	if (!nlh)
1304		return -EMSGSIZE;
1305
1306	hdr = nlmsg_data(nlh);
1307	hdr->rtm_family = AF_MCTP;
1308
1309	/* we use the _len fields as a number of EIDs, rather than
1310	 * a number of bits in the address
1311	 */
1312	hdr->rtm_dst_len = rt->max - rt->min;
1313	hdr->rtm_src_len = 0;
1314	hdr->rtm_tos = 0;
1315	hdr->rtm_table = RT_TABLE_DEFAULT;
1316	hdr->rtm_protocol = RTPROT_STATIC; /* everything is user-defined */
1317	hdr->rtm_scope = RT_SCOPE_LINK; /* TODO: scope in mctp_route? */
1318	hdr->rtm_type = rt->type;
1319
1320	if (nla_put_u8(skb, RTA_DST, rt->min))
1321		goto cancel;
1322
1323	metrics = nla_nest_start_noflag(skb, RTA_METRICS);
1324	if (!metrics)
1325		goto cancel;
1326
1327	if (rt->mtu) {
1328		if (nla_put_u32(skb, RTAX_MTU, rt->mtu))
1329			goto cancel;
1330	}
1331
1332	nla_nest_end(skb, metrics);
1333
1334	if (rt->dev) {
1335		if (nla_put_u32(skb, RTA_OIF, rt->dev->dev->ifindex))
1336			goto cancel;
1337	}
1338
1339	/* TODO: conditional neighbour physaddr? */
1340
1341	nlmsg_end(skb, nlh);
1342
1343	return 0;
1344
1345cancel:
1346	nlmsg_cancel(skb, nlh);
1347	return -EMSGSIZE;
1348}
1349
1350static int mctp_dump_rtinfo(struct sk_buff *skb, struct netlink_callback *cb)
1351{
1352	struct net *net = sock_net(skb->sk);
1353	struct mctp_route *rt;
1354	int s_idx, idx;
1355
1356	/* TODO: allow filtering on route data, possibly under
1357	 * cb->strict_check
1358	 */
1359
1360	/* TODO: change to struct overlay */
1361	s_idx = cb->args[0];
1362	idx = 0;
1363
1364	rcu_read_lock();
1365	list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
1366		if (idx++ < s_idx)
1367			continue;
1368		if (mctp_fill_rtinfo(skb, rt,
1369				     NETLINK_CB(cb->skb).portid,
1370				     cb->nlh->nlmsg_seq,
1371				     RTM_NEWROUTE, NLM_F_MULTI) < 0)
1372			break;
1373	}
1374
1375	rcu_read_unlock();
1376	cb->args[0] = idx;
1377
1378	return skb->len;
1379}
1380
1381/* net namespace implementation */
1382static int __net_init mctp_routes_net_init(struct net *net)
1383{
1384	struct netns_mctp *ns = &net->mctp;
1385
1386	INIT_LIST_HEAD(&ns->routes);
1387	INIT_HLIST_HEAD(&ns->binds);
1388	mutex_init(&ns->bind_lock);
1389	INIT_HLIST_HEAD(&ns->keys);
1390	spin_lock_init(&ns->keys_lock);
1391	WARN_ON(mctp_default_net_set(net, MCTP_INITIAL_DEFAULT_NET));
1392	return 0;
1393}
1394
1395static void __net_exit mctp_routes_net_exit(struct net *net)
1396{
1397	struct mctp_route *rt;
1398
1399	rcu_read_lock();
1400	list_for_each_entry_rcu(rt, &net->mctp.routes, list)
1401		mctp_route_release(rt);
1402	rcu_read_unlock();
1403}
1404
1405static struct pernet_operations mctp_net_ops = {
1406	.init = mctp_routes_net_init,
1407	.exit = mctp_routes_net_exit,
1408};
1409
1410int __init mctp_routes_init(void)
1411{
1412	dev_add_pack(&mctp_packet_type);
1413
1414	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETROUTE,
1415			     NULL, mctp_dump_rtinfo, 0);
1416	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWROUTE,
1417			     mctp_newroute, NULL, 0);
1418	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELROUTE,
1419			     mctp_delroute, NULL, 0);
1420
1421	return register_pernet_subsys(&mctp_net_ops);
1422}
1423
1424void mctp_routes_exit(void)
1425{
1426	unregister_pernet_subsys(&mctp_net_ops);
1427	rtnl_unregister(PF_MCTP, RTM_DELROUTE);
1428	rtnl_unregister(PF_MCTP, RTM_NEWROUTE);
1429	rtnl_unregister(PF_MCTP, RTM_GETROUTE);
1430	dev_remove_pack(&mctp_packet_type);
1431}
1432
1433#if IS_ENABLED(CONFIG_MCTP_TEST)
1434#include "test/route-test.c"
1435#endif
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Management Component Transport Protocol (MCTP) - routing
   4 * implementation.
   5 *
   6 * This is currently based on a simple routing table, with no dst cache. The
   7 * number of routes should stay fairly small, so the lookup cost is small.
   8 *
   9 * Copyright (c) 2021 Code Construct
  10 * Copyright (c) 2021 Google
  11 */
  12
  13#include <linux/idr.h>
  14#include <linux/kconfig.h>
  15#include <linux/mctp.h>
  16#include <linux/netdevice.h>
  17#include <linux/rtnetlink.h>
  18#include <linux/skbuff.h>
  19
  20#include <uapi/linux/if_arp.h>
  21
  22#include <net/mctp.h>
  23#include <net/mctpdevice.h>
  24#include <net/netlink.h>
  25#include <net/sock.h>
  26
  27#include <trace/events/mctp.h>
  28
  29static const unsigned int mctp_message_maxlen = 64 * 1024;
  30static const unsigned long mctp_key_lifetime = 6 * CONFIG_HZ;
  31
  32static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev);
  33
  34/* route output callbacks */
  35static int mctp_route_discard(struct mctp_route *route, struct sk_buff *skb)
  36{
  37	kfree_skb(skb);
  38	return 0;
  39}
  40
  41static struct mctp_sock *mctp_lookup_bind(struct net *net, struct sk_buff *skb)
  42{
  43	struct mctp_skb_cb *cb = mctp_cb(skb);
  44	struct mctp_hdr *mh;
  45	struct sock *sk;
  46	u8 type;
  47
  48	WARN_ON(!rcu_read_lock_held());
  49
  50	/* TODO: look up in skb->cb? */
  51	mh = mctp_hdr(skb);
  52
  53	if (!skb_headlen(skb))
  54		return NULL;
  55
  56	type = (*(u8 *)skb->data) & 0x7f;
  57
  58	sk_for_each_rcu(sk, &net->mctp.binds) {
  59		struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
  60
  61		if (msk->bind_net != MCTP_NET_ANY && msk->bind_net != cb->net)
  62			continue;
  63
  64		if (msk->bind_type != type)
  65			continue;
  66
  67		if (!mctp_address_matches(msk->bind_addr, mh->dest))
  68			continue;
  69
  70		return msk;
  71	}
  72
  73	return NULL;
  74}
  75
  76/* A note on the key allocations.
