2 * VMware vSockets Driver
4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation version 2 and no later version.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 /* Implementation notes:
18 * - There are two kinds of sockets: those created by user action (such as
19 * calling socket(2)) and those created by incoming connection request packets.
21 * - There are two "global" tables, one for bound sockets (sockets that have
22 * specified an address that they are responsible for) and one for connected
23 * sockets (sockets that have established a connection with another socket).
24 * These tables are "global" in that all sockets on the system are placed
25 * within them. - Note, though, that the bound table contains an extra entry
26 * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
27 * that list. The bound table is used solely for lookup of sockets when packets
28 * are received and that's not necessary for SOCK_DGRAM sockets since we create
29 * a datagram handle for each and need not perform a lookup. Keeping SOCK_DGRAM
30 * sockets out of the bound hash buckets will reduce the chance of collisions
31 * when looking for SOCK_STREAM sockets and prevents us from having to check the
32 * socket type in the hash table lookups.
34 * - Sockets created by user action will either be "client" sockets that
35 * initiate a connection or "server" sockets that listen for connections; we do
36 * not support simultaneous connects (two "client" sockets connecting).
38 * - "Server" sockets are referred to as listener sockets throughout this
39 * implementation because they are in the VSOCK_SS_LISTEN state. When a
40 * connection request is received (the second kind of socket mentioned above),
41 * we create a new socket and refer to it as a pending socket. These pending
42 * sockets are placed on the pending connection list of the listener socket.
43 * When future packets are received for the address the listener socket is
44 * bound to, we check if the source of the packet is from one that has an
45 * existing pending connection. If it does, we process the packet for the
46 * pending socket. When that socket reaches the connected state, it is removed
47 * from the listener socket's pending list and enqueued in the listener
48 * socket's accept queue. Callers of accept(2) will accept connected sockets
49 * from the listener socket's accept queue. If the socket cannot be accepted
50 * for some reason then it is marked rejected. Once the connection is
51 * accepted, it is owned by the user process and the responsibility for cleanup
52 * falls with that user process.
54 * - It is possible that these pending sockets will never reach the connected
55 * state; in fact, we may never receive another packet after the connection
56 * request. Because of this, we must schedule a cleanup function to run in the
57 * future, after some amount of time passes where a connection should have been
58 * established. This function ensures that the socket is off all lists so it
59 * cannot be retrieved, then drops all references to the socket so it is cleaned
60 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this
61 * function will also cleanup rejected sockets, those that reach the connected
62 * state but leave it before they have been accepted.
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.
79 #include <linux/types.h>
80 #include <linux/bitops.h>
81 #include <linux/cred.h>
82 #include <linux/init.h>
84 #include <linux/kernel.h>
85 #include <linux/kmod.h>
86 #include <linux/list.h>
87 #include <linux/miscdevice.h>
88 #include <linux/module.h>
89 #include <linux/mutex.h>
90 #include <linux/net.h>
91 #include <linux/poll.h>
92 #include <linux/skbuff.h>
93 #include <linux/smp.h>
94 #include <linux/socket.h>
95 #include <linux/stddef.h>
96 #include <linux/unistd.h>
97 #include <linux/wait.h>
98 #include <linux/workqueue.h>
100 #include <net/af_vsock.h>
102 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
103 static void vsock_sk_destruct(struct sock *sk);
104 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
106 /* Protocol family. */
107 static struct proto vsock_proto = {
109 .owner = THIS_MODULE,
110 .obj_size = sizeof(struct vsock_sock),
113 /* The default peer timeout indicates how long we will wait for a peer response
114 * to a control message.
116 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
118 static const struct vsock_transport *transport;
119 static DEFINE_MUTEX(vsock_register_mutex);
123 /* Get the ID of the local context. This is transport dependent. */
125 int vm_sockets_get_local_cid(void)
127 return transport->get_local_cid();
129 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
133 /* Each bound VSocket is stored in the bind hash table and each connected
134 * VSocket is stored in the connected hash table.
136 * Unbound sockets are all put on the same list attached to the end of the hash
137 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
138 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
139 * represents the list that addr hashes to).
141 * Specifically, we initialize the vsock_bind_table array to a size of
142 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
143 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
144 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
145 * mods with VSOCK_HASH_SIZE to ensure this.
