xref: /openbmc/linux/net/mctp/route.c (revision d839e4ff)
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 
29 static const unsigned int mctp_message_maxlen = 64 * 1024;
30 static const unsigned long mctp_key_lifetime = 6 * CONFIG_HZ;
31 
32 /* route output callbacks */
33 static int mctp_route_discard(struct mctp_route *route, struct sk_buff *skb)
34 {
35 	kfree_skb(skb);
36 	return 0;
37 }
38 
39 static struct mctp_sock *mctp_lookup_bind(struct net *net, struct sk_buff *skb)
40 {
41 	struct mctp_skb_cb *cb = mctp_cb(skb);
42 	struct mctp_hdr *mh;
43 	struct sock *sk;
44 	u8 type;
45 
46 	WARN_ON(!rcu_read_lock_held());
47 
48 	/* TODO: look up in skb->cb? */
49 	mh = mctp_hdr(skb);
50 
51 	if (!skb_headlen(skb))
52 		return NULL;
53 
54 	type = (*(u8 *)skb->data) & 0x7f;
55 
56 	sk_for_each_rcu(sk, &net->mctp.binds) {
57 		struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
58 
59 		if (msk->bind_net != MCTP_NET_ANY && msk->bind_net != cb->net)
60 			continue;
61 
62 		if (msk->bind_type != type)
63 			continue;
64 
65 		if (msk->bind_addr != MCTP_ADDR_ANY &&
66 		    msk->bind_addr != mh->dest)
67 			continue;
68 
69 		return msk;
70 	}
71 
72 	return NULL;
73 }
74 
75 static bool mctp_key_match(struct mctp_sk_key *key, mctp_eid_t local,
76 			   mctp_eid_t peer, u8 tag)
77 {
78 	if (key->local_addr != local)
79 		return false;
80 
81 	if (key->peer_addr != peer)
82 		return false;
83 
84 	if (key->tag != tag)
85 		return false;
86 
87 	return true;
88 }
89 
90 /* returns a key (with key->lock held, and refcounted), or NULL if no such
91  * key exists.
92  */
93 static struct mctp_sk_key *mctp_lookup_key(struct net *net, struct sk_buff *skb,
94 					   mctp_eid_t peer,
95 					   unsigned long *irqflags)
96 	__acquires(&key->lock)
97 {
98 	struct mctp_sk_key *key, *ret;
99 	unsigned long flags;
100 	struct mctp_hdr *mh;
101 	u8 tag;
102 
103 	mh = mctp_hdr(skb);
104 	tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
105 
106 	ret = NULL;
107 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
108 
109 	hlist_for_each_entry(key, &net->mctp.keys, hlist) {
110 		if (!mctp_key_match(key, mh->dest, peer, tag))
111 			continue;
112 
113 		spin_lock(&key->lock);
114 		if (key->valid) {
115 			refcount_inc(&key->refs);
116 			ret = key;
117 			break;
118 		}
119 		spin_unlock(&key->lock);
120 	}
121 
122 	if (ret) {
123 		spin_unlock(&net->mctp.keys_lock);
124 		*irqflags = flags;
125 	} else {
126 		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
127 	}
128 
129 	return ret;
130 }
131 
132 static struct mctp_sk_key *mctp_key_alloc(struct mctp_sock *msk,
133 					  mctp_eid_t local, mctp_eid_t peer,
134 					  u8 tag, gfp_t gfp)
135 {
136 	struct mctp_sk_key *key;
137 
138 	key = kzalloc(sizeof(*key), gfp);
139 	if (!key)
140 		return NULL;
141 
142 	key->peer_addr = peer;
143 	key->local_addr = local;
144 	key->tag = tag;
145 	key->sk = &msk->sk;
146 	key->valid = true;
147 	spin_lock_init(&key->lock);
148 	refcount_set(&key->refs, 1);
149 
150 	return key;
151 }
152 
153 void mctp_key_unref(struct mctp_sk_key *key)
154 {
155 	if (refcount_dec_and_test(&key->refs))
156 		kfree(key);
157 }
158 
159 static int mctp_key_add(struct mctp_sk_key *key, struct mctp_sock *msk)
160 {
161 	struct net *net = sock_net(&msk->sk);
162 	struct mctp_sk_key *tmp;
163 	unsigned long flags;
164 	int rc = 0;
165 
166 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
167 
168 	hlist_for_each_entry(tmp, &net->mctp.keys, hlist) {
169 		if (mctp_key_match(tmp, key->local_addr, key->peer_addr,
170 				   key->tag)) {
171 			spin_lock(&tmp->lock);
172 			if (tmp->valid)
173 				rc = -EEXIST;
174 			spin_unlock(&tmp->lock);
175 			if (rc)
176 				break;
177 		}
178 	}
179 
180 	if (!rc) {
181 		refcount_inc(&key->refs);
182 		key->expiry = jiffies + mctp_key_lifetime;
183 		timer_reduce(&msk->key_expiry, key->expiry);
184 
185 		hlist_add_head(&key->hlist, &net->mctp.keys);
186 		hlist_add_head(&key->sklist, &msk->keys);
187 	}
188 
189 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
190 
191 	return rc;
192 }
193 
194 /* We're done with the key; unset valid and remove from lists. There may still
195  * be outstanding refs on the key though...
