xref: /openbmc/linux/net/vmw_vsock/af_vsock.c (revision 22d55f02)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VMware vSockets Driver
4  *
5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
6  */
7 
8 /* Implementation notes:
9  *
10  * - There are two kinds of sockets: those created by user action (such as
11  * calling socket(2)) and those created by incoming connection request packets.
12  *
13  * - There are two "global" tables, one for bound sockets (sockets that have
14  * specified an address that they are responsible for) and one for connected
15  * sockets (sockets that have established a connection with another socket).
16  * These tables are "global" in that all sockets on the system are placed
17  * within them. - Note, though, that the bound table contains an extra entry
18  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
19  * that list. The bound table is used solely for lookup of sockets when packets
20  * are received and that's not necessary for SOCK_DGRAM sockets since we create
21  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
22  * sockets out of the bound hash buckets will reduce the chance of collisions
23  * when looking for SOCK_STREAM sockets and prevents us from having to check the
24  * socket type in the hash table lookups.
25  *
26  * - Sockets created by user action will either be "client" sockets that
27  * initiate a connection or "server" sockets that listen for connections; we do
28  * not support simultaneous connects (two "client" sockets connecting).
29  *
30  * - "Server" sockets are referred to as listener sockets throughout this
31  * implementation because they are in the TCP_LISTEN state.  When a
32  * connection request is received (the second kind of socket mentioned above),
33  * we create a new socket and refer to it as a pending socket.  These pending
34  * sockets are placed on the pending connection list of the listener socket.
35  * When future packets are received for the address the listener socket is
36  * bound to, we check if the source of the packet is from one that has an
37  * existing pending connection.  If it does, we process the packet for the
38  * pending socket.  When that socket reaches the connected state, it is removed
39  * from the listener socket's pending list and enqueued in the listener
40  * socket's accept queue.  Callers of accept(2) will accept connected sockets
41  * from the listener socket's accept queue.  If the socket cannot be accepted
42  * for some reason then it is marked rejected.  Once the connection is
43  * accepted, it is owned by the user process and the responsibility for cleanup
44  * falls with that user process.
45  *
46  * - It is possible that these pending sockets will never reach the connected
47  * state; in fact, we may never receive another packet after the connection
48  * request.  Because of this, we must schedule a cleanup function to run in the
49  * future, after some amount of time passes where a connection should have been
50  * established.  This function ensures that the socket is off all lists so it
51  * cannot be retrieved, then drops all references to the socket so it is cleaned
52  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
53  * function will also cleanup rejected sockets, those that reach the connected
54  * state but leave it before they have been accepted.
55  *
56  * - Lock ordering for pending or accept queue sockets is:
57  *
58  *     lock_sock(listener);
59  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
60  *
61  * Using explicit nested locking keeps lockdep happy since normally only one
62  * lock of a given class may be taken at a time.
63  *
64  * - Sockets created by user action will be cleaned up when the user process
65  * calls close(2), causing our release implementation to be called. Our release
66  * implementation will perform some cleanup then drop the last reference so our
67  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
68  * perform additional cleanup that's common for both types of sockets.
69  *
70  * - A socket's reference count is what ensures that the structure won't be
71  * freed.  Each entry in a list (such as the "global" bound and connected tables
72  * and the listener socket's pending list and connected queue) ensures a
73  * reference.  When we defer work until process context and pass a socket as our
74  * argument, we must ensure the reference count is increased to ensure the
75  * socket isn't freed before the function is run; the deferred function will
76  * then drop the reference.
77  *
78  * - sk->sk_state uses the TCP state constants because they are widely used by
79  * other address families and exposed to userspace tools like ss(8):
80  *
81  *   TCP_CLOSE - unconnected
82  *   TCP_SYN_SENT - connecting
83  *   TCP_ESTABLISHED - connected
84  *   TCP_CLOSING - disconnecting
85  *   TCP_LISTEN - listening
86  */
87 
88 #include <linux/types.h>
89 #include <linux/bitops.h>
90 #include <linux/cred.h>
91 #include <linux/init.h>
92 #include <linux/io.h>
93 #include <linux/kernel.h>
94 #include <linux/sched/signal.h>
95 #include <linux/kmod.h>
96 #include <linux/list.h>
97 #include <linux/miscdevice.h>
98 #include <linux/module.h>
99 #include <linux/mutex.h>
100 #include <linux/net.h>
101 #include <linux/poll.h>
102 #include <linux/random.h>
103 #include <linux/skbuff.h>
104 #include <linux/smp.h>
105 #include <linux/socket.h>
106 #include <linux/stddef.h>
107 #include <linux/unistd.h>
108 #include <linux/wait.h>
109 #include <linux/workqueue.h>
110 #include <net/sock.h>
111 #include <net/af_vsock.h>
112 
113 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
114 static void vsock_sk_destruct(struct sock *sk);
115 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
116 
117 /* Protocol family. */
118 static struct proto vsock_proto = {
119 	.name = "AF_VSOCK",
120 	.owner = THIS_MODULE,
121 	.obj_size = sizeof(struct vsock_sock),
122 };
123 
124 /* The default peer timeout indicates how long we will wait for a peer response
125  * to a control message.
126  */
127 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
128 
129 static const struct vsock_transport *transport;
130 static DEFINE_MUTEX(vsock_register_mutex);
131 
132 /**** EXPORTS ****/
133 
134 /* Get the ID of the local context.  This is transport dependent. */
135 
136 int vm_sockets_get_local_cid(void)
137 {
138 	return transport->get_local_cid();
139 }
140 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
141 
142 /**** UTILS ****/
143 
144 /* Each bound VSocket is stored in the bind hash table and each connected
145  * VSocket is stored in the connected hash table.
146  *
147  * Unbound sockets are all put on the same list attached to the end of the hash
148  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
149  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
150  * represents the list that addr hashes to).
151  *
152  * Specifically, we initialize the vsock_bind_table array to a size of
153  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
154  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
155  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
156  * mods with VSOCK_HASH_SIZE to ensure this.
