xref: /linux/net/mctp/af_mctp.c (revision 170aafe35cb98e0f3fbacb446ea86389fbce22ea)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Management Component Transport Protocol (MCTP)
4  *
5  * Copyright (c) 2021 Code Construct
6  * Copyright (c) 2021 Google
7  */
8 
9 #include <linux/compat.h>
10 #include <linux/if_arp.h>
11 #include <linux/net.h>
12 #include <linux/mctp.h>
13 #include <linux/module.h>
14 #include <linux/socket.h>
15 
16 #include <net/mctp.h>
17 #include <net/mctpdevice.h>
18 #include <net/sock.h>
19 
20 #define CREATE_TRACE_POINTS
21 #include <trace/events/mctp.h>
22 
23 /* socket implementation */
24 
25 static void mctp_sk_expire_keys(struct timer_list *timer);
26 
27 static int mctp_release(struct socket *sock)
28 {
29 	struct sock *sk = sock->sk;
30 
31 	if (sk) {
32 		sock->sk = NULL;
33 		sk->sk_prot->close(sk, 0);
34 	}
35 
36 	return 0;
37 }
38 
39 /* Generic sockaddr checks, padding checks only so far */
40 static bool mctp_sockaddr_is_ok(const struct sockaddr_mctp *addr)
41 {
42 	return !addr->__smctp_pad0 && !addr->__smctp_pad1;
43 }
44 
45 static bool mctp_sockaddr_ext_is_ok(const struct sockaddr_mctp_ext *addr)
46 {
47 	return !addr->__smctp_pad0[0] &&
48 	       !addr->__smctp_pad0[1] &&
49 	       !addr->__smctp_pad0[2];
50 }
51 
52 static int mctp_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
53 {
54 	struct sock *sk = sock->sk;
55 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
56 	struct sockaddr_mctp *smctp;
57 	int rc;
58 
59 	if (addrlen < sizeof(*smctp))
60 		return -EINVAL;
61 
62 	if (addr->sa_family != AF_MCTP)
63 		return -EAFNOSUPPORT;
64 
65 	if (!capable(CAP_NET_BIND_SERVICE))
66 		return -EACCES;
67 
68 	/* it's a valid sockaddr for MCTP, cast and do protocol checks */
69 	smctp = (struct sockaddr_mctp *)addr;
70 
71 	if (!mctp_sockaddr_is_ok(smctp))
72 		return -EINVAL;
73 
74 	lock_sock(sk);
75 
76 	/* TODO: allow rebind */
77 	if (sk_hashed(sk)) {
78 		rc = -EADDRINUSE;
79 		goto out_release;
80 	}
81 	msk->bind_net = smctp->smctp_network;
82 	msk->bind_addr = smctp->smctp_addr.s_addr;
83 	msk->bind_type = smctp->smctp_type & 0x7f; /* ignore the IC bit */
84 
85 	rc = sk->sk_prot->hash(sk);
86 
87 out_release:
88 	release_sock(sk);
89 
90 	return rc;
91 }
92 
93 static int mctp_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
94 {
95 	DECLARE_SOCKADDR(struct sockaddr_mctp *, addr, msg->msg_name);
96 	int rc, addrlen = msg->msg_namelen;
97 	struct sock *sk = sock->sk;
98 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
99 	struct mctp_skb_cb *cb;
100 	struct mctp_route *rt;
101 	struct sk_buff *skb = NULL;
102 	int hlen;
103 
104 	if (addr) {
105 		const u8 tagbits = MCTP_TAG_MASK | MCTP_TAG_OWNER |
106 			MCTP_TAG_PREALLOC;
107 
108 		if (addrlen < sizeof(struct sockaddr_mctp))
109 			return -EINVAL;
110 		if (addr->smctp_family != AF_MCTP)
111 			return -EINVAL;
112 		if (!mctp_sockaddr_is_ok(addr))
113 			return -EINVAL;
114 		if (addr->smctp_tag & ~tagbits)
115 			return -EINVAL;
116 		/* can't preallocate a non-owned tag */
