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/compat.h>
89	#include <linux/types.h>
90	#include <linux/bitops.h>
91	#include <linux/cred.h>
92	#include <linux/errqueue.h>
93	#include <linux/init.h>
94	#include <linux/io.h>
95	#include <linux/kernel.h>
96	#include <linux/sched/signal.h>
97	#include <linux/kmod.h>
98	#include <linux/list.h>
99	#include <linux/miscdevice.h>
100	#include <linux/module.h>
101	#include <linux/mutex.h>
102	#include <linux/net.h>
103	#include <linux/poll.h>
104	#include <linux/random.h>
105	#include <linux/skbuff.h>
106	#include <linux/smp.h>
107	#include <linux/socket.h>
108	#include <linux/stddef.h>
109	#include <linux/unistd.h>
110	#include <linux/wait.h>
111	#include <linux/workqueue.h>
112	#include <net/sock.h>
113	#include <net/af_vsock.h>
114	#include <uapi/linux/vm_sockets.h>
115
116	static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
117	static void vsock_sk_destruct(struct sock *sk);
118	static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
119
120	/* Protocol family. */
121	struct proto vsock_proto = {
122	.name = "AF_VSOCK",
123	.owner = THIS_MODULE,
124	.obj_size = sizeof(struct vsock_sock),
125	#ifdef CONFIG_BPF_SYSCALL
126	.psock_update_sk_prot = vsock_bpf_update_proto,
127	#endif
128	};
129
130	/* The default peer timeout indicates how long we will wait for a peer response
131	* to a control message.
132	*/
133	#define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
134
135	#define VSOCK_DEFAULT_BUFFER_SIZE (1024 * 256)
136	#define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
137	#define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
138
139	/* Transport used for host->guest communication */
140	static const struct vsock_transport *transport_h2g;
141	/* Transport used for guest->host communication */
142	static const struct vsock_transport *transport_g2h;
143	/* Transport used for DGRAM communication */
144	static const struct vsock_transport *transport_dgram;
145	/* Transport used for local communication */
146	static const struct vsock_transport *transport_local;
147	static DEFINE_MUTEX(vsock_register_mutex);
148
149	/**** UTILS ****/
150
151	/* Each bound VSocket is stored in the bind hash table and each connected
152	* VSocket is stored in the connected hash table.
153	*
154	* Unbound sockets are all put on the same list attached to the end of the hash
155	* table (vsock_unbound_sockets). Bound sockets are added to the hash table in
156	* the bucket that their local address hashes to (vsock_bound_sockets(addr)
157	* represents the list that addr hashes to).
158	*
159	* Specifically, we initialize the vsock_bind_table array to a size of
160	* VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
161	* vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
162	* vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
163	* mods with VSOCK_HASH_SIZE to ensure this.
164	*/
165	#define MAX_PORT_RETRIES 24
166
167	#define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE)
168	#define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
169	#define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
170
171	/* XXX This can probably be implemented in a better way. */
172	#define VSOCK_CONN_HASH(src, dst) \
173	(((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
174	#define vsock_connected_sockets(src, dst) \
175	(&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
176	#define vsock_connected_sockets_vsk(vsk) \
177	vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
178
179	struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
180	EXPORT_SYMBOL_GPL(vsock_bind_table);
181	struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
182	EXPORT_SYMBOL_GPL(vsock_connected_table);
183	DEFINE_SPINLOCK(vsock_table_lock);
184	EXPORT_SYMBOL_GPL(vsock_table_lock);
185
186	/* Autobind this socket to the local address if necessary. */
187	static int vsock_auto_bind(struct vsock_sock *vsk)
188	{
189	struct sock *sk = sk_vsock(vsk);
190	struct sockaddr_vm local_addr;
191
192	if (vsock_addr_bound(addr: &vsk->local_addr))
193	return 0;
194	vsock_addr_init(addr: &local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
195	return __vsock_bind(sk, addr: &local_addr);
196	}
197
198	static void vsock_init_tables(void)
199	{
200	int i;
201
202	for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
203	INIT_LIST_HEAD(list: &vsock_bind_table[i]);
204
205	for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
206	INIT_LIST_HEAD(list: &vsock_connected_table[i]);
207	}
208
209	static void __vsock_insert_bound(struct list_head *list,
210	struct vsock_sock *vsk)
211	{
212	sock_hold(sk: &vsk->sk);
213	list_add(new: &vsk->bound_table, head: list);
214	}
215
216	static void __vsock_insert_connected(struct list_head *list,
217	struct vsock_sock *vsk)
218	{
219	sock_hold(sk: &vsk->sk);
220	list_add(new: &vsk->connected_table, head: list);
221	}
222
223	static void __vsock_remove_bound(struct vsock_sock *vsk)
224	{
225	list_del_init(entry: &vsk->bound_table);
226	sock_put(sk: &vsk->sk);
227	}
228
229	static void __vsock_remove_connected(struct vsock_sock *vsk)
230	{
231	list_del_init(entry: &vsk->connected_table);
232	sock_put(sk: &vsk->sk);
233	}
234
235	static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
236	{
237	struct vsock_sock *vsk;
238
239	list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
240	if (vsock_addr_equals_addr(addr, other: &vsk->local_addr))
241	return sk_vsock(vsk);
242
243	if (addr->svm_port == vsk->local_addr.svm_port &&
244	(vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
245	addr->svm_cid == VMADDR_CID_ANY))
246	return sk_vsock(vsk);
247	}
248
249	return NULL;
250	}
251
252	static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
253	struct sockaddr_vm *dst)
254	{
255	struct vsock_sock *vsk;
256
257	list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
258	connected_table) {
259	if (vsock_addr_equals_addr(addr: src, other: &vsk->remote_addr) &&
260	dst->svm_port == vsk->local_addr.svm_port) {
261	return sk_vsock(vsk);
262	}
263	}
264
265	return NULL;
266	}
267
268	static void vsock_insert_unbound(struct vsock_sock *vsk)
269	{
270	spin_lock_bh(lock: &vsock_table_lock);
271	__vsock_insert_bound(vsock_unbound_sockets, vsk);
272	spin_unlock_bh(lock: &vsock_table_lock);
273	}
274
275	void vsock_insert_connected(struct vsock_sock *vsk)
276	{
277	struct list_head *list = vsock_connected_sockets(
278	&vsk->remote_addr, &vsk->local_addr);
279
280	spin_lock_bh(lock: &vsock_table_lock);
281	__vsock_insert_connected(list, vsk);
282	spin_unlock_bh(lock: &vsock_table_lock);
283	}
284	EXPORT_SYMBOL_GPL(vsock_insert_connected);
285
286	void vsock_remove_bound(struct vsock_sock *vsk)
287	{
288	spin_lock_bh(lock: &vsock_table_lock);
289	if (__vsock_in_bound_table(vsk))
290	__vsock_remove_bound(vsk);
291	spin_unlock_bh(lock: &vsock_table_lock);
292	}
293	EXPORT_SYMBOL_GPL(vsock_remove_bound);
294
295	void vsock_remove_connected(struct vsock_sock *vsk)
296	{
297	spin_lock_bh(lock: &vsock_table_lock);
298	if (__vsock_in_connected_table(vsk))
299	__vsock_remove_connected(vsk);
300	spin_unlock_bh(lock: &vsock_table_lock);
301	}
302	EXPORT_SYMBOL_GPL(vsock_remove_connected);
303
304	struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
305	{
306	struct sock *sk;
307
308	spin_lock_bh(lock: &vsock_table_lock);
309	sk = __vsock_find_bound_socket(addr);
310	if (sk)
311	sock_hold(sk);
312
313	spin_unlock_bh(lock: &vsock_table_lock);
314
315	return sk;
316	}
317	EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
318
319	struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
320	struct sockaddr_vm *dst)
321	{
322	struct sock *sk;
323
324	spin_lock_bh(lock: &vsock_table_lock);
325	sk = __vsock_find_connected_socket(src, dst);
326	if (sk)
327	sock_hold(sk);
328
329	spin_unlock_bh(lock: &vsock_table_lock);
330
331	return sk;
332	}
333	EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
334
335	void vsock_remove_sock(struct vsock_sock *vsk)
336	{
337	vsock_remove_bound(vsk);
338	vsock_remove_connected(vsk);
339	}
340	EXPORT_SYMBOL_GPL(vsock_remove_sock);
341
342	void vsock_for_each_connected_socket(struct vsock_transport *transport,
343	void (*fn)(struct sock *sk))
344	{
345	int i;
346
347	spin_lock_bh(lock: &vsock_table_lock);
348
349	for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
350	struct vsock_sock *vsk;
351	list_for_each_entry(vsk, &vsock_connected_table[i],
352	connected_table) {
353	if (vsk->transport != transport)
354	continue;
355
356	fn(sk_vsock(vsk));
357	}
358	}
359
360	spin_unlock_bh(lock: &vsock_table_lock);
361	}
362	EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
363
364	void vsock_add_pending(struct sock *listener, struct sock *pending)
365	{
366	struct vsock_sock *vlistener;
367	struct vsock_sock *vpending;
368
369	vlistener = vsock_sk(listener);
370	vpending = vsock_sk(pending);
371
372	sock_hold(sk: pending);
373	sock_hold(sk: listener);
374	list_add_tail(new: &vpending->pending_links, head: &vlistener->pending_links);
375	}
376	EXPORT_SYMBOL_GPL(vsock_add_pending);
377
378	void vsock_remove_pending(struct sock *listener, struct sock *pending)
379	{
380	struct vsock_sock *vpending = vsock_sk(pending);
381
382	list_del_init(entry: &vpending->pending_links);
383	sock_put(sk: listener);
384	sock_put(sk: pending);
385	}
386	EXPORT_SYMBOL_GPL(vsock_remove_pending);
387
388	void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
389	{
390	struct vsock_sock *vlistener;
391	struct vsock_sock *vconnected;
392
393	vlistener = vsock_sk(listener);
394	vconnected = vsock_sk(connected);
395
396	sock_hold(sk: connected);
397	sock_hold(sk: listener);
398	list_add_tail(new: &vconnected->accept_queue, head: &vlistener->accept_queue);
399	}
400	EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
401
402	static bool vsock_use_local_transport(unsigned int remote_cid)
403	{
404	if (!transport_local)
405	return false;
406
407	if (remote_cid == VMADDR_CID_LOCAL)
408	return true;
409
410	if (transport_g2h) {
411	return remote_cid == transport_g2h->get_local_cid();
412	} else {
413	return remote_cid == VMADDR_CID_HOST;
414	}
415	}
416
417	static void vsock_deassign_transport(struct vsock_sock *vsk)
418	{
419	if (!vsk->transport)
420	return;
421
422	vsk->transport->destruct(vsk);
423	module_put(module: vsk->transport->module);
424	vsk->transport = NULL;
425	}
426
427	/* Assign a transport to a socket and call the .init transport callback.