157  */
158 #define MAX_PORT_RETRIES        24
159 
160 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
161 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
162 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
163 
164 /* XXX This can probably be implemented in a better way. */
165 #define VSOCK_CONN_HASH(src, dst)				\
166 	(((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
167 #define vsock_connected_sockets(src, dst)		\
168 	(&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
169 #define vsock_connected_sockets_vsk(vsk)				\
170 	vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
171 
172 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
173 EXPORT_SYMBOL_GPL(vsock_bind_table);
174 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
175 EXPORT_SYMBOL_GPL(vsock_connected_table);
176 DEFINE_SPINLOCK(vsock_table_lock);
177 EXPORT_SYMBOL_GPL(vsock_table_lock);
178 
179 /* Autobind this socket to the local address if necessary. */
180 static int vsock_auto_bind(struct vsock_sock *vsk)
181 {
182 	struct sock *sk = sk_vsock(vsk);
183 	struct sockaddr_vm local_addr;
184 
185 	if (vsock_addr_bound(&vsk->local_addr))
186 		return 0;
187 	vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
188 	return __vsock_bind(sk, &local_addr);
189 }
190 
191 static int __init vsock_init_tables(void)
192 {
193 	int i;
194 
195 	for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
196 		INIT_LIST_HEAD(&vsock_bind_table[i]);
197 
198 	for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
199 		INIT_LIST_HEAD(&vsock_connected_table[i]);
200 	return 0;
201 }
202 
203 static void __vsock_insert_bound(struct list_head *list,
204 				 struct vsock_sock *vsk)
205 {
206 	sock_hold(&vsk->sk);
207 	list_add(&vsk->bound_table, list);
208 }
209 
210 static void __vsock_insert_connected(struct list_head *list,
211 				     struct vsock_sock *vsk)
212 {
213 	sock_hold(&vsk->sk);
214 	list_add(&vsk->connected_table, list);
215 }
216 
217 static void __vsock_remove_bound(struct vsock_sock *vsk)
218 {
219 	list_del_init(&vsk->bound_table);
220 	sock_put(&vsk->sk);
221 }
222 
223 static void __vsock_remove_connected(struct vsock_sock *vsk)
224 {
225 	list_del_init(&vsk->connected_table);
226 	sock_put(&vsk->sk);
227 }
228 
229 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
230 {
231 	struct vsock_sock *vsk;
232 
233 	list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
234 		if (addr->svm_port == vsk->local_addr.svm_port)
235 			return sk_vsock(vsk);
236 
237 	return NULL;
238 }
239 
240 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
241 						  struct sockaddr_vm *dst)
242 {
243 	struct vsock_sock *vsk;
244 
245 	list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
246 			    connected_table) {
247 		if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
248 		    dst->svm_port == vsk->local_addr.svm_port) {
249 			return sk_vsock(vsk);
250 		}
251 	}
252 
253 	return NULL;
254 }
255 
256 static void vsock_insert_unbound(struct vsock_sock *vsk)
257 {
258 	spin_lock_bh(&vsock_table_lock);
259 	__vsock_insert_bound(vsock_unbound_sockets, vsk);
260 	spin_unlock_bh(&vsock_table_lock);
261 }
262 
263 void vsock_insert_connected(struct vsock_sock *vsk)
264 {
265 	struct list_head *list = vsock_connected_sockets(
266 		&vsk->remote_addr, &vsk->local_addr);
267 
268 	spin_lock_bh(&vsock_table_lock);
269 	__vsock_insert_connected(list, vsk);
270 	spin_unlock_bh(&vsock_table_lock);
271 }
272 EXPORT_SYMBOL_GPL(vsock_insert_connected);
273 
274 void vsock_remove_bound(struct vsock_sock *vsk)
275 {
276 	spin_lock_bh(&vsock_table_lock);
277 	__vsock_remove_bound(vsk);
278 	spin_unlock_bh(&vsock_table_lock);
279 }
280 EXPORT_SYMBOL_GPL(vsock_remove_bound);
281 
282 void vsock_remove_connected(struct vsock_sock *vsk)
283 {
284 	spin_lock_bh(&vsock_table_lock);
285 	__vsock_remove_connected(vsk);
286 	spin_unlock_bh(&vsock_table_lock);
287 }
288 EXPORT_SYMBOL_GPL(vsock_remove_connected);
289 
290 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
291 {
292 	struct sock *sk;
293 
294 	spin_lock_bh(&vsock_table_lock);
295 	sk = __vsock_find_bound_socket(addr);
296 	if (sk)
297 		sock_hold(sk);
298 
299 	spin_unlock_bh(&vsock_table_lock);
300 
301 	return sk;
302 }
303 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
304 
305 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
306 					 struct sockaddr_vm *dst)
307 {
308 	struct sock *sk;
309 
310 	spin_lock_bh(&vsock_table_lock);
311 	sk = __vsock_find_connected_socket(src, dst);
312 	if (sk)
313 		sock_hold(sk);
314 
315 	spin_unlock_bh(&vsock_table_lock);
316 
317 	return sk;
318 }
319 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
320 
321 static bool vsock_in_bound_table(struct vsock_sock *vsk)
322 {
323 	bool ret;
324 
325 	spin_lock_bh(&vsock_table_lock);
326 	ret = __vsock_in_bound_table(vsk);
327 	spin_unlock_bh(&vsock_table_lock);
328 
329 	return ret;
330 }
331 
332 static bool vsock_in_connected_table(struct vsock_sock *vsk)
333 {
334 	bool ret;
335 
336 	spin_lock_bh(&vsock_table_lock);
337 	ret = __vsock_in_connected_table(vsk);
338 	spin_unlock_bh(&vsock_table_lock);
339 
340 	return ret;
341 }
342 
343 void vsock_remove_sock(struct vsock_sock *vsk)
344 {
345 	if (vsock_in_bound_table(vsk))
346 		vsock_remove_bound(vsk);
347 
348 	if (vsock_in_connected_table(vsk))
349 		vsock_remove_connected(vsk);
350 }
351 EXPORT_SYMBOL_GPL(vsock_remove_sock);
352 
353 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
354 {
355 	int i;
356 
357 	spin_lock_bh(&vsock_table_lock);
358 
359 	for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
360 		struct vsock_sock *vsk;
361 		list_for_each_entry(vsk, &vsock_connected_table[i],
362 				    connected_table)
363 			fn(sk_vsock(vsk));
364 	}
365 
366 	spin_unlock_bh(&vsock_table_lock);
367 }
368 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
369 
370 void vsock_add_pending(struct sock *listener, struct sock *pending)
371 {
372 	struct vsock_sock *vlistener;
373 	struct vsock_sock *vpending;
374 
375 	vlistener = vsock_sk(listener);
376 	vpending = vsock_sk(pending);
377 
378 	sock_hold(pending);
379 	sock_hold(listener);
380 	list_add_tail(&vpending->pending_links, &vlistener->pending_links);
381 }
382 EXPORT_SYMBOL_GPL(vsock_add_pending);
383 
384 void vsock_remove_pending(struct sock *listener, struct sock *pending)
385 {
386 	struct vsock_sock *vpending = vsock_sk(pending);
387 
388 	list_del_init(&vpending->pending_links);
389 	sock_put(listener);
390 	sock_put(pending);
391 }
392 EXPORT_SYMBOL_GPL(vsock_remove_pending);
393 
394 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
395 {
396 	struct vsock_sock *vlistener;
397 	struct vsock_sock *vconnected;
398 
399 	vlistener = vsock_sk(listener);
400 	vconnected = vsock_sk(connected);
401 
402 	sock_hold(connected);
403 	sock_hold(listener);
404 	list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
405 }
406 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
407 
408 static struct sock *vsock_dequeue_accept(struct sock *listener)
409 {
410 	struct vsock_sock *vlistener;
411 	struct vsock_sock *vconnected;
412 
413 	vlistener = vsock_sk(listener);
414 
415 	if (list_empty(&vlistener->accept_queue))
416 		return NULL;
417 
418 	vconnected = list_entry(vlistener->accept_queue.next,
419 				struct vsock_sock, accept_queue);
420 
421 	list_del_init(&vconnected->accept_queue);
422 	sock_put(listener);
423 	/* The caller will need a reference on the connected socket so we let
424 	 * it call sock_put().
425 	 */
426 
427 	return sk_vsock(vconnected);
428 }
429 
430 static bool vsock_is_accept_queue_empty(struct sock *sk)
431 {
432 	struct vsock_sock *vsk = vsock_sk(sk);
433 	return list_empty(&vsk->accept_queue);
434 }
435 
436 static bool vsock_is_pending(struct sock *sk)
437 {
438 	struct vsock_sock *vsk = vsock_sk(sk);
439 	return !list_empty(&vsk->pending_links);
440 }
441 
442 static int vsock_send_shutdown(struct sock *sk, int mode)
443 {
444 	return transport->shutdown(vsock_sk(sk), mode);
445 }
446 
447 static void vsock_pending_work(struct work_struct *work)
448 {
449 	struct sock *sk;
450 	struct sock *listener;
451 	struct vsock_sock *vsk;
452 	bool cleanup;
453 
454 	vsk = container_of(work, struct vsock_sock, pending_work.work);
455 	sk = sk_vsock(vsk);
456 	listener = vsk->listener;
457 	cleanup = true;
458 
459 	lock_sock(listener);
460 	lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
461 
462 	if (vsock_is_pending(sk)) {
463 		vsock_remove_pending(listener, sk);
464 
465 		listener->sk_ack_backlog--;
466 	} else if (!vsk->rejected) {
467 		/* We are not on the pending list and accept() did not reject
468 		 * us, so we must have been accepted by our user process.  We
469 		 * just need to drop our references to the sockets and be on
470 		 * our way.
471 		 */
472 		cleanup = false;
473 		goto out;
474 	}
475 
476 	/* We need to remove ourself from the global connected sockets list so
477 	 * incoming packets can't find this socket, and to reduce the reference
478 	 * count.