117 		if (addr->smctp_tag & MCTP_TAG_PREALLOC &&
118 		    !(addr->smctp_tag & MCTP_TAG_OWNER))
119 			return -EINVAL;
120 
121 	} else {
122 		/* TODO: connect()ed sockets */
123 		return -EDESTADDRREQ;
124 	}
125 
126 	if (!capable(CAP_NET_RAW))
127 		return -EACCES;
128 
129 	if (addr->smctp_network == MCTP_NET_ANY)
130 		addr->smctp_network = mctp_default_net(sock_net(sk));
131 
132 	/* direct addressing */
133 	if (msk->addr_ext && addrlen >= sizeof(struct sockaddr_mctp_ext)) {
134 		DECLARE_SOCKADDR(struct sockaddr_mctp_ext *,
135 				 extaddr, msg->msg_name);
136 		struct net_device *dev;
137 
138 		rc = -EINVAL;
139 		rcu_read_lock();
140 		dev = dev_get_by_index_rcu(sock_net(sk), extaddr->smctp_ifindex);
141 		/* check for correct halen */
142 		if (dev && extaddr->smctp_halen == dev->addr_len) {
143 			hlen = LL_RESERVED_SPACE(dev) + sizeof(struct mctp_hdr);
144 			rc = 0;
145 		}
146 		rcu_read_unlock();
147 		if (rc)
148 			goto err_free;
149 		rt = NULL;
150 	} else {
151 		rt = mctp_route_lookup(sock_net(sk), addr->smctp_network,
152 				       addr->smctp_addr.s_addr);
153 		if (!rt) {
154 			rc = -EHOSTUNREACH;
155 			goto err_free;
156 		}
157 		hlen = LL_RESERVED_SPACE(rt->dev->dev) + sizeof(struct mctp_hdr);
158 	}
159 
160 	skb = sock_alloc_send_skb(sk, hlen + 1 + len,
161 				  msg->msg_flags & MSG_DONTWAIT, &rc);
162 	if (!skb)
163 		return rc;
164 
165 	skb_reserve(skb, hlen);
166 
167 	/* set type as fist byte in payload */
168 	*(u8 *)skb_put(skb, 1) = addr->smctp_type;
169 
170 	rc = memcpy_from_msg((void *)skb_put(skb, len), msg, len);
171 	if (rc < 0)
172 		goto err_free;
173 
174 	/* set up cb */
175 	cb = __mctp_cb(skb);
176 	cb->net = addr->smctp_network;
177 
178 	if (!rt) {
179 		/* fill extended address in cb */
180 		DECLARE_SOCKADDR(struct sockaddr_mctp_ext *,
181 				 extaddr, msg->msg_name);
182 
183 		if (!mctp_sockaddr_ext_is_ok(extaddr) ||
184 		    extaddr->smctp_halen > sizeof(cb->haddr)) {
185 			rc = -EINVAL;
186 			goto err_free;
187 		}
188 
189 		cb->ifindex = extaddr->smctp_ifindex;
190 		/* smctp_halen is checked above */
191 		cb->halen = extaddr->smctp_halen;
192 		memcpy(cb->haddr, extaddr->smctp_haddr, cb->halen);
193 	}
194 
195 	rc = mctp_local_output(sk, rt, skb, addr->smctp_addr.s_addr,
196 			       addr->smctp_tag);
197 
198 	return rc ? : len;
199 
200 err_free:
201 	kfree_skb(skb);
202 	return rc;
203 }
204 
205 static int mctp_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
206 			int flags)
207 {
208 	DECLARE_SOCKADDR(struct sockaddr_mctp *, addr, msg->msg_name);
209 	struct sock *sk = sock->sk;
210 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
211 	struct sk_buff *skb;
212 	size_t msglen;
213 	u8 type;
214 	int rc;
215 
216 	if (flags & ~(MSG_DONTWAIT | MSG_TRUNC | MSG_PEEK))
217 		return -EOPNOTSUPP;
218 
219 	skb = skb_recv_datagram(sk, flags, &rc);
220 	if (!skb)
221 		return rc;
222 
223 	if (!skb->len) {
224 		rc = 0;
225 		goto out_free;
226 	}
227 
228 	/* extract message type, remove from data */
