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/compat.h>
89 #include <linux/types.h>
90 #include <linux/bitops.h>
91 #include <linux/cred.h>
92 #include <linux/init.h>
94 #include <linux/kernel.h>
95 #include <linux/sched/signal.h>
96 #include <linux/kmod.h>
97 #include <linux/list.h>
98 #include <linux/miscdevice.h>
99 #include <linux/module.h>
100 #include <linux/mutex.h>
101 #include <linux/net.h>
102 #include <linux/poll.h>
103 #include <linux/random.h>
104 #include <linux/skbuff.h>
105 #include <linux/smp.h>
106 #include <linux/socket.h>
107 #include <linux/stddef.h>
108 #include <linux/unistd.h>
109 #include <linux/wait.h>
110 #include <linux/workqueue.h>
111 #include <net/sock.h>
112 #include <net/af_vsock.h>
114 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
115 static void vsock_sk_destruct(struct sock *sk);
116 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
118 /* Protocol family. */
119 static struct proto vsock_proto = {
121 .owner = THIS_MODULE,
122 .obj_size = sizeof(struct vsock_sock),
125 /* The default peer timeout indicates how long we will wait for a peer response
126 * to a control message.
128 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
130 #define VSOCK_DEFAULT_BUFFER_SIZE (1024 * 256)
131 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
132 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
134 /* Transport used for host->guest communication */
135 static const struct vsock_transport *transport_h2g;
136 /* Transport used for guest->host communication */
137 static const struct vsock_transport *transport_g2h;
138 /* Transport used for DGRAM communication */
139 static const struct vsock_transport *transport_dgram;
140 /* Transport used for local communication */
141 static const struct vsock_transport *transport_local;
142 static DEFINE_MUTEX(vsock_register_mutex);
146 /* Each bound VSocket is stored in the bind hash table and each connected
147 * VSocket is stored in the connected hash table.
149 * Unbound sockets are all put on the same list attached to the end of the hash
150 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
151 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
152 * represents the list that addr hashes to).
154 * Specifically, we initialize the vsock_bind_table array to a size of
155 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
156 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
157 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
158 * mods with VSOCK_HASH_SIZE to ensure this.
160 #define MAX_PORT_RETRIES 24
162 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE)
163 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
164 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
166 /* XXX This can probably be implemented in a better way. */
167 #define VSOCK_CONN_HASH(src, dst) \
168 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
169 #define vsock_connected_sockets(src, dst) \
170 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
171 #define vsock_connected_sockets_vsk(vsk) \
172 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
174 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
175 EXPORT_SYMBOL_GPL(vsock_bind_table);
176 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
177 EXPORT_SYMBOL_GPL(vsock_connected_table);
178 DEFINE_SPINLOCK(vsock_table_lock);
179 EXPORT_SYMBOL_GPL(vsock_table_lock);
181 /* Autobind this socket to the local address if necessary. */
182 static int vsock_auto_bind(struct vsock_sock *vsk)
184 struct sock *sk = sk_vsock(vsk);
185 struct sockaddr_vm local_addr;
187 if (vsock_addr_bound(&vsk->local_addr))
189 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
190 return __vsock_bind(sk, &local_addr);
193 static void vsock_init_tables(void)
197 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
198 INIT_LIST_HEAD(&vsock_bind_table[i]);
200 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
201 INIT_LIST_HEAD(&vsock_connected_table[i]);
204 static void __vsock_insert_bound(struct list_head *list,
205 struct vsock_sock *vsk)
208 list_add(&vsk->bound_table, list);
211 static void __vsock_insert_connected(struct list_head *list,
212 struct vsock_sock *vsk)
215 list_add(&vsk->connected_table, list);
218 static void __vsock_remove_bound(struct vsock_sock *vsk)
220 list_del_init(&vsk->bound_table);
224 static void __vsock_remove_connected(struct vsock_sock *vsk)
226 list_del_init(&vsk->connected_table);
230 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
232 struct vsock_sock *vsk;
234 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
235 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
236 return sk_vsock(vsk);
238 if (addr->svm_port == vsk->local_addr.svm_port &&
239 (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
240 addr->svm_cid == VMADDR_CID_ANY))
241 return sk_vsock(vsk);
247 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
248 struct sockaddr_vm *dst)
250 struct vsock_sock *vsk;
252 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
254 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
255 dst->svm_port == vsk->local_addr.svm_port) {
256 return sk_vsock(vsk);
263 static void vsock_insert_unbound(struct vsock_sock *vsk)
265 spin_lock_bh(&vsock_table_lock);
266 __vsock_insert_bound(vsock_unbound_sockets, vsk);
267 spin_unlock_bh(&vsock_table_lock);
270 void vsock_insert_connected(struct vsock_sock *vsk)
272 struct list_head *list = vsock_connected_sockets(
273 &vsk->remote_addr, &vsk->local_addr);
275 spin_lock_bh(&vsock_table_lock);
276 __vsock_insert_connected(list, vsk);
277 spin_unlock_bh(&vsock_table_lock);
279 EXPORT_SYMBOL_GPL(vsock_insert_connected);
281 void vsock_remove_bound(struct vsock_sock *vsk)
283 spin_lock_bh(&vsock_table_lock);
284 if (__vsock_in_bound_table(vsk))
285 __vsock_remove_bound(vsk);
286 spin_unlock_bh(&vsock_table_lock);
288 EXPORT_SYMBOL_GPL(vsock_remove_bound);
290 void vsock_remove_connected(struct vsock_sock *vsk)