428	*
429	* Note: for connection oriented socket this must be called when vsk->remote_addr
430	* is set (e.g. during the connect() or when a connection request on a listener
431	* socket is received).
432	* The vsk->remote_addr is used to decide which transport to use:
433	* - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
434	* g2h is not loaded, will use local transport;
435	* - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
436	* includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
437	* - remote CID > VMADDR_CID_HOST will use host->guest transport;
438	*/
439	int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
440	{
441	const struct vsock_transport *new_transport;
442	struct sock *sk = sk_vsock(vsk);
443	unsigned int remote_cid = vsk->remote_addr.svm_cid;
444	__u8 remote_flags;
445	int ret;
446
447	/* If the packet is coming with the source and destination CIDs higher
448	* than VMADDR_CID_HOST, then a vsock channel where all the packets are
449	* forwarded to the host should be established. Then the host will
450	* need to forward the packets to the guest.
451	*
452	* The flag is set on the (listen) receive path (psk is not NULL). On
453	* the connect path the flag can be set by the user space application.
454	*/
455	if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
456	vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
457	vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
458
459	remote_flags = vsk->remote_addr.svm_flags;
460
461	switch (sk->sk_type) {
462	case SOCK_DGRAM:
463	new_transport = transport_dgram;
464	break;
465	case SOCK_STREAM:
466	case SOCK_SEQPACKET:
467	if (vsock_use_local_transport(remote_cid))
468	new_transport = transport_local;
469	else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
470	(remote_flags & VMADDR_FLAG_TO_HOST))
471	new_transport = transport_g2h;
472	else
473	new_transport = transport_h2g;
474	break;
475	default:
476	return -ESOCKTNOSUPPORT;
477	}
478
479	if (vsk->transport) {
480	if (vsk->transport == new_transport)
481	return 0;
482
483	/* transport->release() must be called with sock lock acquired.
484	* This path can only be taken during vsock_connect(), where we
485	* have already held the sock lock. In the other cases, this
486	* function is called on a new socket which is not assigned to
487	* any transport.
488	*/
489	vsk->transport->release(vsk);
490	vsock_deassign_transport(vsk);
491	}
492
493	/* We increase the module refcnt to prevent the transport unloading
494	* while there are open sockets assigned to it.
495	*/
496	if (!new_transport || !try_module_get(module: new_transport->module))
497	return -ENODEV;
498
499	if (sk->sk_type == SOCK_SEQPACKET) {
500	if (!new_transport->seqpacket_allow ||
501	!new_transport->seqpacket_allow(remote_cid)) {
502	module_put(module: new_transport->module);
503	return -ESOCKTNOSUPPORT;
504	}
505	}
506
507	ret = new_transport->init(vsk, psk);
508	if (ret) {
509	module_put(module: new_transport->module);
510	return ret;
511	}
512
513	vsk->transport = new_transport;
514
515	return 0;
516	}
517	EXPORT_SYMBOL_GPL(vsock_assign_transport);
518
519	bool vsock_find_cid(unsigned int cid)
520	{
521	if (transport_g2h && cid == transport_g2h->get_local_cid())
522	return true;
523
524	if (transport_h2g && cid == VMADDR_CID_HOST)
525	return true;
526
527	if (transport_local && cid == VMADDR_CID_LOCAL)
528	return true;
529
530	return false;
531	}
532	EXPORT_SYMBOL_GPL(vsock_find_cid);
533
534	static struct sock *vsock_dequeue_accept(struct sock *listener)
535	{
536	struct vsock_sock *vlistener;
537	struct vsock_sock *vconnected;
538
539	vlistener = vsock_sk(listener);
540
541	if (list_empty(head: &vlistener->accept_queue))
542	return NULL;
543
544	vconnected = list_entry(vlistener->accept_queue.next,
545	struct vsock_sock, accept_queue);
546
547	list_del_init(entry: &vconnected->accept_queue);
548	sock_put(sk: listener);
549	/* The caller will need a reference on the connected socket so we let
550	* it call sock_put().
551	*/
552
553	return sk_vsock(vconnected);
554	}
555
556	static bool vsock_is_accept_queue_empty(struct sock *sk)
557	{
558	struct vsock_sock *vsk = vsock_sk(sk);
559	return list_empty(head: &vsk->accept_queue);
560	}
561
562	static bool vsock_is_pending(struct sock *sk)
563	{
564	struct vsock_sock *vsk = vsock_sk(sk);
565	return !list_empty(head: &vsk->pending_links);
566	}
567
568	static int vsock_send_shutdown(struct sock *sk, int mode)
569	{
570	struct vsock_sock *vsk = vsock_sk(sk);
571
572	if (!vsk->transport)
573	return -ENODEV;
574
575	return vsk->transport->shutdown(vsk, mode);
576	}
577
578	static void vsock_pending_work(struct work_struct *work)
579	{
580	struct sock *sk;
581	struct sock *listener;
582	struct vsock_sock *vsk;
583	bool cleanup;
584
585	vsk = container_of(work, struct vsock_sock, pending_work.work);
586	sk = sk_vsock(vsk);
587	listener = vsk->listener;
588	cleanup = true;
589
590	lock_sock(sk: listener);
591	lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
592
593	if (vsock_is_pending(sk)) {
594	vsock_remove_pending(listener, sk);
595
596	sk_acceptq_removed(sk: listener);
597	} else if (!vsk->rejected) {
598	/* We are not on the pending list and accept() did not reject
599	* us, so we must have been accepted by our user process. We
600	* just need to drop our references to the sockets and be on
601	* our way.
602	*/
603	cleanup = false;
604	goto out;
605	}
606
607	/* We need to remove ourself from the global connected sockets list so
608	* incoming packets can't find this socket, and to reduce the reference
609	* count.
610	*/
611	vsock_remove_connected(vsk);
612
613	sk->sk_state = TCP_CLOSE;
614
615	out:
616	release_sock(sk);
617	release_sock(sk: listener);
618	if (cleanup)
619	sock_put(sk);
620
621	sock_put(sk);
622	sock_put(sk: listener);
623	}
624
625	/**** SOCKET OPERATIONS ****/
626
627	static int __vsock_bind_connectible(struct vsock_sock *vsk,
628	struct sockaddr_vm *addr)
629	{
630	static u32 port;
631	struct sockaddr_vm new_addr;
632
633	if (!port)
634	port = get_random_u32_above(LAST_RESERVED_PORT);
635
636	vsock_addr_init(addr: &new_addr, cid: addr->svm_cid, port: addr->svm_port);
637
638	if (addr->svm_port == VMADDR_PORT_ANY) {
639	bool found = false;
640	unsigned int i;
641
642	for (i = 0; i < MAX_PORT_RETRIES; i++) {
643	if (port <= LAST_RESERVED_PORT)
644	port = LAST_RESERVED_PORT + 1;
645
646	new_addr.svm_port = port++;
647
648	if (!__vsock_find_bound_socket(addr: &new_addr)) {
649	found = true;
650	break;
651	}
652	}
653
654	if (!found)
655	return -EADDRNOTAVAIL;
656	} else {
657	/* If port is in reserved range, ensure caller
658	* has necessary privileges.
659	*/
660	if (addr->svm_port <= LAST_RESERVED_PORT &&
661	!capable(CAP_NET_BIND_SERVICE)) {
662	return -EACCES;
663	}
664
665	if (__vsock_find_bound_socket(addr: &new_addr))
666	return -EADDRINUSE;
667	}
668
669	vsock_addr_init(addr: &vsk->local_addr, cid: new_addr.svm_cid, port: new_addr.svm_port);
670
671	/* Remove connection oriented sockets from the unbound list and add them
672	* to the hash table for easy lookup by its address. The unbound list
673	* is simply an extra entry at the end of the hash table, a trick used
674	* by AF_UNIX.
675	*/
676	__vsock_remove_bound(vsk);
677	__vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
678
679	return 0;
680	}
681
682	static int __vsock_bind_dgram(struct vsock_sock *vsk,
683	struct sockaddr_vm *addr)
684	{
685	return vsk->transport->dgram_bind(vsk, addr);
686	}
687
688	static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
689	{
690	struct vsock_sock *vsk = vsock_sk(sk);
691	int retval;
692
693	/* First ensure this socket isn't already bound. */
694	if (vsock_addr_bound(addr: &vsk->local_addr))
695	return -EINVAL;
696
697	/* Now bind to the provided address or select appropriate values if
698	* none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
699	* like AF_INET prevents binding to a non-local IP address (in most
700	* cases), we only allow binding to a local CID.
