1 // SPDX-License-Identifier: GPL-2.0-only
3 * VMware vSockets Driver
5 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
8 /* Implementation notes:
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.
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.
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).
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.
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.
56 * - Lock ordering for pending or accept queue sockets is:
58 * lock_sock(listener);
59 * lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
61 * Using explicit nested locking keeps lockdep happy since normally only one
62 * lock of a given class may be taken at a time.
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.
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.
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):
81 * TCP_CLOSE - unconnected
82 * TCP_SYN_SENT - connecting
83 * TCP_ESTABLISHED - connected
84 * TCP_CLOSING - disconnecting
85 * TCP_LISTEN - listening
88 #include <linux/types.h>
89 #include <linux/bitops.h>
90 #include <linux/cred.h>
91 #include <linux/init.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>
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);
117 /* Protocol family. */
118 static struct proto vsock_proto = {
120 .owner = THIS_MODULE,
121 .obj_size = sizeof(struct vsock_sock),
124 /* The default peer timeout indicates how long we will wait for a peer response
125 * to a control message.
127 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
129 #define VSOCK_DEFAULT_BUFFER_SIZE (1024 * 256)
130 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
131 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
133 /* Transport used for host->guest communication */
134 static const struct vsock_transport *transport_h2g;
135 /* Transport used for guest->host communication */
136 static const struct vsock_transport *transport_g2h;
137 /* Transport used for DGRAM communication */
138 static const struct vsock_transport *transport_dgram;
139 /* Transport used for local communication */
140 static const struct vsock_transport *transport_local;
141 static DEFINE_MUTEX(vsock_register_mutex);
145 /* Each bound VSocket is stored in the bind hash table and each connected
146 * VSocket is stored in the connected hash table.
148 * Unbound sockets are all put on the same list attached to the end of the hash
149 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
150 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
151 * represents the list that addr hashes to).
153 * Specifically, we initialize the vsock_bind_table array to a size of
154 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
155 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
156 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
157 * mods with VSOCK_HASH_SIZE to ensure this.
159 #define MAX_PORT_RETRIES 24
161 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE)
162 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
163 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
165 /* XXX This can probably be implemented in a better way. */
166 #define VSOCK_CONN_HASH(src, dst) \
167 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
168 #define vsock_connected_sockets(src, dst) \
169 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
170 #define vsock_connected_sockets_vsk(vsk) \
171 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
173 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
174 EXPORT_SYMBOL_GPL(vsock_bind_table);
175 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
176 EXPORT_SYMBOL_GPL(vsock_connected_table);
177 DEFINE_SPINLOCK(vsock_table_lock);
178 EXPORT_SYMBOL_GPL(vsock_table_lock);
180 /* Autobind this socket to the local address if necessary. */
181 static int vsock_auto_bind(struct vsock_sock *vsk)
183 struct sock *sk = sk_vsock(vsk);
184 struct sockaddr_vm local_addr;
186 if (vsock_addr_bound(&vsk->local_addr))
188 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
189 return __vsock_bind(sk, &local_addr);
192 static void vsock_init_tables(void)
196 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
197 INIT_LIST_HEAD(&vsock_bind_table[i]);
199 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
200 INIT_LIST_HEAD(&vsock_connected_table[i]);
203 static void __vsock_insert_bound(struct list_head *list,
204 struct vsock_sock *vsk)
207 list_add(&vsk->bound_table, list);
210 static void __vsock_insert_connected(struct list_head *list,
211 struct vsock_sock *vsk)
214 list_add(&vsk->connected_table, list);
217 static void __vsock_remove_bound(struct vsock_sock *vsk)
219 list_del_init(&vsk->bound_table);
223 static void __vsock_remove_connected(struct vsock_sock *vsk)
225 list_del_init(&vsk->connected_table);
229 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
231 struct vsock_sock *vsk;
233 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
234 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
235 return sk_vsock(vsk);
237 if (addr->svm_port == vsk->local_addr.svm_port &&
238 (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
239 addr->svm_cid == VMADDR_CID_ANY))
240 return sk_vsock(vsk);
246 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
247 struct sockaddr_vm *dst)
249 struct vsock_sock *vsk;
251 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
253 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
254 dst->svm_port == vsk->local_addr.svm_port) {
255 return sk_vsock(vsk);
262 static void vsock_insert_unbound(struct vsock_sock *vsk)
264 spin_lock_bh(&vsock_table_lock);
265 __vsock_insert_bound(vsock_unbound_sockets, vsk);
266 spin_unlock_bh(&vsock_table_lock);
269 void vsock_insert_connected(struct vsock_sock *vsk)
271 struct list_head *list = vsock_connected_sockets(
272 &vsk->remote_addr, &vsk->local_addr);
274 spin_lock_bh(&vsock_table_lock);
275 __vsock_insert_connected(list, vsk);
276 spin_unlock_bh(&vsock_table_lock);
278 EXPORT_SYMBOL_GPL(vsock_insert_connected);
280 void vsock_remove_bound(struct vsock_sock *vsk)