196  */
197 static void __mctp_key_unlock_drop(struct mctp_sk_key *key, struct net *net,
198 				   unsigned long flags)
199 	__releases(&key->lock)
200 {
201 	struct sk_buff *skb;
202 
203 	skb = key->reasm_head;
204 	key->reasm_head = NULL;
205 	key->reasm_dead = true;
206 	key->valid = false;
207 	spin_unlock_irqrestore(&key->lock, flags);
208 
209 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
210 	hlist_del(&key->hlist);
211 	hlist_del(&key->sklist);
212 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
213 
214 	/* one unref for the lists */
215 	mctp_key_unref(key);
216 
217 	/* and one for the local reference */
218 	mctp_key_unref(key);
219 
220 	if (skb)
221 		kfree_skb(skb);
222 
223 }
224 
225 static int mctp_frag_queue(struct mctp_sk_key *key, struct sk_buff *skb)
226 {
227 	struct mctp_hdr *hdr = mctp_hdr(skb);
228 	u8 exp_seq, this_seq;
229 
230 	this_seq = (hdr->flags_seq_tag >> MCTP_HDR_SEQ_SHIFT)
231 		& MCTP_HDR_SEQ_MASK;
232 
233 	if (!key->reasm_head) {
234 		key->reasm_head = skb;
235 		key->reasm_tailp = &(skb_shinfo(skb)->frag_list);
236 		key->last_seq = this_seq;
237 		return 0;
238 	}
239 
240 	exp_seq = (key->last_seq + 1) & MCTP_HDR_SEQ_MASK;
241 
242 	if (this_seq != exp_seq)
243 		return -EINVAL;
244 
245 	if (key->reasm_head->len + skb->len > mctp_message_maxlen)
246 		return -EINVAL;
247 
248 	skb->next = NULL;
249 	skb->sk = NULL;
250 	*key->reasm_tailp = skb;
251 	key->reasm_tailp = &skb->next;
252 
253 	key->last_seq = this_seq;
254 
255 	key->reasm_head->data_len += skb->len;
256 	key->reasm_head->len += skb->len;
257 	key->reasm_head->truesize += skb->truesize;
258 
259 	return 0;
260 }
261 
262 static int mctp_route_input(struct mctp_route *route, struct sk_buff *skb)
263 {
264 	struct net *net = dev_net(skb->dev);
265 	struct mctp_sk_key *key;
266 	struct mctp_sock *msk;
267 	struct mctp_hdr *mh;
268 	unsigned long f;
269 	u8 tag, flags;
270 	int rc;
271 
272 	msk = NULL;
273 	rc = -EINVAL;
274 
275 	/* we may be receiving a locally-routed packet; drop source sk
276 	 * accounting
277 	 */
278 	skb_orphan(skb);
279 
280 	/* ensure we have enough data for a header and a type */
281 	if (skb->len < sizeof(struct mctp_hdr) + 1)
282 		goto out;
283 
284 	/* grab header, advance data ptr */
285 	mh = mctp_hdr(skb);
286 	skb_pull(skb, sizeof(struct mctp_hdr));
287 
288 	if (mh->ver != 1)
289 		goto out;
290 
291 	flags = mh->flags_seq_tag & (MCTP_HDR_FLAG_SOM | MCTP_HDR_FLAG_EOM);
292 	tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
293 
294 	rcu_read_lock();
295 
296 	/* lookup socket / reasm context, exactly matching (src,dest,tag).
297 	 * we hold a ref on the key, and key->lock held.
298 	 */
299 	key = mctp_lookup_key(net, skb, mh->src, &f);
300 
301 	if (flags & MCTP_HDR_FLAG_SOM) {
302 		if (key) {
303 			msk = container_of(key->sk, struct mctp_sock, sk);
304 		} else {
305 			/* first response to a broadcast? do a more general
306 			 * key lookup to find the socket, but don't use this
307 			 * key for reassembly - we'll create a more specific
308 			 * one for future packets if required (ie, !EOM).
309 			 */
310 			key = mctp_lookup_key(net, skb, MCTP_ADDR_ANY, &f);
311 			if (key) {
312 				msk = container_of(key->sk,
313 						   struct mctp_sock, sk);
314 				spin_unlock_irqrestore(&key->lock, f);
315 				mctp_key_unref(key);
316 				key = NULL;
317 			}
318 		}
319 
320 		if (!key && !msk && (tag & MCTP_HDR_FLAG_TO))
321 			msk = mctp_lookup_bind(net, skb);
322 
323 		if (!msk) {
324 			rc = -ENOENT;
325 			goto out_unlock;
326 		}
327 
328 		/* single-packet message? deliver to socket, clean up any
329 		 * pending key.
330 		 */
331 		if (flags & MCTP_HDR_FLAG_EOM) {
332 			sock_queue_rcv_skb(&msk->sk, skb);
333 			if (key) {
334 				/* we've hit a pending reassembly; not much we
335 				 * can do but drop it
336 				 */
337 				trace_mctp_key_release(key,
338 						       MCTP_TRACE_KEY_REPLIED);
339 				__mctp_key_unlock_drop(key, net, f);
340 				key = NULL;
341 			}
342 			rc = 0;
343 			goto out_unlock;
344 		}
345 
346 		/* broadcast response or a bind() - create a key for further
347 		 * packets for this message
348 		 */
349 		if (!key) {
350 			key = mctp_key_alloc(msk, mh->dest, mh->src,
351 					     tag, GFP_ATOMIC);
352 			if (!key) {
353 				rc = -ENOMEM;
354 				goto out_unlock;
355 			}
356 
357 			/* we can queue without the key lock here, as the
358 			 * key isn't observable yet
359 			 */
360 			mctp_frag_queue(key, skb);
361 
362 			/* if the key_add fails, we've raced with another
363 			 * SOM packet with the same src, dest and tag. There's
364 			 * no way to distinguish future packets, so all we
365 			 * can do is drop; we'll free the skb on exit from
366 			 * this function.