  77 *
  78 * struct net->mctp.keys contains our set of currently-allocated keys for
  79 * MCTP tag management. The lookup tuple for these is the peer EID,
  80 * local EID and MCTP tag.
  81 *
  82 * In some cases, the peer EID may be MCTP_EID_ANY: for example, when a
  83 * broadcast message is sent, we may receive responses from any peer EID.
  84 * Because the broadcast dest address is equivalent to ANY, we create
  85 * a key with (local = local-eid, peer = ANY). This allows a match on the
  86 * incoming broadcast responses from any peer.
  87 *
  88 * We perform lookups when packets are received, and when tags are allocated
  89 * in two scenarios:
  90 *
  91 *  - when a packet is sent, with a locally-owned tag: we need to find an
  92 *    unused tag value for the (local, peer) EID pair.
  93 *
  94 *  - when a tag is manually allocated: we need to find an unused tag value
  95 *    for the peer EID, but don't have a specific local EID at that stage.
  96 *
  97 * in the latter case, on successful allocation, we end up with a tag with
  98 * (local = ANY, peer = peer-eid).
  99 *
 100 * So, the key set allows both a local EID of ANY, as well as a peer EID of
 101 * ANY in the lookup tuple. Both may be ANY if we prealloc for a broadcast.
 102 * The matching (in mctp_key_match()) during lookup allows the match value to
 103 * be ANY in either the dest or source addresses.
 104 *
 105 * When allocating (+ inserting) a tag, we need to check for conflicts amongst
 106 * the existing tag set. This requires macthing either exactly on the local
 107 * and peer addresses, or either being ANY.
 108 */
 109
 110static bool mctp_key_match(struct mctp_sk_key *key, unsigned int net,
 111			   mctp_eid_t local, mctp_eid_t peer, u8 tag)
 112{
 113	if (key->net != net)
 114		return false;
 115
 116	if (!mctp_address_matches(key->local_addr, local))
 117		return false;
 118
 119	if (!mctp_address_matches(key->peer_addr, peer))
 120		return false;
 121
 122	if (key->tag != tag)
 123		return false;
 124
 125	return true;
 126}
 127
 128/* returns a key (with key->lock held, and refcounted), or NULL if no such
 129 * key exists.
 130 */
 131static struct mctp_sk_key *mctp_lookup_key(struct net *net, struct sk_buff *skb,
 132					   unsigned int netid, mctp_eid_t peer,
 133					   unsigned long *irqflags)
 134	__acquires(&key->lock)
 135{
 136	struct mctp_sk_key *key, *ret;
 137	unsigned long flags;
 138	struct mctp_hdr *mh;
 139	u8 tag;
 140
 141	mh = mctp_hdr(skb);
 142	tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 143
 144	ret = NULL;
 145	spin_lock_irqsave(&net->mctp.keys_lock, flags);
 146
 147	hlist_for_each_entry(key, &net->mctp.keys, hlist) {
 148		if (!mctp_key_match(key, netid, mh->dest, peer, tag))
 149			continue;
 150
 151		spin_lock(&key->lock);
 152		if (key->valid) {
 153			refcount_inc(&key->refs);
 154			ret = key;
 155			break;
 156		}
 157		spin_unlock(&key->lock);
 158	}
 159
 160	if (ret) {
 161		spin_unlock(&net->mctp.keys_lock);
 162		*irqflags = flags;
 163	} else {
 164		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 165	}
 166
 167	return ret;
 168}
 169
 170static struct mctp_sk_key *mctp_key_alloc(struct mctp_sock *msk,
 171					  unsigned int net,
 172					  mctp_eid_t local, mctp_eid_t peer,
 173					  u8 tag, gfp_t gfp)
 174{
 175	struct mctp_sk_key *key;
 176
 177	key = kzalloc(sizeof(*key), gfp);
 178	if (!key)
 179		return NULL;
 180
 181	key->net = net;
 182	key->peer_addr = peer;
 183	key->local_addr = local;
 184	key->tag = tag;
 185	key->sk = &msk->sk;
 186	key->valid = true;
 187	spin_lock_init(&key->lock);
 188	refcount_set(&key->refs, 1);
 189	sock_hold(key->sk);
 190
 191	return key;
 192}
 193
 194void mctp_key_unref(struct mctp_sk_key *key)
 195{
 196	unsigned long flags;
 197
 198	if (!refcount_dec_and_test(&key->refs))
 199		return;
 200
 201	/* even though no refs exist here, the lock allows us to stay
 202	 * consistent with the locking requirement of mctp_dev_release_key
 203	 */
 204	spin_lock_irqsave(&key->lock, flags);
 205	mctp_dev_release_key(key->dev, key);
 206	spin_unlock_irqrestore(&key->lock, flags);
 207
 208	sock_put(key->sk);
 209	kfree(key);
 210}
 211
 212static int mctp_key_add(struct mctp_sk_key *key, struct mctp_sock *msk)
 213{
 214	struct net *net = sock_net(&msk->sk);
 215	struct mctp_sk_key *tmp;
 216	unsigned long flags;
 217	int rc = 0;
 218
 219	spin_lock_irqsave(&net->mctp.keys_lock, flags);
 220
 221	if (sock_flag(&msk->sk, SOCK_DEAD)) {
 222		rc = -EINVAL;
 223		goto out_unlock;
 224	}
 225
 226	hlist_for_each_entry(tmp, &net->mctp.keys, hlist) {
 227		if (mctp_key_match(tmp, key->net, key->local_addr,
 228				   key->peer_addr, key->tag)) {
 229			spin_lock(&tmp->lock);
 230			if (tmp->valid)
 231				rc = -EEXIST;
 232			spin_unlock(&tmp->lock);
 233			if (rc)
 234				break;
 235		}
 236	}
 237
 238	if (!rc) {
 239		refcount_inc(&key->refs);
 240		key->expiry = jiffies + mctp_key_lifetime;
 241		timer_reduce(&msk->key_expiry, key->expiry);
 242
 243		hlist_add_head(&key->hlist, &net->mctp.keys);
 244		hlist_add_head(&key->sklist, &msk->keys);
 245	}
 246
 247out_unlock:
 248	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 249
 250	return rc;
 251}
 252
 253/* Helper for mctp_route_input().
 254 * We're done with the key; unlock and unref the key.
 255 * For the usual case of automatic expiry we remove the key from lists.
 256 * In the case that manual allocation is set on a key we release the lock
 257 * and local ref, reset reassembly, but don't remove from lists.