282 spin_lock_bh(&vsock_table_lock);
283 if (__vsock_in_bound_table(vsk))
284 __vsock_remove_bound(vsk);
285 spin_unlock_bh(&vsock_table_lock);
287 EXPORT_SYMBOL_GPL(vsock_remove_bound);
289 void vsock_remove_connected(struct vsock_sock *vsk)
291 spin_lock_bh(&vsock_table_lock);
292 if (__vsock_in_connected_table(vsk))
293 __vsock_remove_connected(vsk);
294 spin_unlock_bh(&vsock_table_lock);
296 EXPORT_SYMBOL_GPL(vsock_remove_connected);
298 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
302 spin_lock_bh(&vsock_table_lock);
303 sk = __vsock_find_bound_socket(addr);
307 spin_unlock_bh(&vsock_table_lock);
311 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
313 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
314 struct sockaddr_vm *dst)
318 spin_lock_bh(&vsock_table_lock);
319 sk = __vsock_find_connected_socket(src, dst);
323 spin_unlock_bh(&vsock_table_lock);
327 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
329 void vsock_remove_sock(struct vsock_sock *vsk)
331 vsock_remove_bound(vsk);
332 vsock_remove_connected(vsk);
334 EXPORT_SYMBOL_GPL(vsock_remove_sock);
336 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
340 spin_lock_bh(&vsock_table_lock);
342 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
343 struct vsock_sock *vsk;
344 list_for_each_entry(vsk, &vsock_connected_table[i],
349 spin_unlock_bh(&vsock_table_lock);
351 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
353 void vsock_add_pending(struct sock *listener, struct sock *pending)
355 struct vsock_sock *vlistener;
356 struct vsock_sock *vpending;
358 vlistener = vsock_sk(listener);
359 vpending = vsock_sk(pending);
363 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
365 EXPORT_SYMBOL_GPL(vsock_add_pending);
367 void vsock_remove_pending(struct sock *listener, struct sock *pending)
369 struct vsock_sock *vpending = vsock_sk(pending);
371 list_del_init(&vpending->pending_links);
375 EXPORT_SYMBOL_GPL(vsock_remove_pending);
377 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
379 struct vsock_sock *vlistener;
380 struct vsock_sock *vconnected;
382 vlistener = vsock_sk(listener);
383 vconnected = vsock_sk(connected);
385 sock_hold(connected);
387 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
389 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
391 static bool vsock_use_local_transport(unsigned int remote_cid)
393 if (!transport_local)
396 if (remote_cid == VMADDR_CID_LOCAL)
400 return remote_cid == transport_g2h->get_local_cid();
402 return remote_cid == VMADDR_CID_HOST;
406 static void vsock_deassign_transport(struct vsock_sock *vsk)
411 vsk->transport->destruct(vsk);
412 module_put(vsk->transport->module);
413 vsk->transport = NULL;
416 /* Assign a transport to a socket and call the .init transport callback.
418 * Note: for stream socket this must be called when vsk->remote_addr is set
419 * (e.g. during the connect() or when a connection request on a listener
420 * socket is received).
421 * The vsk->remote_addr is used to decide which transport to use:
422 * - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
423 * g2h is not loaded, will use local transport;
424 * - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
425 * includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
426 * - remote CID > VMADDR_CID_HOST will use host->guest transport;
428 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
430 const struct vsock_transport *new_transport;
431 struct sock *sk = sk_vsock(vsk);
432 unsigned int remote_cid = vsk->remote_addr.svm_cid;
436 /* If the packet is coming with the source and destination CIDs higher
437 * than VMADDR_CID_HOST, then a vsock channel where all the packets are
438 * forwarded to the host should be established. Then the host will
439 * need to forward the packets to the guest.
441 * The flag is set on the (listen) receive path (psk is not NULL). On
442 * the connect path the flag can be set by the user space application.
444 if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
445 vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
446 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
448 remote_flags = vsk->remote_addr.svm_flags;
450 switch (sk->sk_type) {
452 new_transport = transport_dgram;
455 if (vsock_use_local_transport(remote_cid))
456 new_transport = transport_local;
457 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
458 (remote_flags & VMADDR_FLAG_TO_HOST))
459 new_transport = transport_g2h;
461 new_transport = transport_h2g;
464 return -ESOCKTNOSUPPORT;
467 if (vsk->transport) {
468 if (vsk->transport == new_transport)
471 /* transport->release() must be called with sock lock acquired.
472 * This path can only be taken during vsock_stream_connect(),
473 * where we have already held the sock lock.
474 * In the other cases, this function is called on a new socket
475 * which is not assigned to any transport.
477 vsk->transport->release(vsk);
478 vsock_deassign_transport(vsk);
481 /* We increase the module refcnt to prevent the transport unloading
482 * while there are open sockets assigned to it.
484 if (!new_transport || !try_module_get(new_transport->module))
487 ret = new_transport->init(vsk, psk);
489 module_put(new_transport->module);
493 vsk->transport = new_transport;
497 EXPORT_SYMBOL_GPL(vsock_assign_transport);
499 bool vsock_find_cid(unsigned int cid)
501 if (transport_g2h && cid == transport_g2h->get_local_cid())
504 if (transport_h2g && cid == VMADDR_CID_HOST)
507 if (transport_local && cid == VMADDR_CID_LOCAL)
512 EXPORT_SYMBOL_GPL(vsock_find_cid);
514 static struct sock *vsock_dequeue_accept(struct sock *listener)
516 struct vsock_sock *vlistener;
517 struct vsock_sock *vconnected;
519 vlistener = vsock_sk(listener);
521 if (list_empty(&vlistener->accept_queue))
524 vconnected = list_entry(vlistener->accept_queue.next,
525 struct vsock_sock, accept_queue);
527 list_del_init(&vconnected->accept_queue);
529 /* The caller will need a reference on the connected socket so we let
530 * it call sock_put().