292 spin_lock_bh(&vsock_table_lock);
293 if (__vsock_in_connected_table(vsk))
294 __vsock_remove_connected(vsk);
295 spin_unlock_bh(&vsock_table_lock);
297 EXPORT_SYMBOL_GPL(vsock_remove_connected);
299 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
303 spin_lock_bh(&vsock_table_lock);
304 sk = __vsock_find_bound_socket(addr);
308 spin_unlock_bh(&vsock_table_lock);
312 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
314 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
315 struct sockaddr_vm *dst)
319 spin_lock_bh(&vsock_table_lock);
320 sk = __vsock_find_connected_socket(src, dst);
324 spin_unlock_bh(&vsock_table_lock);
328 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
330 void vsock_remove_sock(struct vsock_sock *vsk)
332 vsock_remove_bound(vsk);
333 vsock_remove_connected(vsk);
335 EXPORT_SYMBOL_GPL(vsock_remove_sock);
337 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
341 spin_lock_bh(&vsock_table_lock);
343 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
344 struct vsock_sock *vsk;
345 list_for_each_entry(vsk, &vsock_connected_table[i],
350 spin_unlock_bh(&vsock_table_lock);
352 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
354 void vsock_add_pending(struct sock *listener, struct sock *pending)
356 struct vsock_sock *vlistener;
357 struct vsock_sock *vpending;
359 vlistener = vsock_sk(listener);
360 vpending = vsock_sk(pending);
364 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
366 EXPORT_SYMBOL_GPL(vsock_add_pending);
368 void vsock_remove_pending(struct sock *listener, struct sock *pending)
370 struct vsock_sock *vpending = vsock_sk(pending);
372 list_del_init(&vpending->pending_links);
376 EXPORT_SYMBOL_GPL(vsock_remove_pending);
378 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
380 struct vsock_sock *vlistener;
381 struct vsock_sock *vconnected;
383 vlistener = vsock_sk(listener);
384 vconnected = vsock_sk(connected);
386 sock_hold(connected);
388 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
390 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
392 static bool vsock_use_local_transport(unsigned int remote_cid)
394 if (!transport_local)
397 if (remote_cid == VMADDR_CID_LOCAL)
401 return remote_cid == transport_g2h->get_local_cid();
403 return remote_cid == VMADDR_CID_HOST;
407 static void vsock_deassign_transport(struct vsock_sock *vsk)
412 vsk->transport->destruct(vsk);
413 module_put(vsk->transport->module);
414 vsk->transport = NULL;
417 /* Assign a transport to a socket and call the .init transport callback.
419 * Note: for connection oriented socket this must be called when vsk->remote_addr
420 * is set (e.g. during the connect() or when a connection request on a listener
421 * socket is received).
422 * The vsk->remote_addr is used to decide which transport to use:
423 * - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
424 * g2h is not loaded, will use local transport;
425 * - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
426 * includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
427 * - remote CID > VMADDR_CID_HOST will use host->guest transport;
429 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
431 const struct vsock_transport *new_transport;
432 struct sock *sk = sk_vsock(vsk);
433 unsigned int remote_cid = vsk->remote_addr.svm_cid;
437 /* If the packet is coming with the source and destination CIDs higher
438 * than VMADDR_CID_HOST, then a vsock channel where all the packets are
439 * forwarded to the host should be established. Then the host will
440 * need to forward the packets to the guest.
442 * The flag is set on the (listen) receive path (psk is not NULL). On
443 * the connect path the flag can be set by the user space application.
445 if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
446 vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
447 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
449 remote_flags = vsk->remote_addr.svm_flags;
451 switch (sk->sk_type) {
453 new_transport = transport_dgram;
457 if (vsock_use_local_transport(remote_cid))
458 new_transport = transport_local;
459 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
460 (remote_flags & VMADDR_FLAG_TO_HOST))
461 new_transport = transport_g2h;
463 new_transport = transport_h2g;
466 return -ESOCKTNOSUPPORT;
469 if (vsk->transport) {
470 if (vsk->transport == new_transport)
473 /* transport->release() must be called with sock lock acquired.
474 * This path can only be taken during vsock_connect(), where we
475 * have already held the sock lock. In the other cases, this
476 * function is called on a new socket which is not assigned to
479 vsk->transport->release(vsk);
480 vsock_deassign_transport(vsk);
483 /* We increase the module refcnt to prevent the transport unloading
484 * while there are open sockets assigned to it.
486 if (!new_transport || !try_module_get(new_transport->module))
489 if (sk->sk_type == SOCK_SEQPACKET) {
490 if (!new_transport->seqpacket_allow ||
491 !new_transport->seqpacket_allow(remote_cid)) {
492 module_put(new_transport->module);
493 return -ESOCKTNOSUPPORT;
497 ret = new_transport->init(vsk, psk);
499 module_put(new_transport->module);
503 vsk->transport = new_transport;
507 EXPORT_SYMBOL_GPL(vsock_assign_transport);
509 bool vsock_find_cid(unsigned int cid)
511 if (transport_g2h && cid == transport_g2h->get_local_cid())
514 if (transport_h2g && cid == VMADDR_CID_HOST)
517 if (transport_local && cid == VMADDR_CID_LOCAL)
522 EXPORT_SYMBOL_GPL(vsock_find_cid);
524 static struct sock *vsock_dequeue_accept(struct sock *listener)
526 struct vsock_sock *vlistener;
527 struct vsock_sock *vconnected;
529 vlistener = vsock_sk(listener);
531 if (list_empty(&vlistener->accept_queue))
534 vconnected = list_entry(vlistener->accept_queue.next,
535 struct vsock_sock, accept_queue);
537 list_del_init(&vconnected->accept_queue);
539 /* The caller will need a reference on the connected socket so we let
540 * it call sock_put().