229 	type = *((u8 *)skb->data);
230 	msglen = skb->len - 1;
231 
232 	if (len < msglen)
233 		msg->msg_flags |= MSG_TRUNC;
234 	else
235 		len = msglen;
236 
237 	rc = skb_copy_datagram_msg(skb, 1, msg, len);
238 	if (rc < 0)
239 		goto out_free;
240 
241 	sock_recv_cmsgs(msg, sk, skb);
242 
243 	if (addr) {
244 		struct mctp_skb_cb *cb = mctp_cb(skb);
245 		/* TODO: expand mctp_skb_cb for header fields? */
246 		struct mctp_hdr *hdr = mctp_hdr(skb);
247 
248 		addr = msg->msg_name;
249 		addr->smctp_family = AF_MCTP;
250 		addr->__smctp_pad0 = 0;
251 		addr->smctp_network = cb->net;
252 		addr->smctp_addr.s_addr = hdr->src;
253 		addr->smctp_type = type;
254 		addr->smctp_tag = hdr->flags_seq_tag &
255 					(MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
256 		addr->__smctp_pad1 = 0;
257 		msg->msg_namelen = sizeof(*addr);
258 
259 		if (msk->addr_ext) {
260 			DECLARE_SOCKADDR(struct sockaddr_mctp_ext *, ae,
261 					 msg->msg_name);
262 			msg->msg_namelen = sizeof(*ae);
263 			ae->smctp_ifindex = cb->ifindex;
264 			ae->smctp_halen = cb->halen;
265 			memset(ae->__smctp_pad0, 0x0, sizeof(ae->__smctp_pad0));
266 			memset(ae->smctp_haddr, 0x0, sizeof(ae->smctp_haddr));
267 			memcpy(ae->smctp_haddr, cb->haddr, cb->halen);
268 		}
269 	}
270 
271 	rc = len;
272 
273 	if (flags & MSG_TRUNC)
274 		rc = msglen;
275 
276 out_free:
277 	skb_free_datagram(sk, skb);
278 	return rc;
279 }
280 
281 /* We're done with the key; invalidate, stop reassembly, and remove from lists.
282  */
283 static void __mctp_key_remove(struct mctp_sk_key *key, struct net *net,
284 			      unsigned long flags, unsigned long reason)
285 __releases(&key->lock)
286 __must_hold(&net->mctp.keys_lock)
287 {
288 	struct sk_buff *skb;
289 
290 	trace_mctp_key_release(key, reason);
291 	skb = key->reasm_head;
292 	key->reasm_head = NULL;
293 	key->reasm_dead = true;
294 	key->valid = false;
295 	mctp_dev_release_key(key->dev, key);
296 	spin_unlock_irqrestore(&key->lock, flags);
297 
298 	if (!hlist_unhashed(&key->hlist)) {
299 		hlist_del_init(&key->hlist);
300 		hlist_del_init(&key->sklist);
301 		/* unref for the lists */
302 		mctp_key_unref(key);
303 	}
304 
305 	kfree_skb(skb);
306 }
307 
308 static int mctp_setsockopt(struct socket *sock, int level, int optname,
309 			   sockptr_t optval, unsigned int optlen)
310 {
311 	struct mctp_sock *msk = container_of(sock->sk, struct mctp_sock, sk);
312 	int val;
313 
314 	if (level != SOL_MCTP)
315 		return -EINVAL;
316 
317 	if (optname == MCTP_OPT_ADDR_EXT) {
318 		if (optlen != sizeof(int))
319 			return -EINVAL;
320 		if (copy_from_sockptr(&val, optval, sizeof(int)))
321 			return -EFAULT;
322 		msk->addr_ext = val;
323 		return 0;
324 	}
325 
326 	return -ENOPROTOOPT;
327 }
328 
329 static int mctp_getsockopt(struct socket *sock, int level, int optname,
330 			   char __user *optval, int __user *optlen)
331 {
332 	struct mctp_sock *msk = container_of(sock->sk, struct mctp_sock, sk);
333 	int len, val;
334 
335 	if (level != SOL_MCTP)