479 	 */
480 	if (vsock_in_connected_table(vsk))
481 		vsock_remove_connected(vsk);
482 
483 	sk->sk_state = TCP_CLOSE;
484 
485 out:
486 	release_sock(sk);
487 	release_sock(listener);
488 	if (cleanup)
489 		sock_put(sk);
490 
491 	sock_put(sk);
492 	sock_put(listener);
493 }
494 
495 /**** SOCKET OPERATIONS ****/
496 
497 static int __vsock_bind_stream(struct vsock_sock *vsk,
498 			       struct sockaddr_vm *addr)
499 {
500 	static u32 port;
501 	struct sockaddr_vm new_addr;
502 
503 	if (!port)
504 		port = LAST_RESERVED_PORT + 1 +
505 			prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
506 
507 	vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
508 
509 	if (addr->svm_port == VMADDR_PORT_ANY) {
510 		bool found = false;
511 		unsigned int i;
512 
513 		for (i = 0; i < MAX_PORT_RETRIES; i++) {
514 			if (port <= LAST_RESERVED_PORT)
515 				port = LAST_RESERVED_PORT + 1;
516 
517 			new_addr.svm_port = port++;
518 
519 			if (!__vsock_find_bound_socket(&new_addr)) {
520 				found = true;
521 				break;
522 			}
523 		}
524 
525 		if (!found)
526 			return -EADDRNOTAVAIL;
527 	} else {
528 		/* If port is in reserved range, ensure caller
529 		 * has necessary privileges.
530 		 */
531 		if (addr->svm_port <= LAST_RESERVED_PORT &&
532 		    !capable(CAP_NET_BIND_SERVICE)) {
533 			return -EACCES;
534 		}
535 
536 		if (__vsock_find_bound_socket(&new_addr))
537 			return -EADDRINUSE;
538 	}
539 
540 	vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
541 
542 	/* Remove stream sockets from the unbound list and add them to the hash
543 	 * table for easy lookup by its address.  The unbound list is simply an
544 	 * extra entry at the end of the hash table, a trick used by AF_UNIX.
545 	 */
546 	__vsock_remove_bound(vsk);
547 	__vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
548 
549 	return 0;
550 }
551 
552 static int __vsock_bind_dgram(struct vsock_sock *vsk,
553 			      struct sockaddr_vm *addr)
554 {
555 	return transport->dgram_bind(vsk, addr);
556 }
557 
558 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
559 {
560 	struct vsock_sock *vsk = vsock_sk(sk);
561 	u32 cid;
562 	int retval;
563 
564 	/* First ensure this socket isn't already bound. */
565 	if (vsock_addr_bound(&vsk->local_addr))
566 		return -EINVAL;
567 
568 	/* Now bind to the provided address or select appropriate values if
569 	 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
570 	 * like AF_INET prevents binding to a non-local IP address (in most
571 	 * cases), we only allow binding to the local CID.
572 	 */
573 	cid = transport->get_local_cid();
574 	if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
575 		return -EADDRNOTAVAIL;
576 
577 	switch (sk->sk_socket->type) {
578 	case SOCK_STREAM:
579 		spin_lock_bh(&vsock_table_lock);
580 		retval = __vsock_bind_stream(vsk, addr);
581 		spin_unlock_bh(&vsock_table_lock);
582 		break;
583 
584 	case SOCK_DGRAM:
585 		retval = __vsock_bind_dgram(vsk, addr);
586 		break;
587 
588 	default:
589 		retval = -EINVAL;
590 		break;
591 	}
592 
593 	return retval;
594 }
595 
596 static void vsock_connect_timeout(struct work_struct *work);
597 
598 struct sock *__vsock_create(struct net *net,
599 			    struct socket *sock,
600 			    struct sock *parent,
601 			    gfp_t priority,
602 			    unsigned short type,
603 			    int kern)
604 {
605 	struct sock *sk;
606 	struct vsock_sock *psk;
607 	struct vsock_sock *vsk;
608 
609 	sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
610 	if (!sk)
611 		return NULL;
612 
613 	sock_init_data(sock, sk);
614 
615 	/* sk->sk_type is normally set in sock_init_data, but only if sock is
616 	 * non-NULL. We make sure that our sockets always have a type by
617 	 * setting it here if needed.
618 	 */
619 	if (!sock)
620 		sk->sk_type = type;
621 
622 	vsk = vsock_sk(sk);
623 	vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
624 	vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
625 
626 	sk->sk_destruct = vsock_sk_destruct;
627 	sk->sk_backlog_rcv = vsock_queue_rcv_skb;
628 	sock_reset_flag(sk, SOCK_DONE);
629 
630 	INIT_LIST_HEAD(&vsk->bound_table);
631 	INIT_LIST_HEAD(&vsk->connected_table);
632 	vsk->listener = NULL;
633 	INIT_LIST_HEAD(&vsk->pending_links);
634 	INIT_LIST_HEAD(&vsk->accept_queue);
635 	vsk->rejected = false;
636 	vsk->sent_request = false;
637 	vsk->ignore_connecting_rst = false;
638 	vsk->peer_shutdown = 0;
639 	INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
640 	INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
641 
642 	psk = parent ? vsock_sk(parent) : NULL;
643 	if (parent) {
644 		vsk->trusted = psk->trusted;
645 		vsk->owner = get_cred(psk->owner);
646 		vsk->connect_timeout = psk->connect_timeout;
647 	} else {
648 		vsk->trusted = capable(CAP_NET_ADMIN);
649 		vsk->owner = get_current_cred();
650 		vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
651 	}
652 
653 	if (transport->init(vsk, psk) < 0) {
654 		sk_free(sk);
655 		return NULL;
656 	}
657 
658 	if (sock)
659 		vsock_insert_unbound(vsk);
660 
661 	return sk;
662 }
663 EXPORT_SYMBOL_GPL(__vsock_create);
664 
665 static void __vsock_release(struct sock *sk)
666 {
667 	if (sk) {
668 		struct sk_buff *skb;
669 		struct sock *pending;
670 		struct vsock_sock *vsk;
671 
672 		vsk = vsock_sk(sk);
673 		pending = NULL;	/* Compiler warning. */
674 
675 		transport->release(vsk);
676 
677 		lock_sock(sk);
678 		sock_orphan(sk);
679 		sk->sk_shutdown = SHUTDOWN_MASK;
680 
681 		while ((skb = skb_dequeue(&sk->sk_receive_queue)))
682 			kfree_skb(skb);
683 
684 		/* Clean up any sockets that never were accepted. */
685 		while ((pending = vsock_dequeue_accept(sk)) != NULL) {
686 			__vsock_release(pending);
687 			sock_put(pending);
688 		}
689 
690 		release_sock(sk);
691 		sock_put(sk);
692 	}
693 }
694 
695 static void vsock_sk_destruct(struct sock *sk)
696 {
697 	struct vsock_sock *vsk = vsock_sk(sk);
698 
699 	transport->destruct(vsk);
700 
701 	/* When clearing these addresses, there's no need to set the family and
702 	 * possibly register the address family with the kernel.
703 	 */
704 	vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
705 	vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
706 
707 	put_cred(vsk->owner);
708 }
709 
710 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
711 {
712 	int err;
713 
714 	err = sock_queue_rcv_skb(sk, skb);
715 	if (err)
716 		kfree_skb(skb);
717 
718 	return err;
719 }
720 
721 s64 vsock_stream_has_data(struct vsock_sock *vsk)
722 {
723 	return transport->stream_has_data(vsk);
724 }
725 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
726 
727 s64 vsock_stream_has_space(struct vsock_sock *vsk)
728 {
729 	return transport->stream_has_space(vsk);
730 }
731 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
732 
733 static int vsock_release(struct socket *sock)
734 {
735 	__vsock_release(sock->sk);
736 	sock->sk = NULL;
737 	sock->state = SS_FREE;
738 
739 	return 0;
740 }
741 
742 static int
743 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
744 {
745 	int err;
746 	struct sock *sk;
747 	struct sockaddr_vm *vm_addr;
748 
749 	sk = sock->sk;
750 
751 	if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
752 		return -EINVAL;
753 
754 	lock_sock(sk);
755 	err = __vsock_bind(sk, vm_addr);
756 	release_sock(sk);
757 
758 	return err;
759 }
760 
761 static int vsock_getname(struct socket *sock,
762 			 struct sockaddr *addr, int peer)
763 {
764 	int err;
765 	struct sock *sk;
766 	struct vsock_sock *vsk;
767 	struct sockaddr_vm *vm_addr;
768 
769 	sk = sock->sk;
770 	vsk = vsock_sk(sk);
771 	err = 0;
772 
773 	lock_sock(sk);
774 
775 	if (peer) {
776 		if (sock->state != SS_CONNECTED) {
777 			err = -ENOTCONN;
778 			goto out;
779 		}
780 		vm_addr = &vsk->remote_addr;
781 	} else {
782 		vm_addr = &vsk->local_addr;
783 	}
784 
785 	if (!vm_addr) {
786 		err = -EINVAL;
787 		goto out;
788 	}
789 
790 	/* sys_getsockname() and sys_getpeername() pass us a
791 	 * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
792 	 * that macro is defined in socket.c instead of .h, so we hardcode its
793 	 * value here.
