xref: /linux/net/mctp/af_mctp.c (revision 4ec4b7fc04a7217a6a581ac132bd0fb6abc2f4d5)
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 net *net = sock_net(&msk->sk);
57 	struct sockaddr_mctp *smctp;
58 	int rc;
59 
60 	if (addrlen < sizeof(*smctp))
61 		return -EINVAL;
62 
63 	if (addr->sa_family != AF_MCTP)
64 		return -EAFNOSUPPORT;
65 
66 	if (!capable(CAP_NET_BIND_SERVICE))
67 		return -EACCES;
68 
69 	/* it's a valid sockaddr for MCTP, cast and do protocol checks */
70 	smctp = (struct sockaddr_mctp *)addr;
71 
72 	if (!mctp_sockaddr_is_ok(smctp))
73 		return -EINVAL;
74 
75 	lock_sock(sk);
76 
77 	if (sk_hashed(sk)) {
78 		rc = -EADDRINUSE;
79 		goto out_release;
80 	}
81 
82 	msk->bind_addr = smctp->smctp_addr.s_addr;
83 
84 	/* MCTP_NET_ANY with a specific EID is resolved to the default net
85 	 * at bind() time.
86 	 * For bind_addr=MCTP_ADDR_ANY it is handled specially at route
87 	 * lookup time.
88 	 */
89 	if (smctp->smctp_network == MCTP_NET_ANY &&
90 	    msk->bind_addr != MCTP_ADDR_ANY) {
91 		msk->bind_net = mctp_default_net(net);
92 	} else {
93 		msk->bind_net = smctp->smctp_network;
94 	}
95 
96 	msk->bind_type = smctp->smctp_type & 0x7f; /* ignore the IC bit */
97 
98 	rc = sk->sk_prot->hash(sk);
99 
100 out_release:
101 	release_sock(sk);
102 
103 	return rc;
104 }
105 
106 static int mctp_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
107 {
108 	DECLARE_SOCKADDR(struct sockaddr_mctp *, addr, msg->msg_name);
109 	int rc, addrlen = msg->msg_namelen;
110 	struct sock *sk = sock->sk;
111 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
112 	struct mctp_skb_cb *cb;
113 	struct sk_buff *skb = NULL;
114 	struct mctp_dst dst;
115 	int hlen;
116 
117 	if (addr) {
118 		const u8 tagbits = MCTP_TAG_MASK | MCTP_TAG_OWNER |
119 			MCTP_TAG_PREALLOC;
120 
121 		if (addrlen < sizeof(struct sockaddr_mctp))
122 			return -EINVAL;
123 		if (addr->smctp_family != AF_MCTP)
124 			return -EINVAL;
125 		if (!mctp_sockaddr_is_ok(addr))
126 			return -EINVAL;
127 		if (addr->smctp_tag & ~tagbits)
128 			return -EINVAL;
129 		/* can't preallocate a non-owned tag */
130 		if (addr->smctp_tag & MCTP_TAG_PREALLOC &&
131 		    !(addr->smctp_tag & MCTP_TAG_OWNER))
132 			return -EINVAL;
133 
134 	} else {
135 		/* TODO: connect()ed sockets */
136 		return -EDESTADDRREQ;
137 	}
138 
139 	if (!capable(CAP_NET_RAW))
140 		return -EACCES;
141 
142 	if (addr->smctp_network == MCTP_NET_ANY)
143 		addr->smctp_network = mctp_default_net(sock_net(sk));
144 
145 	/* direct addressing */
146 	if (msk->addr_ext && addrlen >= sizeof(struct sockaddr_mctp_ext)) {
147 		DECLARE_SOCKADDR(struct sockaddr_mctp_ext *,
148 				 extaddr, msg->msg_name);
149 
150 		if (!mctp_sockaddr_ext_is_ok(extaddr))
151 			return -EINVAL;
152 
153 		rc = mctp_dst_from_extaddr(&dst, sock_net(sk),
154 					   extaddr->smctp_ifindex,
155 					   extaddr->smctp_halen,
156 					   extaddr->smctp_haddr);
157 		if (rc)
158 			return rc;
159 
160 	} else {
161 		rc = mctp_route_lookup(sock_net(sk), addr->smctp_network,