147 #define VSOCK_HASH_SIZE 251
148 #define MAX_PORT_RETRIES 24
150 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE)
151 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
152 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
154 /* XXX This can probably be implemented in a better way. */
155 #define VSOCK_CONN_HASH(src, dst) \
156 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
157 #define vsock_connected_sockets(src, dst) \
158 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
159 #define vsock_connected_sockets_vsk(vsk) \
160 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
162 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
163 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
164 static DEFINE_SPINLOCK(vsock_table_lock);
166 /* Autobind this socket to the local address if necessary. */
167 static int vsock_auto_bind(struct vsock_sock *vsk)
169 struct sock *sk = sk_vsock(vsk);
170 struct sockaddr_vm local_addr;
172 if (vsock_addr_bound(&vsk->local_addr))
174 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
175 return __vsock_bind(sk, &local_addr);
178 static void vsock_init_tables(void)
182 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
183 INIT_LIST_HEAD(&vsock_bind_table[i]);
185 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
186 INIT_LIST_HEAD(&vsock_connected_table[i]);
189 static void __vsock_insert_bound(struct list_head *list,
190 struct vsock_sock *vsk)
193 list_add(&vsk->bound_table, list);
196 static void __vsock_insert_connected(struct list_head *list,
197 struct vsock_sock *vsk)
200 list_add(&vsk->connected_table, list);
203 static void __vsock_remove_bound(struct vsock_sock *vsk)
205 list_del_init(&vsk->bound_table);
209 static void __vsock_remove_connected(struct vsock_sock *vsk)
211 list_del_init(&vsk->connected_table);
215 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
217 struct vsock_sock *vsk;
219 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
220 if (addr->svm_port == vsk->local_addr.svm_port)
221 return sk_vsock(vsk);
226 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
227 struct sockaddr_vm *dst)
229 struct vsock_sock *vsk;
231 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
233 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
234 dst->svm_port == vsk->local_addr.svm_port) {
235 return sk_vsock(vsk);
242 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
244 return !list_empty(&vsk->bound_table);
247 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
249 return !list_empty(&vsk->connected_table);
252 static void vsock_insert_unbound(struct vsock_sock *vsk)
254 spin_lock_bh(&vsock_table_lock);
255 __vsock_insert_bound(vsock_unbound_sockets, vsk);
256 spin_unlock_bh(&vsock_table_lock);
259 void vsock_insert_connected(struct vsock_sock *vsk)
261 struct list_head *list = vsock_connected_sockets(
262 &vsk->remote_addr, &vsk->local_addr);
264 spin_lock_bh(&vsock_table_lock);
265 __vsock_insert_connected(list, vsk);
266 spin_unlock_bh(&vsock_table_lock);
268 EXPORT_SYMBOL_GPL(vsock_insert_connected);
270 void vsock_remove_bound(struct vsock_sock *vsk)
272 spin_lock_bh(&vsock_table_lock);
273 __vsock_remove_bound(vsk);
274 spin_unlock_bh(&vsock_table_lock);
276 EXPORT_SYMBOL_GPL(vsock_remove_bound);
278 void vsock_remove_connected(struct vsock_sock *vsk)
280 spin_lock_bh(&vsock_table_lock);
281 __vsock_remove_connected(vsk);
282 spin_unlock_bh(&vsock_table_lock);
284 EXPORT_SYMBOL_GPL(vsock_remove_connected);
286 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
290 spin_lock_bh(&vsock_table_lock);
291 sk = __vsock_find_bound_socket(addr);
295 spin_unlock_bh(&vsock_table_lock);
299 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
301 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
302 struct sockaddr_vm *dst)
306 spin_lock_bh(&vsock_table_lock);
307 sk = __vsock_find_connected_socket(src, dst);
311 spin_unlock_bh(&vsock_table_lock);
315 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
317 static bool vsock_in_bound_table(struct vsock_sock *vsk)
321 spin_lock_bh(&vsock_table_lock);
322 ret = __vsock_in_bound_table(vsk);
323 spin_unlock_bh(&vsock_table_lock);
328 static bool vsock_in_connected_table(struct vsock_sock *vsk)
332 spin_lock_bh(&vsock_table_lock);
333 ret = __vsock_in_connected_table(vsk);
334 spin_unlock_bh(&vsock_table_lock);
339 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
343 spin_lock_bh(&vsock_table_lock);
345 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
346 struct vsock_sock *vsk;
347 list_for_each_entry(vsk, &vsock_connected_table[i],
352 spin_unlock_bh(&vsock_table_lock);
354 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
356 void vsock_add_pending(struct sock *listener, struct sock *pending)
358 struct vsock_sock *vlistener;
359 struct vsock_sock *vpending;
361 vlistener = vsock_sk(listener);
362 vpending = vsock_sk(pending);
366 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
368 EXPORT_SYMBOL_GPL(vsock_add_pending);
370 void vsock_remove_pending(struct sock *listener, struct sock *pending)
372 struct vsock_sock *vpending = vsock_sk(pending);
374 list_del_init(&vpending->pending_links);
378 EXPORT_SYMBOL_GPL(vsock_remove_pending);
380 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
382 struct vsock_sock *vlistener;
383 struct vsock_sock *vconnected;
385 vlistener = vsock_sk(listener);
386 vconnected = vsock_sk(connected);
388 sock_hold(connected);
390 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
392 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
394 static struct sock *vsock_dequeue_accept(struct sock *listener)
396 struct vsock_sock *vlistener;
397 struct vsock_sock *vconnected;
399 vlistener = vsock_sk(listener);
401 if (list_empty(&vlistener->accept_queue))
404 vconnected = list_entry(vlistener->accept_queue.next,
405 struct vsock_sock, accept_queue);
407 list_del_init(&vconnected->accept_queue);
409 /* The caller will need a reference on the connected socket so we let
410 * it call sock_put().