794 	 */
795 	BUILD_BUG_ON(sizeof(*vm_addr) > 128);
796 	memcpy(addr, vm_addr, sizeof(*vm_addr));
797 	err = sizeof(*vm_addr);
798 
799 out:
800 	release_sock(sk);
801 	return err;
802 }
803 
804 static int vsock_shutdown(struct socket *sock, int mode)
805 {
806 	int err;
807 	struct sock *sk;
808 
809 	/* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
810 	 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
811 	 * here like the other address families do.  Note also that the
812 	 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
813 	 * which is what we want.
814 	 */
815 	mode++;
816 
817 	if ((mode & ~SHUTDOWN_MASK) || !mode)
818 		return -EINVAL;
819 
820 	/* If this is a STREAM socket and it is not connected then bail out
821 	 * immediately.  If it is a DGRAM socket then we must first kick the
822 	 * socket so that it wakes up from any sleeping calls, for example
823 	 * recv(), and then afterwards return the error.
824 	 */
825 
826 	sk = sock->sk;
827 	if (sock->state == SS_UNCONNECTED) {
828 		err = -ENOTCONN;
829 		if (sk->sk_type == SOCK_STREAM)
830 			return err;
831 	} else {
832 		sock->state = SS_DISCONNECTING;
833 		err = 0;
834 	}
835 
836 	/* Receive and send shutdowns are treated alike. */
837 	mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
838 	if (mode) {
839 		lock_sock(sk);
840 		sk->sk_shutdown |= mode;
841 		sk->sk_state_change(sk);
842 		release_sock(sk);
843 
844 		if (sk->sk_type == SOCK_STREAM) {
845 			sock_reset_flag(sk, SOCK_DONE);
846 			vsock_send_shutdown(sk, mode);
847 		}
848 	}
849 
850 	return err;
851 }
852 
853 static __poll_t vsock_poll(struct file *file, struct socket *sock,
854 			       poll_table *wait)
855 {
856 	struct sock *sk;
857 	__poll_t mask;
858 	struct vsock_sock *vsk;
859 
860 	sk = sock->sk;
861 	vsk = vsock_sk(sk);
862 
863 	poll_wait(file, sk_sleep(sk), wait);
864 	mask = 0;
865 
866 	if (sk->sk_err)
867 		/* Signify that there has been an error on this socket. */
868 		mask |= EPOLLERR;
869 
870 	/* INET sockets treat local write shutdown and peer write shutdown as a
871 	 * case of EPOLLHUP set.
872 	 */
873 	if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
874 	    ((sk->sk_shutdown & SEND_SHUTDOWN) &&
875 	     (vsk->peer_shutdown & SEND_SHUTDOWN))) {
876 		mask |= EPOLLHUP;
877 	}
878 
879 	if (sk->sk_shutdown & RCV_SHUTDOWN ||
880 	    vsk->peer_shutdown & SEND_SHUTDOWN) {
881 		mask |= EPOLLRDHUP;
882 	}
883 
884 	if (sock->type == SOCK_DGRAM) {
885 		/* For datagram sockets we can read if there is something in
886 		 * the queue and write as long as the socket isn't shutdown for
887 		 * sending.
888 		 */
889 		if (!skb_queue_empty(&sk->sk_receive_queue) ||
890 		    (sk->sk_shutdown & RCV_SHUTDOWN)) {
891 			mask |= EPOLLIN | EPOLLRDNORM;
892 		}
893 
894 		if (!(sk->sk_shutdown & SEND_SHUTDOWN))
895 			mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
896 
897 	} else if (sock->type == SOCK_STREAM) {
898 		lock_sock(sk);
899 
900 		/* Listening sockets that have connections in their accept
901 		 * queue can be read.
902 		 */
903 		if (sk->sk_state == TCP_LISTEN
904 		    && !vsock_is_accept_queue_empty(sk))
905 			mask |= EPOLLIN | EPOLLRDNORM;
906 
907 		/* If there is something in the queue then we can read. */
908 		if (transport->stream_is_active(vsk) &&
909 		    !(sk->sk_shutdown & RCV_SHUTDOWN)) {
910 			bool data_ready_now = false;
911 			int ret = transport->notify_poll_in(
912 					vsk, 1, &data_ready_now);
913 			if (ret < 0) {
914 				mask |= EPOLLERR;
915 			} else {
916 				if (data_ready_now)
917 					mask |= EPOLLIN | EPOLLRDNORM;
918 
919 			}
920 		}
921 
922 		/* Sockets whose connections have been closed, reset, or
923 		 * terminated should also be considered read, and we check the
924 		 * shutdown flag for that.
925 		 */
926 		if (sk->sk_shutdown & RCV_SHUTDOWN ||
927 		    vsk->peer_shutdown & SEND_SHUTDOWN) {
928 			mask |= EPOLLIN | EPOLLRDNORM;
929 		}
930 
931 		/* Connected sockets that can produce data can be written. */
932 		if (sk->sk_state == TCP_ESTABLISHED) {
933 			if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
934 				bool space_avail_now = false;
935 				int ret = transport->notify_poll_out(
936 						vsk, 1, &space_avail_now);
937 				if (ret < 0) {
938 					mask |= EPOLLERR;
939 				} else {
940 					if (space_avail_now)
941 						/* Remove EPOLLWRBAND since INET
942 						 * sockets are not setting it.
943 						 */
944 						mask |= EPOLLOUT | EPOLLWRNORM;
945 
946 				}
947 			}
948 		}
949 
950 		/* Simulate INET socket poll behaviors, which sets
951 		 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
952 		 * but local send is not shutdown.
953 		 */
954 		if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
955 			if (!(sk->sk_shutdown & SEND_SHUTDOWN))
956 				mask |= EPOLLOUT | EPOLLWRNORM;
957 
958 		}
959 
960 		release_sock(sk);
961 	}
962 
963 	return mask;
964 }
965 
966 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
967 			       size_t len)
968 {
969 	int err;
970 	struct sock *sk;
971 	struct vsock_sock *vsk;
972 	struct sockaddr_vm *remote_addr;
973 
974 	if (msg->msg_flags & MSG_OOB)
975 		return -EOPNOTSUPP;
976 
977 	/* For now, MSG_DONTWAIT is always assumed... */
978 	err = 0;
979 	sk = sock->sk;
980 	vsk = vsock_sk(sk);
981 
982 	lock_sock(sk);
983 
984 	err = vsock_auto_bind(vsk);
985 	if (err)
986 		goto out;
987 
988 
989 	/* If the provided message contains an address, use that.  Otherwise
990 	 * fall back on the socket's remote handle (if it has been connected).
991 	 */
992 	if (msg->msg_name &&
993 	    vsock_addr_cast(msg->msg_name, msg->msg_namelen,
994 			    &remote_addr) == 0) {
995 		/* Ensure this address is of the right type and is a valid
996 		 * destination.
997 		 */
998 
999 		if (remote_addr->svm_cid == VMADDR_CID_ANY)
1000 			remote_addr->svm_cid = transport->get_local_cid();
1001 
1002 		if (!vsock_addr_bound(remote_addr)) {
1003 			err = -EINVAL;
1004 			goto out;
1005 		}
1006 	} else if (sock->state == SS_CONNECTED) {
1007 		remote_addr = &vsk->remote_addr;
1008 
1009 		if (remote_addr->svm_cid == VMADDR_CID_ANY)
1010 			remote_addr->svm_cid = transport->get_local_cid();
1011 
1012 		/* XXX Should connect() or this function ensure remote_addr is
1013 		 * bound?