336 		return -EINVAL;
337 
338 	if (get_user(len, optlen))
339 		return -EFAULT;
340 
341 	if (optname == MCTP_OPT_ADDR_EXT) {
342 		if (len != sizeof(int))
343 			return -EINVAL;
344 		val = !!msk->addr_ext;
345 		if (copy_to_user(optval, &val, len))
346 			return -EFAULT;
347 		return 0;
348 	}
349 
350 	return -EINVAL;
351 }
352 
353 /* helpers for reading/writing the tag ioc, handling compatibility across the
354  * two versions, and some basic API error checking
355  */
356 static int mctp_ioctl_tag_copy_from_user(unsigned long arg,
357 					 struct mctp_ioc_tag_ctl2 *ctl,
358 					 bool tagv2)
359 {
360 	struct mctp_ioc_tag_ctl ctl_compat;
361 	unsigned long size;
362 	void *ptr;
363 	int rc;
364 
365 	if (tagv2) {
366 		size = sizeof(*ctl);
367 		ptr = ctl;
368 	} else {
369 		size = sizeof(ctl_compat);
370 		ptr = &ctl_compat;
371 	}
372 
373 	rc = copy_from_user(ptr, (void __user *)arg, size);
374 	if (rc)
375 		return -EFAULT;
376 
377 	if (!tagv2) {
378 		/* compat, using defaults for new fields */
379 		ctl->net = MCTP_INITIAL_DEFAULT_NET;
380 		ctl->peer_addr = ctl_compat.peer_addr;
381 		ctl->local_addr = MCTP_ADDR_ANY;
382 		ctl->flags = ctl_compat.flags;
383 		ctl->tag = ctl_compat.tag;
384 	}
385 
386 	if (ctl->flags)
387 		return -EINVAL;
388 
389 	if (ctl->local_addr != MCTP_ADDR_ANY &&
390 	    ctl->local_addr != MCTP_ADDR_NULL)
391 		return -EINVAL;
392 
393 	return 0;
394 }
395 
396 static int mctp_ioctl_tag_copy_to_user(unsigned long arg,
397 				       struct mctp_ioc_tag_ctl2 *ctl,
398 				       bool tagv2)
399 {
400 	struct mctp_ioc_tag_ctl ctl_compat;
401 	unsigned long size;
402 	void *ptr;
403 	int rc;
404 
405 	if (tagv2) {
406 		ptr = ctl;
407 		size = sizeof(*ctl);
408 	} else {
409 		ctl_compat.peer_addr = ctl->peer_addr;
410 		ctl_compat.tag = ctl->tag;
411 		ctl_compat.flags = ctl->flags;
412 
413 		ptr = &ctl_compat;
414 		size = sizeof(ctl_compat);
415 	}
416 
417 	rc = copy_to_user((void __user *)arg, ptr, size);
418 	if (rc)
419 		return -EFAULT;
420 
421 	return 0;
422 }
423 
424 static int mctp_ioctl_alloctag(struct mctp_sock *msk, bool tagv2,
425 			       unsigned long arg)
426 {
427 	struct net *net = sock_net(&msk->sk);
428 	struct mctp_sk_key *key = NULL;
429 	struct mctp_ioc_tag_ctl2 ctl;
430 	unsigned long flags;
431 	u8 tag;
432 	int rc;
433 
434 	rc = mctp_ioctl_tag_copy_from_user(arg, &ctl, tagv2);
435 	if (rc)
436 		return rc;
437 
438 	if (ctl.tag)
439 		return -EINVAL;
440 
441 	key = mctp_alloc_local_tag(msk, ctl.net, MCTP_ADDR_ANY,
442 				   ctl.peer_addr, true, &tag);
443 	if (IS_ERR(key))
444 		return PTR_ERR(key);
445 
446 	ctl.tag = tag | MCTP_TAG_OWNER | MCTP_TAG_PREALLOC;
447 	rc = mctp_ioctl_tag_copy_to_user(arg, &ctl, tagv2);
448 	if (rc) {
449 		unsigned long fl2;
450 		/* Unwind our key allocation: the keys list lock needs to be
451 		 * taken before the individual key locks, and we need a valid
452 		 * flags value (fl2) to pass to __mctp_key_remove, hence the
453 		 * second spin_lock_irqsave() rather than a plain spin_lock().