 258 */
 259static void __mctp_key_done_in(struct mctp_sk_key *key, struct net *net,
 260			       unsigned long flags, unsigned long reason)
 261__releases(&key->lock)
 262{
 263	struct sk_buff *skb;
 264
 265	trace_mctp_key_release(key, reason);
 266	skb = key->reasm_head;
 267	key->reasm_head = NULL;
 268
 269	if (!key->manual_alloc) {
 270		key->reasm_dead = true;
 271		key->valid = false;
 272		mctp_dev_release_key(key->dev, key);
 273	}
 274	spin_unlock_irqrestore(&key->lock, flags);
 275
 276	if (!key->manual_alloc) {
 277		spin_lock_irqsave(&net->mctp.keys_lock, flags);
 278		if (!hlist_unhashed(&key->hlist)) {
 279			hlist_del_init(&key->hlist);
 280			hlist_del_init(&key->sklist);
 281			mctp_key_unref(key);
 282		}
 283		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 284	}
 285
 286	/* and one for the local reference */
 287	mctp_key_unref(key);
 288
 289	kfree_skb(skb);
 290}
 291
 292#ifdef CONFIG_MCTP_FLOWS
 293static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key)
 294{
 295	struct mctp_flow *flow;
 296
 297	flow = skb_ext_add(skb, SKB_EXT_MCTP);
 298	if (!flow)
 299		return;
 300
 301	refcount_inc(&key->refs);
 302	flow->key = key;
 303}
 304
 305static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev)
 306{
 307	struct mctp_sk_key *key;
 308	struct mctp_flow *flow;
 309
 310	flow = skb_ext_find(skb, SKB_EXT_MCTP);
 311	if (!flow)
 312		return;
 313
 314	key = flow->key;
 315
 316	if (WARN_ON(key->dev && key->dev != dev))
 317		return;
 318
 319	mctp_dev_set_key(dev, key);
 320}
 321#else
 322static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key) {}
 323static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev) {}
 324#endif
 325
 326static int mctp_frag_queue(struct mctp_sk_key *key, struct sk_buff *skb)
 327{
 328	struct mctp_hdr *hdr = mctp_hdr(skb);
 329	u8 exp_seq, this_seq;
 330
 331	this_seq = (hdr->flags_seq_tag >> MCTP_HDR_SEQ_SHIFT)
 332		& MCTP_HDR_SEQ_MASK;
 333
 334	if (!key->reasm_head) {
 335		key->reasm_head = skb;
 336		key->reasm_tailp = &(skb_shinfo(skb)->frag_list);
 337		key->last_seq = this_seq;
 338		return 0;
 339	}
 340
 341	exp_seq = (key->last_seq + 1) & MCTP_HDR_SEQ_MASK;
 342
 343	if (this_seq != exp_seq)
 344		return -EINVAL;
 345
 346	if (key->reasm_head->len + skb->len > mctp_message_maxlen)
 347		return -EINVAL;
 348
 349	skb->next = NULL;
 350	skb->sk = NULL;
 351	*key->reasm_tailp = skb;
 352	key->reasm_tailp = &skb->next;
 353
 354	key->last_seq = this_seq;
 355
 356	key->reasm_head->data_len += skb->len;
 357	key->reasm_head->len += skb->len;
 358	key->reasm_head->truesize += skb->truesize;
 359
 360	return 0;
 361}
 362
 363static int mctp_route_input(struct mctp_route *route, struct sk_buff *skb)
 364{
 365	struct mctp_sk_key *key, *any_key = NULL;
 366	struct net *net = dev_net(skb->dev);
 367	struct mctp_sock *msk;
 368	struct mctp_hdr *mh;
 369	unsigned int netid;
 370	unsigned long f;
 371	u8 tag, flags;
 372	int rc;
 373
 374	msk = NULL;
 375	rc = -EINVAL;
 376
 377	/* we may be receiving a locally-routed packet; drop source sk
 378	 * accounting
 379	 */
 380	skb_orphan(skb);
 381
 382	/* ensure we have enough data for a header and a type */
 383	if (skb->len < sizeof(struct mctp_hdr) + 1)
 384		goto out;
 385
 386	/* grab header, advance data ptr */
 387	mh = mctp_hdr(skb);
 388	netid = mctp_cb(skb)->net;
 389	skb_pull(skb, sizeof(struct mctp_hdr));
 390
 391	if (mh->ver != 1)
 392		goto out;
 393
 394	flags = mh->flags_seq_tag & (MCTP_HDR_FLAG_SOM | MCTP_HDR_FLAG_EOM);
 395	tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 396
 397	rcu_read_lock();
 398
 399	/* lookup socket / reasm context, exactly matching (src,dest,tag).
 400	 * we hold a ref on the key, and key->lock held.
 401	 */
 402	key = mctp_lookup_key(net, skb, netid, mh->src, &f);
 403
 404	if (flags & MCTP_HDR_FLAG_SOM) {
 405		if (key) {
 406			msk = container_of(key->sk, struct mctp_sock, sk);
 407		} else {
 408			/* first response to a broadcast? do a more general
 409			 * key lookup to find the socket, but don't use this
 410			 * key for reassembly - we'll create a more specific
 411			 * one for future packets if required (ie, !EOM).
 412			 *
 413			 * this lookup requires key->peer to be MCTP_ADDR_ANY,
 414			 * it doesn't match just any key->peer.
 415			 */
 416			any_key = mctp_lookup_key(net, skb, netid,
 417						  MCTP_ADDR_ANY, &f);
 418			if (any_key) {
 419				msk = container_of(any_key->sk,
 420						   struct mctp_sock, sk);
 421				spin_unlock_irqrestore(&any_key->lock, f);
 422			}
 423		}
 424
 425		if (!key && !msk && (tag & MCTP_HDR_FLAG_TO))
 426			msk = mctp_lookup_bind(net, skb);
 427
 428		if (!msk) {
 429			rc = -ENOENT;
 430			goto out_unlock;
 431		}
 432
 433		/* single-packet message? deliver to socket, clean up any
 434		 * pending key.
 435		 */
 436		if (flags & MCTP_HDR_FLAG_EOM) {
 437			sock_queue_rcv_skb(&msk->sk, skb);
 438			if (key) {
 439				/* we've hit a pending reassembly; not much we
 440				 * can do but drop it
 441				 */
 442				__mctp_key_done_in(key, net, f,
 443						   MCTP_TRACE_KEY_REPLIED);
 444				key = NULL;
 445			}
 446			rc = 0;
 447			goto out_unlock;
 448		}
 449
 450		/* broadcast response or a bind() - create a key for further
 451		 * packets for this message
 452		 */
 453		if (!key) {
 454			key = mctp_key_alloc(msk, netid, mh->dest, mh->src,
 455					     tag, GFP_ATOMIC);
 456			if (!key) {
 457				rc = -ENOMEM;
 458				goto out_unlock;
 459			}
 460
 461			/* we can queue without the key lock here, as the
 462			 * key isn't observable yet
 463			 */
 464			mctp_frag_queue(key, skb);
 465
 466			/* if the key_add fails, we've raced with another
 467			 * SOM packet with the same src, dest and tag. There's
 468			 * no way to distinguish future packets, so all we
 469			 * can do is drop; we'll free the skb on exit from
 470			 * this function.