1014 		 */
1015 		if (!vsock_addr_bound(&vsk->remote_addr)) {
1016 			err = -EINVAL;
1017 			goto out;
1018 		}
1019 	} else {
1020 		err = -EINVAL;
1021 		goto out;
1022 	}
1023 
1024 	if (!transport->dgram_allow(remote_addr->svm_cid,
1025 				    remote_addr->svm_port)) {
1026 		err = -EINVAL;
1027 		goto out;
1028 	}
1029 
1030 	err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1031 
1032 out:
1033 	release_sock(sk);
1034 	return err;
1035 }
1036 
1037 static int vsock_dgram_connect(struct socket *sock,
1038 			       struct sockaddr *addr, int addr_len, int flags)
1039 {
1040 	int err;
1041 	struct sock *sk;
1042 	struct vsock_sock *vsk;
1043 	struct sockaddr_vm *remote_addr;
1044 
1045 	sk = sock->sk;
1046 	vsk = vsock_sk(sk);
1047 
1048 	err = vsock_addr_cast(addr, addr_len, &remote_addr);
1049 	if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1050 		lock_sock(sk);
1051 		vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1052 				VMADDR_PORT_ANY);
1053 		sock->state = SS_UNCONNECTED;
1054 		release_sock(sk);
1055 		return 0;
1056 	} else if (err != 0)
1057 		return -EINVAL;
1058 
1059 	lock_sock(sk);
1060 
1061 	err = vsock_auto_bind(vsk);
1062 	if (err)
1063 		goto out;
1064 
1065 	if (!transport->dgram_allow(remote_addr->svm_cid,
1066 				    remote_addr->svm_port)) {
1067 		err = -EINVAL;
1068 		goto out;
1069 	}
1070 
1071 	memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1072 	sock->state = SS_CONNECTED;
1073 
1074 out:
1075 	release_sock(sk);
1076 	return err;
1077 }
1078 
1079 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1080 			       size_t len, int flags)
1081 {
1082 	return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1083 }
1084 
1085 static const struct proto_ops vsock_dgram_ops = {
1086 	.family = PF_VSOCK,
1087 	.owner = THIS_MODULE,
1088 	.release = vsock_release,
1089 	.bind = vsock_bind,
1090 	.connect = vsock_dgram_connect,
1091 	.socketpair = sock_no_socketpair,
1092 	.accept = sock_no_accept,
1093 	.getname = vsock_getname,
1094 	.poll = vsock_poll,
1095 	.ioctl = sock_no_ioctl,
1096 	.listen = sock_no_listen,
1097 	.shutdown = vsock_shutdown,
1098 	.setsockopt = sock_no_setsockopt,
1099 	.getsockopt = sock_no_getsockopt,
1100 	.sendmsg = vsock_dgram_sendmsg,
1101 	.recvmsg = vsock_dgram_recvmsg,
1102 	.mmap = sock_no_mmap,
1103 	.sendpage = sock_no_sendpage,
1104 };
1105 
1106 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1107 {
1108 	if (!transport->cancel_pkt)
1109 		return -EOPNOTSUPP;
1110 
1111 	return transport->cancel_pkt(vsk);
1112 }
1113 
1114 static void vsock_connect_timeout(struct work_struct *work)
1115 {
1116 	struct sock *sk;
1117 	struct vsock_sock *vsk;
1118 	int cancel = 0;
1119 
1120 	vsk = container_of(work, struct vsock_sock, connect_work.work);
1121 	sk = sk_vsock(vsk);
1122 
1123 	lock_sock(sk);
1124 	if (sk->sk_state == TCP_SYN_SENT &&
1125 	    (sk->sk_shutdown != SHUTDOWN_MASK)) {
1126 		sk->sk_state = TCP_CLOSE;
1127 		sk->sk_err = ETIMEDOUT;
1128 		sk->sk_error_report(sk);
1129 		cancel = 1;
1130 	}
1131 	release_sock(sk);
1132 	if (cancel)
1133 		vsock_transport_cancel_pkt(vsk);
1134 
1135 	sock_put(sk);
1136 }
1137 
1138 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1139 				int addr_len, int flags)
1140 {
1141 	int err;
1142 	struct sock *sk;
1143 	struct vsock_sock *vsk;
1144 	struct sockaddr_vm *remote_addr;
1145 	long timeout;
1146 	DEFINE_WAIT(wait);
1147 
1148 	err = 0;
1149 	sk = sock->sk;
1150 	vsk = vsock_sk(sk);
1151 
1152 	lock_sock(sk);
1153 
1154 	/* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1155 	switch (sock->state) {
1156 	case SS_CONNECTED:
1157 		err = -EISCONN;
1158 		goto out;
1159 	case SS_DISCONNECTING:
1160 		err = -EINVAL;
1161 		goto out;
1162 	case SS_CONNECTING:
1163 		/* This continues on so we can move sock into the SS_CONNECTED
1164 		 * state once the connection has completed (at which point err
1165 		 * will be set to zero also).  Otherwise, we will either wait
1166 		 * for the connection or return -EALREADY should this be a
1167 		 * non-blocking call.
1168 		 */
1169 		err = -EALREADY;
1170 		break;
1171 	default:
1172 		if ((sk->sk_state == TCP_LISTEN) ||
1173 		    vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1174 			err = -EINVAL;
1175 			goto out;
1176 		}
1177 
1178 		/* The hypervisor and well-known contexts do not have socket
1179 		 * endpoints.
1180 		 */
1181 		if (!transport->stream_allow(remote_addr->svm_cid,
1182 					     remote_addr->svm_port)) {
1183 			err = -ENETUNREACH;
1184 			goto out;
1185 		}
1186 
1187 		/* Set the remote address that we are connecting to. */
1188 		memcpy(&vsk->remote_addr, remote_addr,
1189 		       sizeof(vsk->remote_addr));
1190 
1191 		err = vsock_auto_bind(vsk);
1192 		if (err)
1193 			goto out;
1194 
1195 		sk->sk_state = TCP_SYN_SENT;
1196 
1197 		err = transport->connect(vsk);
1198 		if (err < 0)
1199 			goto out;
1200 
1201 		/* Mark sock as connecting and set the error code to in
1202 		 * progress in case this is a non-blocking connect.
1203 		 */
1204 		sock->state = SS_CONNECTING;
1205 		err = -EINPROGRESS;
1206 	}
1207 
1208 	/* The receive path will handle all communication until we are able to
1209 	 * enter the connected state.  Here we wait for the connection to be
1210 	 * completed or a notification of an error.
1211 	 */
1212 	timeout = vsk->connect_timeout;
1213 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1214 
1215 	while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1216 		if (flags & O_NONBLOCK) {
1217 			/* If we're not going to block, we schedule a timeout
1218 			 * function to generate a timeout on the connection
1219 			 * attempt, in case the peer doesn't respond in a
1220 			 * timely manner. We hold on to the socket until the
1221 			 * timeout fires.