543 return sk_vsock(vconnected);
546 static bool vsock_is_accept_queue_empty(struct sock *sk)
548 struct vsock_sock *vsk = vsock_sk(sk);
549 return list_empty(&vsk->accept_queue);
552 static bool vsock_is_pending(struct sock *sk)
554 struct vsock_sock *vsk = vsock_sk(sk);
555 return !list_empty(&vsk->pending_links);
558 static int vsock_send_shutdown(struct sock *sk, int mode)
560 struct vsock_sock *vsk = vsock_sk(sk);
565 return vsk->transport->shutdown(vsk, mode);
568 static void vsock_pending_work(struct work_struct *work)
571 struct sock *listener;
572 struct vsock_sock *vsk;
575 vsk = container_of(work, struct vsock_sock, pending_work.work);
577 listener = vsk->listener;
581 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
583 if (vsock_is_pending(sk)) {
584 vsock_remove_pending(listener, sk);
586 sk_acceptq_removed(listener);
587 } else if (!vsk->rejected) {
588 /* We are not on the pending list and accept() did not reject
589 * us, so we must have been accepted by our user process. We
590 * just need to drop our references to the sockets and be on
597 /* We need to remove ourself from the global connected sockets list so
598 * incoming packets can't find this socket, and to reduce the reference
601 vsock_remove_connected(vsk);
603 sk->sk_state = TCP_CLOSE;
607 release_sock(listener);
615 /**** SOCKET OPERATIONS ****/
617 static int __vsock_bind_connectible(struct vsock_sock *vsk,
618 struct sockaddr_vm *addr)
621 struct sockaddr_vm new_addr;
624 port = LAST_RESERVED_PORT + 1 +
625 prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
627 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
629 if (addr->svm_port == VMADDR_PORT_ANY) {
633 for (i = 0; i < MAX_PORT_RETRIES; i++) {
634 if (port <= LAST_RESERVED_PORT)
635 port = LAST_RESERVED_PORT + 1;
637 new_addr.svm_port = port++;
639 if (!__vsock_find_bound_socket(&new_addr)) {
646 return -EADDRNOTAVAIL;
648 /* If port is in reserved range, ensure caller
649 * has necessary privileges.
651 if (addr->svm_port <= LAST_RESERVED_PORT &&
652 !capable(CAP_NET_BIND_SERVICE)) {
656 if (__vsock_find_bound_socket(&new_addr))
660 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
662 /* Remove connection oriented sockets from the unbound list and add them
663 * to the hash table for easy lookup by its address. The unbound list
664 * is simply an extra entry at the end of the hash table, a trick used
667 __vsock_remove_bound(vsk);
668 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
673 static int __vsock_bind_dgram(struct vsock_sock *vsk,
674 struct sockaddr_vm *addr)
676 return vsk->transport->dgram_bind(vsk, addr);
679 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
681 struct vsock_sock *vsk = vsock_sk(sk);
684 /* First ensure this socket isn't already bound. */
685 if (vsock_addr_bound(&vsk->local_addr))
688 /* Now bind to the provided address or select appropriate values if
689 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
690 * like AF_INET prevents binding to a non-local IP address (in most
691 * cases), we only allow binding to a local CID.
693 if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
694 return -EADDRNOTAVAIL;
696 switch (sk->sk_socket->type) {
699 spin_lock_bh(&vsock_table_lock);
700 retval = __vsock_bind_connectible(vsk, addr);
701 spin_unlock_bh(&vsock_table_lock);
705 retval = __vsock_bind_dgram(vsk, addr);
716 static void vsock_connect_timeout(struct work_struct *work);
718 static struct sock *__vsock_create(struct net *net,
726 struct vsock_sock *psk;
727 struct vsock_sock *vsk;
729 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
733 sock_init_data(sock, sk);
735 /* sk->sk_type is normally set in sock_init_data, but only if sock is
736 * non-NULL. We make sure that our sockets always have a type by
737 * setting it here if needed.
743 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
744 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
746 sk->sk_destruct = vsock_sk_destruct;
747 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
748 sock_reset_flag(sk, SOCK_DONE);
750 INIT_LIST_HEAD(&vsk->bound_table);
751 INIT_LIST_HEAD(&vsk->connected_table);
752 vsk->listener = NULL;
753 INIT_LIST_HEAD(&vsk->pending_links);
754 INIT_LIST_HEAD(&vsk->accept_queue);
755 vsk->rejected = false;
756 vsk->sent_request = false;
757 vsk->ignore_connecting_rst = false;
758 vsk->peer_shutdown = 0;
759 INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
760 INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
762 psk = parent ? vsock_sk(parent) : NULL;
764 vsk->trusted = psk->trusted;
765 vsk->owner = get_cred(psk->owner);
766 vsk->connect_timeout = psk->connect_timeout;
767 vsk->buffer_size = psk->buffer_size;
768 vsk->buffer_min_size = psk->buffer_min_size;
769 vsk->buffer_max_size = psk->buffer_max_size;
770 security_sk_clone(parent, sk);
772 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
773 vsk->owner = get_current_cred();
774 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
775 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
776 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
777 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
783 static bool sock_type_connectible(u16 type)
785 return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
788 static void __vsock_release(struct sock *sk, int level)
791 struct sock *pending;
792 struct vsock_sock *vsk;
795 pending = NULL; /* Compiler warning. */
797 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
798 * version to avoid the warning "possible recursive locking
799 * detected". When "level" is 0, lock_sock_nested(sk, level)
800 * is the same as lock_sock(sk).