701	*/
702	if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
703	return -EADDRNOTAVAIL;
704
705	switch (sk->sk_socket->type) {
706	case SOCK_STREAM:
707	case SOCK_SEQPACKET:
708	spin_lock_bh(lock: &vsock_table_lock);
709	retval = __vsock_bind_connectible(vsk, addr);
710	spin_unlock_bh(lock: &vsock_table_lock);
711	break;
712
713	case SOCK_DGRAM:
714	retval = __vsock_bind_dgram(vsk, addr);
715	break;
716
717	default:
718	retval = -EINVAL;
719	break;
720	}
721
722	return retval;
723	}
724
725	static void vsock_connect_timeout(struct work_struct *work);
726
727	static struct sock *__vsock_create(struct net *net,
728	struct socket *sock,
729	struct sock *parent,
730	gfp_t priority,
731	unsigned short type,
732	int kern)
733	{
734	struct sock *sk;
735	struct vsock_sock *psk;
736	struct vsock_sock *vsk;
737
738	sk = sk_alloc(net, AF_VSOCK, priority, prot: &vsock_proto, kern);
739	if (!sk)
740	return NULL;
741
742	sock_init_data(sock, sk);
743
744	/* sk->sk_type is normally set in sock_init_data, but only if sock is
745	* non-NULL. We make sure that our sockets always have a type by
746	* setting it here if needed.
747	*/
748	if (!sock)
749	sk->sk_type = type;
750
751	vsk = vsock_sk(sk);
752	vsock_addr_init(addr: &vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
753	vsock_addr_init(addr: &vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
754
755	sk->sk_destruct = vsock_sk_destruct;
756	sk->sk_backlog_rcv = vsock_queue_rcv_skb;
757	sock_reset_flag(sk, flag: SOCK_DONE);
758
759	INIT_LIST_HEAD(list: &vsk->bound_table);
760	INIT_LIST_HEAD(list: &vsk->connected_table);
761	vsk->listener = NULL;
762	INIT_LIST_HEAD(list: &vsk->pending_links);
763	INIT_LIST_HEAD(list: &vsk->accept_queue);
764	vsk->rejected = false;
765	vsk->sent_request = false;
766	vsk->ignore_connecting_rst = false;
767	vsk->peer_shutdown = 0;
768	INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
769	INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
770
771	psk = parent ? vsock_sk(parent) : NULL;
772	if (parent) {
773	vsk->trusted = psk->trusted;
774	vsk->owner = get_cred(cred: psk->owner);
775	vsk->connect_timeout = psk->connect_timeout;
776	vsk->buffer_size = psk->buffer_size;
777	vsk->buffer_min_size = psk->buffer_min_size;
778	vsk->buffer_max_size = psk->buffer_max_size;
779	security_sk_clone(sk: parent, newsk: sk);
780	} else {
781	vsk->trusted = ns_capable_noaudit(ns: &init_user_ns, CAP_NET_ADMIN);
782	vsk->owner = get_current_cred();
783	vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
784	vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
785	vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
786	vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
787	}
788
789	return sk;
790	}
791
792	static bool sock_type_connectible(u16 type)
793	{
794	return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
795	}
796
797	static void __vsock_release(struct sock *sk, int level)
798	{
799	if (sk) {
800	struct sock *pending;
801	struct vsock_sock *vsk;
802
803	vsk = vsock_sk(sk);
804	pending = NULL; /* Compiler warning. */
805
806	/* When "level" is SINGLE_DEPTH_NESTING, use the nested
807	* version to avoid the warning "possible recursive locking
808	* detected". When "level" is 0, lock_sock_nested(sk, level)
809	* is the same as lock_sock(sk).
810	*/
811	lock_sock_nested(sk, subclass: level);
812
813	if (vsk->transport)
814	vsk->transport->release(vsk);
815	else if (sock_type_connectible(type: sk->sk_type))
816	vsock_remove_sock(vsk);
817
818	sock_orphan(sk);
819	sk->sk_shutdown = SHUTDOWN_MASK;
820
821	skb_queue_purge(list: &sk->sk_receive_queue);
822
823	/* Clean up any sockets that never were accepted. */
824	while ((pending = vsock_dequeue_accept(listener: sk)) != NULL) {
825	__vsock_release(sk: pending, SINGLE_DEPTH_NESTING);
826	sock_put(sk: pending);
827	}
828
829	release_sock(sk);
830	sock_put(sk);
831	}
832	}
833
834	static void vsock_sk_destruct(struct sock *sk)
835	{
836	struct vsock_sock *vsk = vsock_sk(sk);
837
838	vsock_deassign_transport(vsk);
839
840	/* When clearing these addresses, there's no need to set the family and
841	* possibly register the address family with the kernel.
842	*/
843	vsock_addr_init(addr: &vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
844	vsock_addr_init(addr: &vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
845
846	put_cred(cred: vsk->owner);
847	}
848
849	static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
850	{
851	int err;
852
853	err = sock_queue_rcv_skb(sk, skb);
854	if (err)
855	kfree_skb(skb);
856
857	return err;
858	}
859
860	struct sock *vsock_create_connected(struct sock *parent)
861	{
862	return __vsock_create(net: sock_net(sk: parent), NULL, parent, GFP_KERNEL,
863	type: parent->sk_type, kern: 0);
864	}
865	EXPORT_SYMBOL_GPL(vsock_create_connected);
866
867	s64 vsock_stream_has_data(struct vsock_sock *vsk)
868	{
869	return vsk->transport->stream_has_data(vsk);
870	}
871	EXPORT_SYMBOL_GPL(vsock_stream_has_data);
872
873	s64 vsock_connectible_has_data(struct vsock_sock *vsk)
874	{
875	struct sock *sk = sk_vsock(vsk);
876
877	if (sk->sk_type == SOCK_SEQPACKET)
878	return vsk->transport->seqpacket_has_data(vsk);
879	else
880	return vsock_stream_has_data(vsk);
881	}
882	EXPORT_SYMBOL_GPL(vsock_connectible_has_data);
883
884	s64 vsock_stream_has_space(struct vsock_sock *vsk)
885	{
886	return vsk->transport->stream_has_space(vsk);
887	}
888	EXPORT_SYMBOL_GPL(vsock_stream_has_space);
889
890	void vsock_data_ready(struct sock *sk)
891	{
892	struct vsock_sock *vsk = vsock_sk(sk);
893
894	if (vsock_stream_has_data(vsk) >= sk->sk_rcvlowat ||
895	sock_flag(sk, flag: SOCK_DONE))
896	sk->sk_data_ready(sk);
897	}
898	EXPORT_SYMBOL_GPL(vsock_data_ready);
899
900	static int vsock_release(struct socket *sock)
901	{
902	__vsock_release(sk: sock->sk, level: 0);
903	sock->sk = NULL;
904	sock->state = SS_FREE;
905
906	return 0;
907	}
908
909	static int
910	vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
911	{
912	int err;
913	struct sock *sk;
914	struct sockaddr_vm *vm_addr;
915
916	sk = sock->sk;
917
918	if (vsock_addr_cast(addr, len: addr_len, out_addr: &vm_addr) != 0)
919	return -EINVAL;
920
921	lock_sock(sk);
922	err = __vsock_bind(sk, addr: vm_addr);
923	release_sock(sk);
924
925	return err;
926	}
927
928	static int vsock_getname(struct socket *sock,
929	struct sockaddr *addr, int peer)
930	{
931	int err;
932	struct sock *sk;
933	struct vsock_sock *vsk;
934	struct sockaddr_vm *vm_addr;
935
936	sk = sock->sk;
937	vsk = vsock_sk(sk);
938	err = 0;
939
940	lock_sock(sk);
941
942	if (peer) {
943	if (sock->state != SS_CONNECTED) {
944	err = -ENOTCONN;
945	goto out;
946	}
947	vm_addr = &vsk->remote_addr;
948	} else {
949	vm_addr = &vsk->local_addr;
950	}
951
952	if (!vm_addr) {
953	err = -EINVAL;
954	goto out;
955	}
956
957	/* sys_getsockname() and sys_getpeername() pass us a
958	* MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
959	* that macro is defined in socket.c instead of .h, so we hardcode its
960	* value here.
961	*/
962	BUILD_BUG_ON(sizeof(*vm_addr) > 128);
963	memcpy(addr, vm_addr, sizeof(*vm_addr));
964	err = sizeof(*vm_addr);
965
966	out:
967	release_sock(sk);
968	return err;
969	}
970
971	static int vsock_shutdown(struct socket *sock, int mode)
972	{
973	int err;
974	struct sock *sk;
975
976	/* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
977	* RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
978	* here like the other address families do. Note also that the
979	* increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
980	* which is what we want.
981	*/
982	mode++;
983
984	if ((mode & ~SHUTDOWN_MASK) || !mode)
985	return -EINVAL;
986
987	/* If this is a connection oriented socket and it is not connected then
988	* bail out immediately. If it is a DGRAM socket then we must first
989	* kick the socket so that it wakes up from any sleeping calls, for
990	* example recv(), and then afterwards return the error.
991	*/
992
993	sk = sock->sk;
994
995	lock_sock(sk);
996	if (sock->state == SS_UNCONNECTED) {
997	err = -ENOTCONN;
998	if (sock_type_connectible(type: sk->sk_type))
999	goto out;
1000	} else {
1001	sock->state = SS_DISCONNECTING;
1002	err = 0;
1003	}
1004
1005	/* Receive and send shutdowns are treated alike. */
1006	mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
1007	if (mode) {
1008	sk->sk_shutdown |= mode;
1009	sk->sk_state_change(sk);
1010
1011	if (sock_type_connectible(type: sk->sk_type)) {
1012	sock_reset_flag(sk, flag: SOCK_DONE);
1013	vsock_send_shutdown(sk, mode);
1014	}
1015	}
1016
1017	out:
1018	release_sock(sk);
1019	return err;
1020	}
1021
1022	static __poll_t vsock_poll(struct file *file, struct socket *sock,
1023	poll_table *wait)
1024	{
1025	struct sock *sk;
1026	__poll_t mask;
1027	struct vsock_sock *vsk;
1028
1029	sk = sock->sk;
1030	vsk = vsock_sk(sk);
1031
1032	poll_wait(filp: file, wait_address: sk_sleep(sk), p: wait);
1033	mask = 0;
1034
1035	if (sk->sk_err || !skb_queue_empty_lockless(list: &sk->sk_error_queue))
1036	/* Signify that there has been an error on this socket. */
1037	mask |= EPOLLERR;
1038
1039	/* INET sockets treat local write shutdown and peer write shutdown as a
1040	* case of EPOLLHUP set.