 471			 */
 472			rc = mctp_key_add(key, msk);
 473			if (!rc)
 474				trace_mctp_key_acquire(key);
 475
 476			/* we don't need to release key->lock on exit, so
 477			 * clean up here and suppress the unlock via
 478			 * setting to NULL
 479			 */
 480			mctp_key_unref(key);
 481			key = NULL;
 482
 483		} else {
 484			if (key->reasm_head || key->reasm_dead) {
 485				/* duplicate start? drop everything */
 486				__mctp_key_done_in(key, net, f,
 487						   MCTP_TRACE_KEY_INVALIDATED);
 488				rc = -EEXIST;
 489				key = NULL;
 490			} else {
 491				rc = mctp_frag_queue(key, skb);
 492			}
 493		}
 494
 495	} else if (key) {
 496		/* this packet continues a previous message; reassemble
 497		 * using the message-specific key
 498		 */
 499
 500		/* we need to be continuing an existing reassembly... */
 501		if (!key->reasm_head)
 502			rc = -EINVAL;
 503		else
 504			rc = mctp_frag_queue(key, skb);
 505
 506		/* end of message? deliver to socket, and we're done with
 507		 * the reassembly/response key
 508		 */
 509		if (!rc && flags & MCTP_HDR_FLAG_EOM) {
 510			sock_queue_rcv_skb(key->sk, key->reasm_head);
 511			key->reasm_head = NULL;
 512			__mctp_key_done_in(key, net, f, MCTP_TRACE_KEY_REPLIED);
 513			key = NULL;
 514		}
 515
 516	} else {
 517		/* not a start, no matching key */
 518		rc = -ENOENT;
 519	}
 520
 521out_unlock:
 522	rcu_read_unlock();
 523	if (key) {
 524		spin_unlock_irqrestore(&key->lock, f);
 525		mctp_key_unref(key);
 526	}
 527	if (any_key)
 528		mctp_key_unref(any_key);
 529out:
 530	if (rc)
 531		kfree_skb(skb);
 532	return rc;
 533}
 534
 535static unsigned int mctp_route_mtu(struct mctp_route *rt)
 536{
 537	return rt->mtu ?: READ_ONCE(rt->dev->dev->mtu);
 538}
 539
 540static int mctp_route_output(struct mctp_route *route, struct sk_buff *skb)
 541{
 542	struct mctp_skb_cb *cb = mctp_cb(skb);
 543	struct mctp_hdr *hdr = mctp_hdr(skb);
 544	char daddr_buf[MAX_ADDR_LEN];
 545	char *daddr = NULL;
 546	unsigned int mtu;
 547	int rc;
 548
 549	skb->protocol = htons(ETH_P_MCTP);
 550
 551	mtu = READ_ONCE(skb->dev->mtu);
 552	if (skb->len > mtu) {
 553		kfree_skb(skb);
 554		return -EMSGSIZE;
 555	}
 556
 557	if (cb->ifindex) {
 558		/* direct route; use the hwaddr we stashed in sendmsg */
 559		if (cb->halen != skb->dev->addr_len) {
 560			/* sanity check, sendmsg should have already caught this */
 561			kfree_skb(skb);
 562			return -EMSGSIZE;
 563		}
 564		daddr = cb->haddr;
 565	} else {
 566		/* If lookup fails let the device handle daddr==NULL */
 567		if (mctp_neigh_lookup(route->dev, hdr->dest, daddr_buf) == 0)
 568			daddr = daddr_buf;
 569	}
 570
 571	rc = dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
 572			     daddr, skb->dev->dev_addr, skb->len);
 573	if (rc < 0) {
 574		kfree_skb(skb);
 575		return -EHOSTUNREACH;
 576	}
 577
 578	mctp_flow_prepare_output(skb, route->dev);
 579
 580	rc = dev_queue_xmit(skb);
 581	if (rc)
 582		rc = net_xmit_errno(rc);
 583
 584	return rc;
 585}
 586
 587/* route alloc/release */
 588static void mctp_route_release(struct mctp_route *rt)
 589{
 590	if (refcount_dec_and_test(&rt->refs)) {
 591		mctp_dev_put(rt->dev);
 592		kfree_rcu(rt, rcu);
 593	}
 594}
 595
 596/* returns a route with the refcount at 1 */
 597static struct mctp_route *mctp_route_alloc(void)
 598{
 599	struct mctp_route *rt;
 600
 601	rt = kzalloc(sizeof(*rt), GFP_KERNEL);
 602	if (!rt)
 603		return NULL;
 604
 605	INIT_LIST_HEAD(&rt->list);
 606	refcount_set(&rt->refs, 1);
 607	rt->output = mctp_route_discard;
 608
 609	return rt;
 610}
 611
 612unsigned int mctp_default_net(struct net *net)
 613{
 614	return READ_ONCE(net->mctp.default_net);
 615}
 616
 617int mctp_default_net_set(struct net *net, unsigned int index)
 618{
 619	if (index == 0)
 620		return -EINVAL;
 621	WRITE_ONCE(net->mctp.default_net, index);
 622	return 0;
 623}
 624
 625/* tag management */
 626static void mctp_reserve_tag(struct net *net, struct mctp_sk_key *key,
 627			     struct mctp_sock *msk)
 628{
 629	struct netns_mctp *mns = &net->mctp;
 630
 631	lockdep_assert_held(&mns->keys_lock);
 632
 633	key->expiry = jiffies + mctp_key_lifetime;
 634	timer_reduce(&msk->key_expiry, key->expiry);
 635
 636	/* we hold the net->key_lock here, allowing updates to both
 637	 * then net and sk
 638	 */
 639	hlist_add_head_rcu(&key->hlist, &mns->keys);
 640	hlist_add_head_rcu(&key->sklist, &msk->keys);
 641	refcount_inc(&key->refs);
 642}
 643
 644/* Allocate a locally-owned tag value for (local, peer), and reserve
 645 * it for the socket msk
 646 */
 647struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
 648					 unsigned int netid,
 649					 mctp_eid_t local, mctp_eid_t peer,
 650					 bool manual, u8 *tagp)
 651{
 652	struct net *net = sock_net(&msk->sk);
 653	struct netns_mctp *mns = &net->mctp;
 654	struct mctp_sk_key *key, *tmp;
 655	unsigned long flags;
 656	u8 tagbits;
 657
 658	/* for NULL destination EIDs, we may get a response from any peer */
 659	if (peer == MCTP_ADDR_NULL)
 660		peer = MCTP_ADDR_ANY;
 661
 662	/* be optimistic, alloc now */
 663	key = mctp_key_alloc(msk, netid, local, peer, 0, GFP_KERNEL);
 664	if (!key)
 665		return ERR_PTR(-ENOMEM);
 666
 667	/* 8 possible tag values */
 668	tagbits = 0xff;
 669
 670	spin_lock_irqsave(&mns->keys_lock, flags);
 671
 672	/* Walk through the existing keys, looking for potential conflicting
 673	 * tags. If we find a conflict, clear that bit from tagbits
 674	 */
 675	hlist_for_each_entry(tmp, &mns->keys, hlist) {
 676		/* We can check the lookup fields (*_addr, tag) without the
 677		 * lock held, they don't change over the lifetime of the key.
 678		 */
 679
 680		/* tags are net-specific */
 681		if (tmp->net != netid)
 682			continue;
 683
 684		/* if we don't own the tag, it can't conflict */
 685		if (tmp->tag & MCTP_HDR_FLAG_TO)
 686			continue;
 687
 688		/* Since we're avoiding conflicting entries, match peer and
 689		 * local addresses, including with a wildcard on ANY. See
 690		 * 'A note on key allocations' for background.