367 			 */
368 			rc = mctp_key_add(key, msk);
369 			if (rc)
370 				kfree(key);
371 
372 			trace_mctp_key_acquire(key);
373 
374 			/* we don't need to release key->lock on exit */
375 			key = NULL;
376 
377 		} else {
378 			if (key->reasm_head || key->reasm_dead) {
379 				/* duplicate start? drop everything */
380 				trace_mctp_key_release(key,
381 						       MCTP_TRACE_KEY_INVALIDATED);
382 				__mctp_key_unlock_drop(key, net, f);
383 				rc = -EEXIST;
384 				key = NULL;
385 			} else {
386 				rc = mctp_frag_queue(key, skb);
387 			}
388 		}
389 
390 	} else if (key) {
391 		/* this packet continues a previous message; reassemble
392 		 * using the message-specific key
393 		 */
394 
395 		/* we need to be continuing an existing reassembly... */
396 		if (!key->reasm_head)
397 			rc = -EINVAL;
398 		else
399 			rc = mctp_frag_queue(key, skb);
400 
401 		/* end of message? deliver to socket, and we're done with
402 		 * the reassembly/response key
403 		 */
404 		if (!rc && flags & MCTP_HDR_FLAG_EOM) {
405 			sock_queue_rcv_skb(key->sk, key->reasm_head);
406 			key->reasm_head = NULL;
407 			trace_mctp_key_release(key, MCTP_TRACE_KEY_REPLIED);
408 			__mctp_key_unlock_drop(key, net, f);
409 			key = NULL;
410 		}
411 
412 	} else {
413 		/* not a start, no matching key */
414 		rc = -ENOENT;
415 	}
416 
417 out_unlock:
418 	rcu_read_unlock();
419 	if (key) {
420 		spin_unlock_irqrestore(&key->lock, f);
421 		mctp_key_unref(key);
422 	}
423 out:
424 	if (rc)
425 		kfree_skb(skb);
426 	return rc;
427 }
428 
429 static unsigned int mctp_route_mtu(struct mctp_route *rt)
430 {
431 	return rt->mtu ?: READ_ONCE(rt->dev->dev->mtu);
432 }
433 
434 static int mctp_route_output(struct mctp_route *route, struct sk_buff *skb)
435 {
436 	struct mctp_hdr *hdr = mctp_hdr(skb);
437 	char daddr_buf[MAX_ADDR_LEN];
438 	char *daddr = NULL;
439 	unsigned int mtu;
440 	int rc;
441 
442 	skb->protocol = htons(ETH_P_MCTP);
443 
444 	mtu = READ_ONCE(skb->dev->mtu);
445 	if (skb->len > mtu) {
446 		kfree_skb(skb);
447 		return -EMSGSIZE;
448 	}
449 
450 	/* If lookup fails let the device handle daddr==NULL */
451 	if (mctp_neigh_lookup(route->dev, hdr->dest, daddr_buf) == 0)
452 		daddr = daddr_buf;
453 
454 	rc = dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
455 			     daddr, skb->dev->dev_addr, skb->len);
456 	if (rc) {
457 		kfree_skb(skb);
458 		return -EHOSTUNREACH;
459 	}
460 
461 	rc = dev_queue_xmit(skb);
462 	if (rc)
463 		rc = net_xmit_errno(rc);
464 
465 	return rc;
466 }
467 
468 /* route alloc/release */
469 static void mctp_route_release(struct mctp_route *rt)
470 {
471 	if (refcount_dec_and_test(&rt->refs)) {
472 		mctp_dev_put(rt->dev);
473 		kfree_rcu(rt, rcu);
474 	}
475 }
476 
477 /* returns a route with the refcount at 1 */
478 static struct mctp_route *mctp_route_alloc(void)
479 {
480 	struct mctp_route *rt;
481 
482 	rt = kzalloc(sizeof(*rt), GFP_KERNEL);
483 	if (!rt)
484 		return NULL;
485 
486 	INIT_LIST_HEAD(&rt->list);
487 	refcount_set(&rt->refs, 1);
488 	rt->output = mctp_route_discard;
489 
490 	return rt;
491 }
492 
493 unsigned int mctp_default_net(struct net *net)
494 {
495 	return READ_ONCE(net->mctp.default_net);
496 }
497 
498 int mctp_default_net_set(struct net *net, unsigned int index)
499 {
500 	if (index == 0)
501 		return -EINVAL;
502 	WRITE_ONCE(net->mctp.default_net, index);
503 	return 0;
504 }
505 
506 /* tag management */
507 static void mctp_reserve_tag(struct net *net, struct mctp_sk_key *key,
508 			     struct mctp_sock *msk)
509 {
510 	struct netns_mctp *mns = &net->mctp;
511 
512 	lockdep_assert_held(&mns->keys_lock);
513 
514 	key->expiry = jiffies + mctp_key_lifetime;
515 	timer_reduce(&msk->key_expiry, key->expiry);
516 
517 	/* we hold the net->key_lock here, allowing updates to both
518 	 * then net and sk
519 	 */
520 	hlist_add_head_rcu(&key->hlist, &mns->keys);
521 	hlist_add_head_rcu(&key->sklist, &msk->keys);
522 	refcount_inc(&key->refs);
523 }
524 