533 return sk_vsock(vconnected);
536 static bool vsock_is_accept_queue_empty(struct sock *sk)
538 struct vsock_sock *vsk = vsock_sk(sk);
539 return list_empty(&vsk->accept_queue);
542 static bool vsock_is_pending(struct sock *sk)
544 struct vsock_sock *vsk = vsock_sk(sk);
545 return !list_empty(&vsk->pending_links);
548 static int vsock_send_shutdown(struct sock *sk, int mode)
550 struct vsock_sock *vsk = vsock_sk(sk);
555 return vsk->transport->shutdown(vsk, mode);
558 static void vsock_pending_work(struct work_struct *work)
561 struct sock *listener;
562 struct vsock_sock *vsk;
565 vsk = container_of(work, struct vsock_sock, pending_work.work);
567 listener = vsk->listener;
571 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
573 if (vsock_is_pending(sk)) {
574 vsock_remove_pending(listener, sk);
576 sk_acceptq_removed(listener);
577 } else if (!vsk->rejected) {
578 /* We are not on the pending list and accept() did not reject
579 * us, so we must have been accepted by our user process. We
580 * just need to drop our references to the sockets and be on
587 /* We need to remove ourself from the global connected sockets list so
588 * incoming packets can't find this socket, and to reduce the reference
591 vsock_remove_connected(vsk);
593 sk->sk_state = TCP_CLOSE;
597 release_sock(listener);
605 /**** SOCKET OPERATIONS ****/
607 static int __vsock_bind_stream(struct vsock_sock *vsk,
608 struct sockaddr_vm *addr)
611 struct sockaddr_vm new_addr;
614 port = LAST_RESERVED_PORT + 1 +
615 prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
617 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
619 if (addr->svm_port == VMADDR_PORT_ANY) {
623 for (i = 0; i < MAX_PORT_RETRIES; i++) {
624 if (port <= LAST_RESERVED_PORT)
625 port = LAST_RESERVED_PORT + 1;
627 new_addr.svm_port = port++;
629 if (!__vsock_find_bound_socket(&new_addr)) {
636 return -EADDRNOTAVAIL;
638 /* If port is in reserved range, ensure caller
639 * has necessary privileges.
641 if (addr->svm_port <= LAST_RESERVED_PORT &&
642 !capable(CAP_NET_BIND_SERVICE)) {
646 if (__vsock_find_bound_socket(&new_addr))
650 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
652 /* Remove stream sockets from the unbound list and add them to the hash
653 * table for easy lookup by its address. The unbound list is simply an
654 * extra entry at the end of the hash table, a trick used by AF_UNIX.
656 __vsock_remove_bound(vsk);
657 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
662 static int __vsock_bind_dgram(struct vsock_sock *vsk,
663 struct sockaddr_vm *addr)
665 return vsk->transport->dgram_bind(vsk, addr);
668 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
670 struct vsock_sock *vsk = vsock_sk(sk);
673 /* First ensure this socket isn't already bound. */
674 if (vsock_addr_bound(&vsk->local_addr))
677 /* Now bind to the provided address or select appropriate values if
678 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
679 * like AF_INET prevents binding to a non-local IP address (in most
680 * cases), we only allow binding to a local CID.
682 if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
683 return -EADDRNOTAVAIL;
685 switch (sk->sk_socket->type) {
687 spin_lock_bh(&vsock_table_lock);
688 retval = __vsock_bind_stream(vsk, addr);
689 spin_unlock_bh(&vsock_table_lock);
693 retval = __vsock_bind_dgram(vsk, addr);
704 static void vsock_connect_timeout(struct work_struct *work);
706 static struct sock *__vsock_create(struct net *net,
714 struct vsock_sock *psk;
715 struct vsock_sock *vsk;
717 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
721 sock_init_data(sock, sk);
723 /* sk->sk_type is normally set in sock_init_data, but only if sock is
724 * non-NULL. We make sure that our sockets always have a type by
725 * setting it here if needed.
731 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
732 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
734 sk->sk_destruct = vsock_sk_destruct;
735 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
736 sock_reset_flag(sk, SOCK_DONE);
738 INIT_LIST_HEAD(&vsk->bound_table);
739 INIT_LIST_HEAD(&vsk->connected_table);
740 vsk->listener = NULL;
741 INIT_LIST_HEAD(&vsk->pending_links);
742 INIT_LIST_HEAD(&vsk->accept_queue);
743 vsk->rejected = false;
744 vsk->sent_request = false;
745 vsk->ignore_connecting_rst = false;
746 vsk->peer_shutdown = 0;
747 INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
748 INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
750 psk = parent ? vsock_sk(parent) : NULL;
752 vsk->trusted = psk->trusted;
753 vsk->owner = get_cred(psk->owner);
754 vsk->connect_timeout = psk->connect_timeout;
755 vsk->buffer_size = psk->buffer_size;
756 vsk->buffer_min_size = psk->buffer_min_size;
757 vsk->buffer_max_size = psk->buffer_max_size;
759 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
760 vsk->owner = get_current_cred();
761 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
762 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
763 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
764 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
770 static void __vsock_release(struct sock *sk, int level)
773 struct sock *pending;
774 struct vsock_sock *vsk;
777 pending = NULL; /* Compiler warning. */
779 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
780 * version to avoid the warning "possible recursive locking
781 * detected". When "level" is 0, lock_sock_nested(sk, level)
782 * is the same as lock_sock(sk).