 691		 */
 692		if (peer != MCTP_ADDR_ANY &&
 693		    !mctp_address_matches(tmp->peer_addr, peer))
 694			continue;
 695
 696		if (local != MCTP_ADDR_ANY &&
 697		    !mctp_address_matches(tmp->local_addr, local))
 698			continue;
 699
 700		spin_lock(&tmp->lock);
 701		/* key must still be valid. If we find a match, clear the
 702		 * potential tag value
 703		 */
 704		if (tmp->valid)
 705			tagbits &= ~(1 << tmp->tag);
 706		spin_unlock(&tmp->lock);
 707
 708		if (!tagbits)
 709			break;
 710	}
 711
 712	if (tagbits) {
 713		key->tag = __ffs(tagbits);
 714		mctp_reserve_tag(net, key, msk);
 715		trace_mctp_key_acquire(key);
 716
 717		key->manual_alloc = manual;
 718		*tagp = key->tag;
 719	}
 720
 721	spin_unlock_irqrestore(&mns->keys_lock, flags);
 722
 723	if (!tagbits) {
 724		mctp_key_unref(key);
 725		return ERR_PTR(-EBUSY);
 726	}
 727
 728	return key;
 729}
 730
 731static struct mctp_sk_key *mctp_lookup_prealloc_tag(struct mctp_sock *msk,
 732						    unsigned int netid,
 733						    mctp_eid_t daddr,
 734						    u8 req_tag, u8 *tagp)
 735{
 736	struct net *net = sock_net(&msk->sk);
 737	struct netns_mctp *mns = &net->mctp;
 738	struct mctp_sk_key *key, *tmp;
 739	unsigned long flags;
 740
 741	req_tag &= ~(MCTP_TAG_PREALLOC | MCTP_TAG_OWNER);
 742	key = NULL;
 743
 744	spin_lock_irqsave(&mns->keys_lock, flags);
 745
 746	hlist_for_each_entry(tmp, &mns->keys, hlist) {
 747		if (tmp->net != netid)
 748			continue;
 749
 750		if (tmp->tag != req_tag)
 751			continue;
 752
 753		if (!mctp_address_matches(tmp->peer_addr, daddr))
 754			continue;
 755
 756		if (!tmp->manual_alloc)
 757			continue;
 758
 759		spin_lock(&tmp->lock);
 760		if (tmp->valid) {
 761			key = tmp;
 762			refcount_inc(&key->refs);
 763			spin_unlock(&tmp->lock);
 764			break;
 765		}
 766		spin_unlock(&tmp->lock);
 767	}
 768	spin_unlock_irqrestore(&mns->keys_lock, flags);
 769
 770	if (!key)
 771		return ERR_PTR(-ENOENT);
 772
 773	if (tagp)
 774		*tagp = key->tag;
 775
 776	return key;
 777}
 778
 779/* routing lookups */
 780static bool mctp_rt_match_eid(struct mctp_route *rt,
 781			      unsigned int net, mctp_eid_t eid)
 782{
 783	return READ_ONCE(rt->dev->net) == net &&
 784		rt->min <= eid && rt->max >= eid;
 785}
 786
 787/* compares match, used for duplicate prevention */
 788static bool mctp_rt_compare_exact(struct mctp_route *rt1,
 789				  struct mctp_route *rt2)
 790{
 791	ASSERT_RTNL();
 792	return rt1->dev->net == rt2->dev->net &&
 793		rt1->min == rt2->min &&
 794		rt1->max == rt2->max;
 795}
 796
 797struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
 798				     mctp_eid_t daddr)
 799{
 800	struct mctp_route *tmp, *rt = NULL;
 801
 802	rcu_read_lock();
 803
 804	list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 805		/* TODO: add metrics */
 806		if (mctp_rt_match_eid(tmp, dnet, daddr)) {
 807			if (refcount_inc_not_zero(&tmp->refs)) {
 808				rt = tmp;
 809				break;
 810			}
 811		}
 812	}
 813
 814	rcu_read_unlock();
 815
 816	return rt;
 817}
 818
 819static struct mctp_route *mctp_route_lookup_null(struct net *net,
 820						 struct net_device *dev)
 821{
 822	struct mctp_route *tmp, *rt = NULL;
 823
 824	rcu_read_lock();
 825
 826	list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 827		if (tmp->dev->dev == dev && tmp->type == RTN_LOCAL &&
 828		    refcount_inc_not_zero(&tmp->refs)) {
 829			rt = tmp;
 830			break;
 831		}
 832	}
 833
 834	rcu_read_unlock();
 835
 836	return rt;
 837}
 838
 839static int mctp_do_fragment_route(struct mctp_route *rt, struct sk_buff *skb,
 840				  unsigned int mtu, u8 tag)
 841{
 842	const unsigned int hlen = sizeof(struct mctp_hdr);
 843	struct mctp_hdr *hdr, *hdr2;
 844	unsigned int pos, size, headroom;
 845	struct sk_buff *skb2;
 846	int rc;
 847	u8 seq;
 848
 849	hdr = mctp_hdr(skb);
 850	seq = 0;
 851	rc = 0;
 852
 853	if (mtu < hlen + 1) {
 854		kfree_skb(skb);
 855		return -EMSGSIZE;
 856	}
 857
 858	/* keep same headroom as the original skb */
 859	headroom = skb_headroom(skb);
 860
 861	/* we've got the header */
 862	skb_pull(skb, hlen);
 863
 864	for (pos = 0; pos < skb->len;) {
 865		/* size of message payload */
 866		size = min(mtu - hlen, skb->len - pos);
 867
 868		skb2 = alloc_skb(headroom + hlen + size, GFP_KERNEL);
 869		if (!skb2) {
 870			rc = -ENOMEM;
 871			break;
 872		}
 873
 874		/* generic skb copy */
 875		skb2->protocol = skb->protocol;
 876		skb2->priority = skb->priority;
 877		skb2->dev = skb->dev;
 878		memcpy(skb2->cb, skb->cb, sizeof(skb2->cb));
 879
 880		if (skb->sk)
 881			skb_set_owner_w(skb2, skb->sk);
 882
 883		/* establish packet */
 884		skb_reserve(skb2, headroom);
 885		skb_reset_network_header(skb2);
 886		skb_put(skb2, hlen + size);
 887		skb2->transport_header = skb2->network_header + hlen;
 888
 889		/* copy header fields, calculate SOM/EOM flags & seq */
 890		hdr2 = mctp_hdr(skb2);
 891		hdr2->ver = hdr->ver;
 892		hdr2->dest = hdr->dest;
 893		hdr2->src = hdr->src;
 894		hdr2->flags_seq_tag = tag &
 895			(MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 896
 897		if (pos == 0)
 898			hdr2->flags_seq_tag |= MCTP_HDR_FLAG_SOM;
 899
 900		if (pos + size == skb->len)
 901			hdr2->flags_seq_tag |= MCTP_HDR_FLAG_EOM;
 902
