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