784 lock_sock_nested(sk, level);
787 vsk->transport->release(vsk);
788 else if (sk->sk_type == SOCK_STREAM)
789 vsock_remove_sock(vsk);
792 sk->sk_shutdown = SHUTDOWN_MASK;
794 skb_queue_purge(&sk->sk_receive_queue);
796 /* Clean up any sockets that never were accepted. */
797 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
798 __vsock_release(pending, SINGLE_DEPTH_NESTING);
807 static void vsock_sk_destruct(struct sock *sk)
809 struct vsock_sock *vsk = vsock_sk(sk);
811 vsock_deassign_transport(vsk);
813 /* When clearing these addresses, there's no need to set the family and
814 * possibly register the address family with the kernel.
816 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
817 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
819 put_cred(vsk->owner);
822 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
826 err = sock_queue_rcv_skb(sk, skb);
833 struct sock *vsock_create_connected(struct sock *parent)
835 return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
838 EXPORT_SYMBOL_GPL(vsock_create_connected);
840 s64 vsock_stream_has_data(struct vsock_sock *vsk)
842 return vsk->transport->stream_has_data(vsk);
844 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
846 s64 vsock_stream_has_space(struct vsock_sock *vsk)
848 return vsk->transport->stream_has_space(vsk);
850 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
852 static int vsock_release(struct socket *sock)
854 __vsock_release(sock->sk, 0);
856 sock->state = SS_FREE;
862 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
866 struct sockaddr_vm *vm_addr;
870 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
874 err = __vsock_bind(sk, vm_addr);
880 static int vsock_getname(struct socket *sock,
881 struct sockaddr *addr, int peer)
885 struct vsock_sock *vsk;
886 struct sockaddr_vm *vm_addr;
895 if (sock->state != SS_CONNECTED) {
899 vm_addr = &vsk->remote_addr;
901 vm_addr = &vsk->local_addr;
909 /* sys_getsockname() and sys_getpeername() pass us a
910 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
911 * that macro is defined in socket.c instead of .h, so we hardcode its
914 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
915 memcpy(addr, vm_addr, sizeof(*vm_addr));
916 err = sizeof(*vm_addr);
923 static int vsock_shutdown(struct socket *sock, int mode)
928 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
929 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
930 * here like the other address families do. Note also that the
931 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
932 * which is what we want.
936 if ((mode & ~SHUTDOWN_MASK) || !mode)
939 /* If this is a STREAM socket and it is not connected then bail out
940 * immediately. If it is a DGRAM socket then we must first kick the
941 * socket so that it wakes up from any sleeping calls, for example
942 * recv(), and then afterwards return the error.
946 if (sock->state == SS_UNCONNECTED) {
948 if (sk->sk_type == SOCK_STREAM)
951 sock->state = SS_DISCONNECTING;
955 /* Receive and send shutdowns are treated alike. */
956 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
959 sk->sk_shutdown |= mode;
960 sk->sk_state_change(sk);
963 if (sk->sk_type == SOCK_STREAM) {
964 sock_reset_flag(sk, SOCK_DONE);
965 vsock_send_shutdown(sk, mode);
972 static __poll_t vsock_poll(struct file *file, struct socket *sock,
977 struct vsock_sock *vsk;
982 poll_wait(file, sk_sleep(sk), wait);
986 /* Signify that there has been an error on this socket. */
989 /* INET sockets treat local write shutdown and peer write shutdown as a
990 * case of EPOLLHUP set.
992 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
993 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
994 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
998 if (sk->sk_shutdown & RCV_SHUTDOWN ||
999 vsk->peer_shutdown & SEND_SHUTDOWN) {
1003 if (sock->type == SOCK_DGRAM) {
1004 /* For datagram sockets we can read if there is something in
1005 * the queue and write as long as the socket isn't shutdown for
1008 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1009 (sk->sk_shutdown & RCV_SHUTDOWN)) {
1010 mask |= EPOLLIN | EPOLLRDNORM;
1013 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1014 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1016 } else if (sock->type == SOCK_STREAM) {
1017 const struct vsock_transport *transport = vsk->transport;
1020 /* Listening sockets that have connections in their accept
1021 * queue can be read.
1023 if (sk->sk_state == TCP_LISTEN
1024 && !vsock_is_accept_queue_empty(sk))
1025 mask |= EPOLLIN | EPOLLRDNORM;
1027 /* If there is something in the queue then we can read. */
1028 if (transport && transport->stream_is_active(vsk) &&
1029 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1030 bool data_ready_now = false;
1031 int ret = transport->notify_poll_in(
1032 vsk, 1, &data_ready_now);
1037 mask |= EPOLLIN | EPOLLRDNORM;
1042 /* Sockets whose connections have been closed, reset, or
1043 * terminated should also be considered read, and we check the
1044 * shutdown flag for that.