1041	*/
1042	if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1043	((sk->sk_shutdown & SEND_SHUTDOWN) &&
1044	(vsk->peer_shutdown & SEND_SHUTDOWN))) {
1045	mask |= EPOLLHUP;
1046	}
1047
1048	if (sk->sk_shutdown & RCV_SHUTDOWN ||
1049	vsk->peer_shutdown & SEND_SHUTDOWN) {
1050	mask |= EPOLLRDHUP;
1051	}
1052
1053	if (sock->type == SOCK_DGRAM) {
1054	/* For datagram sockets we can read if there is something in
1055	* the queue and write as long as the socket isn't shutdown for
1056	* sending.
1057	*/
1058	if (!skb_queue_empty_lockless(list: &sk->sk_receive_queue) ||
1059	(sk->sk_shutdown & RCV_SHUTDOWN)) {
1060	mask |= EPOLLIN | EPOLLRDNORM;
1061	}
1062
1063	if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1064	mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1065
1066	} else if (sock_type_connectible(type: sk->sk_type)) {
1067	const struct vsock_transport *transport;
1068
1069	lock_sock(sk);
1070
1071	transport = vsk->transport;
1072
1073	/* Listening sockets that have connections in their accept
1074	* queue can be read.
1075	*/
1076	if (sk->sk_state == TCP_LISTEN
1077	&& !vsock_is_accept_queue_empty(sk))
1078	mask |= EPOLLIN | EPOLLRDNORM;
1079
1080	/* If there is something in the queue then we can read. */
1081	if (transport && transport->stream_is_active(vsk) &&
1082	!(sk->sk_shutdown & RCV_SHUTDOWN)) {
1083	bool data_ready_now = false;
1084	int target = sock_rcvlowat(sk, waitall: 0, INT_MAX);
1085	int ret = transport->notify_poll_in(
1086	vsk, target, &data_ready_now);
1087	if (ret < 0) {
1088	mask |= EPOLLERR;
1089	} else {
1090	if (data_ready_now)
1091	mask |= EPOLLIN | EPOLLRDNORM;
1092
1093	}
1094	}
1095
1096	/* Sockets whose connections have been closed, reset, or
1097	* terminated should also be considered read, and we check the
1098	* shutdown flag for that.
1099	*/
1100	if (sk->sk_shutdown & RCV_SHUTDOWN ||
1101	vsk->peer_shutdown & SEND_SHUTDOWN) {
1102	mask |= EPOLLIN | EPOLLRDNORM;
1103	}
1104
1105	/* Connected sockets that can produce data can be written. */
1106	if (transport && sk->sk_state == TCP_ESTABLISHED) {
1107	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1108	bool space_avail_now = false;
1109	int ret = transport->notify_poll_out(
1110	vsk, 1, &space_avail_now);
1111	if (ret < 0) {
1112	mask |= EPOLLERR;
1113	} else {
1114	if (space_avail_now)
1115	/* Remove EPOLLWRBAND since INET
1116	* sockets are not setting it.
1117	*/
1118	mask |= EPOLLOUT | EPOLLWRNORM;
1119
1120	}
1121	}
1122	}
1123
1124	/* Simulate INET socket poll behaviors, which sets
1125	* EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1126	* but local send is not shutdown.
1127	*/
1128	if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1129	if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1130	mask |= EPOLLOUT | EPOLLWRNORM;
1131
1132	}
1133
1134	release_sock(sk);
1135	}
1136
1137	return mask;
1138	}
1139
1140	static int vsock_read_skb(struct sock *sk, skb_read_actor_t read_actor)
1141	{
1142	struct vsock_sock *vsk = vsock_sk(sk);
1143
1144	return vsk->transport->read_skb(vsk, read_actor);
1145	}
1146
1147	static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1148	size_t len)
1149	{
1150	int err;
1151	struct sock *sk;
1152	struct vsock_sock *vsk;
1153	struct sockaddr_vm *remote_addr;
1154	const struct vsock_transport *transport;
1155
1156	if (msg->msg_flags & MSG_OOB)
1157	return -EOPNOTSUPP;
1158
1159	/* For now, MSG_DONTWAIT is always assumed... */
1160	err = 0;
1161	sk = sock->sk;
1162	vsk = vsock_sk(sk);
1163
1164	lock_sock(sk);
1165
1166	transport = vsk->transport;
1167
1168	err = vsock_auto_bind(vsk);
1169	if (err)
1170	goto out;
1171
1172
1173	/* If the provided message contains an address, use that. Otherwise
1174	* fall back on the socket's remote handle (if it has been connected).
1175	*/
1176	if (msg->msg_name &&
1177	vsock_addr_cast(addr: msg->msg_name, len: msg->msg_namelen,
1178	out_addr: &remote_addr) == 0) {
1179	/* Ensure this address is of the right type and is a valid
1180	* destination.
1181	*/
1182
1183	if (remote_addr->svm_cid == VMADDR_CID_ANY)
1184	remote_addr->svm_cid = transport->get_local_cid();
1185
1186	if (!vsock_addr_bound(addr: remote_addr)) {
1187	err = -EINVAL;
1188	goto out;
1189	}
1190	} else if (sock->state == SS_CONNECTED) {
1191	remote_addr = &vsk->remote_addr;
1192
1193	if (remote_addr->svm_cid == VMADDR_CID_ANY)
1194	remote_addr->svm_cid = transport->get_local_cid();
1195
1196	/* XXX Should connect() or this function ensure remote_addr is
1197	* bound?
1198	*/
1199	if (!vsock_addr_bound(addr: &vsk->remote_addr)) {
1200	err = -EINVAL;
1201	goto out;
1202	}
1203	} else {
1204	err = -EINVAL;
1205	goto out;
1206	}
1207
1208	if (!transport->dgram_allow(remote_addr->svm_cid,
1209	remote_addr->svm_port)) {
1210	err = -EINVAL;
1211	goto out;
1212	}
1213
1214	err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1215
1216	out:
1217	release_sock(sk);
1218	return err;
1219	}
1220
1221	static int vsock_dgram_connect(struct socket *sock,
1222	struct sockaddr *addr, int addr_len, int flags)
1223	{
1224	int err;
1225	struct sock *sk;
1226	struct vsock_sock *vsk;
1227	struct sockaddr_vm *remote_addr;
1228
1229	sk = sock->sk;
1230	vsk = vsock_sk(sk);
1231
1232	err = vsock_addr_cast(addr, len: addr_len, out_addr: &remote_addr);
1233	if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1234	lock_sock(sk);
1235	vsock_addr_init(addr: &vsk->remote_addr, VMADDR_CID_ANY,
1236	VMADDR_PORT_ANY);
1237	sock->state = SS_UNCONNECTED;
1238	release_sock(sk);
1239	return 0;
1240	} else if (err != 0)
1241	return -EINVAL;
1242
1243	lock_sock(sk);
1244
1245	err = vsock_auto_bind(vsk);
1246	if (err)
1247	goto out;
1248
1249	if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1250	remote_addr->svm_port)) {
1251	err = -EINVAL;
1252	goto out;
1253	}
1254
1255	memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1256	sock->state = SS_CONNECTED;
1257
1258	/* sock map disallows redirection of non-TCP sockets with sk_state !=
1259	* TCP_ESTABLISHED (see sock_map_redirect_allowed()), so we set
1260	* TCP_ESTABLISHED here to allow redirection of connected vsock dgrams.
1261	*
1262	* This doesn't seem to be abnormal state for datagram sockets, as the
1263	* same approach can be see in other datagram socket types as well
1264	* (such as unix sockets).