802 lock_sock_nested(sk, level);
805 vsk->transport->release(vsk);
806 else if (sock_type_connectible(sk->sk_type))
807 vsock_remove_sock(vsk);
810 sk->sk_shutdown = SHUTDOWN_MASK;
812 skb_queue_purge(&sk->sk_receive_queue);
814 /* Clean up any sockets that never were accepted. */
815 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
816 __vsock_release(pending, SINGLE_DEPTH_NESTING);
825 static void vsock_sk_destruct(struct sock *sk)
827 struct vsock_sock *vsk = vsock_sk(sk);
829 vsock_deassign_transport(vsk);
831 /* When clearing these addresses, there's no need to set the family and
832 * possibly register the address family with the kernel.
834 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
835 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
837 put_cred(vsk->owner);
840 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
844 err = sock_queue_rcv_skb(sk, skb);
851 struct sock *vsock_create_connected(struct sock *parent)
853 return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
856 EXPORT_SYMBOL_GPL(vsock_create_connected);
858 s64 vsock_stream_has_data(struct vsock_sock *vsk)
860 return vsk->transport->stream_has_data(vsk);
862 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
864 static s64 vsock_connectible_has_data(struct vsock_sock *vsk)
866 struct sock *sk = sk_vsock(vsk);
868 if (sk->sk_type == SOCK_SEQPACKET)
869 return vsk->transport->seqpacket_has_data(vsk);
871 return vsock_stream_has_data(vsk);
874 s64 vsock_stream_has_space(struct vsock_sock *vsk)
876 return vsk->transport->stream_has_space(vsk);
878 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
880 static int vsock_release(struct socket *sock)
882 __vsock_release(sock->sk, 0);
884 sock->state = SS_FREE;
890 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
894 struct sockaddr_vm *vm_addr;
898 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
902 err = __vsock_bind(sk, vm_addr);
908 static int vsock_getname(struct socket *sock,
909 struct sockaddr *addr, int peer)
913 struct vsock_sock *vsk;
914 struct sockaddr_vm *vm_addr;
923 if (sock->state != SS_CONNECTED) {
927 vm_addr = &vsk->remote_addr;
929 vm_addr = &vsk->local_addr;
937 /* sys_getsockname() and sys_getpeername() pass us a
938 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
939 * that macro is defined in socket.c instead of .h, so we hardcode its
942 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
943 memcpy(addr, vm_addr, sizeof(*vm_addr));
944 err = sizeof(*vm_addr);
951 static int vsock_shutdown(struct socket *sock, int mode)
956 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
957 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
958 * here like the other address families do. Note also that the
959 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
960 * which is what we want.
964 if ((mode & ~SHUTDOWN_MASK) || !mode)
967 /* If this is a connection oriented socket and it is not connected then
968 * bail out immediately. If it is a DGRAM socket then we must first
969 * kick the socket so that it wakes up from any sleeping calls, for
970 * example recv(), and then afterwards return the error.
976 if (sock->state == SS_UNCONNECTED) {
978 if (sock_type_connectible(sk->sk_type))
981 sock->state = SS_DISCONNECTING;
985 /* Receive and send shutdowns are treated alike. */
986 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
988 sk->sk_shutdown |= mode;
989 sk->sk_state_change(sk);
991 if (sock_type_connectible(sk->sk_type)) {
992 sock_reset_flag(sk, SOCK_DONE);
993 vsock_send_shutdown(sk, mode);
1002 static __poll_t vsock_poll(struct file *file, struct socket *sock,
1007 struct vsock_sock *vsk;
1012 poll_wait(file, sk_sleep(sk), wait);
1016 /* Signify that there has been an error on this socket. */
1019 /* INET sockets treat local write shutdown and peer write shutdown as a
1020 * case of EPOLLHUP set.
1022 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1023 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1024 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1028 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1029 vsk->peer_shutdown & SEND_SHUTDOWN) {
1033 if (sock->type == SOCK_DGRAM) {
1034 /* For datagram sockets we can read if there is something in
1035 * the queue and write as long as the socket isn't shutdown for
1038 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1039 (sk->sk_shutdown & RCV_SHUTDOWN)) {
1040 mask |= EPOLLIN | EPOLLRDNORM;
1043 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1044 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1046 } else if (sock_type_connectible(sk->sk_type)) {
1047 const struct vsock_transport *transport;
1051 transport = vsk->transport;
1053 /* Listening sockets that have connections in their accept
1054 * queue can be read.
1056 if (sk->sk_state == TCP_LISTEN
1057 && !vsock_is_accept_queue_empty(sk))
1058 mask |= EPOLLIN | EPOLLRDNORM;
1060 /* If there is something in the queue then we can read. */
1061 if (transport && transport->stream_is_active(vsk) &&
1062 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1063 bool data_ready_now = false;
1064 int ret = transport->notify_poll_in(
1065 vsk, 1, &data_ready_now);
1070 mask |= EPOLLIN | EPOLLRDNORM;
1075 /* Sockets whose connections have been closed, reset, or
1076 * terminated should also be considered read, and we check the
1077 * shutdown flag for that.
1079 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1080 vsk->peer_shutdown & SEND_SHUTDOWN) {
1081 mask |= EPOLLIN | EPOLLRDNORM;
1084 /* Connected sockets that can produce data can be written. */
1085 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1086 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1087 bool space_avail_now = false;
1088 int ret = transport->notify_poll_out(
1089 vsk, 1, &space_avail_now);
1093 if (space_avail_now)
1094 /* Remove EPOLLWRBAND since INET
1095 * sockets are not setting it.
1097 mask |= EPOLLOUT | EPOLLWRNORM;
1103 /* Simulate INET socket poll behaviors, which sets
1104 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1105 * but local send is not shutdown.