454 		 */
455 		spin_lock_irqsave(&net->mctp.keys_lock, flags);
456 		spin_lock_irqsave(&key->lock, fl2);
457 		__mctp_key_remove(key, net, fl2, MCTP_TRACE_KEY_DROPPED);
458 		mctp_key_unref(key);
459 		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
460 		return rc;
461 	}
462 
463 	mctp_key_unref(key);
464 	return 0;
465 }
466 
467 static int mctp_ioctl_droptag(struct mctp_sock *msk, bool tagv2,
468 			      unsigned long arg)
469 {
470 	struct net *net = sock_net(&msk->sk);
471 	struct mctp_ioc_tag_ctl2 ctl;
472 	unsigned long flags, fl2;
473 	struct mctp_sk_key *key;
474 	struct hlist_node *tmp;
475 	int rc;
476 	u8 tag;
477 
478 	rc = mctp_ioctl_tag_copy_from_user(arg, &ctl, tagv2);
479 	if (rc)
480 		return rc;
481 
482 	/* Must be a local tag, TO set, preallocated */
483 	if ((ctl.tag & ~MCTP_TAG_MASK) != (MCTP_TAG_OWNER | MCTP_TAG_PREALLOC))
484 		return -EINVAL;
485 
486 	tag = ctl.tag & MCTP_TAG_MASK;
487 	rc = -EINVAL;
488 
489 	if (ctl.peer_addr == MCTP_ADDR_NULL)
490 		ctl.peer_addr = MCTP_ADDR_ANY;
491 
492 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
493 	hlist_for_each_entry_safe(key, tmp, &msk->keys, sklist) {
494 		/* we do an irqsave here, even though we know the irq state,
495 		 * so we have the flags to pass to __mctp_key_remove
496 		 */
497 		spin_lock_irqsave(&key->lock, fl2);
498 		if (key->manual_alloc &&
499 		    ctl.net == key->net &&
500 		    ctl.peer_addr == key->peer_addr &&
501 		    tag == key->tag) {
502 			__mctp_key_remove(key, net, fl2,
503 					  MCTP_TRACE_KEY_DROPPED);
504 			rc = 0;
505 		} else {
506 			spin_unlock_irqrestore(&key->lock, fl2);
507 		}
508 	}
509 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
510 
511 	return rc;
512 }
513 
514 static int mctp_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
515 {
516 	struct mctp_sock *msk = container_of(sock->sk, struct mctp_sock, sk);
517 	bool tagv2 = false;
518 
519 	switch (cmd) {
520 	case SIOCMCTPALLOCTAG2:
521 	case SIOCMCTPALLOCTAG:
522 		tagv2 = cmd == SIOCMCTPALLOCTAG2;
523 		return mctp_ioctl_alloctag(msk, tagv2, arg);
524 	case SIOCMCTPDROPTAG:
525 	case SIOCMCTPDROPTAG2:
526 		tagv2 = cmd == SIOCMCTPDROPTAG2;
527 		return mctp_ioctl_droptag(msk, tagv2, arg);
528 	}
529 
530 	return -EINVAL;
531 }
532 
533 #ifdef CONFIG_COMPAT
534 static int mctp_compat_ioctl(struct socket *sock, unsigned int cmd,
535 			     unsigned long arg)
536 {
537 	void __user *argp = compat_ptr(arg);
538 
539 	switch (cmd) {
540 	/* These have compatible ptr layouts */
541 	case SIOCMCTPALLOCTAG:
542 	case SIOCMCTPDROPTAG:
543 		return mctp_ioctl(sock, cmd, (unsigned long)argp);
544 	}
545 
546 	return -ENOIOCTLCMD;
547 }
548 #endif
549 
550 static const struct proto_ops mctp_dgram_ops = {
551 	.family		= PF_MCTP,
552 	.release	= mctp_release,
553 	.bind		= mctp_bind,
554 	.connect	= sock_no_connect,
555 	.socketpair	= sock_no_socketpair,
556 	.accept		= sock_no_accept,
557 	.getname	= sock_no_getname,