413 return sk_vsock(vconnected);
416 static bool vsock_is_accept_queue_empty(struct sock *sk)
418 struct vsock_sock *vsk = vsock_sk(sk);
419 return list_empty(&vsk->accept_queue);
422 static bool vsock_is_pending(struct sock *sk)
424 struct vsock_sock *vsk = vsock_sk(sk);
425 return !list_empty(&vsk->pending_links);
428 static int vsock_send_shutdown(struct sock *sk, int mode)
430 return transport->shutdown(vsock_sk(sk), mode);
433 void vsock_pending_work(struct work_struct *work)
436 struct sock *listener;
437 struct vsock_sock *vsk;
440 vsk = container_of(work, struct vsock_sock, dwork.work);
442 listener = vsk->listener;
448 if (vsock_is_pending(sk)) {
449 vsock_remove_pending(listener, sk);
450 } else if (!vsk->rejected) {
451 /* We are not on the pending list and accept() did not reject
452 * us, so we must have been accepted by our user process. We
453 * just need to drop our references to the sockets and be on
460 listener->sk_ack_backlog--;
462 /* We need to remove ourself from the global connected sockets list so
463 * incoming packets can't find this socket, and to reduce the reference
466 if (vsock_in_connected_table(vsk))
467 vsock_remove_connected(vsk);
469 sk->sk_state = SS_FREE;
473 release_sock(listener);
480 EXPORT_SYMBOL_GPL(vsock_pending_work);
482 /**** SOCKET OPERATIONS ****/
484 static int __vsock_bind_stream(struct vsock_sock *vsk,
485 struct sockaddr_vm *addr)
487 static u32 port = LAST_RESERVED_PORT + 1;
488 struct sockaddr_vm new_addr;
490 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
492 if (addr->svm_port == VMADDR_PORT_ANY) {
496 for (i = 0; i < MAX_PORT_RETRIES; i++) {
497 if (port <= LAST_RESERVED_PORT)
498 port = LAST_RESERVED_PORT + 1;
500 new_addr.svm_port = port++;
502 if (!__vsock_find_bound_socket(&new_addr)) {
509 return -EADDRNOTAVAIL;
511 /* If port is in reserved range, ensure caller
512 * has necessary privileges.
514 if (addr->svm_port <= LAST_RESERVED_PORT &&
515 !capable(CAP_NET_BIND_SERVICE)) {
519 if (__vsock_find_bound_socket(&new_addr))
523 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
525 /* Remove stream sockets from the unbound list and add them to the hash
526 * table for easy lookup by its address. The unbound list is simply an
527 * extra entry at the end of the hash table, a trick used by AF_UNIX.
529 __vsock_remove_bound(vsk);
530 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
535 static int __vsock_bind_dgram(struct vsock_sock *vsk,
536 struct sockaddr_vm *addr)
538 return transport->dgram_bind(vsk, addr);
541 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
543 struct vsock_sock *vsk = vsock_sk(sk);
547 /* First ensure this socket isn't already bound. */
548 if (vsock_addr_bound(&vsk->local_addr))
551 /* Now bind to the provided address or select appropriate values if
552 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
553 * like AF_INET prevents binding to a non-local IP address (in most
554 * cases), we only allow binding to the local CID.