1107 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1108 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1109 mask |= EPOLLOUT | EPOLLWRNORM;
1119 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1124 struct vsock_sock *vsk;
1125 struct sockaddr_vm *remote_addr;
1126 const struct vsock_transport *transport;
1128 if (msg->msg_flags & MSG_OOB)
1131 /* For now, MSG_DONTWAIT is always assumed... */
1138 transport = vsk->transport;
1140 err = vsock_auto_bind(vsk);
1145 /* If the provided message contains an address, use that. Otherwise
1146 * fall back on the socket's remote handle (if it has been connected).
1148 if (msg->msg_name &&
1149 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1150 &remote_addr) == 0) {
1151 /* Ensure this address is of the right type and is a valid
1155 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1156 remote_addr->svm_cid = transport->get_local_cid();
1158 if (!vsock_addr_bound(remote_addr)) {
1162 } else if (sock->state == SS_CONNECTED) {
1163 remote_addr = &vsk->remote_addr;
1165 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1166 remote_addr->svm_cid = transport->get_local_cid();
1168 /* XXX Should connect() or this function ensure remote_addr is
1171 if (!vsock_addr_bound(&vsk->remote_addr)) {
1180 if (!transport->dgram_allow(remote_addr->svm_cid,
1181 remote_addr->svm_port)) {
1186 err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1193 static int vsock_dgram_connect(struct socket *sock,
1194 struct sockaddr *addr, int addr_len, int flags)
1198 struct vsock_sock *vsk;
1199 struct sockaddr_vm *remote_addr;
1204 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1205 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1207 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1209 sock->state = SS_UNCONNECTED;
1212 } else if (err != 0)
1217 err = vsock_auto_bind(vsk);
1221 if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1222 remote_addr->svm_port)) {
1227 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1228 sock->state = SS_CONNECTED;
1235 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1236 size_t len, int flags)
1238 struct vsock_sock *vsk = vsock_sk(sock->sk);
1240 return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1243 static const struct proto_ops vsock_dgram_ops = {
1245 .owner = THIS_MODULE,
1246 .release = vsock_release,
1248 .connect = vsock_dgram_connect,
1249 .socketpair = sock_no_socketpair,
1250 .accept = sock_no_accept,
1251 .getname = vsock_getname,
1253 .ioctl = sock_no_ioctl,
1254 .listen = sock_no_listen,
1255 .shutdown = vsock_shutdown,
1256 .sendmsg = vsock_dgram_sendmsg,
1257 .recvmsg = vsock_dgram_recvmsg,
1258 .mmap = sock_no_mmap,
1259 .sendpage = sock_no_sendpage,
1262 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1264 const struct vsock_transport *transport = vsk->transport;
1266 if (!transport || !transport->cancel_pkt)
1269 return transport->cancel_pkt(vsk);
1272 static void vsock_connect_timeout(struct work_struct *work)
1275 struct vsock_sock *vsk;
1277 vsk = container_of(work, struct vsock_sock, connect_work.work);
1281 if (sk->sk_state == TCP_SYN_SENT &&
1282 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1283 sk->sk_state = TCP_CLOSE;
1284 sk->sk_err = ETIMEDOUT;
1285 sk_error_report(sk);
1286 vsock_transport_cancel_pkt(vsk);
1293 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1294 int addr_len, int flags)
1298 struct vsock_sock *vsk;
1299 const struct vsock_transport *transport;
1300 struct sockaddr_vm *remote_addr;
1310 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1311 switch (sock->state) {
1315 case SS_DISCONNECTING:
1319 /* This continues on so we can move sock into the SS_CONNECTED
1320 * state once the connection has completed (at which point err
1321 * will be set to zero also). Otherwise, we will either wait
1322 * for the connection or return -EALREADY should this be a
1323 * non-blocking call.
1326 if (flags & O_NONBLOCK)
1330 if ((sk->sk_state == TCP_LISTEN) ||
1331 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1336 /* Set the remote address that we are connecting to. */
1337 memcpy(&vsk->remote_addr, remote_addr,
1338 sizeof(vsk->remote_addr));
1340 err = vsock_assign_transport(vsk, NULL);
1344 transport = vsk->transport;
1346 /* The hypervisor and well-known contexts do not have socket
1350 !transport->stream_allow(remote_addr->svm_cid,
1351 remote_addr->svm_port)) {
1356 err = vsock_auto_bind(vsk);
1360 sk->sk_state = TCP_SYN_SENT;
1362 err = transport->connect(vsk);
1366 /* Mark sock as connecting and set the error code to in
1367 * progress in case this is a non-blocking connect.
1369 sock->state = SS_CONNECTING;
1373 /* The receive path will handle all communication until we are able to
1374 * enter the connected state. Here we wait for the connection to be
1375 * completed or a notification of an error.
1377 timeout = vsk->connect_timeout;
1378 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1380 while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1381 if (flags & O_NONBLOCK) {
1382 /* If we're not going to block, we schedule a timeout
1383 * function to generate a timeout on the connection
1384 * attempt, in case the peer doesn't respond in a
1385 * timely manner. We hold on to the socket until the
1389 schedule_delayed_work(&vsk->connect_work, timeout);
1391 /* Skip ahead to preserve error code set above. */
1396 timeout = schedule_timeout(timeout);
1399 if (signal_pending(current)) {
1400 err = sock_intr_errno(timeout);
1401 sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1402 sock->state = SS_UNCONNECTED;
1403 vsock_transport_cancel_pkt(vsk);
1405 } else if (timeout == 0) {
1407 sk->sk_state = TCP_CLOSE;
1408 sock->state = SS_UNCONNECTED;
1409 vsock_transport_cancel_pkt(vsk);
1413 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1418 sk->sk_state = TCP_CLOSE;
1419 sock->state = SS_UNCONNECTED;
1425 finish_wait(sk_sleep(sk), &wait);
1431 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1434 struct sock *listener;
1436 struct sock *connected;
1437 struct vsock_sock *vconnected;
1442 listener = sock->sk;
1444 lock_sock(listener);
1446 if (!sock_type_connectible(sock->type)) {
1451 if (listener->sk_state != TCP_LISTEN) {
1456 /* Wait for children sockets to appear; these are the new sockets
1457 * created upon connection establishment.