525 /* Allocate a locally-owned tag value for (saddr, daddr), and reserve
526  * it for the socket msk
527  */
528 static int mctp_alloc_local_tag(struct mctp_sock *msk,
529 				mctp_eid_t saddr, mctp_eid_t daddr, u8 *tagp)
530 {
531 	struct net *net = sock_net(&msk->sk);
532 	struct netns_mctp *mns = &net->mctp;
533 	struct mctp_sk_key *key, *tmp;
534 	unsigned long flags;
535 	int rc = -EAGAIN;
536 	u8 tagbits;
537 
538 	/* for NULL destination EIDs, we may get a response from any peer */
539 	if (daddr == MCTP_ADDR_NULL)
540 		daddr = MCTP_ADDR_ANY;
541 
542 	/* be optimistic, alloc now */
543 	key = mctp_key_alloc(msk, saddr, daddr, 0, GFP_KERNEL);
544 	if (!key)
545 		return -ENOMEM;
546 
547 	/* 8 possible tag values */
548 	tagbits = 0xff;
549 
550 	spin_lock_irqsave(&mns->keys_lock, flags);
551 
552 	/* Walk through the existing keys, looking for potential conflicting
553 	 * tags. If we find a conflict, clear that bit from tagbits
554 	 */
555 	hlist_for_each_entry(tmp, &mns->keys, hlist) {
556 		/* We can check the lookup fields (*_addr, tag) without the
557 		 * lock held, they don't change over the lifetime of the key.
558 		 */
559 
560 		/* if we don't own the tag, it can't conflict */
561 		if (tmp->tag & MCTP_HDR_FLAG_TO)
562 			continue;
563 
564 		if (!((tmp->peer_addr == daddr ||
565 		       tmp->peer_addr == MCTP_ADDR_ANY) &&
566 		       tmp->local_addr == saddr))
567 			continue;
568 
569 		spin_lock(&tmp->lock);
570 		/* key must still be valid. If we find a match, clear the
571 		 * potential tag value
572 		 */
573 		if (tmp->valid)
574 			tagbits &= ~(1 << tmp->tag);
575 		spin_unlock(&tmp->lock);
576 
577 		if (!tagbits)
578 			break;
579 	}
580 
581 	if (tagbits) {
582 		key->tag = __ffs(tagbits);
583 		mctp_reserve_tag(net, key, msk);
584 		trace_mctp_key_acquire(key);
585 
586 		*tagp = key->tag;
587 		rc = 0;
588 	}
589 
590 	spin_unlock_irqrestore(&mns->keys_lock, flags);
591 
592 	if (!tagbits)
593 		kfree(key);
594 
595 	return rc;
596 }
597 
598 /* routing lookups */
599 static bool mctp_rt_match_eid(struct mctp_route *rt,
600 			      unsigned int net, mctp_eid_t eid)
601 {
602 	return READ_ONCE(rt->dev->net) == net &&
603 		rt->min <= eid && rt->max >= eid;
604 }
605 
606 /* compares match, used for duplicate prevention */
607 static bool mctp_rt_compare_exact(struct mctp_route *rt1,
608 				  struct mctp_route *rt2)
609 {
610 	ASSERT_RTNL();
611 	return rt1->dev->net == rt2->dev->net &&
612 		rt1->min == rt2->min &&
613 		rt1->max == rt2->max;
614 }
615 
616 struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
617 				     mctp_eid_t daddr)
618 {
619 	struct mctp_route *tmp, *rt = NULL;
620 
621 	list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
622 		/* TODO: add metrics */
623 		if (mctp_rt_match_eid(tmp, dnet, daddr)) {
624 			if (refcount_inc_not_zero(&tmp->refs)) {
625 				rt = tmp;
626 				break;
627 			}
628 		}
629 	}
630 
631 	return rt;
632 }
633 
634 static struct mctp_route *mctp_route_lookup_null(struct net *net,
635 						 struct net_device *dev)
636 {
637 	struct mctp_route *rt;
638 
639 	list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
640 		if (rt->dev->dev == dev && rt->type == RTN_LOCAL &&
641 		    refcount_inc_not_zero(&rt->refs))
642 			return rt;
643 	}
644 
645 	return NULL;
646 }
647 
648 /* sends a skb to rt and releases the route. */
649 int mctp_do_route(struct mctp_route *rt, struct sk_buff *skb)
650 {
651 	int rc;
652 
653 	rc = rt->output(rt, skb);
654 	mctp_route_release(rt);
655 	return rc;
656 }
657 
658 static int mctp_do_fragment_route(struct mctp_route *rt, struct sk_buff *skb,
659 				  unsigned int mtu, u8 tag)
660 {
661 	const unsigned int hlen = sizeof(struct mctp_hdr);
662 	struct mctp_hdr *hdr, *hdr2;
663 	unsigned int pos, size;
664 	struct sk_buff *skb2;
665 	int rc;
666 	u8 seq;
667 
668 	hdr = mctp_hdr(skb);
669 	seq = 0;
670 	rc = 0;
671 
672 	if (mtu < hlen + 1) {
673 		kfree_skb(skb);
674 		return -EMSGSIZE;