162 				       addr->smctp_addr.s_addr, &dst);
163 		if (rc)
164 			return rc;
165 	}
166 
167 	hlen = LL_RESERVED_SPACE(dst.dev->dev) + sizeof(struct mctp_hdr);
168 
169 	skb = sock_alloc_send_skb(sk, hlen + 1 + len,
170 				  msg->msg_flags & MSG_DONTWAIT, &rc);
171 	if (!skb)
172 		goto err_release_dst;
173 
174 	skb_reserve(skb, hlen);
175 
176 	/* set type as fist byte in payload */
177 	*(u8 *)skb_put(skb, 1) = addr->smctp_type;
178 
179 	rc = memcpy_from_msg((void *)skb_put(skb, len), msg, len);
180 	if (rc < 0)
181 		goto err_free;
182 
183 	/* set up cb */
184 	cb = __mctp_cb(skb);
185 	cb->net = addr->smctp_network;
186 
187 	rc = mctp_local_output(sk, &dst, skb, addr->smctp_addr.s_addr,
188 			       addr->smctp_tag);
189 
190 	mctp_dst_release(&dst);
191 	return rc ? : len;
192 
193 err_free:
194 	kfree_skb(skb);
195 err_release_dst:
196 	mctp_dst_release(&dst);
197 	return rc;
198 }
199 
200 static int mctp_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
201 			int flags)
202 {
203 	DECLARE_SOCKADDR(struct sockaddr_mctp *, addr, msg->msg_name);
204 	struct sock *sk = sock->sk;
205 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
206 	struct sk_buff *skb;
207 	size_t msglen;
208 	u8 type;
209 	int rc;
210 
211 	if (flags & ~(MSG_DONTWAIT | MSG_TRUNC | MSG_PEEK))
212 		return -EOPNOTSUPP;
213 
214 	skb = skb_recv_datagram(sk, flags, &rc);
215 	if (!skb)
216 		return rc;
217 
218 	if (!skb->len) {
219 		rc = 0;
220 		goto out_free;
221 	}
222 
223 	/* extract message type, remove from data */
224 	type = *((u8 *)skb->data);
225 	msglen = skb->len - 1;
226 
227 	if (len < msglen)
228 		msg->msg_flags |= MSG_TRUNC;
229 	else
230 		len = msglen;
231 
232 	rc = skb_copy_datagram_msg(skb, 1, msg, len);
233 	if (rc < 0)
234 		goto out_free;
235 
236 	sock_recv_cmsgs(msg, sk, skb);
237 
238 	if (addr) {
239 		struct mctp_skb_cb *cb = mctp_cb(skb);
240 		/* TODO: expand mctp_skb_cb for header fields? */
241 		struct mctp_hdr *hdr = mctp_hdr(skb);
242 
243 		addr = msg->msg_name;
244 		addr->smctp_family = AF_MCTP;
245 		addr->__smctp_pad0 = 0;
246 		addr->smctp_network = cb->net;
247 		addr->smctp_addr.s_addr = hdr->src;
248 		addr->smctp_type = type;
249 		addr->smctp_tag = hdr->flags_seq_tag &
250 					(MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
251 		addr->__smctp_pad1 = 0;
252 		msg->msg_namelen = sizeof(*addr);
253 
254 		if (msk->addr_ext) {
255 			DECLARE_SOCKADDR(struct sockaddr_mctp_ext *, ae,
256 					 msg->msg_name);
257 			msg->msg_namelen = sizeof(*ae);
258 			ae->smctp_ifindex = cb->ifindex;
259 			ae->smctp_halen = cb->halen;
260 			memset(ae->__smctp_pad0, 0x0, sizeof(ae->__smctp_pad0));
261 			memset(ae->smctp_haddr, 0x0, sizeof(ae->smctp_haddr));
262 			memcpy(ae->smctp_haddr, cb->haddr, cb->halen);
263 		}
264 	}
265 
266 	rc = len;
267 
268 	if (flags & MSG_TRUNC)
269 		rc = msglen;
270 
271 out_free:
272 	skb_free_datagram(sk, skb);
273 	return rc;
274 }
275 
276 /* We're done with the key; invalidate, stop reassembly, and remove from lists.