1222 			 */
1223 			sock_hold(sk);
1224 			schedule_delayed_work(&vsk->connect_work, timeout);
1225 
1226 			/* Skip ahead to preserve error code set above. */
1227 			goto out_wait;
1228 		}
1229 
1230 		release_sock(sk);
1231 		timeout = schedule_timeout(timeout);
1232 		lock_sock(sk);
1233 
1234 		if (signal_pending(current)) {
1235 			err = sock_intr_errno(timeout);
1236 			sk->sk_state = TCP_CLOSE;
1237 			sock->state = SS_UNCONNECTED;
1238 			vsock_transport_cancel_pkt(vsk);
1239 			goto out_wait;
1240 		} else if (timeout == 0) {
1241 			err = -ETIMEDOUT;
1242 			sk->sk_state = TCP_CLOSE;
1243 			sock->state = SS_UNCONNECTED;
1244 			vsock_transport_cancel_pkt(vsk);
1245 			goto out_wait;
1246 		}
1247 
1248 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1249 	}
1250 
1251 	if (sk->sk_err) {
1252 		err = -sk->sk_err;
1253 		sk->sk_state = TCP_CLOSE;
1254 		sock->state = SS_UNCONNECTED;
1255 	} else {
1256 		err = 0;
1257 	}
1258 
1259 out_wait:
1260 	finish_wait(sk_sleep(sk), &wait);
1261 out:
1262 	release_sock(sk);
1263 	return err;
1264 }
1265 
1266 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1267 			bool kern)
1268 {
1269 	struct sock *listener;
1270 	int err;
1271 	struct sock *connected;
1272 	struct vsock_sock *vconnected;
1273 	long timeout;
1274 	DEFINE_WAIT(wait);
1275 
1276 	err = 0;
1277 	listener = sock->sk;
1278 
1279 	lock_sock(listener);
1280 
1281 	if (sock->type != SOCK_STREAM) {
1282 		err = -EOPNOTSUPP;
1283 		goto out;
1284 	}
1285 
1286 	if (listener->sk_state != TCP_LISTEN) {
1287 		err = -EINVAL;
1288 		goto out;
1289 	}
1290 
1291 	/* Wait for children sockets to appear; these are the new sockets
1292 	 * created upon connection establishment.
1293 	 */
1294 	timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1295 	prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1296 
1297 	while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1298 	       listener->sk_err == 0) {
1299 		release_sock(listener);
1300 		timeout = schedule_timeout(timeout);
1301 		finish_wait(sk_sleep(listener), &wait);
1302 		lock_sock(listener);
1303 
1304 		if (signal_pending(current)) {
1305 			err = sock_intr_errno(timeout);
1306 			goto out;
1307 		} else if (timeout == 0) {
1308 			err = -EAGAIN;
1309 			goto out;
1310 		}
1311 
1312 		prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1313 	}
1314 	finish_wait(sk_sleep(listener), &wait);
1315 
1316 	if (listener->sk_err)
1317 		err = -listener->sk_err;
1318 
1319 	if (connected) {
1320 		listener->sk_ack_backlog--;
1321 
1322 		lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1323 		vconnected = vsock_sk(connected);
1324 
1325 		/* If the listener socket has received an error, then we should
1326 		 * reject this socket and return.  Note that we simply mark the
1327 		 * socket rejected, drop our reference, and let the cleanup
1328 		 * function handle the cleanup; the fact that we found it in
1329 		 * the listener's accept queue guarantees that the cleanup
1330 		 * function hasn't run yet.
1331 		 */
1332 		if (err) {
1333 			vconnected->rejected = true;
1334 		} else {
1335 			newsock->state = SS_CONNECTED;
1336 			sock_graft(connected, newsock);
1337 		}
1338 
1339 		release_sock(connected);
1340 		sock_put(connected);
1341 	}
1342 
1343 out:
1344 	release_sock(listener);
1345 	return err;
1346 }
1347 
1348 static int vsock_listen(struct socket *sock, int backlog)
1349 {
1350 	int err;
1351 	struct sock *sk;
1352 	struct vsock_sock *vsk;
1353 
1354 	sk = sock->sk;
1355 
1356 	lock_sock(sk);
1357 
1358 	if (sock->type != SOCK_STREAM) {
1359 		err = -EOPNOTSUPP;
1360 		goto out;
1361 	}
1362 
1363 	if (sock->state != SS_UNCONNECTED) {
1364 		err = -EINVAL;
1365 		goto out;
1366 	}
1367 
1368 	vsk = vsock_sk(sk);
1369 
1370 	if (!vsock_addr_bound(&vsk->local_addr)) {
1371 		err = -EINVAL;
1372 		goto out;
1373 	}
1374 
1375 	sk->sk_max_ack_backlog = backlog;
1376 	sk->sk_state = TCP_LISTEN;
1377 
1378 	err = 0;
1379 
1380 out:
1381 	release_sock(sk);
1382 	return err;
1383 }
1384 
1385 static int vsock_stream_setsockopt(struct socket *sock,
1386 				   int level,
1387 				   int optname,
1388 				   char __user *optval,
1389 				   unsigned int optlen)
1390 {
1391 	int err;
1392 	struct sock *sk;
1393 	struct vsock_sock *vsk;
1394 	u64 val;
1395 
1396 	if (level != AF_VSOCK)
1397 		return -ENOPROTOOPT;
1398 
1399 #define COPY_IN(_v)                                       \
1400 	do {						  \
1401 		if (optlen < sizeof(_v)) {		  \
1402 			err = -EINVAL;			  \
1403 			goto exit;			  \
1404 		}					  \
1405 		if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {	\
1406 			err = -EFAULT;					\
1407 			goto exit;					\
1408 		}							\
1409 	} while (0)
1410 
1411 	err = 0;
1412 	sk = sock->sk;
1413 	vsk = vsock_sk(sk);
1414 
1415 	lock_sock(sk);
1416 
1417 	switch (optname) {
1418 	case SO_VM_SOCKETS_BUFFER_SIZE:
1419 		COPY_IN(val);
1420 		transport->set_buffer_size(vsk, val);
1421 		break;
1422 
1423 	case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1424 		COPY_IN(val);
1425 		transport->set_max_buffer_size(vsk, val);
1426 		break;
1427 
1428 	case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1429 		COPY_IN(val);
1430 		transport->set_min_buffer_size(vsk, val);
1431 		break;
1432 
1433 	case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1434 		struct __kernel_old_timeval tv;
1435 		COPY_IN(tv);
1436 		if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1437 		    tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1438 			vsk->connect_timeout = tv.tv_sec * HZ +
1439 			    DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1440 			if (vsk->connect_timeout == 0)
1441 				vsk->connect_timeout =
1442 				    VSOCK_DEFAULT_CONNECT_TIMEOUT;
1443 
1444 		} else {
1445 			err = -ERANGE;
1446 		}
1447 		break;
1448 	}
1449 
1450 	default:
1451 		err = -ENOPROTOOPT;
1452 		break;
1453 	}
1454 
1455 #undef COPY_IN
1456 
1457 exit:
1458 	release_sock(sk);
1459 	return err;
1460 }
1461 
1462 static int vsock_stream_getsockopt(struct socket *sock,
1463 				   int level, int optname,
1464 				   char __user *optval,
1465 				   int __user *optlen)
1466 {
1467 	int err;
1468 	int len;
1469 	struct sock *sk;
1470 	struct vsock_sock *vsk;
1471 	u64 val;
1472 
1473 	if (level != AF_VSOCK)
1474 		return -ENOPROTOOPT;
1475 
1476 	err = get_user(len, optlen);
1477 	if (err != 0)
1478 		return err;
1479 
1480 #define COPY_OUT(_v)                            \
1481 	do {					\
1482 		if (len < sizeof(_v))		\
1483 			return -EINVAL;		\
1484 						\
1485 		len = sizeof(_v);		\
1486 		if (copy_to_user(optval, &_v, len) != 0)	\
1487 			return -EFAULT;				\
1488 								\
1489 	} while (0)
1490 
1491 	err = 0;
1492 	sk = sock->sk;
1493 	vsk = vsock_sk(sk);
1494 
1495 	switch (optname) {
1496 	case SO_VM_SOCKETS_BUFFER_SIZE:
1497 		val = transport->get_buffer_size(vsk);
1498 		COPY_OUT(val);
1499 		break;
1500 
1501 	case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1502 		val = transport->get_max_buffer_size(vsk);
1503 		COPY_OUT(val);
1504 		break;
1505 
1506 	case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1507 		val = transport->get_min_buffer_size(vsk);
1508 		COPY_OUT(val);
1509 		break;
1510 
1511 	case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1512 		struct __kernel_old_timeval tv;
1513 		tv.tv_sec = vsk->connect_timeout / HZ;
1514 		tv.tv_usec =
1515 		    (vsk->connect_timeout -
1516 		     tv.tv_sec * HZ) * (1000000 / HZ);
1517 		COPY_OUT(tv);
1518 		break;
1519 	}
1520 	default:
1521 		return -ENOPROTOOPT;
1522 	}
1523 
1524 	err = put_user(len, optlen);
1525 	if (err != 0)
1526 		return -EFAULT;
1527 
1528 #undef COPY_OUT
1529 
1530 	return 0;
1531 }
1532 
1533 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1534 				size_t len)
1535 {
1536 	struct sock *sk;
1537 	struct vsock_sock *vsk;
1538 	ssize_t total_written;
1539 	long timeout;
1540 	int err;
1541 	struct vsock_transport_send_notify_data send_data;
1542 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1543 
1544 	sk = sock->sk;
1545 	vsk = vsock_sk(sk);
1546 	total_written = 0;
1547 	err = 0;
1548 
1549 	if (msg->msg_flags & MSG_OOB)
1550 		return -EOPNOTSUPP;
1551 
1552 	lock_sock(sk);
1553 
1554 	/* Callers should not provide a destination with stream sockets. */
1555 	if (msg->msg_namelen) {
1556 		err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1557 		goto out;
1558 	}
1559 
1560 	/* Send data only if both sides are not shutdown in the direction. */
1561 	if (sk->sk_shutdown & SEND_SHUTDOWN ||
1562 	    vsk->peer_shutdown & RCV_SHUTDOWN) {
1563 		err = -EPIPE;
1564 		goto out;
1565 	}
1566 
1567 	if (sk->sk_state != TCP_ESTABLISHED ||
1568 	    !vsock_addr_bound(&vsk->local_addr)) {
1569 		err = -ENOTCONN;
1570 		goto out;
1571 	}
1572 
1573 	if (!vsock_addr_bound(&vsk->remote_addr)) {
1574 		err = -EDESTADDRREQ;
1575 		goto out;
1576 	}
1577 
1578 	/* Wait for room in the produce queue to enqueue our user's data. */
1579 	timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1580 
1581 	err = transport->notify_send_init(vsk, &send_data);
1582 	if (err < 0)
1583 		goto out;
1584 
1585 	while (total_written < len) {
1586 		ssize_t written;
1587 
1588 		add_wait_queue(sk_sleep(sk), &wait);
1589 		while (vsock_stream_has_space(vsk) == 0 &&
1590 		       sk->sk_err == 0 &&
1591 		       !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1592 		       !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1593 
1594 			/* Don't wait for non-blocking sockets. */
1595 			if (timeout == 0) {
1596 				err = -EAGAIN;
1597 				remove_wait_queue(sk_sleep(sk), &wait);
1598 				goto out_err;
1599 			}
1600 
1601 			err = transport->notify_send_pre_block(vsk, &send_data);
1602 			if (err < 0) {
1603 				remove_wait_queue(sk_sleep(sk), &wait);
1604 				goto out_err;
1605 			}
1606 
1607 			release_sock(sk);
1608 			timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1609 			lock_sock(sk);
1610 			if (signal_pending(current)) {
1611 				err = sock_intr_errno(timeout);
1612 				remove_wait_queue(sk_sleep(sk), &wait);
1613 				goto out_err;
1614 			} else if (timeout == 0) {
1615 				err = -EAGAIN;
1616 				remove_wait_queue(sk_sleep(sk), &wait);
1617 				goto out_err;
1618 			}
1619 		}
1620 		remove_wait_queue(sk_sleep(sk), &wait);
1621 
1622 		/* These checks occur both as part of and after the loop
1623 		 * conditional since we need to check before and after
1624 		 * sleeping.
1625 		 */
1626 		if (sk->sk_err) {
1627 			err = -sk->sk_err;
1628 			goto out_err;
1629 		} else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1630 			   (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1631 			err = -EPIPE;
1632 			goto out_err;
1633 		}
1634 
1635 		err = transport->notify_send_pre_enqueue(vsk, &send_data);
1636 		if (err < 0)
1637 			goto out_err;
1638 
1639 		/* Note that enqueue will only write as many bytes as are free
1640 		 * in the produce queue, so we don't need to ensure len is
1641 		 * smaller than the queue size.  It is the caller's
1642 		 * responsibility to check how many bytes we were able to send.
1643 		 */
1644 
1645 		written = transport->stream_enqueue(
1646 				vsk, msg,
1647 				len - total_written);
1648 		if (written < 0) {
1649 			err = -ENOMEM;
1650 			goto out_err;
1651 		}
1652 
1653 		total_written += written;
1654 
1655 		err = transport->notify_send_post_enqueue(
1656 				vsk, written, &send_data);
1657 		if (err < 0)
1658 			goto out_err;
1659 
1660 	}
1661 
1662 out_err:
1663 	if (total_written > 0)
1664 		err = total_written;
1665 out:
1666 	release_sock(sk);
1667 	return err;
1668 }
1669 
1670 
1671 static int
1672 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1673 		     int flags)
1674 {
1675 	struct sock *sk;
1676 	struct vsock_sock *vsk;
1677 	int err;
1678 	size_t target;
1679 	ssize_t copied;
1680 	long timeout;
1681 	struct vsock_transport_recv_notify_data recv_data;
1682 
1683 	DEFINE_WAIT(wait);
1684 
1685 	sk = sock->sk;
1686 	vsk = vsock_sk(sk);
1687 	err = 0;
1688 
1689 	lock_sock(sk);
1690 
1691 	if (sk->sk_state != TCP_ESTABLISHED) {
1692 		/* Recvmsg is supposed to return 0 if a peer performs an
1693 		 * orderly shutdown. Differentiate between that case and when a
1694 		 * peer has not connected or a local shutdown occured with the
1695 		 * SOCK_DONE flag.
1696 		 */
1697 		if (sock_flag(sk, SOCK_DONE))
1698 			err = 0;
1699 		else
1700 			err = -ENOTCONN;
1701 
1702 		goto out;
1703 	}
1704 
1705 	if (flags & MSG_OOB) {
1706 		err = -EOPNOTSUPP;
1707 		goto out;
1708 	}
1709 
1710 	/* We don't check peer_shutdown flag here since peer may actually shut
1711 	 * down, but there can be data in the queue that a local socket can
1712 	 * receive.
1713 	 */
1714 	if (sk->sk_shutdown & RCV_SHUTDOWN) {
1715 		err = 0;
1716 		goto out;
1717 	}
1718 
1719 	/* It is valid on Linux to pass in a zero-length receive buffer.  This
1720 	 * is not an error.  We may as well bail out now.
1721 	 */
1722 	if (!len) {
1723 		err = 0;
1724 		goto out;
1725 	}
1726 
1727 	/* We must not copy less than target bytes into the user's buffer
1728 	 * before returning successfully, so we wait for the consume queue to
1729 	 * have that much data to consume before dequeueing.  Note that this
1730 	 * makes it impossible to handle cases where target is greater than the
1731 	 * queue size.
1732 	 */
1733 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1734 	if (target >= transport->stream_rcvhiwat(vsk)) {
1735 		err = -ENOMEM;
1736 		goto out;
1737 	}
1738 	timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1739 	copied = 0;
1740 
1741 	err = transport->notify_recv_init(vsk, target, &recv_data);
1742 	if (err < 0)
1743 		goto out;
1744 
1745 
1746 	while (1) {
1747 		s64 ready;
1748 
1749 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1750 		ready = vsock_stream_has_data(vsk);
1751 
1752 		if (ready == 0) {
1753 			if (sk->sk_err != 0 ||
1754 			    (sk->sk_shutdown & RCV_SHUTDOWN) ||
1755 			    (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1756 				finish_wait(sk_sleep(sk), &wait);
1757 				break;
1758 			}
1759 			/* Don't wait for non-blocking sockets. */
1760 			if (timeout == 0) {
1761 				err = -EAGAIN;
1762 				finish_wait(sk_sleep(sk), &wait);
1763 				break;
1764 			}
1765 
1766 			err = transport->notify_recv_pre_block(
1767 					vsk, target, &recv_data);
1768 			if (err < 0) {
1769 				finish_wait(sk_sleep(sk), &wait);
1770 				break;
1771 			}
1772 			release_sock(sk);
1773 			timeout = schedule_timeout(timeout);
1774 			lock_sock(sk);
1775 
1776 			if (signal_pending(current)) {
1777 				err = sock_intr_errno(timeout);
1778 				finish_wait(sk_sleep(sk), &wait);
1779 				break;
1780 			} else if (timeout == 0) {
1781 				err = -EAGAIN;
1782 				finish_wait(sk_sleep(sk), &wait);
1783 				break;
1784 			}
1785 		} else {
1786 			ssize_t read;
1787 
1788 			finish_wait(sk_sleep(sk), &wait);
1789 
1790 			if (ready < 0) {
1791 				/* Invalid queue pair content. XXX This should
1792 				* be changed to a connection reset in a later
1793 				* change.