1046 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1047 vsk->peer_shutdown & SEND_SHUTDOWN) {
1048 mask |= EPOLLIN | EPOLLRDNORM;
1051 /* Connected sockets that can produce data can be written. */
1052 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1053 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1054 bool space_avail_now = false;
1055 int ret = transport->notify_poll_out(
1056 vsk, 1, &space_avail_now);
1060 if (space_avail_now)
1061 /* Remove EPOLLWRBAND since INET
1062 * sockets are not setting it.
1064 mask |= EPOLLOUT | EPOLLWRNORM;
1070 /* Simulate INET socket poll behaviors, which sets
1071 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1072 * but local send is not shutdown.
1074 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1075 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1076 mask |= EPOLLOUT | EPOLLWRNORM;
1086 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1091 struct vsock_sock *vsk;
1092 struct sockaddr_vm *remote_addr;
1093 const struct vsock_transport *transport;
1095 if (msg->msg_flags & MSG_OOB)
1098 /* For now, MSG_DONTWAIT is always assumed... */
1102 transport = vsk->transport;
1106 err = vsock_auto_bind(vsk);
1111 /* If the provided message contains an address, use that. Otherwise
1112 * fall back on the socket's remote handle (if it has been connected).
1114 if (msg->msg_name &&
1115 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1116 &remote_addr) == 0) {
1117 /* Ensure this address is of the right type and is a valid
1121 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1122 remote_addr->svm_cid = transport->get_local_cid();
1124 if (!vsock_addr_bound(remote_addr)) {
1128 } else if (sock->state == SS_CONNECTED) {
1129 remote_addr = &vsk->remote_addr;
1131 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1132 remote_addr->svm_cid = transport->get_local_cid();
1134 /* XXX Should connect() or this function ensure remote_addr is
1137 if (!vsock_addr_bound(&vsk->remote_addr)) {
1146 if (!transport->dgram_allow(remote_addr->svm_cid,
1147 remote_addr->svm_port)) {
1152 err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1159 static int vsock_dgram_connect(struct socket *sock,
1160 struct sockaddr *addr, int addr_len, int flags)
1164 struct vsock_sock *vsk;
1165 struct sockaddr_vm *remote_addr;
1170 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1171 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1173 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1175 sock->state = SS_UNCONNECTED;
1178 } else if (err != 0)
1183 err = vsock_auto_bind(vsk);
1187 if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1188 remote_addr->svm_port)) {
1193 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1194 sock->state = SS_CONNECTED;
1201 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1202 size_t len, int flags)
1204 struct vsock_sock *vsk = vsock_sk(sock->sk);
1206 return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1209 static const struct proto_ops vsock_dgram_ops = {
1211 .owner = THIS_MODULE,
1212 .release = vsock_release,
1214 .connect = vsock_dgram_connect,
1215 .socketpair = sock_no_socketpair,
1216 .accept = sock_no_accept,
1217 .getname = vsock_getname,
1219 .ioctl = sock_no_ioctl,
1220 .listen = sock_no_listen,
1221 .shutdown = vsock_shutdown,
1222 .sendmsg = vsock_dgram_sendmsg,
1223 .recvmsg = vsock_dgram_recvmsg,
1224 .mmap = sock_no_mmap,
1225 .sendpage = sock_no_sendpage,
1228 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1230 const struct vsock_transport *transport = vsk->transport;
1232 if (!transport->cancel_pkt)
1235 return transport->cancel_pkt(vsk);
1238 static void vsock_connect_timeout(struct work_struct *work)
1241 struct vsock_sock *vsk;
1244 vsk = container_of(work, struct vsock_sock, connect_work.work);
1248 if (sk->sk_state == TCP_SYN_SENT &&
1249 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1250 sk->sk_state = TCP_CLOSE;
1251 sk->sk_err = ETIMEDOUT;
1252 sk->sk_error_report(sk);
1257 vsock_transport_cancel_pkt(vsk);
1262 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1263 int addr_len, int flags)
1267 struct vsock_sock *vsk;
1268 const struct vsock_transport *transport;
1269 struct sockaddr_vm *remote_addr;
1279 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1280 switch (sock->state) {
1284 case SS_DISCONNECTING:
1288 /* This continues on so we can move sock into the SS_CONNECTED
1289 * state once the connection has completed (at which point err
1290 * will be set to zero also). Otherwise, we will either wait
1291 * for the connection or return -EALREADY should this be a
1292 * non-blocking call.
1297 if ((sk->sk_state == TCP_LISTEN) ||
1298 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1303 /* Set the remote address that we are connecting to. */
1304 memcpy(&vsk->remote_addr, remote_addr,
1305 sizeof(vsk->remote_addr));
1307 err = vsock_assign_transport(vsk, NULL);
1311 transport = vsk->transport;
1313 /* The hypervisor and well-known contexts do not have socket
1317 !transport->stream_allow(remote_addr->svm_cid,
1318 remote_addr->svm_port)) {
1323 err = vsock_auto_bind(vsk);
1327 sk->sk_state = TCP_SYN_SENT;
1329 err = transport->connect(vsk);
1333 /* Mark sock as connecting and set the error code to in
1334 * progress in case this is a non-blocking connect.
1336 sock->state = SS_CONNECTING;
1340 /* The receive path will handle all communication until we are able to
1341 * enter the connected state. Here we wait for the connection to be
1342 * completed or a notification of an error.