277  */
278 static void __mctp_key_remove(struct mctp_sk_key *key, struct net *net,
279 			      unsigned long flags, unsigned long reason)
280 __releases(&key->lock)
281 __must_hold(&net->mctp.keys_lock)
282 {
283 	struct sk_buff *skb;
284 
285 	trace_mctp_key_release(key, reason);
286 	skb = key->reasm_head;
287 	key->reasm_head = NULL;
288 	key->reasm_dead = true;
289 	key->valid = false;
290 	mctp_dev_release_key(key->dev, key);
291 	spin_unlock_irqrestore(&key->lock, flags);
292 
293 	if (!hlist_unhashed(&key->hlist)) {
294 		hlist_del_init(&key->hlist);
295 		hlist_del_init(&key->sklist);
296 		/* unref for the lists */
297 		mctp_key_unref(key);
298 	}
299 
300 	kfree_skb(skb);
301 }
302 
303 static int mctp_setsockopt(struct socket *sock, int level, int optname,
304 			   sockptr_t optval, unsigned int optlen)
305 {
306 	struct mctp_sock *msk = container_of(sock->sk, struct mctp_sock, sk);
307 	int val;
308 
309 	if (level != SOL_MCTP)
310 		return -EINVAL;
311 
312 	if (optname == MCTP_OPT_ADDR_EXT) {
313 		if (optlen != sizeof(int))
314 			return -EINVAL;
315 		if (copy_from_sockptr(&val, optval, sizeof(int)))
316 			return -EFAULT;
317 		msk->addr_ext = val;
318 		return 0;
319 	}
320 
321 	return -ENOPROTOOPT;
322 }
323 
324 static int mctp_getsockopt(struct socket *sock, int level, int optname,
325 			   char __user *optval, int __user *optlen)
326 {
327 	struct mctp_sock *msk = container_of(sock->sk, struct mctp_sock, sk);
328 	int len, val;
329 
330 	if (level != SOL_MCTP)
331 		return -EINVAL;
332 
333 	if (get_user(len, optlen))
334 		return -EFAULT;
335 
336 	if (optname == MCTP_OPT_ADDR_EXT) {
337 		if (len != sizeof(int))
338 			return -EINVAL;
339 		val = !!msk->addr_ext;
340 		if (copy_to_user(optval, &val, len))
341 			return -EFAULT;
342 		return 0;
343 	}
344 
345 	return -EINVAL;
346 }
347 
348 /* helpers for reading/writing the tag ioc, handling compatibility across the
349  * two versions, and some basic API error checking
350  */
351 static int mctp_ioctl_tag_copy_from_user(unsigned long arg,
352 					 struct mctp_ioc_tag_ctl2 *ctl,
353 					 bool tagv2)
354 {
355 	struct mctp_ioc_tag_ctl ctl_compat;
356 	unsigned long size;
357 	void *ptr;
358 	int rc;
359 
360 	if (tagv2) {
361 		size = sizeof(*ctl);
362 		ptr = ctl;
363 	} else {
364 		size = sizeof(ctl_compat);
365 		ptr = &ctl_compat;
366 	}
367 
368 	rc = copy_from_user(ptr, (void __user *)arg, size);
369 	if (rc)
370 		return -EFAULT;
371 
372 	if (!tagv2) {
373 		/* compat, using defaults for new fields */
374 		ctl->net = MCTP_INITIAL_DEFAULT_NET;
375 		ctl->peer_addr = ctl_compat.peer_addr;
376 		ctl->local_addr = MCTP_ADDR_ANY;
377 		ctl->flags = ctl_compat.flags;
378 		ctl->tag = ctl_compat.tag;
379 	}
380 
381 	if (ctl->flags)
382 		return -EINVAL;
383 
384 	if (ctl->local_addr != MCTP_ADDR_ANY &&