 903		hdr2->flags_seq_tag |= seq << MCTP_HDR_SEQ_SHIFT;
 904
 905		/* copy message payload */
 906		skb_copy_bits(skb, pos, skb_transport_header(skb2), size);
 907
 908		/* we need to copy the extensions, for MCTP flow data */
 909		skb_ext_copy(skb2, skb);
 910
 911		/* do route */
 912		rc = rt->output(rt, skb2);
 913		if (rc)
 914			break;
 915
 916		seq = (seq + 1) & MCTP_HDR_SEQ_MASK;
 917		pos += size;
 918	}
 919
 920	consume_skb(skb);
 921	return rc;
 922}
 923
 924int mctp_local_output(struct sock *sk, struct mctp_route *rt,
 925		      struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag)
 926{
 927	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
 928	struct mctp_skb_cb *cb = mctp_cb(skb);
 929	struct mctp_route tmp_rt = {0};
 930	struct mctp_sk_key *key;
 931	struct mctp_hdr *hdr;
 932	unsigned long flags;
 933	unsigned int netid;
 934	unsigned int mtu;
 935	mctp_eid_t saddr;
 936	bool ext_rt;
 937	int rc;
 938	u8 tag;
 939
 940	rc = -ENODEV;
 941
 942	if (rt) {
 943		ext_rt = false;
 944		if (WARN_ON(!rt->dev))
 945			goto out_release;
 946
 947	} else if (cb->ifindex) {
 948		struct net_device *dev;
 949
 950		ext_rt = true;
 951		rt = &tmp_rt;
 952
 953		rcu_read_lock();
 954		dev = dev_get_by_index_rcu(sock_net(sk), cb->ifindex);
 955		if (!dev) {
 956			rcu_read_unlock();
 957			goto out_free;
 958		}
 959		rt->dev = __mctp_dev_get(dev);
 960		rcu_read_unlock();
 961
 962		if (!rt->dev)
 963			goto out_release;
 964
 965		/* establish temporary route - we set up enough to keep
 966		 * mctp_route_output happy
 967		 */
 968		rt->output = mctp_route_output;
 969		rt->mtu = 0;
 970
 971	} else {
 972		rc = -EINVAL;
 973		goto out_free;
 974	}
 975
 976	spin_lock_irqsave(&rt->dev->addrs_lock, flags);
 977	if (rt->dev->num_addrs == 0) {
 978		rc = -EHOSTUNREACH;
 979	} else {
 980		/* use the outbound interface's first address as our source */
 981		saddr = rt->dev->addrs[0];
 982		rc = 0;
 983	}
 984	spin_unlock_irqrestore(&rt->dev->addrs_lock, flags);
 985	netid = READ_ONCE(rt->dev->net);
 986
 987	if (rc)
 988		goto out_release;
 989
 990	if (req_tag & MCTP_TAG_OWNER) {
 991		if (req_tag & MCTP_TAG_PREALLOC)
 992			key = mctp_lookup_prealloc_tag(msk, netid, daddr,
 993						       req_tag, &tag);
 994		else
 995			key = mctp_alloc_local_tag(msk, netid, saddr, daddr,
 996						   false, &tag);
 997
 998		if (IS_ERR(key)) {
 999			rc = PTR_ERR(key);
1000			goto out_release;
1001		}
1002		mctp_skb_set_flow(skb, key);
1003		/* done with the key in this scope */
1004		mctp_key_unref(key);
1005		tag |= MCTP_HDR_FLAG_TO;
1006	} else {
1007		key = NULL;
1008		tag = req_tag & MCTP_TAG_MASK;
1009	}
1010
1011	skb->protocol = htons(ETH_P_MCTP);
1012	skb->priority = 0;
1013	skb_reset_transport_header(skb);
1014	skb_push(skb, sizeof(struct mctp_hdr));
1015	skb_reset_network_header(skb);
1016	skb->dev = rt->dev->dev;
1017
1018	/* cb->net will have been set on initial ingress */
1019	cb->src = saddr;
1020
1021	/* set up common header fields */
1022	hdr = mctp_hdr(skb);
1023	hdr->ver = 1;
1024	hdr->dest = daddr;
1025	hdr->src = saddr;
1026
1027	mtu = mctp_route_mtu(rt);
1028
1029	if (skb->len + sizeof(struct mctp_hdr) <= mtu) {
1030		hdr->flags_seq_tag = MCTP_HDR_FLAG_SOM |
1031			MCTP_HDR_FLAG_EOM | tag;
1032		rc = rt->output(rt, skb);
1033	} else {
1034		rc = mctp_do_fragment_route(rt, skb, mtu, tag);
1035	}
1036
1037	/* route output functions consume the skb, even on error */
1038	skb = NULL;
1039
1040out_release:
1041	if (!ext_rt)
1042		mctp_route_release(rt);
1043
1044	mctp_dev_put(tmp_rt.dev);
1045
1046out_free:
1047	kfree_skb(skb);
1048	return rc;
1049}
1050
1051/* route management */
1052static int mctp_route_add(struct mctp_dev *mdev, mctp_eid_t daddr_start,
1053			  unsigned int daddr_extent, unsigned int mtu,
1054			  unsigned char type)
1055{
1056	int (*rtfn)(struct mctp_route *rt, struct sk_buff *skb);
1057	struct net *net = dev_net(mdev->dev);
1058	struct mctp_route *rt, *ert;
1059
1060	if (!mctp_address_unicast(daddr_start))
1061		return -EINVAL;
1062
1063	if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
1064		return -EINVAL;
1065
1066	switch (type) {
1067	case RTN_LOCAL:
1068		rtfn = mctp_route_input;
1069		break;
1070	case RTN_UNICAST:
1071		rtfn = mctp_route_output;
1072		break;
1073	default:
1074		return -EINVAL;
1075	}
1076
1077	rt = mctp_route_alloc();
1078	if (!rt)
1079		return -ENOMEM;
1080
1081	rt->min = daddr_start;
1082	rt->max = daddr_start + daddr_extent;
1083	rt->mtu = mtu;
1084	rt->dev = mdev;
1085	mctp_dev_hold(rt->dev);
1086	rt->type = type;
1087	rt->output = rtfn;
1088
1089	ASSERT_RTNL();
1090	/* Prevent duplicate identical routes. */
1091	list_for_each_entry(ert, &net->mctp.routes, list) {
1092		if (mctp_rt_compare_exact(rt, ert)) {
1093			mctp_route_release(rt);
1094			return -EEXIST;
1095		}
1096	}
1097
1098	list_add_rcu(&rt->list, &net->mctp.routes);
1099
1100	return 0;
1101}
1102
1103static int mctp_route_remove(struct mctp_dev *mdev, mctp_eid_t daddr_start,
1104			     unsigned int daddr_extent, unsigned char type)
1105{
1106	struct net *net = dev_net(mdev->dev);
1107	struct mctp_route *rt, *tmp;
1108	mctp_eid_t daddr_end;
1109	bool dropped;
1110
1111	if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
1112		return -EINVAL;
1113