1344 timeout = vsk->connect_timeout;
1345 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1347 while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1348 if (flags & O_NONBLOCK) {
1349 /* If we're not going to block, we schedule a timeout
1350 * function to generate a timeout on the connection
1351 * attempt, in case the peer doesn't respond in a
1352 * timely manner. We hold on to the socket until the
1356 schedule_delayed_work(&vsk->connect_work, timeout);
1358 /* Skip ahead to preserve error code set above. */
1363 timeout = schedule_timeout(timeout);
1366 if (signal_pending(current)) {
1367 err = sock_intr_errno(timeout);
1368 sk->sk_state = TCP_CLOSE;
1369 sock->state = SS_UNCONNECTED;
1370 vsock_transport_cancel_pkt(vsk);
1372 } else if (timeout == 0) {
1374 sk->sk_state = TCP_CLOSE;
1375 sock->state = SS_UNCONNECTED;
1376 vsock_transport_cancel_pkt(vsk);
1380 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1385 sk->sk_state = TCP_CLOSE;
1386 sock->state = SS_UNCONNECTED;
1392 finish_wait(sk_sleep(sk), &wait);
1398 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1401 struct sock *listener;
1403 struct sock *connected;
1404 struct vsock_sock *vconnected;
1409 listener = sock->sk;
1411 lock_sock(listener);
1413 if (sock->type != SOCK_STREAM) {
1418 if (listener->sk_state != TCP_LISTEN) {
1423 /* Wait for children sockets to appear; these are the new sockets
1424 * created upon connection establishment.
1426 timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1427 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1429 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1430 listener->sk_err == 0) {
1431 release_sock(listener);
1432 timeout = schedule_timeout(timeout);
1433 finish_wait(sk_sleep(listener), &wait);
1434 lock_sock(listener);
1436 if (signal_pending(current)) {
1437 err = sock_intr_errno(timeout);
1439 } else if (timeout == 0) {
1444 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1446 finish_wait(sk_sleep(listener), &wait);
1448 if (listener->sk_err)
1449 err = -listener->sk_err;
1452 sk_acceptq_removed(listener);
1454 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1455 vconnected = vsock_sk(connected);
1457 /* If the listener socket has received an error, then we should
1458 * reject this socket and return. Note that we simply mark the
1459 * socket rejected, drop our reference, and let the cleanup
1460 * function handle the cleanup; the fact that we found it in
1461 * the listener's accept queue guarantees that the cleanup
1462 * function hasn't run yet.
1465 vconnected->rejected = true;
1467 newsock->state = SS_CONNECTED;
1468 sock_graft(connected, newsock);
1471 release_sock(connected);
1472 sock_put(connected);
1476 release_sock(listener);
1480 static int vsock_listen(struct socket *sock, int backlog)
1484 struct vsock_sock *vsk;
1490 if (sock->type != SOCK_STREAM) {
1495 if (sock->state != SS_UNCONNECTED) {
1502 if (!vsock_addr_bound(&vsk->local_addr)) {
1507 sk->sk_max_ack_backlog = backlog;
1508 sk->sk_state = TCP_LISTEN;
1517 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1518 const struct vsock_transport *transport,
1521 if (val > vsk->buffer_max_size)
1522 val = vsk->buffer_max_size;
1524 if (val < vsk->buffer_min_size)
1525 val = vsk->buffer_min_size;
1527 if (val != vsk->buffer_size &&
1528 transport && transport->notify_buffer_size)
1529 transport->notify_buffer_size(vsk, &val);
1531 vsk->buffer_size = val;
1534 static int vsock_stream_setsockopt(struct socket *sock,
1538 unsigned int optlen)
1542 struct vsock_sock *vsk;
1543 const struct vsock_transport *transport;
1546 if (level != AF_VSOCK)
1547 return -ENOPROTOOPT;
1549 #define COPY_IN(_v) \
1551 if (optlen < sizeof(_v)) { \
1555 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) { \
1564 transport = vsk->transport;
1569 case SO_VM_SOCKETS_BUFFER_SIZE:
1571 vsock_update_buffer_size(vsk, transport, val);
1574 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1576 vsk->buffer_max_size = val;
1577 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1580 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1582 vsk->buffer_min_size = val;
1583 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1586 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1587 struct __kernel_old_timeval tv;
1589 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1590 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1591 vsk->connect_timeout = tv.tv_sec * HZ +
1592 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1593 if (vsk->connect_timeout == 0)
1594 vsk->connect_timeout =
1595 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1615 static int vsock_stream_getsockopt(struct socket *sock,
1616 int level, int optname,
1617 char __user *optval,
1623 struct vsock_sock *vsk;
1626 if (level != AF_VSOCK)
1627 return -ENOPROTOOPT;
1629 err = get_user(len, optlen);
1633 #define COPY_OUT(_v) \
1635 if (len < sizeof(_v)) \
1639 if (copy_to_user(optval, &_v, len) != 0) \
1649 case SO_VM_SOCKETS_BUFFER_SIZE:
1650 val = vsk->buffer_size;
1654 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1655 val = vsk->buffer_max_size;