385 	    ctl->local_addr != MCTP_ADDR_NULL)
386 		return -EINVAL;
387 
388 	return 0;
389 }
390 
391 static int mctp_ioctl_tag_copy_to_user(unsigned long arg,
392 				       struct mctp_ioc_tag_ctl2 *ctl,
393 				       bool tagv2)
394 {
395 	struct mctp_ioc_tag_ctl ctl_compat;
396 	unsigned long size;
397 	void *ptr;
398 	int rc;
399 
400 	if (tagv2) {
401 		ptr = ctl;
402 		size = sizeof(*ctl);
403 	} else {
404 		ctl_compat.peer_addr = ctl->peer_addr;
405 		ctl_compat.tag = ctl->tag;
406 		ctl_compat.flags = ctl->flags;
407 
408 		ptr = &ctl_compat;
409 		size = sizeof(ctl_compat);
410 	}
411 
412 	rc = copy_to_user((void __user *)arg, ptr, size);
413 	if (rc)
414 		return -EFAULT;
415 
416 	return 0;
417 }
418 
419 static int mctp_ioctl_alloctag(struct mctp_sock *msk, bool tagv2,
420 			       unsigned long arg)
421 {
422 	struct net *net = sock_net(&msk->sk);
423 	struct mctp_sk_key *key = NULL;
424 	struct mctp_ioc_tag_ctl2 ctl;
425 	unsigned long flags;
426 	u8 tag;
427 	int rc;
428 
429 	rc = mctp_ioctl_tag_copy_from_user(arg, &ctl, tagv2);
430 	if (rc)
431 		return rc;
432 
433 	if (ctl.tag)
434 		return -EINVAL;
435 
436 	key = mctp_alloc_local_tag(msk, ctl.net, MCTP_ADDR_ANY,
437 				   ctl.peer_addr, true, &tag);
438 	if (IS_ERR(key))
439 		return PTR_ERR(key);
440 
441 	ctl.tag = tag | MCTP_TAG_OWNER | MCTP_TAG_PREALLOC;
442 	rc = mctp_ioctl_tag_copy_to_user(arg, &ctl, tagv2);
443 	if (rc) {
444 		unsigned long fl2;
445 		/* Unwind our key allocation: the keys list lock needs to be
446 		 * taken before the individual key locks, and we need a valid
447 		 * flags value (fl2) to pass to __mctp_key_remove, hence the
448 		 * second spin_lock_irqsave() rather than a plain spin_lock().
449 		 */
450 		spin_lock_irqsave(&net->mctp.keys_lock, flags);
451 		spin_lock_irqsave(&key->lock, fl2);
452 		__mctp_key_remove(key, net, fl2, MCTP_TRACE_KEY_DROPPED);
453 		mctp_key_unref(key);
454 		spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
455 		return rc;
456 	}
457 
458 	mctp_key_unref(key);
459 	return 0;
460 }
461 
462 static int mctp_ioctl_droptag(struct mctp_sock *msk, bool tagv2,
463 			      unsigned long arg)
464 {
465 	struct net *net = sock_net(&msk->sk);
466 	struct mctp_ioc_tag_ctl2 ctl;
467 	unsigned long flags, fl2;
468 	struct mctp_sk_key *key;
469 	struct hlist_node *tmp;
470 	int rc;
471 	u8 tag;
472 
473 	rc = mctp_ioctl_tag_copy_from_user(arg, &ctl, tagv2);
474 	if (rc)
475 		return rc;
476 
477 	/* Must be a local tag, TO set, preallocated */
478 	if ((ctl.tag & ~MCTP_TAG_MASK) != (MCTP_TAG_OWNER | MCTP_TAG_PREALLOC))
479 		return -EINVAL;
480 
481 	tag = ctl.tag & MCTP_TAG_MASK;
482 	rc = -EINVAL;
483 
484 	if (ctl.peer_addr == MCTP_ADDR_NULL)
485 		ctl.peer_addr = MCTP_ADDR_ANY;
486 
487 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