1265	*/
1266	sk->sk_state = TCP_ESTABLISHED;
1267
1268	out:
1269	release_sock(sk);
1270	return err;
1271	}
1272
1273	int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1274	size_t len, int flags)
1275	{
1276	#ifdef CONFIG_BPF_SYSCALL
1277	const struct proto *prot;
1278	#endif
1279	struct vsock_sock *vsk;
1280	struct sock *sk;
1281
1282	sk = sock->sk;
1283	vsk = vsock_sk(sk);
1284
1285	#ifdef CONFIG_BPF_SYSCALL
1286	prot = READ_ONCE(sk->sk_prot);
1287	if (prot != &vsock_proto)
1288	return prot->recvmsg(sk, msg, len, flags, NULL);
1289	#endif
1290
1291	return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1292	}
1293	EXPORT_SYMBOL_GPL(vsock_dgram_recvmsg);
1294
1295	static const struct proto_ops vsock_dgram_ops = {
1296	.family = PF_VSOCK,
1297	.owner = THIS_MODULE,
1298	.release = vsock_release,
1299	.bind = vsock_bind,
1300	.connect = vsock_dgram_connect,
1301	.socketpair = sock_no_socketpair,
1302	.accept = sock_no_accept,
1303	.getname = vsock_getname,
1304	.poll = vsock_poll,
1305	.ioctl = sock_no_ioctl,
1306	.listen = sock_no_listen,
1307	.shutdown = vsock_shutdown,
1308	.sendmsg = vsock_dgram_sendmsg,
1309	.recvmsg = vsock_dgram_recvmsg,
1310	.mmap = sock_no_mmap,
1311	.read_skb = vsock_read_skb,
1312	};
1313
1314	static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1315	{
1316	const struct vsock_transport *transport = vsk->transport;
1317
1318	if (!transport || !transport->cancel_pkt)
1319	return -EOPNOTSUPP;
1320
1321	return transport->cancel_pkt(vsk);
1322	}
1323
1324	static void vsock_connect_timeout(struct work_struct *work)
1325	{
1326	struct sock *sk;
1327	struct vsock_sock *vsk;
1328
1329	vsk = container_of(work, struct vsock_sock, connect_work.work);
1330	sk = sk_vsock(vsk);
1331
1332	lock_sock(sk);
1333	if (sk->sk_state == TCP_SYN_SENT &&
1334	(sk->sk_shutdown != SHUTDOWN_MASK)) {
1335	sk->sk_state = TCP_CLOSE;
1336	sk->sk_socket->state = SS_UNCONNECTED;
1337	sk->sk_err = ETIMEDOUT;
1338	sk_error_report(sk);
1339	vsock_transport_cancel_pkt(vsk);
1340	}
1341	release_sock(sk);
1342
1343	sock_put(sk);
1344	}
1345
1346	static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1347	int addr_len, int flags)
1348	{
1349	int err;
1350	struct sock *sk;
1351	struct vsock_sock *vsk;
1352	const struct vsock_transport *transport;
1353	struct sockaddr_vm *remote_addr;
1354	long timeout;
1355	DEFINE_WAIT(wait);
1356
1357	err = 0;
1358	sk = sock->sk;
1359	vsk = vsock_sk(sk);
1360
1361	lock_sock(sk);
1362
1363	/* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1364	switch (sock->state) {
1365	case SS_CONNECTED:
1366	err = -EISCONN;
1367	goto out;
1368	case SS_DISCONNECTING:
1369	err = -EINVAL;
1370	goto out;
1371	case SS_CONNECTING:
1372	/* This continues on so we can move sock into the SS_CONNECTED
1373	* state once the connection has completed (at which point err
1374	* will be set to zero also). Otherwise, we will either wait
1375	* for the connection or return -EALREADY should this be a
1376	* non-blocking call.
1377	*/
1378	err = -EALREADY;
1379	if (flags & O_NONBLOCK)
1380	goto out;
1381	break;
1382	default:
1383	if ((sk->sk_state == TCP_LISTEN) ||
1384	vsock_addr_cast(addr, len: addr_len, out_addr: &remote_addr) != 0) {
1385	err = -EINVAL;
1386	goto out;
1387	}
1388
1389	/* Set the remote address that we are connecting to. */
1390	memcpy(&vsk->remote_addr, remote_addr,
1391	sizeof(vsk->remote_addr));
1392
1393	err = vsock_assign_transport(vsk, NULL);
1394	if (err)
1395	goto out;
1396
1397	transport = vsk->transport;
1398
1399	/* The hypervisor and well-known contexts do not have socket
1400	* endpoints.
1401	*/
1402	if (!transport ||
1403	!transport->stream_allow(remote_addr->svm_cid,
1404	remote_addr->svm_port)) {
1405	err = -ENETUNREACH;
1406	goto out;
1407	}
1408
1409	if (vsock_msgzerocopy_allow(t: transport)) {
1410	set_bit(SOCK_SUPPORT_ZC, addr: &sk->sk_socket->flags);
1411	} else if (sock_flag(sk, flag: SOCK_ZEROCOPY)) {
1412	/* If this option was set before 'connect()',
1413	* when transport was unknown, check that this
1414	* feature is supported here.
1415	*/
1416	err = -EOPNOTSUPP;
1417	goto out;
1418	}
1419
1420	err = vsock_auto_bind(vsk);
1421	if (err)
1422	goto out;
1423
1424	sk->sk_state = TCP_SYN_SENT;
1425
1426	err = transport->connect(vsk);
1427	if (err < 0)
1428	goto out;
1429
1430	/* Mark sock as connecting and set the error code to in
1431	* progress in case this is a non-blocking connect.
1432	*/
1433	sock->state = SS_CONNECTING;
1434	err = -EINPROGRESS;
1435	}
1436
1437	/* The receive path will handle all communication until we are able to
1438	* enter the connected state. Here we wait for the connection to be
1439	* completed or a notification of an error.
1440	*/
1441	timeout = vsk->connect_timeout;
1442	prepare_to_wait(wq_head: sk_sleep(sk), wq_entry: &wait, TASK_INTERRUPTIBLE);
1443
1444	while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1445	if (flags & O_NONBLOCK) {
1446	/* If we're not going to block, we schedule a timeout
1447	* function to generate a timeout on the connection
1448	* attempt, in case the peer doesn't respond in a
1449	* timely manner. We hold on to the socket until the
1450	* timeout fires.
1451	*/
1452	sock_hold(sk);
1453
1454	/* If the timeout function is already scheduled,
1455	* reschedule it, then ungrab the socket refcount to
1456	* keep it balanced.
1457	*/
1458	if (mod_delayed_work(wq: system_wq, dwork: &vsk->connect_work,
1459	delay: timeout))
1460	sock_put(sk);
1461
1462	/* Skip ahead to preserve error code set above. */
1463	goto out_wait;
1464	}
1465
1466	release_sock(sk);
1467	timeout = schedule_timeout(timeout);
1468	lock_sock(sk);
1469
1470	if (signal_pending(current)) {
1471	err = sock_intr_errno(timeo: timeout);
1472	sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1473	sock->state = SS_UNCONNECTED;
1474	vsock_transport_cancel_pkt(vsk);
1475	vsock_remove_connected(vsk);
1476	goto out_wait;
1477	} else if ((sk->sk_state != TCP_ESTABLISHED) && (timeout == 0)) {
1478	err = -ETIMEDOUT;
1479	sk->sk_state = TCP_CLOSE;
1480	sock->state = SS_UNCONNECTED;
1481	vsock_transport_cancel_pkt(vsk);
1482	goto out_wait;
1483	}
1484
1485	prepare_to_wait(wq_head: sk_sleep(sk), wq_entry: &wait, TASK_INTERRUPTIBLE);
1486	}
1487
1488	if (sk->sk_err) {
1489	err = -sk->sk_err;
1490	sk->sk_state = TCP_CLOSE;
1491	sock->state = SS_UNCONNECTED;
1492	} else {
1493	err = 0;
1494	}
1495
1496	out_wait:
1497	finish_wait(wq_head: sk_sleep(sk), wq_entry: &wait);
1498	out:
1499	release_sock(sk);
1500	return err;
1501	}
1502
1503	static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1504	bool kern)
1505	{
1506	struct sock *listener;
1507	int err;
1508	struct sock *connected;
1509	struct vsock_sock *vconnected;
1510	long timeout;
1511	DEFINE_WAIT(wait);
1512
1513	err = 0;
1514	listener = sock->sk;
1515
1516	lock_sock(sk: listener);
1517
1518	if (!sock_type_connectible(type: sock->type)) {
1519	err = -EOPNOTSUPP;
1520	goto out;
1521	}
1522
1523	if (listener->sk_state != TCP_LISTEN) {
1524	err = -EINVAL;
1525	goto out;
1526	}
1527
1528	/* Wait for children sockets to appear; these are the new sockets
1529	* created upon connection establishment.
1530	*/
1531	timeout = sock_rcvtimeo(sk: listener, noblock: flags & O_NONBLOCK);
1532	prepare_to_wait(wq_head: sk_sleep(sk: listener), wq_entry: &wait, TASK_INTERRUPTIBLE);
1533
1534	while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1535	listener->sk_err == 0) {
1536	release_sock(sk: listener);
1537	timeout = schedule_timeout(timeout);
1538	finish_wait(wq_head: sk_sleep(sk: listener), wq_entry: &wait);
1539	lock_sock(sk: listener);
1540
1541	if (signal_pending(current)) {
1542	err = sock_intr_errno(timeo: timeout);
1543	goto out;
1544	} else if (timeout == 0) {
1545	err = -EAGAIN;
1546	goto out;
1547	}
1548
1549	prepare_to_wait(wq_head: sk_sleep(sk: listener), wq_entry: &wait, TASK_INTERRUPTIBLE);
1550	}
1551	finish_wait(wq_head: sk_sleep(sk: listener), wq_entry: &wait);
1552
1553	if (listener->sk_err)
1554	err = -listener->sk_err;
1555
1556	if (connected) {
1557	sk_acceptq_removed(sk: listener);
1558
1559	lock_sock_nested(sk: connected, SINGLE_DEPTH_NESTING);
1560	vconnected = vsock_sk(connected);
1561
1562	/* If the listener socket has received an error, then we should
1563	* reject this socket and return. Note that we simply mark the
1564	* socket rejected, drop our reference, and let the cleanup
1565	* function handle the cleanup; the fact that we found it in
1566	* the listener's accept queue guarantees that the cleanup
1567	* function hasn't run yet.