556 cid = transport->get_local_cid();
557 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
558 return -EADDRNOTAVAIL;
560 switch (sk->sk_socket->type) {
562 spin_lock_bh(&vsock_table_lock);
563 retval = __vsock_bind_stream(vsk, addr);
564 spin_unlock_bh(&vsock_table_lock);
568 retval = __vsock_bind_dgram(vsk, addr);
579 struct sock *__vsock_create(struct net *net,
587 struct vsock_sock *psk;
588 struct vsock_sock *vsk;
590 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
594 sock_init_data(sock, sk);
596 /* sk->sk_type is normally set in sock_init_data, but only if sock is
597 * non-NULL. We make sure that our sockets always have a type by
598 * setting it here if needed.
604 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
605 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
607 sk->sk_destruct = vsock_sk_destruct;
608 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
610 sock_reset_flag(sk, SOCK_DONE);
612 INIT_LIST_HEAD(&vsk->bound_table);
613 INIT_LIST_HEAD(&vsk->connected_table);
614 vsk->listener = NULL;
615 INIT_LIST_HEAD(&vsk->pending_links);
616 INIT_LIST_HEAD(&vsk->accept_queue);
617 vsk->rejected = false;
618 vsk->sent_request = false;
619 vsk->ignore_connecting_rst = false;
620 vsk->peer_shutdown = 0;
622 psk = parent ? vsock_sk(parent) : NULL;
624 vsk->trusted = psk->trusted;
625 vsk->owner = get_cred(psk->owner);
626 vsk->connect_timeout = psk->connect_timeout;
628 vsk->trusted = capable(CAP_NET_ADMIN);
629 vsk->owner = get_current_cred();
630 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
633 if (transport->init(vsk, psk) < 0) {
639 vsock_insert_unbound(vsk);
643 EXPORT_SYMBOL_GPL(__vsock_create);
645 static void __vsock_release(struct sock *sk)
649 struct sock *pending;
650 struct vsock_sock *vsk;
653 pending = NULL; /* Compiler warning. */
655 if (vsock_in_bound_table(vsk))
656 vsock_remove_bound(vsk);
658 if (vsock_in_connected_table(vsk))
659 vsock_remove_connected(vsk);
661 transport->release(vsk);
665 sk->sk_shutdown = SHUTDOWN_MASK;
667 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
670 /* Clean up any sockets that never were accepted. */
671 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
672 __vsock_release(pending);
681 static void vsock_sk_destruct(struct sock *sk)
683 struct vsock_sock *vsk = vsock_sk(sk);
685 transport->destruct(vsk);
687 /* When clearing these addresses, there's no need to set the family and
688 * possibly register the address family with the kernel.
690 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
691 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
693 put_cred(vsk->owner);
696 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
700 err = sock_queue_rcv_skb(sk, skb);
707 s64 vsock_stream_has_data(struct vsock_sock *vsk)
709 return transport->stream_has_data(vsk);
711 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
713 s64 vsock_stream_has_space(struct vsock_sock *vsk)
715 return transport->stream_has_space(vsk);
717 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
719 static int vsock_release(struct socket *sock)
721 __vsock_release(sock->sk);
723 sock->state = SS_FREE;
729 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
733 struct sockaddr_vm *vm_addr;
737 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
741 err = __vsock_bind(sk, vm_addr);
747 static int vsock_getname(struct socket *sock,
748 struct sockaddr *addr, int *addr_len, int peer)
752 struct vsock_sock *vsk;
753 struct sockaddr_vm *vm_addr;
762 if (sock->state != SS_CONNECTED) {
766 vm_addr = &vsk->remote_addr;
768 vm_addr = &vsk->local_addr;
776 /* sys_getsockname() and sys_getpeername() pass us a
777 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
778 * that macro is defined in socket.c instead of .h, so we hardcode its
781 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
782 memcpy(addr, vm_addr, sizeof(*vm_addr));
783 *addr_len = sizeof(*vm_addr);
790 static int vsock_shutdown(struct socket *sock, int mode)
795 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
796 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
797 * here like the other address families do. Note also that the
798 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
799 * which is what we want.
803 if ((mode & ~SHUTDOWN_MASK) || !mode)
806 /* If this is a STREAM socket and it is not connected then bail out
807 * immediately. If it is a DGRAM socket then we must first kick the
808 * socket so that it wakes up from any sleeping calls, for example
809 * recv(), and then afterwards return the error.