1794 				*/
1795 
1796 				err = -ENOMEM;
1797 				goto out;
1798 			}
1799 
1800 			err = transport->notify_recv_pre_dequeue(
1801 					vsk, target, &recv_data);
1802 			if (err < 0)
1803 				break;
1804 
1805 			read = transport->stream_dequeue(
1806 					vsk, msg,
1807 					len - copied, flags);
1808 			if (read < 0) {
1809 				err = -ENOMEM;
1810 				break;
1811 			}
1812 
1813 			copied += read;
1814 
1815 			err = transport->notify_recv_post_dequeue(
1816 					vsk, target, read,
1817 					!(flags & MSG_PEEK), &recv_data);
1818 			if (err < 0)
1819 				goto out;
1820 
1821 			if (read >= target || flags & MSG_PEEK)
1822 				break;
1823 
1824 			target -= read;
1825 		}
1826 	}
1827 
1828 	if (sk->sk_err)
1829 		err = -sk->sk_err;
1830 	else if (sk->sk_shutdown & RCV_SHUTDOWN)
1831 		err = 0;
1832 
1833 	if (copied > 0)
1834 		err = copied;
1835 
1836 out:
1837 	release_sock(sk);
1838 	return err;
1839 }
1840 
1841 static const struct proto_ops vsock_stream_ops = {
1842 	.family = PF_VSOCK,
1843 	.owner = THIS_MODULE,
1844 	.release = vsock_release,
1845 	.bind = vsock_bind,
1846 	.connect = vsock_stream_connect,
1847 	.socketpair = sock_no_socketpair,
1848 	.accept = vsock_accept,
1849 	.getname = vsock_getname,
1850 	.poll = vsock_poll,
1851 	.ioctl = sock_no_ioctl,
1852 	.listen = vsock_listen,
1853 	.shutdown = vsock_shutdown,
1854 	.setsockopt = vsock_stream_setsockopt,
1855 	.getsockopt = vsock_stream_getsockopt,
1856 	.sendmsg = vsock_stream_sendmsg,
1857 	.recvmsg = vsock_stream_recvmsg,
1858 	.mmap = sock_no_mmap,
1859 	.sendpage = sock_no_sendpage,
1860 };
1861 
1862 static int vsock_create(struct net *net, struct socket *sock,
1863 			int protocol, int kern)
1864 {
1865 	if (!sock)
1866 		return -EINVAL;
1867 
1868 	if (protocol && protocol != PF_VSOCK)
1869 		return -EPROTONOSUPPORT;
1870 
1871 	switch (sock->type) {
1872 	case SOCK_DGRAM:
1873 		sock->ops = &vsock_dgram_ops;
1874 		break;
1875 	case SOCK_STREAM:
1876 		sock->ops = &vsock_stream_ops;
1877 		break;
1878 	default:
1879 		return -ESOCKTNOSUPPORT;
1880 	}
1881 
1882 	sock->state = SS_UNCONNECTED;
1883 
1884 	return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1885 }
1886 
1887 static const struct net_proto_family vsock_family_ops = {
1888 	.family = AF_VSOCK,
1889 	.create = vsock_create,
1890 	.owner = THIS_MODULE,
1891 };
1892 
1893 static long vsock_dev_do_ioctl(struct file *filp,
1894 			       unsigned int cmd, void __user *ptr)
1895 {
1896 	u32 __user *p = ptr;
1897 	int retval = 0;
1898 
1899 	switch (cmd) {
1900 	case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1901 		if (put_user(transport->get_local_cid(), p) != 0)
1902 			retval = -EFAULT;
1903 		break;
1904 
1905 	default:
1906 		pr_err("Unknown ioctl %d\n", cmd);
1907 		retval = -EINVAL;
1908 	}
1909 
1910 	return retval;
1911 }
1912 
1913 static long vsock_dev_ioctl(struct file *filp,
1914 			    unsigned int cmd, unsigned long arg)
1915 {
1916 	return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1917 }
1918 
1919 #ifdef CONFIG_COMPAT
1920 static long vsock_dev_compat_ioctl(struct file *filp,
1921 				   unsigned int cmd, unsigned long arg)
1922 {
1923 	return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1924 }
1925 #endif
1926 
1927 static const struct file_operations vsock_device_ops = {
1928 	.owner		= THIS_MODULE,
1929 	.unlocked_ioctl	= vsock_dev_ioctl,
1930 #ifdef CONFIG_COMPAT
1931 	.compat_ioctl	= vsock_dev_compat_ioctl,
1932 #endif
1933 	.open		= nonseekable_open,
1934 };
1935 
1936 static struct miscdevice vsock_device = {
1937 	.name		= "vsock",
1938 	.fops		= &vsock_device_ops,
1939 };
1940 
1941 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1942 {
1943 	int err = mutex_lock_interruptible(&vsock_register_mutex);
1944 
1945 	if (err)
1946 		return err;
1947 
1948 	if (transport) {
1949 		err = -EBUSY;
1950 		goto err_busy;
1951 	}
1952 
1953 	/* Transport must be the owner of the protocol so that it can't
1954 	 * unload while there are open sockets.
1955 	 */
1956 	vsock_proto.owner = owner;
1957 	transport = t;
1958 
1959 	vsock_device.minor = MISC_DYNAMIC_MINOR;
1960 	err = misc_register(&vsock_device);
1961 	if (err) {
1962 		pr_err("Failed to register misc device\n");
1963 		goto err_reset_transport;
1964 	}
1965 
1966 	err = proto_register(&vsock_proto, 1);	/* we want our slab */
1967 	if (err) {
1968 		pr_err("Cannot register vsock protocol\n");
1969 		goto err_deregister_misc;
1970 	}
1971 
1972 	err = sock_register(&vsock_family_ops);
1973 	if (err) {
1974 		pr_err("could not register af_vsock (%d) address family: %d\n",
1975 		       AF_VSOCK, err);
1976 		goto err_unregister_proto;
1977 	}
1978 
1979 	mutex_unlock(&vsock_register_mutex);
1980 	return 0;
1981 
1982 err_unregister_proto:
1983 	proto_unregister(&vsock_proto);
1984 err_deregister_misc:
1985 	misc_deregister(&vsock_device);
1986 err_reset_transport:
1987 	transport = NULL;
1988 err_busy:
1989 	mutex_unlock(&vsock_register_mutex);
1990 	return err;
1991 }
1992 EXPORT_SYMBOL_GPL(__vsock_core_init);
1993 
1994 void vsock_core_exit(void)
1995 {
1996 	mutex_lock(&vsock_register_mutex);
1997 
1998 	misc_deregister(&vsock_device);
1999 	sock_unregister(AF_VSOCK);
2000 	proto_unregister(&vsock_proto);
2001 
2002 	/* We do not want the assignment below re-ordered. */
2003 	mb();
2004 	transport = NULL;
2005 
2006 	mutex_unlock(&vsock_register_mutex);
2007 }
2008 EXPORT_SYMBOL_GPL(vsock_core_exit);
2009 
2010 const struct vsock_transport *vsock_core_get_transport(void)
2011 {
2012 	/* vsock_register_mutex not taken since only the transport uses this
2013 	 * function and only while registered.
2014 	 */
2015 	return transport;
2016 }
2017 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2018 
2019 static void __exit vsock_exit(void)
2020 {
2021 	/* Do nothing.  This function makes this module removable. */
2022 }
2023 
2024 module_init(vsock_init_tables);
2025 module_exit(vsock_exit);
2026 
2027 MODULE_AUTHOR("VMware, Inc.");
2028 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2029 MODULE_VERSION("1.0.2.0-k");
2030 MODULE_LICENSE("GPL v2");
2031