1459 timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1460 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1462 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1463 listener->sk_err == 0) {
1464 release_sock(listener);
1465 timeout = schedule_timeout(timeout);
1466 finish_wait(sk_sleep(listener), &wait);
1467 lock_sock(listener);
1469 if (signal_pending(current)) {
1470 err = sock_intr_errno(timeout);
1472 } else if (timeout == 0) {
1477 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1479 finish_wait(sk_sleep(listener), &wait);
1481 if (listener->sk_err)
1482 err = -listener->sk_err;
1485 sk_acceptq_removed(listener);
1487 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1488 vconnected = vsock_sk(connected);
1490 /* If the listener socket has received an error, then we should
1491 * reject this socket and return. Note that we simply mark the
1492 * socket rejected, drop our reference, and let the cleanup
1493 * function handle the cleanup; the fact that we found it in
1494 * the listener's accept queue guarantees that the cleanup
1495 * function hasn't run yet.
1498 vconnected->rejected = true;
1500 newsock->state = SS_CONNECTED;
1501 sock_graft(connected, newsock);
1504 release_sock(connected);
1505 sock_put(connected);
1509 release_sock(listener);
1513 static int vsock_listen(struct socket *sock, int backlog)
1517 struct vsock_sock *vsk;
1523 if (!sock_type_connectible(sk->sk_type)) {
1528 if (sock->state != SS_UNCONNECTED) {
1535 if (!vsock_addr_bound(&vsk->local_addr)) {
1540 sk->sk_max_ack_backlog = backlog;
1541 sk->sk_state = TCP_LISTEN;
1550 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1551 const struct vsock_transport *transport,
1554 if (val > vsk->buffer_max_size)
1555 val = vsk->buffer_max_size;
1557 if (val < vsk->buffer_min_size)
1558 val = vsk->buffer_min_size;
1560 if (val != vsk->buffer_size &&
1561 transport && transport->notify_buffer_size)
1562 transport->notify_buffer_size(vsk, &val);
1564 vsk->buffer_size = val;
1567 static int vsock_connectible_setsockopt(struct socket *sock,
1571 unsigned int optlen)
1575 struct vsock_sock *vsk;
1576 const struct vsock_transport *transport;
1579 if (level != AF_VSOCK)
1580 return -ENOPROTOOPT;
1582 #define COPY_IN(_v) \
1584 if (optlen < sizeof(_v)) { \
1588 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) { \
1600 transport = vsk->transport;
1603 case SO_VM_SOCKETS_BUFFER_SIZE:
1605 vsock_update_buffer_size(vsk, transport, val);
1608 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1610 vsk->buffer_max_size = val;
1611 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1614 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1616 vsk->buffer_min_size = val;
1617 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1620 case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1621 case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: {
1622 struct __kernel_sock_timeval tv;
1624 err = sock_copy_user_timeval(&tv, optval, optlen,
1625 optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1628 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1629 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1630 vsk->connect_timeout = tv.tv_sec * HZ +
1631 DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ));
1632 if (vsk->connect_timeout == 0)
1633 vsk->connect_timeout =
1634 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1654 static int vsock_connectible_getsockopt(struct socket *sock,
1655 int level, int optname,
1656 char __user *optval,
1659 struct sock *sk = sock->sk;
1660 struct vsock_sock *vsk = vsock_sk(sk);
1664 struct old_timeval32 tm32;
1665 struct __kernel_old_timeval tm;
1666 struct __kernel_sock_timeval stm;
1669 int lv = sizeof(v.val64);
1672 if (level != AF_VSOCK)
1673 return -ENOPROTOOPT;
1675 if (get_user(len, optlen))
1678 memset(&v, 0, sizeof(v));
1681 case SO_VM_SOCKETS_BUFFER_SIZE:
1682 v.val64 = vsk->buffer_size;
1685 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1686 v.val64 = vsk->buffer_max_size;
1689 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1690 v.val64 = vsk->buffer_min_size;
1693 case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1694 case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD:
1695 lv = sock_get_timeout(vsk->connect_timeout, &v,
1696 optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1700 return -ENOPROTOOPT;
1707 if (copy_to_user(optval, &v, len))
1710 if (put_user(len, optlen))
1716 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1720 struct vsock_sock *vsk;
1721 const struct vsock_transport *transport;
1722 ssize_t total_written;
1725 struct vsock_transport_send_notify_data send_data;
1726 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1733 if (msg->msg_flags & MSG_OOB)
1738 transport = vsk->transport;
1740 /* Callers should not provide a destination with connection oriented
1743 if (msg->msg_namelen) {
1744 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1748 /* Send data only if both sides are not shutdown in the direction. */
1749 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1750 vsk->peer_shutdown & RCV_SHUTDOWN) {
1755 if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1756 !vsock_addr_bound(&vsk->local_addr)) {
1761 if (!vsock_addr_bound(&vsk->remote_addr)) {
1762 err = -EDESTADDRREQ;
1766 /* Wait for room in the produce queue to enqueue our user's data. */
1767 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1769 err = transport->notify_send_init(vsk, &send_data);
1773 while (total_written < len) {
1776 add_wait_queue(sk_sleep(sk), &wait);
1777 while (vsock_stream_has_space(vsk) == 0 &&
1779 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1780 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1782 /* Don't wait for non-blocking sockets. */
1785 remove_wait_queue(sk_sleep(sk), &wait);
1789 err = transport->notify_send_pre_block(vsk, &send_data);
1791 remove_wait_queue(sk_sleep(sk), &wait);
1796 timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1798 if (signal_pending(current)) {
1799 err = sock_intr_errno(timeout);
1800 remove_wait_queue(sk_sleep(sk), &wait);
1802 } else if (timeout == 0) {
1804 remove_wait_queue(sk_sleep(sk), &wait);
1808 remove_wait_queue(sk_sleep(sk), &wait);
1810 /* These checks occur both as part of and after the loop
1811 * conditional since we need to check before and after
1817 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1818 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1823 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1827 /* Note that enqueue will only write as many bytes as are free
1828 * in the produce queue, so we don't need to ensure len is
1829 * smaller than the queue size. It is the caller's
1830 * responsibility to check how many bytes we were able to send.