813 if (sock->state == SS_UNCONNECTED) {
815 if (sk->sk_type == SOCK_STREAM)
818 sock->state = SS_DISCONNECTING;
822 /* Receive and send shutdowns are treated alike. */
823 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
826 sk->sk_shutdown |= mode;
827 sk->sk_state_change(sk);
830 if (sk->sk_type == SOCK_STREAM) {
831 sock_reset_flag(sk, SOCK_DONE);
832 vsock_send_shutdown(sk, mode);
839 static unsigned int vsock_poll(struct file *file, struct socket *sock,
844 struct vsock_sock *vsk;
849 poll_wait(file, sk_sleep(sk), wait);
853 /* Signify that there has been an error on this socket. */
856 /* INET sockets treat local write shutdown and peer write shutdown as a
857 * case of POLLHUP set.
859 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
860 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
861 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
865 if (sk->sk_shutdown & RCV_SHUTDOWN ||
866 vsk->peer_shutdown & SEND_SHUTDOWN) {
870 if (sock->type == SOCK_DGRAM) {
871 /* For datagram sockets we can read if there is something in
872 * the queue and write as long as the socket isn't shutdown for
875 if (!skb_queue_empty(&sk->sk_receive_queue) ||
876 (sk->sk_shutdown & RCV_SHUTDOWN)) {
877 mask |= POLLIN | POLLRDNORM;
880 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
881 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
883 } else if (sock->type == SOCK_STREAM) {
886 /* Listening sockets that have connections in their accept
889 if (sk->sk_state == VSOCK_SS_LISTEN
890 && !vsock_is_accept_queue_empty(sk))
891 mask |= POLLIN | POLLRDNORM;
893 /* If there is something in the queue then we can read. */
894 if (transport->stream_is_active(vsk) &&
895 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
896 bool data_ready_now = false;
897 int ret = transport->notify_poll_in(
898 vsk, 1, &data_ready_now);
903 mask |= POLLIN | POLLRDNORM;
908 /* Sockets whose connections have been closed, reset, or
909 * terminated should also be considered read, and we check the
910 * shutdown flag for that.
912 if (sk->sk_shutdown & RCV_SHUTDOWN ||
913 vsk->peer_shutdown & SEND_SHUTDOWN) {
914 mask |= POLLIN | POLLRDNORM;
917 /* Connected sockets that can produce data can be written. */
918 if (sk->sk_state == SS_CONNECTED) {
919 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
920 bool space_avail_now = false;
921 int ret = transport->notify_poll_out(
922 vsk, 1, &space_avail_now);
927 /* Remove POLLWRBAND since INET
928 * sockets are not setting it.
930 mask |= POLLOUT | POLLWRNORM;
936 /* Simulate INET socket poll behaviors, which sets
937 * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
938 * but local send is not shutdown.
940 if (sk->sk_state == SS_UNCONNECTED) {
941 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
942 mask |= POLLOUT | POLLWRNORM;
952 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
957 struct vsock_sock *vsk;
958 struct sockaddr_vm *remote_addr;
960 if (msg->msg_flags & MSG_OOB)
963 /* For now, MSG_DONTWAIT is always assumed... */
970 err = vsock_auto_bind(vsk);
975 /* If the provided message contains an address, use that. Otherwise
976 * fall back on the socket's remote handle (if it has been connected).
979 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
980 &remote_addr) == 0) {
981 /* Ensure this address is of the right type and is a valid
985 if (remote_addr->svm_cid == VMADDR_CID_ANY)
986 remote_addr->svm_cid = transport->get_local_cid();
988 if (!vsock_addr_bound(remote_addr)) {
992 } else if (sock->state == SS_CONNECTED) {
993 remote_addr = &vsk->remote_addr;
995 if (remote_addr->svm_cid == VMADDR_CID_ANY)
996 remote_addr->svm_cid = transport->get_local_cid();
998 /* XXX Should connect() or this function ensure remote_addr is
1001 if (!vsock_addr_bound(&vsk->remote_addr)) {
1010 if (!transport->dgram_allow(remote_addr->svm_cid,
1011 remote_addr->svm_port)) {
1016 err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1023 static int vsock_dgram_connect(struct socket *sock,
1024 struct sockaddr *addr, int addr_len, int flags)
1028 struct vsock_sock *vsk;
1029 struct sockaddr_vm *remote_addr;
1034 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1035 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1037 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1039 sock->state = SS_UNCONNECTED;