488 	hlist_for_each_entry_safe(key, tmp, &msk->keys, sklist) {
489 		/* we do an irqsave here, even though we know the irq state,
490 		 * so we have the flags to pass to __mctp_key_remove
491 		 */
492 		spin_lock_irqsave(&key->lock, fl2);
493 		if (key->manual_alloc &&
494 		    ctl.net == key->net &&
495 		    ctl.peer_addr == key->peer_addr &&
496 		    tag == key->tag) {
497 			__mctp_key_remove(key, net, fl2,
498 					  MCTP_TRACE_KEY_DROPPED);
499 			rc = 0;
500 		} else {
501 			spin_unlock_irqrestore(&key->lock, fl2);
502 		}
503 	}
504 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
505 
506 	return rc;
507 }
508 
509 static int mctp_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
510 {
511 	struct mctp_sock *msk = container_of(sock->sk, struct mctp_sock, sk);
512 	bool tagv2 = false;
513 
514 	switch (cmd) {
515 	case SIOCMCTPALLOCTAG2:
516 	case SIOCMCTPALLOCTAG:
517 		tagv2 = cmd == SIOCMCTPALLOCTAG2;
518 		return mctp_ioctl_alloctag(msk, tagv2, arg);
519 	case SIOCMCTPDROPTAG:
520 	case SIOCMCTPDROPTAG2:
521 		tagv2 = cmd == SIOCMCTPDROPTAG2;
522 		return mctp_ioctl_droptag(msk, tagv2, arg);
523 	}
524 
525 	return -EINVAL;
526 }
527 
528 #ifdef CONFIG_COMPAT
529 static int mctp_compat_ioctl(struct socket *sock, unsigned int cmd,
530 			     unsigned long arg)
531 {
532 	void __user *argp = compat_ptr(arg);
533 
534 	switch (cmd) {
535 	/* These have compatible ptr layouts */
536 	case SIOCMCTPALLOCTAG:
537 	case SIOCMCTPDROPTAG:
538 		return mctp_ioctl(sock, cmd, (unsigned long)argp);
539 	}
540 
541 	return -ENOIOCTLCMD;
542 }
543 #endif
544 
545 static const struct proto_ops mctp_dgram_ops = {
546 	.family		= PF_MCTP,
547 	.release	= mctp_release,
548 	.bind		= mctp_bind,
549 	.connect	= sock_no_connect,
550 	.socketpair	= sock_no_socketpair,
551 	.accept		= sock_no_accept,
552 	.getname	= sock_no_getname,
553 	.poll		= datagram_poll,
554 	.ioctl		= mctp_ioctl,
555 	.gettstamp	= sock_gettstamp,
556 	.listen		= sock_no_listen,
557 	.shutdown	= sock_no_shutdown,
558 	.setsockopt	= mctp_setsockopt,
559 	.getsockopt	= mctp_getsockopt,
560 	.sendmsg	= mctp_sendmsg,
561 	.recvmsg	= mctp_recvmsg,
562 	.mmap		= sock_no_mmap,
563 #ifdef CONFIG_COMPAT
564 	.compat_ioctl	= mctp_compat_ioctl,
565 #endif
566 };
567 
568 static void mctp_sk_expire_keys(struct timer_list *timer)
569 {
570 	struct mctp_sock *msk = container_of(timer, struct mctp_sock,
571 					     key_expiry);
572 	struct net *net = sock_net(&msk->sk);
573 	unsigned long next_expiry, flags, fl2;
574 	struct mctp_sk_key *key;
575 	struct hlist_node *tmp;
576 	bool next_expiry_valid = false;
577 
578 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
579 
580 	hlist_for_each_entry_safe(key, tmp, &msk->keys, sklist) {
581 		/* don't expire. manual_alloc is immutable, no locking
582 		 * required.