1114	daddr_end = daddr_start + daddr_extent;
1115	dropped = false;
1116
1117	ASSERT_RTNL();
1118
1119	list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1120		if (rt->dev == mdev &&
1121		    rt->min == daddr_start && rt->max == daddr_end &&
1122		    rt->type == type) {
1123			list_del_rcu(&rt->list);
1124			/* TODO: immediate RTM_DELROUTE */
1125			mctp_route_release(rt);
1126			dropped = true;
1127		}
1128	}
1129
1130	return dropped ? 0 : -ENOENT;
1131}
1132
1133int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr)
1134{
1135	return mctp_route_add(mdev, addr, 0, 0, RTN_LOCAL);
1136}
1137
1138int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr)
1139{
1140	return mctp_route_remove(mdev, addr, 0, RTN_LOCAL);
1141}
1142
1143/* removes all entries for a given device */
1144void mctp_route_remove_dev(struct mctp_dev *mdev)
1145{
1146	struct net *net = dev_net(mdev->dev);
1147	struct mctp_route *rt, *tmp;
1148
1149	ASSERT_RTNL();
1150	list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1151		if (rt->dev == mdev) {
1152			list_del_rcu(&rt->list);
1153			/* TODO: immediate RTM_DELROUTE */
1154			mctp_route_release(rt);
1155		}
1156	}
1157}
1158
1159/* Incoming packet-handling */
1160
1161static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,
1162				struct packet_type *pt,
1163				struct net_device *orig_dev)
1164{
1165	struct net *net = dev_net(dev);
1166	struct mctp_dev *mdev;
1167	struct mctp_skb_cb *cb;
1168	struct mctp_route *rt;
1169	struct mctp_hdr *mh;
1170
1171	rcu_read_lock();
1172	mdev = __mctp_dev_get(dev);
1173	rcu_read_unlock();
1174	if (!mdev) {
1175		/* basic non-data sanity checks */
1176		goto err_drop;
1177	}
1178
1179	if (!pskb_may_pull(skb, sizeof(struct mctp_hdr)))
1180		goto err_drop;
1181
1182	skb_reset_transport_header(skb);
1183	skb_reset_network_header(skb);
1184
1185	/* We have enough for a header; decode and route */
1186	mh = mctp_hdr(skb);
1187	if (mh->ver < MCTP_VER_MIN || mh->ver > MCTP_VER_MAX)
1188		goto err_drop;
1189
1190	/* source must be valid unicast or null; drop reserved ranges and
1191	 * broadcast
1192	 */
1193	if (!(mctp_address_unicast(mh->src) || mctp_address_null(mh->src)))
1194		goto err_drop;
1195
1196	/* dest address: as above, but allow broadcast */
1197	if (!(mctp_address_unicast(mh->dest) || mctp_address_null(mh->dest) ||
1198	      mctp_address_broadcast(mh->dest)))
1199		goto err_drop;
1200
1201	/* MCTP drivers must populate halen/haddr */
1202	if (dev->type == ARPHRD_MCTP) {
1203		cb = mctp_cb(skb);
1204	} else {
1205		cb = __mctp_cb(skb);
1206		cb->halen = 0;
1207	}
1208	cb->net = READ_ONCE(mdev->net);
1209	cb->ifindex = dev->ifindex;
1210
1211	rt = mctp_route_lookup(net, cb->net, mh->dest);
1212
1213	/* NULL EID, but addressed to our physical address */
1214	if (!rt && mh->dest == MCTP_ADDR_NULL && skb->pkt_type == PACKET_HOST)
1215		rt = mctp_route_lookup_null(net, dev);
1216
1217	if (!rt)
1218		goto err_drop;
1219
1220	rt->output(rt, skb);
1221	mctp_route_release(rt);
1222	mctp_dev_put(mdev);
1223
1224	return NET_RX_SUCCESS;
1225
1226err_drop:
1227	kfree_skb(skb);
1228	mctp_dev_put(mdev);
1229	return NET_RX_DROP;
1230}
1231
1232static struct packet_type mctp_packet_type = {
1233	.type = cpu_to_be16(ETH_P_MCTP),
1234	.func = mctp_pkttype_receive,
1235};
1236
1237/* netlink interface */
1238
1239static const struct nla_policy rta_mctp_policy[RTA_MAX + 1] = {
1240	[RTA_DST]		= { .type = NLA_U8 },
1241	[RTA_METRICS]		= { .type = NLA_NESTED },
1242	[RTA_OIF]		= { .type = NLA_U32 },
1243};
1244
1245/* Common part for RTM_NEWROUTE and RTM_DELROUTE parsing.
1246 * tb must hold RTA_MAX+1 elements.
1247 */
1248static int mctp_route_nlparse(struct sk_buff *skb, struct nlmsghdr *nlh,
1249			      struct netlink_ext_ack *extack,
1250			      struct nlattr **tb, struct rtmsg **rtm,
1251			      struct mctp_dev **mdev, mctp_eid_t *daddr_start)
1252{
1253	struct net *net = sock_net(skb->sk);
1254	struct net_device *dev;
1255	unsigned int ifindex;
1256	int rc;
1257
1258	rc = nlmsg_parse(nlh, sizeof(struct rtmsg), tb, RTA_MAX,
1259			 rta_mctp_policy, extack);
1260	if (rc < 0) {
1261		NL_SET_ERR_MSG(extack, "incorrect format");
1262		return rc;
1263	}
1264
1265	if (!tb[RTA_DST]) {
1266		NL_SET_ERR_MSG(extack, "dst EID missing");
1267		return -EINVAL;
1268	}
1269	*daddr_start = nla_get_u8(tb[RTA_DST]);
1270
1271	if (!tb[RTA_OIF]) {
1272		NL_SET_ERR_MSG(extack, "ifindex missing");
1273		return -EINVAL;
1274	}
1275	ifindex = nla_get_u32(tb[RTA_OIF]);
1276
1277	*rtm = nlmsg_data(nlh);
1278	if ((*rtm)->rtm_family != AF_MCTP) {
1279		NL_SET_ERR_MSG(extack, "route family must be AF_MCTP");
1280		return -EINVAL;
1281	}
1282
1283	dev = __dev_get_by_index(net, ifindex);
1284	if (!dev) {
1285		NL_SET_ERR_MSG(extack, "bad ifindex");
1286		return -ENODEV;
1287	}
1288	*mdev = mctp_dev_get_rtnl(dev);
1289	if (!*mdev)
1290		return -ENODEV;
1291
1292	if (dev->flags & IFF_LOOPBACK) {
1293		NL_SET_ERR_MSG(extack, "no routes to loopback");
1294		return -EINVAL;
1295	}
1296
1297	return 0;
1298}
1299
1300static const struct nla_policy rta_metrics_policy[RTAX_MAX + 1] = {
1301	[RTAX_MTU]		= { .type = NLA_U32 },
1302};
1303
1304static int mctp_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1305			 struct netlink_ext_ack *extack)