1042 } else if (err != 0)
1047 err = vsock_auto_bind(vsk);
1051 if (!transport->dgram_allow(remote_addr->svm_cid,
1052 remote_addr->svm_port)) {
1057 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1058 sock->state = SS_CONNECTED;
1065 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1066 size_t len, int flags)
1068 return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1071 static const struct proto_ops vsock_dgram_ops = {
1073 .owner = THIS_MODULE,
1074 .release = vsock_release,
1076 .connect = vsock_dgram_connect,
1077 .socketpair = sock_no_socketpair,
1078 .accept = sock_no_accept,
1079 .getname = vsock_getname,
1081 .ioctl = sock_no_ioctl,
1082 .listen = sock_no_listen,
1083 .shutdown = vsock_shutdown,
1084 .setsockopt = sock_no_setsockopt,
1085 .getsockopt = sock_no_getsockopt,
1086 .sendmsg = vsock_dgram_sendmsg,
1087 .recvmsg = vsock_dgram_recvmsg,
1088 .mmap = sock_no_mmap,
1089 .sendpage = sock_no_sendpage,
1092 static void vsock_connect_timeout(struct work_struct *work)
1095 struct vsock_sock *vsk;
1097 vsk = container_of(work, struct vsock_sock, dwork.work);
1101 if (sk->sk_state == SS_CONNECTING &&
1102 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1103 sk->sk_state = SS_UNCONNECTED;
1104 sk->sk_err = ETIMEDOUT;
1105 sk->sk_error_report(sk);
1112 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1113 int addr_len, int flags)
1117 struct vsock_sock *vsk;
1118 struct sockaddr_vm *remote_addr;
1128 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1129 switch (sock->state) {
1133 case SS_DISCONNECTING:
1137 /* This continues on so we can move sock into the SS_CONNECTED
1138 * state once the connection has completed (at which point err
1139 * will be set to zero also). Otherwise, we will either wait
1140 * for the connection or return -EALREADY should this be a
1141 * non-blocking call.
1146 if ((sk->sk_state == VSOCK_SS_LISTEN) ||
1147 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1152 /* The hypervisor and well-known contexts do not have socket
1155 if (!transport->stream_allow(remote_addr->svm_cid,
1156 remote_addr->svm_port)) {
1161 /* Set the remote address that we are connecting to. */
1162 memcpy(&vsk->remote_addr, remote_addr,
1163 sizeof(vsk->remote_addr));
1165 err = vsock_auto_bind(vsk);
1169 sk->sk_state = SS_CONNECTING;
1171 err = transport->connect(vsk);
1175 /* Mark sock as connecting and set the error code to in
1176 * progress in case this is a non-blocking connect.
1178 sock->state = SS_CONNECTING;
1182 /* The receive path will handle all communication until we are able to
1183 * enter the connected state. Here we wait for the connection to be
1184 * completed or a notification of an error.
1186 timeout = vsk->connect_timeout;
1187 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1189 while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1190 if (flags & O_NONBLOCK) {
1191 /* If we're not going to block, we schedule a timeout
1192 * function to generate a timeout on the connection
1193 * attempt, in case the peer doesn't respond in a
1194 * timely manner. We hold on to the socket until the
1198 INIT_DELAYED_WORK(&vsk->dwork,
1199 vsock_connect_timeout);
1200 schedule_delayed_work(&vsk->dwork, timeout);
1202 /* Skip ahead to preserve error code set above. */
1207 timeout = schedule_timeout(timeout);
1210 if (signal_pending(current)) {
1211 err = sock_intr_errno(timeout);
1212 sk->sk_state = SS_UNCONNECTED;
1213 sock->state = SS_UNCONNECTED;
1215 } else if (timeout == 0) {
1217 sk->sk_state = SS_UNCONNECTED;
1218 sock->state = SS_UNCONNECTED;
1222 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1227 sk->sk_state = SS_UNCONNECTED;
1228 sock->state = SS_UNCONNECTED;
1234 finish_wait(sk_sleep(sk), &wait);
1240 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1242 struct sock *listener;
1244 struct sock *connected;
1245 struct vsock_sock *vconnected;
1250 listener = sock->sk;
1252 lock_sock(listener);
1254 if (sock->type != SOCK_STREAM) {
1259 if (listener->sk_state != VSOCK_SS_LISTEN) {
1264 /* Wait for children sockets to appear; these are the new sockets
1265 * created upon connection establishment.