675 	}
676 
677 	/* we've got the header */
678 	skb_pull(skb, hlen);
679 
680 	for (pos = 0; pos < skb->len;) {
681 		/* size of message payload */
682 		size = min(mtu - hlen, skb->len - pos);
683 
684 		skb2 = alloc_skb(MCTP_HEADER_MAXLEN + hlen + size, GFP_KERNEL);
685 		if (!skb2) {
686 			rc = -ENOMEM;
687 			break;
688 		}
689 
690 		/* generic skb copy */
691 		skb2->protocol = skb->protocol;
692 		skb2->priority = skb->priority;
693 		skb2->dev = skb->dev;
694 		memcpy(skb2->cb, skb->cb, sizeof(skb2->cb));
695 
696 		if (skb->sk)
697 			skb_set_owner_w(skb2, skb->sk);
698 
699 		/* establish packet */
700 		skb_reserve(skb2, MCTP_HEADER_MAXLEN);
701 		skb_reset_network_header(skb2);
702 		skb_put(skb2, hlen + size);
703 		skb2->transport_header = skb2->network_header + hlen;
704 
705 		/* copy header fields, calculate SOM/EOM flags & seq */
706 		hdr2 = mctp_hdr(skb2);
707 		hdr2->ver = hdr->ver;
708 		hdr2->dest = hdr->dest;
709 		hdr2->src = hdr->src;
710 		hdr2->flags_seq_tag = tag &
711 			(MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
712 
713 		if (pos == 0)
714 			hdr2->flags_seq_tag |= MCTP_HDR_FLAG_SOM;
715 
716 		if (pos + size == skb->len)
717 			hdr2->flags_seq_tag |= MCTP_HDR_FLAG_EOM;
718 
719 		hdr2->flags_seq_tag |= seq << MCTP_HDR_SEQ_SHIFT;
720 
721 		/* copy message payload */
722 		skb_copy_bits(skb, pos, skb_transport_header(skb2), size);
723 
724 		/* do route, but don't drop the rt reference */
725 		rc = rt->output(rt, skb2);
726 		if (rc)
727 			break;
728 
729 		seq = (seq + 1) & MCTP_HDR_SEQ_MASK;
730 		pos += size;
731 	}
732 
733 	mctp_route_release(rt);
734 	consume_skb(skb);
735 	return rc;
736 }
737 
738 int mctp_local_output(struct sock *sk, struct mctp_route *rt,
739 		      struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag)
740 {
741 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
742 	struct mctp_skb_cb *cb = mctp_cb(skb);
743 	struct mctp_hdr *hdr;
744 	unsigned long flags;
745 	unsigned int mtu;
746 	mctp_eid_t saddr;
747 	int rc;
748 	u8 tag;
749 
750 	if (WARN_ON(!rt->dev))
751 		return -EINVAL;
752 
753 	spin_lock_irqsave(&rt->dev->addrs_lock, flags);
754 	if (rt->dev->num_addrs == 0) {
755 		rc = -EHOSTUNREACH;
756 	} else {
757 		/* use the outbound interface's first address as our source */
758 		saddr = rt->dev->addrs[0];
759 		rc = 0;
760 	}
761 	spin_unlock_irqrestore(&rt->dev->addrs_lock, flags);
762 
763 	if (rc)
764 		return rc;
765 
766 	if (req_tag & MCTP_HDR_FLAG_TO) {
767 		rc = mctp_alloc_local_tag(msk, saddr, daddr, &tag);
768 		if (rc)
769 			return rc;
770 		tag |= MCTP_HDR_FLAG_TO;
771 	} else {
772 		tag = req_tag;
773 	}
774 
775 
776 	skb->protocol = htons(ETH_P_MCTP);
777 	skb->priority = 0;
778 	skb_reset_transport_header(skb);
779 	skb_push(skb, sizeof(struct mctp_hdr));
780 	skb_reset_network_header(skb);
781 	skb->dev = rt->dev->dev;
782 
783 	/* cb->net will have been set on initial ingress */
784 	cb->src = saddr;
785 
786 	/* set up common header fields */
787 	hdr = mctp_hdr(skb);
788 	hdr->ver = 1;
789 	hdr->dest = daddr;
790 	hdr->src = saddr;
791 
792 	mtu = mctp_route_mtu(rt);
793 
794 	if (skb->len + sizeof(struct mctp_hdr) <= mtu) {
795 		hdr->flags_seq_tag = MCTP_HDR_FLAG_SOM | MCTP_HDR_FLAG_EOM |
796 			tag;
797 		return mctp_do_route(rt, skb);
798 	} else {
799 		return mctp_do_fragment_route(rt, skb, mtu, tag);
800 	}
801 }
802 
803 /* route management */
804 static int mctp_route_add(struct mctp_dev *mdev, mctp_eid_t daddr_start,
805 			  unsigned int daddr_extent, unsigned int mtu,
806 			  unsigned char type)
807 {
808 	int (*rtfn)(struct mctp_route *rt, struct sk_buff *skb);
809 	struct net *net = dev_net(mdev->dev);
810 	struct mctp_route *rt, *ert;
811 
812 	if (!mctp_address_ok(daddr_start))
813 		return -EINVAL;
814 
815 	if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
816 		return -EINVAL;
817 
818 	switch (type) {
819 	case RTN_LOCAL:
820 		rtfn = mctp_route_input;
821 		break;