1659 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1660 val = vsk->buffer_min_size;
1664 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1665 struct __kernel_old_timeval tv;
1666 tv.tv_sec = vsk->connect_timeout / HZ;
1668 (vsk->connect_timeout -
1669 tv.tv_sec * HZ) * (1000000 / HZ);
1674 return -ENOPROTOOPT;
1677 err = put_user(len, optlen);
1686 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1690 struct vsock_sock *vsk;
1691 const struct vsock_transport *transport;
1692 ssize_t total_written;
1695 struct vsock_transport_send_notify_data send_data;
1696 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1700 transport = vsk->transport;
1704 if (msg->msg_flags & MSG_OOB)
1709 /* Callers should not provide a destination with stream sockets. */
1710 if (msg->msg_namelen) {
1711 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1715 /* Send data only if both sides are not shutdown in the direction. */
1716 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1717 vsk->peer_shutdown & RCV_SHUTDOWN) {
1722 if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1723 !vsock_addr_bound(&vsk->local_addr)) {
1728 if (!vsock_addr_bound(&vsk->remote_addr)) {
1729 err = -EDESTADDRREQ;
1733 /* Wait for room in the produce queue to enqueue our user's data. */
1734 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1736 err = transport->notify_send_init(vsk, &send_data);
1740 while (total_written < len) {
1743 add_wait_queue(sk_sleep(sk), &wait);
1744 while (vsock_stream_has_space(vsk) == 0 &&
1746 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1747 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1749 /* Don't wait for non-blocking sockets. */
1752 remove_wait_queue(sk_sleep(sk), &wait);
1756 err = transport->notify_send_pre_block(vsk, &send_data);
1758 remove_wait_queue(sk_sleep(sk), &wait);
1763 timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1765 if (signal_pending(current)) {
1766 err = sock_intr_errno(timeout);
1767 remove_wait_queue(sk_sleep(sk), &wait);
1769 } else if (timeout == 0) {
1771 remove_wait_queue(sk_sleep(sk), &wait);
1775 remove_wait_queue(sk_sleep(sk), &wait);
1777 /* These checks occur both as part of and after the loop
1778 * conditional since we need to check before and after
1784 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1785 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1790 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1794 /* Note that enqueue will only write as many bytes as are free
1795 * in the produce queue, so we don't need to ensure len is
1796 * smaller than the queue size. It is the caller's
1797 * responsibility to check how many bytes we were able to send.
1800 written = transport->stream_enqueue(
1802 len - total_written);
1808 total_written += written;
1810 err = transport->notify_send_post_enqueue(
1811 vsk, written, &send_data);
1818 if (total_written > 0)
1819 err = total_written;
1827 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1831 struct vsock_sock *vsk;
1832 const struct vsock_transport *transport;
1837 struct vsock_transport_recv_notify_data recv_data;
1843 transport = vsk->transport;
1848 if (!transport || sk->sk_state != TCP_ESTABLISHED) {
1849 /* Recvmsg is supposed to return 0 if a peer performs an
1850 * orderly shutdown. Differentiate between that case and when a
1851 * peer has not connected or a local shutdown occured with the
1854 if (sock_flag(sk, SOCK_DONE))
1862 if (flags & MSG_OOB) {
1867 /* We don't check peer_shutdown flag here since peer may actually shut
1868 * down, but there can be data in the queue that a local socket can
1871 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1876 /* It is valid on Linux to pass in a zero-length receive buffer. This
1877 * is not an error. We may as well bail out now.
1884 /* We must not copy less than target bytes into the user's buffer
1885 * before returning successfully, so we wait for the consume queue to
1886 * have that much data to consume before dequeueing. Note that this
1887 * makes it impossible to handle cases where target is greater than the
1890 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1891 if (target >= transport->stream_rcvhiwat(vsk)) {
1895 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1898 err = transport->notify_recv_init(vsk, target, &recv_data);
1906 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1907 ready = vsock_stream_has_data(vsk);
1910 if (sk->sk_err != 0 ||
1911 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1912 (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1913 finish_wait(sk_sleep(sk), &wait);
1916 /* Don't wait for non-blocking sockets. */
1919 finish_wait(sk_sleep(sk), &wait);
1923 err = transport->notify_recv_pre_block(
1924 vsk, target, &recv_data);
1926 finish_wait(sk_sleep(sk), &wait);
1930 timeout = schedule_timeout(timeout);
1933 if (signal_pending(current)) {
1934 err = sock_intr_errno(timeout);
1935 finish_wait(sk_sleep(sk), &wait);
1937 } else if (timeout == 0) {
1939 finish_wait(sk_sleep(sk), &wait);
1945 finish_wait(sk_sleep(sk), &wait);
1948 /* Invalid queue pair content. XXX This should