1267 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1268 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1270 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1271 listener->sk_err == 0) {
1272 release_sock(listener);
1273 timeout = schedule_timeout(timeout);
1274 finish_wait(sk_sleep(listener), &wait);
1275 lock_sock(listener);
1277 if (signal_pending(current)) {
1278 err = sock_intr_errno(timeout);
1280 } else if (timeout == 0) {
1285 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1287 finish_wait(sk_sleep(listener), &wait);
1289 if (listener->sk_err)
1290 err = -listener->sk_err;
1293 listener->sk_ack_backlog--;
1295 lock_sock(connected);
1296 vconnected = vsock_sk(connected);
1298 /* If the listener socket has received an error, then we should
1299 * reject this socket and return. Note that we simply mark the
1300 * socket rejected, drop our reference, and let the cleanup
1301 * function handle the cleanup; the fact that we found it in
1302 * the listener's accept queue guarantees that the cleanup
1303 * function hasn't run yet.
1306 vconnected->rejected = true;
1308 newsock->state = SS_CONNECTED;
1309 sock_graft(connected, newsock);
1312 release_sock(connected);
1313 sock_put(connected);
1317 release_sock(listener);
1321 static int vsock_listen(struct socket *sock, int backlog)
1325 struct vsock_sock *vsk;
1331 if (sock->type != SOCK_STREAM) {
1336 if (sock->state != SS_UNCONNECTED) {
1343 if (!vsock_addr_bound(&vsk->local_addr)) {
1348 sk->sk_max_ack_backlog = backlog;
1349 sk->sk_state = VSOCK_SS_LISTEN;
1358 static int vsock_stream_setsockopt(struct socket *sock,
1361 char __user *optval,
1362 unsigned int optlen)
1366 struct vsock_sock *vsk;
1369 if (level != AF_VSOCK)
1370 return -ENOPROTOOPT;
1372 #define COPY_IN(_v) \
1374 if (optlen < sizeof(_v)) { \
1378 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \
1391 case SO_VM_SOCKETS_BUFFER_SIZE:
1393 transport->set_buffer_size(vsk, val);
1396 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1398 transport->set_max_buffer_size(vsk, val);
1401 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1403 transport->set_min_buffer_size(vsk, val);
1406 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1409 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1410 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1411 vsk->connect_timeout = tv.tv_sec * HZ +
1412 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1413 if (vsk->connect_timeout == 0)
1414 vsk->connect_timeout =
1415 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1435 static int vsock_stream_getsockopt(struct socket *sock,
1436 int level, int optname,
1437 char __user *optval,
1443 struct vsock_sock *vsk;
1446 if (level != AF_VSOCK)
1447 return -ENOPROTOOPT;
1449 err = get_user(len, optlen);
1453 #define COPY_OUT(_v) \
1455 if (len < sizeof(_v)) \
1459 if (copy_to_user(optval, &_v, len) != 0) \
1469 case SO_VM_SOCKETS_BUFFER_SIZE:
1470 val = transport->get_buffer_size(vsk);
1474 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1475 val = transport->get_max_buffer_size(vsk);
1479 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1480 val = transport->get_min_buffer_size(vsk);
1484 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1486 tv.tv_sec = vsk->connect_timeout / HZ;
1488 (vsk->connect_timeout -
1489 tv.tv_sec * HZ) * (1000000 / HZ);
1494 return -ENOPROTOOPT;
1497 err = put_user(len, optlen);
1506 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1510 struct vsock_sock *vsk;
1511 ssize_t total_written;
1514 struct vsock_transport_send_notify_data send_data;
1523 if (msg->msg_flags & MSG_OOB)
1528 /* Callers should not provide a destination with stream sockets. */
1529 if (msg->msg_namelen) {
1530 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1534 /* Send data only if both sides are not shutdown in the direction. */
1535 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1536 vsk->peer_shutdown & RCV_SHUTDOWN) {
1541 if (sk->sk_state != SS_CONNECTED ||
1542 !vsock_addr_bound(&vsk->local_addr)) {
1547 if (!vsock_addr_bound(&vsk->remote_addr)) {
1548 err = -EDESTADDRREQ;
1552 /* Wait for room in the produce queue to enqueue our user's data. */
1553 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1555 err = transport->notify_send_init(vsk, &send_data);
1560 while (total_written < len) {
1563 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1564 while (vsock_stream_has_space(vsk) == 0 &&
1566 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1567 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1569 /* Don't wait for non-blocking sockets. */
1572 finish_wait(sk_sleep(sk), &wait);
1576 err = transport->notify_send_pre_block(vsk, &send_data);
1578 finish_wait(sk_sleep(sk), &wait);
1583 timeout = schedule_timeout(timeout);
1585 if (signal_pending(current)) {
1586 err = sock_intr_errno(timeout);
1587 finish_wait(sk_sleep(sk), &wait);
1589 } else if (timeout == 0) {
1591 finish_wait(sk_sleep(sk), &wait);
1595 prepare_to_wait(sk_sleep(sk), &wait,
1596 TASK_INTERRUPTIBLE);
1598 finish_wait(sk_sleep(sk), &wait);
1600 /* These checks occur both as part of and after the loop
1601 * conditional since we need to check before and after
1607 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1608 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1613 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1617 /* Note that enqueue will only write as many bytes as are free
1618 * in the produce queue, so we don't need to ensure len is
1619 * smaller than the queue size. It is the caller's
1620 * responsibility to check how many bytes we were able to send.