558 	.poll		= datagram_poll,
559 	.ioctl		= mctp_ioctl,
560 	.gettstamp	= sock_gettstamp,
561 	.listen		= sock_no_listen,
562 	.shutdown	= sock_no_shutdown,
563 	.setsockopt	= mctp_setsockopt,
564 	.getsockopt	= mctp_getsockopt,
565 	.sendmsg	= mctp_sendmsg,
566 	.recvmsg	= mctp_recvmsg,
567 	.mmap		= sock_no_mmap,
568 #ifdef CONFIG_COMPAT
569 	.compat_ioctl	= mctp_compat_ioctl,
570 #endif
571 };
572 
573 static void mctp_sk_expire_keys(struct timer_list *timer)
574 {
575 	struct mctp_sock *msk = container_of(timer, struct mctp_sock,
576 					     key_expiry);
577 	struct net *net = sock_net(&msk->sk);
578 	unsigned long next_expiry, flags, fl2;
579 	struct mctp_sk_key *key;
580 	struct hlist_node *tmp;
581 	bool next_expiry_valid = false;
582 
583 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
584 
585 	hlist_for_each_entry_safe(key, tmp, &msk->keys, sklist) {
586 		/* don't expire. manual_alloc is immutable, no locking
587 		 * required.
588 		 */
589 		if (key->manual_alloc)
590 			continue;
591 
592 		spin_lock_irqsave(&key->lock, fl2);
593 		if (!time_after_eq(key->expiry, jiffies)) {
594 			__mctp_key_remove(key, net, fl2,
595 					  MCTP_TRACE_KEY_TIMEOUT);
596 			continue;
597 		}
598 
599 		if (next_expiry_valid) {
600 			if (time_before(key->expiry, next_expiry))
601 				next_expiry = key->expiry;
602 		} else {
603 			next_expiry = key->expiry;
604 			next_expiry_valid = true;
605 		}
606 		spin_unlock_irqrestore(&key->lock, fl2);
607 	}
608 
609 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
610 
611 	if (next_expiry_valid)
612 		mod_timer(timer, next_expiry);
613 }
614 
615 static int mctp_sk_init(struct sock *sk)
616 {
617 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
618 
619 	INIT_HLIST_HEAD(&msk->keys);
620 	timer_setup(&msk->key_expiry, mctp_sk_expire_keys, 0);
621 	return 0;
622 }
623 
624 static void mctp_sk_close(struct sock *sk, long timeout)
625 {
626 	sk_common_release(sk);
627 }
628 
629 static int mctp_sk_hash(struct sock *sk)
630 {
631 	struct net *net = sock_net(sk);
632 
633 	mutex_lock(&net->mctp.bind_lock);
634 	sk_add_node_rcu(sk, &net->mctp.binds);
635 	mutex_unlock(&net->mctp.bind_lock);
636 
637 	return 0;
638 }
639 
640 static void mctp_sk_unhash(struct sock *sk)
641 {
642 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
643 	struct net *net = sock_net(sk);
644 	unsigned long flags, fl2;
645 	struct mctp_sk_key *key;
646 	struct hlist_node *tmp;
647 
648 	/* remove from any type-based binds */
649 	mutex_lock(&net->mctp.bind_lock);
650 	sk_del_node_init_rcu(sk);
651 	mutex_unlock(&net->mctp.bind_lock);
652 
653 	/* remove tag allocations */
654 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
655 	hlist_for_each_entry_safe(key, tmp, &msk->keys, sklist) {
656 		spin_lock_irqsave(&key->lock, fl2);
657 		__mctp_key_remove(key, net, fl2, MCTP_TRACE_KEY_CLOSED);
658 	}
659 	sock_set_flag(sk, SOCK_DEAD);
660 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