1949 * be changed to a connection reset in a later
1957 err = transport->notify_recv_pre_dequeue(
1958 vsk, target, &recv_data);
1962 read = transport->stream_dequeue(
1964 len - copied, flags);
1972 err = transport->notify_recv_post_dequeue(
1974 !(flags & MSG_PEEK), &recv_data);
1978 if (read >= target || flags & MSG_PEEK)
1987 else if (sk->sk_shutdown & RCV_SHUTDOWN)
1998 static const struct proto_ops vsock_stream_ops = {
2000 .owner = THIS_MODULE,
2001 .release = vsock_release,
2003 .connect = vsock_stream_connect,
2004 .socketpair = sock_no_socketpair,
2005 .accept = vsock_accept,
2006 .getname = vsock_getname,
2008 .ioctl = sock_no_ioctl,
2009 .listen = vsock_listen,
2010 .shutdown = vsock_shutdown,
2011 .setsockopt = vsock_stream_setsockopt,
2012 .getsockopt = vsock_stream_getsockopt,
2013 .sendmsg = vsock_stream_sendmsg,
2014 .recvmsg = vsock_stream_recvmsg,
2015 .mmap = sock_no_mmap,
2016 .sendpage = sock_no_sendpage,
2019 static int vsock_create(struct net *net, struct socket *sock,
2020 int protocol, int kern)
2022 struct vsock_sock *vsk;
2029 if (protocol && protocol != PF_VSOCK)
2030 return -EPROTONOSUPPORT;
2032 switch (sock->type) {
2034 sock->ops = &vsock_dgram_ops;
2037 sock->ops = &vsock_stream_ops;
2040 return -ESOCKTNOSUPPORT;
2043 sock->state = SS_UNCONNECTED;
2045 sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2051 if (sock->type == SOCK_DGRAM) {
2052 ret = vsock_assign_transport(vsk, NULL);
2059 vsock_insert_unbound(vsk);
2064 static const struct net_proto_family vsock_family_ops = {
2066 .create = vsock_create,
2067 .owner = THIS_MODULE,
2070 static long vsock_dev_do_ioctl(struct file *filp,
2071 unsigned int cmd, void __user *ptr)
2073 u32 __user *p = ptr;
2074 u32 cid = VMADDR_CID_ANY;
2078 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2079 /* To be compatible with the VMCI behavior, we prioritize the
2080 * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2083 cid = transport_g2h->get_local_cid();
2084 else if (transport_h2g)
2085 cid = transport_h2g->get_local_cid();
2087 if (put_user(cid, p) != 0)
2092 retval = -ENOIOCTLCMD;
2098 static long vsock_dev_ioctl(struct file *filp,
2099 unsigned int cmd, unsigned long arg)
2101 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2104 #ifdef CONFIG_COMPAT
2105 static long vsock_dev_compat_ioctl(struct file *filp,
2106 unsigned int cmd, unsigned long arg)
2108 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2112 static const struct file_operations vsock_device_ops = {
2113 .owner = THIS_MODULE,
2114 .unlocked_ioctl = vsock_dev_ioctl,
2115 #ifdef CONFIG_COMPAT
2116 .compat_ioctl = vsock_dev_compat_ioctl,
2118 .open = nonseekable_open,
2121 static struct miscdevice vsock_device = {
2123 .fops = &vsock_device_ops,
2126 static int __init vsock_init(void)
2130 vsock_init_tables();
2132 vsock_proto.owner = THIS_MODULE;
2133 vsock_device.minor = MISC_DYNAMIC_MINOR;
2134 err = misc_register(&vsock_device);
2136 pr_err("Failed to register misc device\n");
2137 goto err_reset_transport;
2140 err = proto_register(&vsock_proto, 1); /* we want our slab */
2142 pr_err("Cannot register vsock protocol\n");
2143 goto err_deregister_misc;
2146 err = sock_register(&vsock_family_ops);
2148 pr_err("could not register af_vsock (%d) address family: %d\n",
2150 goto err_unregister_proto;
2155 err_unregister_proto:
2156 proto_unregister(&vsock_proto);
2157 err_deregister_misc:
2158 misc_deregister(&vsock_device);
2159 err_reset_transport:
2163 static void __exit vsock_exit(void)
2165 misc_deregister(&vsock_device);
2166 sock_unregister(AF_VSOCK);
2167 proto_unregister(&vsock_proto);
2170 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2172 return vsk->transport;
2174 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2176 int vsock_core_register(const struct vsock_transport *t, int features)
2178 const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2179 int err = mutex_lock_interruptible(&vsock_register_mutex);
2184 t_h2g = transport_h2g;
2185 t_g2h = transport_g2h;
2186 t_dgram = transport_dgram;
2187 t_local = transport_local;
2189 if (features & VSOCK_TRANSPORT_F_H2G) {
2197 if (features & VSOCK_TRANSPORT_F_G2H) {
2205 if (features & VSOCK_TRANSPORT_F_DGRAM) {
2213 if (features & VSOCK_TRANSPORT_F_LOCAL) {
2221 transport_h2g = t_h2g;
2222 transport_g2h = t_g2h;
2223 transport_dgram = t_dgram;
2224 transport_local = t_local;
2227 mutex_unlock(&vsock_register_mutex);
2230 EXPORT_SYMBOL_GPL(vsock_core_register);
2232 void vsock_core_unregister(const struct vsock_transport *t)
2234 mutex_lock(&vsock_register_mutex);
2236 if (transport_h2g == t)
2237 transport_h2g = NULL;
2239 if (transport_g2h == t)
2240 transport_g2h = NULL;
2242 if (transport_dgram == t)
2243 transport_dgram = NULL;
2245 if (transport_local == t)
2246 transport_local = NULL;
2248 mutex_unlock(&vsock_register_mutex);
2250 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2252 module_init(vsock_init);
2253 module_exit(vsock_exit);
2255 MODULE_AUTHOR("VMware, Inc.");
2256 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2257 MODULE_VERSION("1.0.2.0-k");
2258 MODULE_LICENSE("GPL v2");