583 		 */
584 		if (key->manual_alloc)
585 			continue;
586 
587 		spin_lock_irqsave(&key->lock, fl2);
588 		if (!time_after_eq(key->expiry, jiffies)) {
589 			__mctp_key_remove(key, net, fl2,
590 					  MCTP_TRACE_KEY_TIMEOUT);
591 			continue;
592 		}
593 
594 		if (next_expiry_valid) {
595 			if (time_before(key->expiry, next_expiry))
596 				next_expiry = key->expiry;
597 		} else {
598 			next_expiry = key->expiry;
599 			next_expiry_valid = true;
600 		}
601 		spin_unlock_irqrestore(&key->lock, fl2);
602 	}
603 
604 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
605 
606 	if (next_expiry_valid)
607 		mod_timer(timer, next_expiry);
608 }
609 
610 static int mctp_sk_init(struct sock *sk)
611 {
612 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
613 
614 	INIT_HLIST_HEAD(&msk->keys);
615 	timer_setup(&msk->key_expiry, mctp_sk_expire_keys, 0);
616 	return 0;
617 }
618 
619 static void mctp_sk_close(struct sock *sk, long timeout)
620 {
621 	sk_common_release(sk);
622 }
623 
624 static int mctp_sk_hash(struct sock *sk)
625 {
626 	struct net *net = sock_net(sk);
627 	struct sock *existing;
628 	struct mctp_sock *msk;
629 	int rc;
630 
631 	msk = container_of(sk, struct mctp_sock, sk);
632 
633 	/* Bind lookup runs under RCU, remain live during that. */
634 	sock_set_flag(sk, SOCK_RCU_FREE);
635 
636 	mutex_lock(&net->mctp.bind_lock);
637 
638 	/* Prevent duplicate binds. */
639 	sk_for_each(existing, &net->mctp.binds) {
640 		struct mctp_sock *mex =
641 			container_of(existing, struct mctp_sock, sk);
642 
643 		if (mex->bind_type == msk->bind_type &&
644 		    mex->bind_addr == msk->bind_addr &&
645 		    mex->bind_net == msk->bind_net) {
646 			rc = -EADDRINUSE;
647 			goto out;
648 		}
649 	}
650 
651 	sk_add_node_rcu(sk, &net->mctp.binds);
652 	rc = 0;
653 
654 out:
655 	mutex_unlock(&net->mctp.bind_lock);
656 	return rc;
657 }
658 
659 static void mctp_sk_unhash(struct sock *sk)
660 {
661 	struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
662 	struct net *net = sock_net(sk);
663 	unsigned long flags, fl2;
664 	struct mctp_sk_key *key;
665 	struct hlist_node *tmp;
666 
667 	/* remove from any type-based binds */
668 	mutex_lock(&net->mctp.bind_lock);
669 	sk_del_node_init_rcu(sk);
670 	mutex_unlock(&net->mctp.bind_lock);
671 
672 	/* remove tag allocations */
673 	spin_lock_irqsave(&net->mctp.keys_lock, flags);
674 	hlist_for_each_entry_safe(key, tmp, &msk->keys, sklist) {
675 		spin_lock_irqsave(&key->lock, fl2);
676 		__mctp_key_remove(key, net, fl2, MCTP_TRACE_KEY_CLOSED);
677 	}
678 	sock_set_flag(sk, SOCK_DEAD);
679 	spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
680 
681 	/* Since there are no more tag allocations (we have removed all of the
682 	 * keys), stop any pending expiry events. the timer cannot be re-queued
683 	 * as the sk is no longer observable
684 	 */
685 	timer_delete_sync(&msk->key_expiry);