1568	*/
1569	if (err) {
1570	vconnected->rejected = true;
1571	} else {
1572	newsock->state = SS_CONNECTED;
1573	sock_graft(sk: connected, parent: newsock);
1574	if (vsock_msgzerocopy_allow(t: vconnected->transport))
1575	set_bit(SOCK_SUPPORT_ZC,
1576	addr: &connected->sk_socket->flags);
1577	}
1578
1579	release_sock(sk: connected);
1580	sock_put(sk: connected);
1581	}
1582
1583	out:
1584	release_sock(sk: listener);
1585	return err;
1586	}
1587
1588	static int vsock_listen(struct socket *sock, int backlog)
1589	{
1590	int err;
1591	struct sock *sk;
1592	struct vsock_sock *vsk;
1593
1594	sk = sock->sk;
1595
1596	lock_sock(sk);
1597
1598	if (!sock_type_connectible(type: sk->sk_type)) {
1599	err = -EOPNOTSUPP;
1600	goto out;
1601	}
1602
1603	if (sock->state != SS_UNCONNECTED) {
1604	err = -EINVAL;
1605	goto out;
1606	}
1607
1608	vsk = vsock_sk(sk);
1609
1610	if (!vsock_addr_bound(addr: &vsk->local_addr)) {
1611	err = -EINVAL;
1612	goto out;
1613	}
1614
1615	sk->sk_max_ack_backlog = backlog;
1616	sk->sk_state = TCP_LISTEN;
1617
1618	err = 0;
1619
1620	out:
1621	release_sock(sk);
1622	return err;
1623	}
1624
1625	static void vsock_update_buffer_size(struct vsock_sock *vsk,
1626	const struct vsock_transport *transport,
1627	u64 val)
1628	{
1629	if (val > vsk->buffer_max_size)
1630	val = vsk->buffer_max_size;
1631
1632	if (val < vsk->buffer_min_size)
1633	val = vsk->buffer_min_size;
1634
1635	if (val != vsk->buffer_size &&
1636	transport && transport->notify_buffer_size)
1637	transport->notify_buffer_size(vsk, &val);
1638
1639	vsk->buffer_size = val;
1640	}
1641
1642	static int vsock_connectible_setsockopt(struct socket *sock,
1643	int level,
1644	int optname,
1645	sockptr_t optval,
1646	unsigned int optlen)
1647	{
1648	int err;
1649	struct sock *sk;
1650	struct vsock_sock *vsk;
1651	const struct vsock_transport *transport;
1652	u64 val;
1653
1654	if (level != AF_VSOCK && level != SOL_SOCKET)
1655	return -ENOPROTOOPT;
1656
1657	#define COPY_IN(_v) \
1658	do { \
1659	if (optlen < sizeof(_v)) { \
1660	err = -EINVAL; \
1661	goto exit; \
1662	} \
1663	if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) { \
1664	err = -EFAULT; \
1665	goto exit; \
1666	} \
1667	} while (0)
1668
1669	err = 0;
1670	sk = sock->sk;
1671	vsk = vsock_sk(sk);
1672
1673	lock_sock(sk);
1674
1675	transport = vsk->transport;
1676
1677	if (level == SOL_SOCKET) {
1678	int zerocopy;
1679
1680	if (optname != SO_ZEROCOPY) {
1681	release_sock(sk);
1682	return sock_setsockopt(sock, level, op: optname, optval, optlen);
1683	}
1684
1685	/* Use 'int' type here, because variable to
1686	* set this option usually has this type.
1687	*/
1688	COPY_IN(zerocopy);
1689
1690	if (zerocopy < 0 || zerocopy > 1) {
1691	err = -EINVAL;
1692	goto exit;
1693	}
1694
1695	if (transport && !vsock_msgzerocopy_allow(t: transport)) {
1696	err = -EOPNOTSUPP;
1697	goto exit;
1698	}
1699
1700	sock_valbool_flag(sk, bit: SOCK_ZEROCOPY, valbool: zerocopy);
1701	goto exit;
1702	}
1703
1704	switch (optname) {
1705	case SO_VM_SOCKETS_BUFFER_SIZE:
1706	COPY_IN(val);
1707	vsock_update_buffer_size(vsk, transport, val);
1708	break;
1709
1710	case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1711	COPY_IN(val);
1712	vsk->buffer_max_size = val;
1713	vsock_update_buffer_size(vsk, transport, val: vsk->buffer_size);
1714	break;
1715
1716	case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1717	COPY_IN(val);
1718	vsk->buffer_min_size = val;
1719	vsock_update_buffer_size(vsk, transport, val: vsk->buffer_size);
1720	break;
1721
1722	case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1723	case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: {
1724	struct __kernel_sock_timeval tv;
1725
1726	err = sock_copy_user_timeval(tv: &tv, optval, optlen,
1727	old_timeval: optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1728	if (err)
1729	break;
1730	if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1731	tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1732	vsk->connect_timeout = tv.tv_sec * HZ +
1733	DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ));
1734	if (vsk->connect_timeout == 0)
1735	vsk->connect_timeout =
1736	VSOCK_DEFAULT_CONNECT_TIMEOUT;
1737
1738	} else {
1739	err = -ERANGE;
1740	}
1741	break;
1742	}
1743
1744	default:
1745	err = -ENOPROTOOPT;
1746	break;
1747	}
1748
1749	#undef COPY_IN
1750
1751	exit:
1752	release_sock(sk);
1753	return err;
1754	}
1755
1756	static int vsock_connectible_getsockopt(struct socket *sock,
1757	int level, int optname,
1758	char __user *optval,
1759	int __user *optlen)
1760	{
1761	struct sock *sk = sock->sk;
1762	struct vsock_sock *vsk = vsock_sk(sk);
1763
1764	union {
1765	u64 val64;
1766	struct old_timeval32 tm32;
1767	struct __kernel_old_timeval tm;
1768	struct __kernel_sock_timeval stm;
1769	} v;
1770
1771	int lv = sizeof(v.val64);
1772	int len;
1773
1774	if (level != AF_VSOCK)
1775	return -ENOPROTOOPT;
1776
1777	if (get_user(len, optlen))
1778	return -EFAULT;
1779
1780	memset(&v, 0, sizeof(v));
1781
1782	switch (optname) {
1783	case SO_VM_SOCKETS_BUFFER_SIZE:
1784	v.val64 = vsk->buffer_size;
1785	break;
1786
1787	case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1788	v.val64 = vsk->buffer_max_size;
1789	break;
1790
1791	case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1792	v.val64 = vsk->buffer_min_size;
1793	break;
1794
1795	case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1796	case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD:
1797	lv = sock_get_timeout(timeo: vsk->connect_timeout, optval: &v,
1798	old_timeval: optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1799	break;
1800
1801	default:
1802	return -ENOPROTOOPT;
1803	}
1804
1805	if (len < lv)
1806	return -EINVAL;
1807	if (len > lv)
1808	len = lv;
1809	if (copy_to_user(to: optval, from: &v, n: len))
1810	return -EFAULT;
1811
1812	if (put_user(len, optlen))
1813	return -EFAULT;
1814
1815	return 0;
1816	}
1817
1818	static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1819	size_t len)
1820	{
1821	struct sock *sk;
1822	struct vsock_sock *vsk;
1823	const struct vsock_transport *transport;
1824	ssize_t total_written;
1825	long timeout;
1826	int err;
1827	struct vsock_transport_send_notify_data send_data;
1828	DEFINE_WAIT_FUNC(wait, woken_wake_function);
1829
1830	sk = sock->sk;
1831	vsk = vsock_sk(sk);
1832	total_written = 0;
1833	err = 0;
1834
1835	if (msg->msg_flags & MSG_OOB)
1836	return -EOPNOTSUPP;
1837
1838	lock_sock(sk);
1839
1840	transport = vsk->transport;
1841
1842	/* Callers should not provide a destination with connection oriented
1843	* sockets.
1844	*/
1845	if (msg->msg_namelen) {
1846	err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1847	goto out;
1848	}
1849
1850	/* Send data only if both sides are not shutdown in the direction. */
1851	if (sk->sk_shutdown & SEND_SHUTDOWN ||
1852	vsk->peer_shutdown & RCV_SHUTDOWN) {
1853	err = -EPIPE;
1854	goto out;
1855	}
1856
1857	if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1858	!vsock_addr_bound(addr: &vsk->local_addr)) {
1859	err = -ENOTCONN;
1860	goto out;
1861	}
1862
1863	if (!vsock_addr_bound(addr: &vsk->remote_addr)) {
1864	err = -EDESTADDRREQ;
1865	goto out;
1866	}
1867
1868	if (msg->msg_flags & MSG_ZEROCOPY &&
1869	!vsock_msgzerocopy_allow(t: transport)) {
1870	err = -EOPNOTSUPP;
1871	goto out;
1872	}
1873
1874	/* Wait for room in the produce queue to enqueue our user's data. */
1875	timeout = sock_sndtimeo(sk, noblock: msg->msg_flags & MSG_DONTWAIT);
1876
1877	err = transport->notify_send_init(vsk, &send_data);
1878	if (err < 0)
1879	goto out;
1880
1881	while (total_written < len) {
1882	ssize_t written;
1883
1884	add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1885	while (vsock_stream_has_space(vsk) == 0 &&
1886	sk->sk_err == 0 &&
1887	!(sk->sk_shutdown & SEND_SHUTDOWN) &&
1888	!(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1889
1890	/* Don't wait for non-blocking sockets. */
1891	if (timeout == 0) {
1892	err = -EAGAIN;
1893	remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1894	goto out_err;
1895	}
1896
1897	err = transport->notify_send_pre_block(vsk, &send_data);
1898	if (err < 0) {
1899	remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1900	goto out_err;
1901	}
1902
1903	release_sock(sk);
1904	timeout = wait_woken(wq_entry: &wait, TASK_INTERRUPTIBLE, timeout);
1905	lock_sock(sk);
1906	if (signal_pending(current)) {
1907	err = sock_intr_errno(timeo: timeout);
1908	remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1909	goto out_err;
1910	} else if (timeout == 0) {
1911	err = -EAGAIN;
1912	remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1913	goto out_err;
1914	}
1915	}
1916	remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
1917
1918	/* These checks occur both as part of and after the loop
1919	* conditional since we need to check before and after
1920	* sleeping.
1921	*/
1922	if (sk->sk_err) {
1923	err = -sk->sk_err;
1924	goto out_err;
1925	} else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1926	(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1927	err = -EPIPE;
1928	goto out_err;
1929	}
1930
1931	err = transport->notify_send_pre_enqueue(vsk, &send_data);
1932	if (err < 0)
1933	goto out_err;
1934
1935	/* Note that enqueue will only write as many bytes as are free
1936	* in the produce queue, so we don't need to ensure len is
1937	* smaller than the queue size. It is the caller's
1938	* responsibility to check how many bytes we were able to send.