1833 if (sk->sk_type == SOCK_SEQPACKET) {
1834 written = transport->seqpacket_enqueue(vsk,
1835 msg, len - total_written);
1837 written = transport->stream_enqueue(vsk,
1838 msg, len - total_written);
1845 total_written += written;
1847 err = transport->notify_send_post_enqueue(
1848 vsk, written, &send_data);
1855 if (total_written > 0) {
1856 /* Return number of written bytes only if:
1857 * 1) SOCK_STREAM socket.
1858 * 2) SOCK_SEQPACKET socket when whole buffer is sent.
1860 if (sk->sk_type == SOCK_STREAM || total_written == len)
1861 err = total_written;
1868 static int vsock_connectible_wait_data(struct sock *sk,
1869 struct wait_queue_entry *wait,
1871 struct vsock_transport_recv_notify_data *recv_data,
1874 const struct vsock_transport *transport;
1875 struct vsock_sock *vsk;
1881 transport = vsk->transport;
1883 while ((data = vsock_connectible_has_data(vsk)) == 0) {
1884 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
1886 if (sk->sk_err != 0 ||
1887 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1888 (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1892 /* Don't wait for non-blocking sockets. */
1899 err = transport->notify_recv_pre_block(vsk, target, recv_data);
1905 timeout = schedule_timeout(timeout);
1908 if (signal_pending(current)) {
1909 err = sock_intr_errno(timeout);
1911 } else if (timeout == 0) {
1917 finish_wait(sk_sleep(sk), wait);
1922 /* Internal transport error when checking for available
1923 * data. XXX This should be changed to a connection
1924 * reset in a later change.
1932 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
1933 size_t len, int flags)
1935 struct vsock_transport_recv_notify_data recv_data;
1936 const struct vsock_transport *transport;
1937 struct vsock_sock *vsk;
1946 transport = vsk->transport;
1948 /* We must not copy less than target bytes into the user's buffer
1949 * before returning successfully, so we wait for the consume queue to
1950 * have that much data to consume before dequeueing. Note that this
1951 * makes it impossible to handle cases where target is greater than the
1954 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1955 if (target >= transport->stream_rcvhiwat(vsk)) {
1959 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1962 err = transport->notify_recv_init(vsk, target, &recv_data);
1970 err = vsock_connectible_wait_data(sk, &wait, timeout,
1971 &recv_data, target);
1975 err = transport->notify_recv_pre_dequeue(vsk, target,
1980 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
1988 err = transport->notify_recv_post_dequeue(vsk, target, read,
1989 !(flags & MSG_PEEK), &recv_data);
1993 if (read >= target || flags & MSG_PEEK)
2001 else if (sk->sk_shutdown & RCV_SHUTDOWN)
2011 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2012 size_t len, int flags)
2014 const struct vsock_transport *transport;
2015 struct vsock_sock *vsk;
2022 transport = vsk->transport;
2024 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2026 err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0);
2030 msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2039 } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2042 /* User sets MSG_TRUNC, so return real length of
2045 if (flags & MSG_TRUNC)
2048 err = len - msg_data_left(msg);
2050 /* Always set MSG_TRUNC if real length of packet is
2051 * bigger than user's buffer.
2054 msg->msg_flags |= MSG_TRUNC;
2062 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2066 struct vsock_sock *vsk;
2067 const struct vsock_transport *transport;
2078 transport = vsk->transport;
2080 if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2081 /* Recvmsg is supposed to return 0 if a peer performs an
2082 * orderly shutdown. Differentiate between that case and when a
2083 * peer has not connected or a local shutdown occurred with the
2086 if (sock_flag(sk, SOCK_DONE))
2094 if (flags & MSG_OOB) {
2099 /* We don't check peer_shutdown flag here since peer may actually shut
2100 * down, but there can be data in the queue that a local socket can
2103 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2108 /* It is valid on Linux to pass in a zero-length receive buffer. This
2109 * is not an error. We may as well bail out now.