686 }
687 
688 static void mctp_sk_destruct(struct sock *sk)
689 {
690 	skb_queue_purge(&sk->sk_receive_queue);
691 }
692 
693 static struct proto mctp_proto = {
694 	.name		= "MCTP",
695 	.owner		= THIS_MODULE,
696 	.obj_size	= sizeof(struct mctp_sock),
697 	.init		= mctp_sk_init,
698 	.close		= mctp_sk_close,
699 	.hash		= mctp_sk_hash,
700 	.unhash		= mctp_sk_unhash,
701 };
702 
703 static int mctp_pf_create(struct net *net, struct socket *sock,
704 			  int protocol, int kern)
705 {
706 	const struct proto_ops *ops;
707 	struct proto *proto;
708 	struct sock *sk;
709 	int rc;
710 
711 	if (protocol)
712 		return -EPROTONOSUPPORT;
713 
714 	/* only datagram sockets are supported */
715 	if (sock->type != SOCK_DGRAM)
716 		return -ESOCKTNOSUPPORT;
717 
718 	proto = &mctp_proto;
719 	ops = &mctp_dgram_ops;
720 
721 	sock->state = SS_UNCONNECTED;
722 	sock->ops = ops;
723 
724 	sk = sk_alloc(net, PF_MCTP, GFP_KERNEL, proto, kern);
725 	if (!sk)
726 		return -ENOMEM;
727 
728 	sock_init_data(sock, sk);
729 	sk->sk_destruct = mctp_sk_destruct;
730 
731 	rc = 0;
732 	if (sk->sk_prot->init)
733 		rc = sk->sk_prot->init(sk);
734 
735 	if (rc)
736 		goto err_sk_put;
737 
738 	return 0;
739 
740 err_sk_put:
741 	sock_orphan(sk);
742 	sock_put(sk);
743 	return rc;
744 }
745 
746 static struct net_proto_family mctp_pf = {
747 	.family = PF_MCTP,
748 	.create = mctp_pf_create,
749 	.owner = THIS_MODULE,
750 };
751 
752 static __init int mctp_init(void)
753 {
754 	int rc;
755 
756 	/* ensure our uapi tag definitions match the header format */
757 	BUILD_BUG_ON(MCTP_TAG_OWNER != MCTP_HDR_FLAG_TO);
758 	BUILD_BUG_ON(MCTP_TAG_MASK != MCTP_HDR_TAG_MASK);
759 
760 	pr_info("mctp: management component transport protocol core\n");
761 
762 	rc = sock_register(&mctp_pf);
763 	if (rc)
764 		return rc;
765 
766 	rc = proto_register(&mctp_proto, 0);
767 	if (rc)
768 		goto err_unreg_sock;
769 
770 	rc = mctp_routes_init();
771 	if (rc)
772 		goto err_unreg_proto;
773 
774 	rc = mctp_neigh_init();
775 	if (rc)
776 		goto err_unreg_routes;
777 
778 	rc = mctp_device_init();
779 	if (rc)
780 		goto err_unreg_neigh;
781 
782 	return 0;
783 
784 err_unreg_neigh:
785 	mctp_neigh_exit();
786 err_unreg_routes:
787 	mctp_routes_exit();
788 err_unreg_proto:
789 	proto_unregister(&mctp_proto);
790 err_unreg_sock:
791 	sock_unregister(PF_MCTP);
792 
793 	return rc;
794 }
795 
796 static __exit void mctp_exit(void)
797 {
798 	mctp_device_exit();
799 	mctp_neigh_exit();
800 	mctp_routes_exit();
801 	proto_unregister(&mctp_proto);
802 	sock_unregister(PF_MCTP);
803 }
804 
805 subsys_initcall(mctp_init);
806 module_exit(mctp_exit);
807 
808 MODULE_DESCRIPTION("MCTP core");
809 MODULE_AUTHOR("Jeremy Kerr <jk@codeconstruct.com.au>");
810 
811 MODULE_ALIAS_NETPROTO(PF_MCTP);
812 
813 #if IS_ENABLED(CONFIG_MCTP_TEST)
814 #include "test/sock-test.c"
815 #endif
816