822 	case RTN_UNICAST:
823 		rtfn = mctp_route_output;
824 		break;
825 	default:
826 		return -EINVAL;
827 	}
828 
829 	rt = mctp_route_alloc();
830 	if (!rt)
831 		return -ENOMEM;
832 
833 	rt->min = daddr_start;
834 	rt->max = daddr_start + daddr_extent;
835 	rt->mtu = mtu;
836 	rt->dev = mdev;
837 	mctp_dev_hold(rt->dev);
838 	rt->type = type;
839 	rt->output = rtfn;
840 
841 	ASSERT_RTNL();
842 	/* Prevent duplicate identical routes. */
843 	list_for_each_entry(ert, &net->mctp.routes, list) {
844 		if (mctp_rt_compare_exact(rt, ert)) {
845 			mctp_route_release(rt);
846 			return -EEXIST;
847 		}
848 	}
849 
850 	list_add_rcu(&rt->list, &net->mctp.routes);
851 
852 	return 0;
853 }
854 
855 static int mctp_route_remove(struct mctp_dev *mdev, mctp_eid_t daddr_start,
856 			     unsigned int daddr_extent, unsigned char type)
857 {
858 	struct net *net = dev_net(mdev->dev);
859 	struct mctp_route *rt, *tmp;
860 	mctp_eid_t daddr_end;
861 	bool dropped;
862 
863 	if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
864 		return -EINVAL;
865 
866 	daddr_end = daddr_start + daddr_extent;
867 	dropped = false;
868 
869 	ASSERT_RTNL();
870 
871 	list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
872 		if (rt->dev == mdev &&
873 		    rt->min == daddr_start && rt->max == daddr_end &&
874 		    rt->type == type) {
875 			list_del_rcu(&rt->list);
876 			/* TODO: immediate RTM_DELROUTE */
877 			mctp_route_release(rt);
878 			dropped = true;
879 		}
880 	}
881 
882 	return dropped ? 0 : -ENOENT;
883 }
884 
885 int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr)
886 {
887 	return mctp_route_add(mdev, addr, 0, 0, RTN_LOCAL);
888 }
889 
890 int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr)
891 {
892 	return mctp_route_remove(mdev, addr, 0, RTN_LOCAL);
893 }
894 
895 /* removes all entries for a given device */
896 void mctp_route_remove_dev(struct mctp_dev *mdev)
897 {
898 	struct net *net = dev_net(mdev->dev);
899 	struct mctp_route *rt, *tmp;
900 
901 	ASSERT_RTNL();
902 	list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
903 		if (rt->dev == mdev) {
904 			list_del_rcu(&rt->list);
905 			/* TODO: immediate RTM_DELROUTE */
906 			mctp_route_release(rt);
907 		}
908 	}
909 }
910 
911 /* Incoming packet-handling */
912 
913 static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,
914 				struct packet_type *pt,
915 				struct net_device *orig_dev)
916 {
917 	struct net *net = dev_net(dev);
918 	struct mctp_dev *mdev;
919 	struct mctp_skb_cb *cb;
920 	struct mctp_route *rt;
921 	struct mctp_hdr *mh;
922 
923 	rcu_read_lock();
924 	mdev = __mctp_dev_get(dev);
925 	rcu_read_unlock();
926 	if (!mdev) {
927 		/* basic non-data sanity checks */
928 		goto err_drop;
929 	}
930 
931 	if (!pskb_may_pull(skb, sizeof(struct mctp_hdr)))
932 		goto err_drop;
933 
934 	skb_reset_transport_header(skb);
935 	skb_reset_network_header(skb);
936 
937 	/* We have enough for a header; decode and route */
938 	mh = mctp_hdr(skb);
939 	if (mh->ver < MCTP_VER_MIN || mh->ver > MCTP_VER_MAX)
940 		goto err_drop;
941 
942 	cb = __mctp_cb(skb);
943 	cb->net = READ_ONCE(mdev->net);
944 
945 	rt = mctp_route_lookup(net, cb->net, mh->dest);
946 
947 	/* NULL EID, but addressed to our physical address */
948 	if (!rt && mh->dest == MCTP_ADDR_NULL && skb->pkt_type == PACKET_HOST)
949 		rt = mctp_route_lookup_null(net, dev);
950 
951 	if (!rt)
952 		goto err_drop;
953 
954 	mctp_do_route(rt, skb);
955 
956 	return NET_RX_SUCCESS;
957 
958 err_drop:
959 	kfree_skb(skb);
960 	return NET_RX_DROP;
961 }
962 
963 static struct packet_type mctp_packet_type = {
964 	.type = cpu_to_be16(ETH_P_MCTP),
965 	.func = mctp_pkttype_receive,
966 };
967 
968 /* netlink interface */
969 
970 static const struct nla_policy rta_mctp_policy[RTA_MAX + 1] = {
971 	[RTA_DST]		= { .type = NLA_U8 },
972 	[RTA_METRICS]		= { .type = NLA_NESTED },
973 	[RTA_OIF]		= { .type = NLA_U32 },
974 };
975 
976 /* Common part for RTM_NEWROUTE and RTM_DELROUTE parsing.