2116 if (sk->sk_type == SOCK_STREAM)
2117 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2119 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2126 static const struct proto_ops vsock_stream_ops = {
2128 .owner = THIS_MODULE,
2129 .release = vsock_release,
2131 .connect = vsock_connect,
2132 .socketpair = sock_no_socketpair,
2133 .accept = vsock_accept,
2134 .getname = vsock_getname,
2136 .ioctl = sock_no_ioctl,
2137 .listen = vsock_listen,
2138 .shutdown = vsock_shutdown,
2139 .setsockopt = vsock_connectible_setsockopt,
2140 .getsockopt = vsock_connectible_getsockopt,
2141 .sendmsg = vsock_connectible_sendmsg,
2142 .recvmsg = vsock_connectible_recvmsg,
2143 .mmap = sock_no_mmap,
2144 .sendpage = sock_no_sendpage,
2147 static const struct proto_ops vsock_seqpacket_ops = {
2149 .owner = THIS_MODULE,
2150 .release = vsock_release,
2152 .connect = vsock_connect,
2153 .socketpair = sock_no_socketpair,
2154 .accept = vsock_accept,
2155 .getname = vsock_getname,
2157 .ioctl = sock_no_ioctl,
2158 .listen = vsock_listen,
2159 .shutdown = vsock_shutdown,
2160 .setsockopt = vsock_connectible_setsockopt,
2161 .getsockopt = vsock_connectible_getsockopt,
2162 .sendmsg = vsock_connectible_sendmsg,
2163 .recvmsg = vsock_connectible_recvmsg,
2164 .mmap = sock_no_mmap,
2165 .sendpage = sock_no_sendpage,
2168 static int vsock_create(struct net *net, struct socket *sock,
2169 int protocol, int kern)
2171 struct vsock_sock *vsk;
2178 if (protocol && protocol != PF_VSOCK)
2179 return -EPROTONOSUPPORT;
2181 switch (sock->type) {
2183 sock->ops = &vsock_dgram_ops;
2186 sock->ops = &vsock_stream_ops;
2188 case SOCK_SEQPACKET:
2189 sock->ops = &vsock_seqpacket_ops;
2192 return -ESOCKTNOSUPPORT;
2195 sock->state = SS_UNCONNECTED;
2197 sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2203 if (sock->type == SOCK_DGRAM) {
2204 ret = vsock_assign_transport(vsk, NULL);
2211 vsock_insert_unbound(vsk);
2216 static const struct net_proto_family vsock_family_ops = {
2218 .create = vsock_create,
2219 .owner = THIS_MODULE,
2222 static long vsock_dev_do_ioctl(struct file *filp,
2223 unsigned int cmd, void __user *ptr)
2225 u32 __user *p = ptr;
2226 u32 cid = VMADDR_CID_ANY;
2230 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2231 /* To be compatible with the VMCI behavior, we prioritize the
2232 * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2235 cid = transport_g2h->get_local_cid();
2236 else if (transport_h2g)
2237 cid = transport_h2g->get_local_cid();
2239 if (put_user(cid, p) != 0)
2244 retval = -ENOIOCTLCMD;
2250 static long vsock_dev_ioctl(struct file *filp,
2251 unsigned int cmd, unsigned long arg)
2253 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2256 #ifdef CONFIG_COMPAT
2257 static long vsock_dev_compat_ioctl(struct file *filp,
2258 unsigned int cmd, unsigned long arg)
2260 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2264 static const struct file_operations vsock_device_ops = {
2265 .owner = THIS_MODULE,
2266 .unlocked_ioctl = vsock_dev_ioctl,
2267 #ifdef CONFIG_COMPAT
2268 .compat_ioctl = vsock_dev_compat_ioctl,
2270 .open = nonseekable_open,
2273 static struct miscdevice vsock_device = {
2275 .fops = &vsock_device_ops,
2278 static int __init vsock_init(void)
2282 vsock_init_tables();
2284 vsock_proto.owner = THIS_MODULE;
2285 vsock_device.minor = MISC_DYNAMIC_MINOR;
2286 err = misc_register(&vsock_device);
2288 pr_err("Failed to register misc device\n");
2289 goto err_reset_transport;
2292 err = proto_register(&vsock_proto, 1); /* we want our slab */
2294 pr_err("Cannot register vsock protocol\n");
2295 goto err_deregister_misc;
2298 err = sock_register(&vsock_family_ops);
2300 pr_err("could not register af_vsock (%d) address family: %d\n",
2302 goto err_unregister_proto;
2307 err_unregister_proto:
2308 proto_unregister(&vsock_proto);
2309 err_deregister_misc:
2310 misc_deregister(&vsock_device);
2311 err_reset_transport:
2315 static void __exit vsock_exit(void)
2317 misc_deregister(&vsock_device);
2318 sock_unregister(AF_VSOCK);
2319 proto_unregister(&vsock_proto);
2322 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2324 return vsk->transport;
2326 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2328 int vsock_core_register(const struct vsock_transport *t, int features)
2330 const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2331 int err = mutex_lock_interruptible(&vsock_register_mutex);
2336 t_h2g = transport_h2g;
2337 t_g2h = transport_g2h;
2338 t_dgram = transport_dgram;
2339 t_local = transport_local;
2341 if (features & VSOCK_TRANSPORT_F_H2G) {
2349 if (features & VSOCK_TRANSPORT_F_G2H) {
2357 if (features & VSOCK_TRANSPORT_F_DGRAM) {
2365 if (features & VSOCK_TRANSPORT_F_LOCAL) {
2373 transport_h2g = t_h2g;
2374 transport_g2h = t_g2h;
2375 transport_dgram = t_dgram;
2376 transport_local = t_local;
2379 mutex_unlock(&vsock_register_mutex);
2382 EXPORT_SYMBOL_GPL(vsock_core_register);
2384 void vsock_core_unregister(const struct vsock_transport *t)
2386 mutex_lock(&vsock_register_mutex);
2388 if (transport_h2g == t)
2389 transport_h2g = NULL;
2391 if (transport_g2h == t)
2392 transport_g2h = NULL;
2394 if (transport_dgram == t)
2395 transport_dgram = NULL;
2397 if (transport_local == t)
2398 transport_local = NULL;
2400 mutex_unlock(&vsock_register_mutex);
2402 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2404 module_init(vsock_init);
2405 module_exit(vsock_exit);
2407 MODULE_AUTHOR("VMware, Inc.");
2408 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2409 MODULE_VERSION("1.0.2.0-k");
2410 MODULE_LICENSE("GPL v2");