1939	*/
1940
1941	if (sk->sk_type == SOCK_SEQPACKET) {
1942	written = transport->seqpacket_enqueue(vsk,
1943	msg, len - total_written);
1944	} else {
1945	written = transport->stream_enqueue(vsk,
1946	msg, len - total_written);
1947	}
1948
1949	if (written < 0) {
1950	err = written;
1951	goto out_err;
1952	}
1953
1954	total_written += written;
1955
1956	err = transport->notify_send_post_enqueue(
1957	vsk, written, &send_data);
1958	if (err < 0)
1959	goto out_err;
1960
1961	}
1962
1963	out_err:
1964	if (total_written > 0) {
1965	/* Return number of written bytes only if:
1966	* 1) SOCK_STREAM socket.
1967	* 2) SOCK_SEQPACKET socket when whole buffer is sent.
1968	*/
1969	if (sk->sk_type == SOCK_STREAM || total_written == len)
1970	err = total_written;
1971	}
1972	out:
1973	if (sk->sk_type == SOCK_STREAM)
1974	err = sk_stream_error(sk, flags: msg->msg_flags, err);
1975
1976	release_sock(sk);
1977	return err;
1978	}
1979
1980	static int vsock_connectible_wait_data(struct sock *sk,
1981	struct wait_queue_entry *wait,
1982	long timeout,
1983	struct vsock_transport_recv_notify_data *recv_data,
1984	size_t target)
1985	{
1986	const struct vsock_transport *transport;
1987	struct vsock_sock *vsk;
1988	s64 data;
1989	int err;
1990
1991	vsk = vsock_sk(sk);
1992	err = 0;
1993	transport = vsk->transport;
1994
1995	while (1) {
1996	prepare_to_wait(wq_head: sk_sleep(sk), wq_entry: wait, TASK_INTERRUPTIBLE);
1997	data = vsock_connectible_has_data(vsk);
1998	if (data != 0)
1999	break;
2000
2001	if (sk->sk_err != 0 ||
2002	(sk->sk_shutdown & RCV_SHUTDOWN) ||
2003	(vsk->peer_shutdown & SEND_SHUTDOWN)) {
2004	break;
2005	}
2006
2007	/* Don't wait for non-blocking sockets. */
2008	if (timeout == 0) {
2009	err = -EAGAIN;
2010	break;
2011	}
2012
2013	if (recv_data) {
2014	err = transport->notify_recv_pre_block(vsk, target, recv_data);
2015	if (err < 0)
2016	break;
2017	}
2018
2019	release_sock(sk);
2020	timeout = schedule_timeout(timeout);
2021	lock_sock(sk);
2022
2023	if (signal_pending(current)) {
2024	err = sock_intr_errno(timeo: timeout);
2025	break;
2026	} else if (timeout == 0) {
2027	err = -EAGAIN;
2028	break;
2029	}
2030	}
2031
2032	finish_wait(wq_head: sk_sleep(sk), wq_entry: wait);
2033
2034	if (err)
2035	return err;
2036
2037	/* Internal transport error when checking for available
2038	* data. XXX This should be changed to a connection
2039	* reset in a later change.
2040	*/
2041	if (data < 0)
2042	return -ENOMEM;
2043
2044	return data;
2045	}
2046
2047	static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2048	size_t len, int flags)
2049	{
2050	struct vsock_transport_recv_notify_data recv_data;
2051	const struct vsock_transport *transport;
2052	struct vsock_sock *vsk;
2053	ssize_t copied;
2054	size_t target;
2055	long timeout;
2056	int err;
2057
2058	DEFINE_WAIT(wait);
2059
2060	vsk = vsock_sk(sk);
2061	transport = vsk->transport;
2062
2063	/* We must not copy less than target bytes into the user's buffer
2064	* before returning successfully, so we wait for the consume queue to
2065	* have that much data to consume before dequeueing. Note that this
2066	* makes it impossible to handle cases where target is greater than the
2067	* queue size.
2068	*/
2069	target = sock_rcvlowat(sk, waitall: flags & MSG_WAITALL, len);
2070	if (target >= transport->stream_rcvhiwat(vsk)) {
2071	err = -ENOMEM;
2072	goto out;
2073	}
2074	timeout = sock_rcvtimeo(sk, noblock: flags & MSG_DONTWAIT);
2075	copied = 0;
2076
2077	err = transport->notify_recv_init(vsk, target, &recv_data);
2078	if (err < 0)
2079	goto out;
2080
2081
2082	while (1) {
2083	ssize_t read;
2084
2085	err = vsock_connectible_wait_data(sk, wait: &wait, timeout,
2086	recv_data: &recv_data, target);
2087	if (err <= 0)
2088	break;
2089
2090	err = transport->notify_recv_pre_dequeue(vsk, target,
2091	&recv_data);
2092	if (err < 0)
2093	break;
2094
2095	read = transport->stream_dequeue(vsk, msg, len - copied, flags);
2096	if (read < 0) {
2097	err = read;
2098	break;
2099	}
2100
2101	copied += read;
2102
2103	err = transport->notify_recv_post_dequeue(vsk, target, read,
2104	!(flags & MSG_PEEK), &recv_data);
2105	if (err < 0)
2106	goto out;
2107
2108	if (read >= target || flags & MSG_PEEK)
2109	break;
2110
2111	target -= read;
2112	}
2113
2114	if (sk->sk_err)
2115	err = -sk->sk_err;
2116	else if (sk->sk_shutdown & RCV_SHUTDOWN)
2117	err = 0;
2118
2119	if (copied > 0)
2120	err = copied;
2121
2122	out:
2123	return err;
2124	}
2125
2126	static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2127	size_t len, int flags)
2128	{
2129	const struct vsock_transport *transport;
2130	struct vsock_sock *vsk;
2131	ssize_t msg_len;
2132	long timeout;
2133	int err = 0;
2134	DEFINE_WAIT(wait);
2135
2136	vsk = vsock_sk(sk);
2137	transport = vsk->transport;
2138
2139	timeout = sock_rcvtimeo(sk, noblock: flags & MSG_DONTWAIT);
2140
2141	err = vsock_connectible_wait_data(sk, wait: &wait, timeout, NULL, target: 0);
2142	if (err <= 0)
2143	goto out;
2144
2145	msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2146
2147	if (msg_len < 0) {
2148	err = msg_len;
2149	goto out;
2150	}
2151
2152	if (sk->sk_err) {
2153	err = -sk->sk_err;
2154	} else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2155	err = 0;
2156	} else {
2157	/* User sets MSG_TRUNC, so return real length of
2158	* packet.
2159	*/
2160	if (flags & MSG_TRUNC)
2161	err = msg_len;
2162	else
2163	err = len - msg_data_left(msg);
2164
2165	/* Always set MSG_TRUNC if real length of packet is
2166	* bigger than user's buffer.
2167	*/
2168	if (msg_len > len)
2169	msg->msg_flags |= MSG_TRUNC;
2170	}
2171
2172	out:
2173	return err;
2174	}
2175
2176	int
2177	vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2178	int flags)
2179	{
2180	struct sock *sk;
2181	struct vsock_sock *vsk;
2182	const struct vsock_transport *transport;
2183	#ifdef CONFIG_BPF_SYSCALL
2184	const struct proto *prot;
2185	#endif
2186	int err;
2187
2188	sk = sock->sk;
2189
2190	if (unlikely(flags & MSG_ERRQUEUE))
2191	return sock_recv_errqueue(sk, msg, len, SOL_VSOCK, VSOCK_RECVERR);
2192
2193	vsk = vsock_sk(sk);
2194	err = 0;
2195
2196	lock_sock(sk);
2197
2198	transport = vsk->transport;
2199
2200	if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2201	/* Recvmsg is supposed to return 0 if a peer performs an
2202	* orderly shutdown. Differentiate between that case and when a
2203	* peer has not connected or a local shutdown occurred with the
2204	* SOCK_DONE flag.
2205	*/
2206	if (sock_flag(sk, flag: SOCK_DONE))
2207	err = 0;
2208	else
2209	err = -ENOTCONN;
2210
2211	goto out;
2212	}
2213
2214	if (flags & MSG_OOB) {
2215	err = -EOPNOTSUPP;
2216	goto out;
2217	}
2218
2219	/* We don't check peer_shutdown flag here since peer may actually shut
2220	* down, but there can be data in the queue that a local socket can
2221	* receive.
2222	*/
2223	if (sk->sk_shutdown & RCV_SHUTDOWN) {
2224	err = 0;
2225	goto out;
2226	}
2227
2228	/* It is valid on Linux to pass in a zero-length receive buffer. This
2229	* is not an error. We may as well bail out now.