977  * tb must hold RTA_MAX+1 elements.
978  */
979 static int mctp_route_nlparse(struct sk_buff *skb, struct nlmsghdr *nlh,
980 			      struct netlink_ext_ack *extack,
981 			      struct nlattr **tb, struct rtmsg **rtm,
982 			      struct mctp_dev **mdev, mctp_eid_t *daddr_start)
983 {
984 	struct net *net = sock_net(skb->sk);
985 	struct net_device *dev;
986 	unsigned int ifindex;
987 	int rc;
988 
989 	rc = nlmsg_parse(nlh, sizeof(struct rtmsg), tb, RTA_MAX,
990 			 rta_mctp_policy, extack);
991 	if (rc < 0) {
992 		NL_SET_ERR_MSG(extack, "incorrect format");
993 		return rc;
994 	}
995 
996 	if (!tb[RTA_DST]) {
997 		NL_SET_ERR_MSG(extack, "dst EID missing");
998 		return -EINVAL;
999 	}
1000 	*daddr_start = nla_get_u8(tb[RTA_DST]);
1001 
1002 	if (!tb[RTA_OIF]) {
1003 		NL_SET_ERR_MSG(extack, "ifindex missing");
1004 		return -EINVAL;
1005 	}
1006 	ifindex = nla_get_u32(tb[RTA_OIF]);
1007 
1008 	*rtm = nlmsg_data(nlh);
1009 	if ((*rtm)->rtm_family != AF_MCTP) {
1010 		NL_SET_ERR_MSG(extack, "route family must be AF_MCTP");
1011 		return -EINVAL;
1012 	}
1013 
1014 	dev = __dev_get_by_index(net, ifindex);
1015 	if (!dev) {
1016 		NL_SET_ERR_MSG(extack, "bad ifindex");
1017 		return -ENODEV;
1018 	}
1019 	*mdev = mctp_dev_get_rtnl(dev);
1020 	if (!*mdev)
1021 		return -ENODEV;
1022 
1023 	if (dev->flags & IFF_LOOPBACK) {
1024 		NL_SET_ERR_MSG(extack, "no routes to loopback");
1025 		return -EINVAL;
1026 	}
1027 
1028 	return 0;
1029 }
1030 
1031 static const struct nla_policy rta_metrics_policy[RTAX_MAX + 1] = {
1032 	[RTAX_MTU]		= { .type = NLA_U32 },
1033 };
1034 
1035 static int mctp_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1036 			 struct netlink_ext_ack *extack)
1037 {
1038 	struct nlattr *tb[RTA_MAX + 1];
1039 	struct nlattr *tbx[RTAX_MAX + 1];
1040 	mctp_eid_t daddr_start;
1041 	struct mctp_dev *mdev;
1042 	struct rtmsg *rtm;
1043 	unsigned int mtu;
1044 	int rc;
1045 
1046 	rc = mctp_route_nlparse(skb, nlh, extack, tb,
1047 				&rtm, &mdev, &daddr_start);
1048 	if (rc < 0)
1049 		return rc;
1050 
1051 	if (rtm->rtm_type != RTN_UNICAST) {
1052 		NL_SET_ERR_MSG(extack, "rtm_type must be RTN_UNICAST");
1053 		return -EINVAL;
1054 	}
1055 
1056 	mtu = 0;
1057 	if (tb[RTA_METRICS]) {
1058 		rc = nla_parse_nested(tbx, RTAX_MAX, tb[RTA_METRICS],
1059 				      rta_metrics_policy, NULL);
1060 		if (rc < 0)
1061 			return rc;
1062 		if (tbx[RTAX_MTU])
1063 			mtu = nla_get_u32(tbx[RTAX_MTU]);
1064 	}
1065 
1066 	if (rtm->rtm_type != RTN_UNICAST)
1067 		return -EINVAL;
1068 
1069 	rc = mctp_route_add(mdev, daddr_start, rtm->rtm_dst_len, mtu,
1070 			    rtm->rtm_type);
1071 	return rc;
1072 }
1073 
1074 static int mctp_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1075 			 struct netlink_ext_ack *extack)
1076 {
1077 	struct nlattr *tb[RTA_MAX + 1];
1078 	mctp_eid_t daddr_start;
1079 	struct mctp_dev *mdev;
1080 	struct rtmsg *rtm;
1081 	int rc;
1082 
1083 	rc = mctp_route_nlparse(skb, nlh, extack, tb,
1084 				&rtm, &mdev, &daddr_start);
1085 	if (rc < 0)
1086 		return rc;
1087 
1088 	/* we only have unicast routes */
1089 	if (rtm->rtm_type != RTN_UNICAST)
1090 		return -EINVAL;
1091 
1092 	rc = mctp_route_remove(mdev, daddr_start, rtm->rtm_dst_len, RTN_UNICAST);
1093 	return rc;
1094 }
1095 
1096 static int mctp_fill_rtinfo(struct sk_buff *skb, struct mctp_route *rt,
1097 			    u32 portid, u32 seq, int event, unsigned int flags)
1098 {
1099 	struct nlmsghdr *nlh;
1100 	struct rtmsg *hdr;
1101 	void *metrics;
1102 
1103 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*hdr), flags);
1104 	if (!nlh)
1105 		return -EMSGSIZE;