1306{
1307	struct nlattr *tb[RTA_MAX + 1];
1308	struct nlattr *tbx[RTAX_MAX + 1];
1309	mctp_eid_t daddr_start;
1310	struct mctp_dev *mdev;
1311	struct rtmsg *rtm;
1312	unsigned int mtu;
1313	int rc;
1314
1315	rc = mctp_route_nlparse(skb, nlh, extack, tb,
1316				&rtm, &mdev, &daddr_start);
1317	if (rc < 0)
1318		return rc;
1319
1320	if (rtm->rtm_type != RTN_UNICAST) {
1321		NL_SET_ERR_MSG(extack, "rtm_type must be RTN_UNICAST");
1322		return -EINVAL;
1323	}
1324
1325	mtu = 0;
1326	if (tb[RTA_METRICS]) {
1327		rc = nla_parse_nested(tbx, RTAX_MAX, tb[RTA_METRICS],
1328				      rta_metrics_policy, NULL);
1329		if (rc < 0)
1330			return rc;
1331		if (tbx[RTAX_MTU])
1332			mtu = nla_get_u32(tbx[RTAX_MTU]);
1333	}
1334
1335	rc = mctp_route_add(mdev, daddr_start, rtm->rtm_dst_len, mtu,
1336			    rtm->rtm_type);
1337	return rc;
1338}
1339
1340static int mctp_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1341			 struct netlink_ext_ack *extack)
1342{
1343	struct nlattr *tb[RTA_MAX + 1];
1344	mctp_eid_t daddr_start;
1345	struct mctp_dev *mdev;
1346	struct rtmsg *rtm;
1347	int rc;
1348
1349	rc = mctp_route_nlparse(skb, nlh, extack, tb,
1350				&rtm, &mdev, &daddr_start);
1351	if (rc < 0)
1352		return rc;
1353
1354	/* we only have unicast routes */
1355	if (rtm->rtm_type != RTN_UNICAST)
1356		return -EINVAL;
1357
1358	rc = mctp_route_remove(mdev, daddr_start, rtm->rtm_dst_len, RTN_UNICAST);
1359	return rc;
1360}
1361
1362static int mctp_fill_rtinfo(struct sk_buff *skb, struct mctp_route *rt,
1363			    u32 portid, u32 seq, int event, unsigned int flags)
1364{
1365	struct nlmsghdr *nlh;
1366	struct rtmsg *hdr;
1367	void *metrics;
1368
1369	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*hdr), flags);
1370	if (!nlh)
1371		return -EMSGSIZE;
1372
1373	hdr = nlmsg_data(nlh);
1374	hdr->rtm_family = AF_MCTP;
1375
1376	/* we use the _len fields as a number of EIDs, rather than
1377	 * a number of bits in the address
1378	 */
1379	hdr->rtm_dst_len = rt->max - rt->min;
1380	hdr->rtm_src_len = 0;
1381	hdr->rtm_tos = 0;
1382	hdr->rtm_table = RT_TABLE_DEFAULT;
1383	hdr->rtm_protocol = RTPROT_STATIC; /* everything is user-defined */
1384	hdr->rtm_scope = RT_SCOPE_LINK; /* TODO: scope in mctp_route? */
1385	hdr->rtm_type = rt->type;
1386
1387	if (nla_put_u8(skb, RTA_DST, rt->min))
1388		goto cancel;
1389
1390	metrics = nla_nest_start_noflag(skb, RTA_METRICS);
1391	if (!metrics)
1392		goto cancel;
1393
1394	if (rt->mtu) {
1395		if (nla_put_u32(skb, RTAX_MTU, rt->mtu))
1396			goto cancel;
1397	}
1398
1399	nla_nest_end(skb, metrics);
1400
1401	if (rt->dev) {
1402		if (nla_put_u32(skb, RTA_OIF, rt->dev->dev->ifindex))
1403			goto cancel;
1404	}
1405
1406	/* TODO: conditional neighbour physaddr? */
1407
1408	nlmsg_end(skb, nlh);
1409
1410	return 0;
1411
1412cancel:
1413	nlmsg_cancel(skb, nlh);
1414	return -EMSGSIZE;
1415}
1416
1417static int mctp_dump_rtinfo(struct sk_buff *skb, struct netlink_callback *cb)
1418{
1419	struct net *net = sock_net(skb->sk);
1420	struct mctp_route *rt;
1421	int s_idx, idx;
1422
1423	/* TODO: allow filtering on route data, possibly under
1424	 * cb->strict_check
1425	 */
1426
1427	/* TODO: change to struct overlay */
1428	s_idx = cb->args[0];
1429	idx = 0;
1430
1431	rcu_read_lock();
1432	list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
1433		if (idx++ < s_idx)
1434			continue;
1435		if (mctp_fill_rtinfo(skb, rt,
1436				     NETLINK_CB(cb->skb).portid,
1437				     cb->nlh->nlmsg_seq,
1438				     RTM_NEWROUTE, NLM_F_MULTI) < 0)
1439			break;
1440	}
1441
1442	rcu_read_unlock();
1443	cb->args[0] = idx;
1444
1445	return skb->len;
1446}
1447
1448/* net namespace implementation */
1449static int __net_init mctp_routes_net_init(struct net *net)
1450{
1451	struct netns_mctp *ns = &net->mctp;
1452
1453	INIT_LIST_HEAD(&ns->routes);
1454	INIT_HLIST_HEAD(&ns->binds);
1455	mutex_init(&ns->bind_lock);
1456	INIT_HLIST_HEAD(&ns->keys);
1457	spin_lock_init(&ns->keys_lock);
1458	WARN_ON(mctp_default_net_set(net, MCTP_INITIAL_DEFAULT_NET));
1459	return 0;
1460}
1461
1462static void __net_exit mctp_routes_net_exit(struct net *net)
1463{
1464	struct mctp_route *rt;
1465
1466	rcu_read_lock();
1467	list_for_each_entry_rcu(rt, &net->mctp.routes, list)
1468		mctp_route_release(rt);
1469	rcu_read_unlock();
1470}
1471
1472static struct pernet_operations mctp_net_ops = {
1473	.init = mctp_routes_net_init,
1474	.exit = mctp_routes_net_exit,
1475};
1476
1477int __init mctp_routes_init(void)
1478{
1479	dev_add_pack(&mctp_packet_type);
1480
1481	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETROUTE,
1482			     NULL, mctp_dump_rtinfo, 0);
1483	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWROUTE,
1484			     mctp_newroute, NULL, 0);
1485	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELROUTE,
1486			     mctp_delroute, NULL, 0);
1487
1488	return register_pernet_subsys(&mctp_net_ops);
1489}
1490
1491void mctp_routes_exit(void)
1492{
1493	unregister_pernet_subsys(&mctp_net_ops);
1494	rtnl_unregister(PF_MCTP, RTM_DELROUTE);
1495	rtnl_unregister(PF_MCTP, RTM_NEWROUTE);
1496	rtnl_unregister(PF_MCTP, RTM_GETROUTE);
1497	dev_remove_pack(&mctp_packet_type);
1498}
1499
1500#if IS_ENABLED(CONFIG_MCTP_TEST)
1501#include "test/route-test.c"
1502#endif