2230	*/
2231	if (!len) {
2232	err = 0;
2233	goto out;
2234	}
2235
2236	#ifdef CONFIG_BPF_SYSCALL
2237	prot = READ_ONCE(sk->sk_prot);
2238	if (prot != &vsock_proto) {
2239	release_sock(sk);
2240	return prot->recvmsg(sk, msg, len, flags, NULL);
2241	}
2242	#endif
2243
2244	if (sk->sk_type == SOCK_STREAM)
2245	err = __vsock_stream_recvmsg(sk, msg, len, flags);
2246	else
2247	err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2248
2249	out:
2250	release_sock(sk);
2251	return err;
2252	}
2253	EXPORT_SYMBOL_GPL(vsock_connectible_recvmsg);
2254
2255	static int vsock_set_rcvlowat(struct sock *sk, int val)
2256	{
2257	const struct vsock_transport *transport;
2258	struct vsock_sock *vsk;
2259
2260	vsk = vsock_sk(sk);
2261
2262	if (val > vsk->buffer_size)
2263	return -EINVAL;
2264
2265	transport = vsk->transport;
2266
2267	if (transport && transport->set_rcvlowat)
2268	return transport->set_rcvlowat(vsk, val);
2269
2270	WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
2271	return 0;
2272	}
2273
2274	static const struct proto_ops vsock_stream_ops = {
2275	.family = PF_VSOCK,
2276	.owner = THIS_MODULE,
2277	.release = vsock_release,
2278	.bind = vsock_bind,
2279	.connect = vsock_connect,
2280	.socketpair = sock_no_socketpair,
2281	.accept = vsock_accept,
2282	.getname = vsock_getname,
2283	.poll = vsock_poll,
2284	.ioctl = sock_no_ioctl,
2285	.listen = vsock_listen,
2286	.shutdown = vsock_shutdown,
2287	.setsockopt = vsock_connectible_setsockopt,
2288	.getsockopt = vsock_connectible_getsockopt,
2289	.sendmsg = vsock_connectible_sendmsg,
2290	.recvmsg = vsock_connectible_recvmsg,
2291	.mmap = sock_no_mmap,
2292	.set_rcvlowat = vsock_set_rcvlowat,
2293	.read_skb = vsock_read_skb,
2294	};
2295
2296	static const struct proto_ops vsock_seqpacket_ops = {
2297	.family = PF_VSOCK,
2298	.owner = THIS_MODULE,
2299	.release = vsock_release,
2300	.bind = vsock_bind,
2301	.connect = vsock_connect,
2302	.socketpair = sock_no_socketpair,
2303	.accept = vsock_accept,
2304	.getname = vsock_getname,
2305	.poll = vsock_poll,
2306	.ioctl = sock_no_ioctl,
2307	.listen = vsock_listen,
2308	.shutdown = vsock_shutdown,
2309	.setsockopt = vsock_connectible_setsockopt,
2310	.getsockopt = vsock_connectible_getsockopt,
2311	.sendmsg = vsock_connectible_sendmsg,
2312	.recvmsg = vsock_connectible_recvmsg,
2313	.mmap = sock_no_mmap,
2314	.read_skb = vsock_read_skb,
2315	};
2316
2317	static int vsock_create(struct net *net, struct socket *sock,
2318	int protocol, int kern)
2319	{
2320	struct vsock_sock *vsk;
2321	struct sock *sk;
2322	int ret;
2323
2324	if (!sock)
2325	return -EINVAL;
2326
2327	if (protocol && protocol != PF_VSOCK)
2328	return -EPROTONOSUPPORT;
2329
2330	switch (sock->type) {
2331	case SOCK_DGRAM:
2332	sock->ops = &vsock_dgram_ops;
2333	break;
2334	case SOCK_STREAM:
2335	sock->ops = &vsock_stream_ops;
2336	break;
2337	case SOCK_SEQPACKET:
2338	sock->ops = &vsock_seqpacket_ops;
2339	break;
2340	default:
2341	return -ESOCKTNOSUPPORT;
2342	}
2343
2344	sock->state = SS_UNCONNECTED;
2345
2346	sk = __vsock_create(net, sock, NULL, GFP_KERNEL, type: 0, kern);
2347	if (!sk)
2348	return -ENOMEM;
2349
2350	vsk = vsock_sk(sk);
2351
2352	if (sock->type == SOCK_DGRAM) {
2353	ret = vsock_assign_transport(vsk, NULL);
2354	if (ret < 0) {
2355	sock_put(sk);
2356	return ret;
2357	}
2358	}
2359
2360	/* SOCK_DGRAM doesn't have 'setsockopt' callback set in its
2361	* proto_ops, so there is no handler for custom logic.
2362	*/
2363	if (sock_type_connectible(type: sock->type))
2364	set_bit(SOCK_CUSTOM_SOCKOPT, addr: &sk->sk_socket->flags);
2365
2366	vsock_insert_unbound(vsk);
2367
2368	return 0;
2369	}
2370
2371	static const struct net_proto_family vsock_family_ops = {
2372	.family = AF_VSOCK,
2373	.create = vsock_create,
2374	.owner = THIS_MODULE,
2375	};
2376
2377	static long vsock_dev_do_ioctl(struct file *filp,
2378	unsigned int cmd, void __user *ptr)
2379	{
2380	u32 __user *p = ptr;
2381	u32 cid = VMADDR_CID_ANY;
2382	int retval = 0;
2383
2384	switch (cmd) {
2385	case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2386	/* To be compatible with the VMCI behavior, we prioritize the
2387	* guest CID instead of well-know host CID (VMADDR_CID_HOST).
2388	*/
2389	if (transport_g2h)
2390	cid = transport_g2h->get_local_cid();
2391	else if (transport_h2g)
2392	cid = transport_h2g->get_local_cid();
2393
2394	if (put_user(cid, p) != 0)
2395	retval = -EFAULT;
2396	break;
2397
2398	default:
2399	retval = -ENOIOCTLCMD;
2400	}
2401
2402	return retval;
2403	}
2404
2405	static long vsock_dev_ioctl(struct file *filp,
2406	unsigned int cmd, unsigned long arg)
2407	{
2408	return vsock_dev_do_ioctl(filp, cmd, ptr: (void __user *)arg);
2409	}
2410
2411	#ifdef CONFIG_COMPAT
2412	static long vsock_dev_compat_ioctl(struct file *filp,
2413	unsigned int cmd, unsigned long arg)
2414	{
2415	return vsock_dev_do_ioctl(filp, cmd, ptr: compat_ptr(uptr: arg));
2416	}
2417	#endif
2418
2419	static const struct file_operations vsock_device_ops = {
2420	.owner = THIS_MODULE,
2421	.unlocked_ioctl = vsock_dev_ioctl,
2422	#ifdef CONFIG_COMPAT
2423	.compat_ioctl = vsock_dev_compat_ioctl,
2424	#endif
2425	.open = nonseekable_open,
2426	};
2427
2428	static struct miscdevice vsock_device = {
2429	.name = "vsock",
2430	.fops = &vsock_device_ops,
2431	};
2432
2433	static int __init vsock_init(void)
2434	{
2435	int err = 0;
2436
2437	vsock_init_tables();
2438
2439	vsock_proto.owner = THIS_MODULE;
2440	vsock_device.minor = MISC_DYNAMIC_MINOR;
2441	err = misc_register(misc: &vsock_device);
2442	if (err) {
2443	pr_err("Failed to register misc device\n");
2444	goto err_reset_transport;
2445	}
2446
2447	err = proto_register(prot: &vsock_proto, alloc_slab: 1); /* we want our slab */
2448	if (err) {
2449	pr_err("Cannot register vsock protocol\n");
2450	goto err_deregister_misc;
2451	}
2452
2453	err = sock_register(fam: &vsock_family_ops);
2454	if (err) {
2455	pr_err("could not register af_vsock (%d) address family: %d\n",
2456	AF_VSOCK, err);
2457	goto err_unregister_proto;
2458	}
2459
2460	vsock_bpf_build_proto();
2461
2462	return 0;
2463
2464	err_unregister_proto:
2465	proto_unregister(prot: &vsock_proto);
2466	err_deregister_misc:
2467	misc_deregister(misc: &vsock_device);
2468	err_reset_transport:
2469	return err;
2470	}
2471
2472	static void __exit vsock_exit(void)
2473	{
2474	misc_deregister(misc: &vsock_device);
2475	sock_unregister(AF_VSOCK);
2476	proto_unregister(prot: &vsock_proto);
2477	}
2478
2479	const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2480	{
2481	return vsk->transport;
2482	}
2483	EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2484
2485	int vsock_core_register(const struct vsock_transport *t, int features)
2486	{
2487	const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2488	int err = mutex_lock_interruptible(&vsock_register_mutex);
2489
2490	if (err)
2491	return err;
2492
2493	t_h2g = transport_h2g;
2494	t_g2h = transport_g2h;
2495	t_dgram = transport_dgram;
2496	t_local = transport_local;
2497
2498	if (features & VSOCK_TRANSPORT_F_H2G) {
2499	if (t_h2g) {
2500	err = -EBUSY;
2501	goto err_busy;
2502	}
2503	t_h2g = t;
2504	}
2505
2506	if (features & VSOCK_TRANSPORT_F_G2H) {
2507	if (t_g2h) {
2508	err = -EBUSY;
2509	goto err_busy;
2510	}
2511	t_g2h = t;
2512	}
2513
2514	if (features & VSOCK_TRANSPORT_F_DGRAM) {
2515	if (t_dgram) {
2516	err = -EBUSY;
2517	goto err_busy;
2518	}
2519	t_dgram = t;
2520	}
2521
2522	if (features & VSOCK_TRANSPORT_F_LOCAL) {
2523	if (t_local) {
2524	err = -EBUSY;
2525	goto err_busy;
2526	}
2527	t_local = t;
2528	}
2529
2530	transport_h2g = t_h2g;
2531	transport_g2h = t_g2h;
2532	transport_dgram = t_dgram;
2533	transport_local = t_local;
2534
2535	err_busy:
2536	mutex_unlock(lock: &vsock_register_mutex);
2537	return err;
2538	}
2539	EXPORT_SYMBOL_GPL(vsock_core_register);
2540
2541	void vsock_core_unregister(const struct vsock_transport *t)
2542	{
2543	mutex_lock(&vsock_register_mutex);
2544
2545	if (transport_h2g == t)
2546	transport_h2g = NULL;
2547
2548	if (transport_g2h == t)
2549	transport_g2h = NULL;
2550
2551	if (transport_dgram == t)
2552	transport_dgram = NULL;
2553
2554	if (transport_local == t)
2555	transport_local = NULL;
2556
2557	mutex_unlock(lock: &vsock_register_mutex);
2558	}
2559	EXPORT_SYMBOL_GPL(vsock_core_unregister);
2560
2561	module_init(vsock_init);
2562	module_exit(vsock_exit);
2563
2564	MODULE_AUTHOR("VMware, Inc.");
2565	MODULE_DESCRIPTION("VMware Virtual Socket Family");
2566	MODULE_VERSION("1.0.2.0-k");
2567	MODULE_LICENSE("GPL v2");
2568