1106 
1107 	hdr = nlmsg_data(nlh);
1108 	hdr->rtm_family = AF_MCTP;
1109 
1110 	/* we use the _len fields as a number of EIDs, rather than
1111 	 * a number of bits in the address
1112 	 */
1113 	hdr->rtm_dst_len = rt->max - rt->min;
1114 	hdr->rtm_src_len = 0;
1115 	hdr->rtm_tos = 0;
1116 	hdr->rtm_table = RT_TABLE_DEFAULT;
1117 	hdr->rtm_protocol = RTPROT_STATIC; /* everything is user-defined */
1118 	hdr->rtm_scope = RT_SCOPE_LINK; /* TODO: scope in mctp_route? */
1119 	hdr->rtm_type = rt->type;
1120 
1121 	if (nla_put_u8(skb, RTA_DST, rt->min))
1122 		goto cancel;
1123 
1124 	metrics = nla_nest_start_noflag(skb, RTA_METRICS);
1125 	if (!metrics)
1126 		goto cancel;
1127 
1128 	if (rt->mtu) {
1129 		if (nla_put_u32(skb, RTAX_MTU, rt->mtu))
1130 			goto cancel;
1131 	}
1132 
1133 	nla_nest_end(skb, metrics);
1134 
1135 	if (rt->dev) {
1136 		if (nla_put_u32(skb, RTA_OIF, rt->dev->dev->ifindex))
1137 			goto cancel;
1138 	}
1139 
1140 	/* TODO: conditional neighbour physaddr? */
1141 
1142 	nlmsg_end(skb, nlh);
1143 
1144 	return 0;
1145 
1146 cancel:
1147 	nlmsg_cancel(skb, nlh);
1148 	return -EMSGSIZE;
1149 }
1150 
1151 static int mctp_dump_rtinfo(struct sk_buff *skb, struct netlink_callback *cb)
1152 {
1153 	struct net *net = sock_net(skb->sk);
1154 	struct mctp_route *rt;
1155 	int s_idx, idx;
1156 
1157 	/* TODO: allow filtering on route data, possibly under
1158 	 * cb->strict_check
1159 	 */
1160 
1161 	/* TODO: change to struct overlay */
1162 	s_idx = cb->args[0];
1163 	idx = 0;
1164 
1165 	rcu_read_lock();
1166 	list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
1167 		if (idx++ < s_idx)
1168 			continue;
1169 		if (mctp_fill_rtinfo(skb, rt,
1170 				     NETLINK_CB(cb->skb).portid,
1171 				     cb->nlh->nlmsg_seq,
1172 				     RTM_NEWROUTE, NLM_F_MULTI) < 0)
1173 			break;
1174 	}
1175 
1176 	rcu_read_unlock();
1177 	cb->args[0] = idx;
1178 
1179 	return skb->len;
1180 }
1181 
1182 /* net namespace implementation */
1183 static int __net_init mctp_routes_net_init(struct net *net)
1184 {
1185 	struct netns_mctp *ns = &net->mctp;
1186 
1187 	INIT_LIST_HEAD(&ns->routes);
1188 	INIT_HLIST_HEAD(&ns->binds);
1189 	mutex_init(&ns->bind_lock);
1190 	INIT_HLIST_HEAD(&ns->keys);
1191 	spin_lock_init(&ns->keys_lock);
1192 	WARN_ON(mctp_default_net_set(net, MCTP_INITIAL_DEFAULT_NET));
1193 	return 0;
1194 }
1195 
1196 static void __net_exit mctp_routes_net_exit(struct net *net)
1197 {
1198 	struct mctp_route *rt;
1199 
1200 	rcu_read_lock();
1201 	list_for_each_entry_rcu(rt, &net->mctp.routes, list)
1202 		mctp_route_release(rt);
1203 	rcu_read_unlock();
1204 }
1205 
1206 static struct pernet_operations mctp_net_ops = {
1207 	.init = mctp_routes_net_init,
1208 	.exit = mctp_routes_net_exit,
1209 };
1210 
1211 int __init mctp_routes_init(void)
1212 {
1213 	dev_add_pack(&mctp_packet_type);
1214 
1215 	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETROUTE,
1216 			     NULL, mctp_dump_rtinfo, 0);
1217 	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWROUTE,
1218 			     mctp_newroute, NULL, 0);
1219 	rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELROUTE,
1220 			     mctp_delroute, NULL, 0);
1221 
1222 	return register_pernet_subsys(&mctp_net_ops);
1223 }
1224 
1225 void __exit mctp_routes_exit(void)
1226 {
1227 	unregister_pernet_subsys(&mctp_net_ops);
1228 	rtnl_unregister(PF_MCTP, RTM_DELROUTE);
1229 	rtnl_unregister(PF_MCTP, RTM_NEWROUTE);
1230 	rtnl_unregister(PF_MCTP, RTM_GETROUTE);
1231 	dev_remove_pack(&mctp_packet_type);
1232 }
1233 
1234 #if IS_ENABLED(CONFIG_MCTP_TEST)
1235 #include "test/route-test.c"
1236 #endif
1237