661 
662 	/* Since there are no more tag allocations (we have removed all of the
663 	 * keys), stop any pending expiry events. the timer cannot be re-queued
664 	 * as the sk is no longer observable
665 	 */
666 	del_timer_sync(&msk->key_expiry);
667 }
668 
669 static void mctp_sk_destruct(struct sock *sk)
670 {
671 	skb_queue_purge(&sk->sk_receive_queue);
672 }
673 
674 static struct proto mctp_proto = {
675 	.name		= "MCTP",
676 	.owner		= THIS_MODULE,
677 	.obj_size	= sizeof(struct mctp_sock),
678 	.init		= mctp_sk_init,
679 	.close		= mctp_sk_close,
680 	.hash		= mctp_sk_hash,
681 	.unhash		= mctp_sk_unhash,
682 };
683 
684 static int mctp_pf_create(struct net *net, struct socket *sock,
685 			  int protocol, int kern)
686 {
687 	const struct proto_ops *ops;
688 	struct proto *proto;
689 	struct sock *sk;
690 	int rc;
691 
692 	if (protocol)
693 		return -EPROTONOSUPPORT;
694 
695 	/* only datagram sockets are supported */
696 	if (sock->type != SOCK_DGRAM)
697 		return -ESOCKTNOSUPPORT;
698 
699 	proto = &mctp_proto;
700 	ops = &mctp_dgram_ops;
701 
702 	sock->state = SS_UNCONNECTED;
703 	sock->ops = ops;
704 
705 	sk = sk_alloc(net, PF_MCTP, GFP_KERNEL, proto, kern);
706 	if (!sk)
707 		return -ENOMEM;
708 
709 	sock_init_data(sock, sk);
710 	sk->sk_destruct = mctp_sk_destruct;
711 
712 	rc = 0;
713 	if (sk->sk_prot->init)
714 		rc = sk->sk_prot->init(sk);
715 
716 	if (rc)
717 		goto err_sk_put;
718 
719 	return 0;
720 
721 err_sk_put:
722 	sock_orphan(sk);
723 	sock_put(sk);
724 	return rc;
725 }
726 
727 static struct net_proto_family mctp_pf = {
728 	.family = PF_MCTP,
729 	.create = mctp_pf_create,
730 	.owner = THIS_MODULE,
731 };
732 
733 static __init int mctp_init(void)
734 {
735 	int rc;
736 
737 	/* ensure our uapi tag definitions match the header format */
738 	BUILD_BUG_ON(MCTP_TAG_OWNER != MCTP_HDR_FLAG_TO);
739 	BUILD_BUG_ON(MCTP_TAG_MASK != MCTP_HDR_TAG_MASK);
740 
741 	pr_info("mctp: management component transport protocol core\n");
742 
743 	rc = sock_register(&mctp_pf);
744 	if (rc)
745 		return rc;
746 
747 	rc = proto_register(&mctp_proto, 0);
748 	if (rc)
749 		goto err_unreg_sock;
750 
751 	rc = mctp_routes_init();
752 	if (rc)
753 		goto err_unreg_proto;
754 
755 	rc = mctp_neigh_init();
756 	if (rc)
757 		goto err_unreg_routes;
758 
759 	mctp_device_init();
760 
761 	return 0;
762 
763 err_unreg_routes:
764 	mctp_routes_exit();
765 err_unreg_proto:
766 	proto_unregister(&mctp_proto);
767 err_unreg_sock:
768 	sock_unregister(PF_MCTP);
769 
770 	return rc;
771 }
772 
773 static __exit void mctp_exit(void)
774 {
775 	mctp_device_exit();
776 	mctp_neigh_exit();
777 	mctp_routes_exit();
778 	proto_unregister(&mctp_proto);
779 	sock_unregister(PF_MCTP);
780 }
781 
782 subsys_initcall(mctp_init);
783 module_exit(mctp_exit);
784 
785 MODULE_DESCRIPTION("MCTP core");
786 MODULE_AUTHOR("Jeremy Kerr <jk@codeconstruct.com.au>");
787 
788 MODULE_ALIAS_NETPROTO(PF_MCTP);
789