1623 written = transport->stream_enqueue(
1625 len - total_written);
1631 total_written += written;
1633 err = transport->notify_send_post_enqueue(
1634 vsk, written, &send_data);
1641 if (total_written > 0)
1642 err = total_written;
1650 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1654 struct vsock_sock *vsk;
1659 struct vsock_transport_recv_notify_data recv_data;
1669 if (sk->sk_state != SS_CONNECTED) {
1670 /* Recvmsg is supposed to return 0 if a peer performs an
1671 * orderly shutdown. Differentiate between that case and when a
1672 * peer has not connected or a local shutdown occured with the
1675 if (sock_flag(sk, SOCK_DONE))
1683 if (flags & MSG_OOB) {
1688 /* We don't check peer_shutdown flag here since peer may actually shut
1689 * down, but there can be data in the queue that a local socket can
1692 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1697 /* It is valid on Linux to pass in a zero-length receive buffer. This
1698 * is not an error. We may as well bail out now.
1705 /* We must not copy less than target bytes into the user's buffer
1706 * before returning successfully, so we wait for the consume queue to
1707 * have that much data to consume before dequeueing. Note that this
1708 * makes it impossible to handle cases where target is greater than the
1711 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1712 if (target >= transport->stream_rcvhiwat(vsk)) {
1716 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1719 err = transport->notify_recv_init(vsk, target, &recv_data);
1727 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1728 ready = vsock_stream_has_data(vsk);
1731 if (sk->sk_err != 0 ||
1732 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1733 (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1734 finish_wait(sk_sleep(sk), &wait);
1737 /* Don't wait for non-blocking sockets. */
1740 finish_wait(sk_sleep(sk), &wait);
1744 err = transport->notify_recv_pre_block(
1745 vsk, target, &recv_data);
1747 finish_wait(sk_sleep(sk), &wait);
1751 timeout = schedule_timeout(timeout);
1754 if (signal_pending(current)) {
1755 err = sock_intr_errno(timeout);
1756 finish_wait(sk_sleep(sk), &wait);
1758 } else if (timeout == 0) {
1760 finish_wait(sk_sleep(sk), &wait);
1766 finish_wait(sk_sleep(sk), &wait);
1769 /* Invalid queue pair content. XXX This should
1770 * be changed to a connection reset in a later
1778 err = transport->notify_recv_pre_dequeue(
1779 vsk, target, &recv_data);
1783 read = transport->stream_dequeue(
1785 len - copied, flags);
1793 err = transport->notify_recv_post_dequeue(
1795 !(flags & MSG_PEEK), &recv_data);
1799 if (read >= target || flags & MSG_PEEK)
1808 else if (sk->sk_shutdown & RCV_SHUTDOWN)
1819 static const struct proto_ops vsock_stream_ops = {
1821 .owner = THIS_MODULE,
1822 .release = vsock_release,
1824 .connect = vsock_stream_connect,
1825 .socketpair = sock_no_socketpair,
1826 .accept = vsock_accept,
1827 .getname = vsock_getname,
1829 .ioctl = sock_no_ioctl,
1830 .listen = vsock_listen,
1831 .shutdown = vsock_shutdown,
1832 .setsockopt = vsock_stream_setsockopt,
1833 .getsockopt = vsock_stream_getsockopt,
1834 .sendmsg = vsock_stream_sendmsg,
1835 .recvmsg = vsock_stream_recvmsg,
1836 .mmap = sock_no_mmap,
1837 .sendpage = sock_no_sendpage,
1840 static int vsock_create(struct net *net, struct socket *sock,
1841 int protocol, int kern)
1846 if (protocol && protocol != PF_VSOCK)
1847 return -EPROTONOSUPPORT;
1849 switch (sock->type) {
1851 sock->ops = &vsock_dgram_ops;
1854 sock->ops = &vsock_stream_ops;
1857 return -ESOCKTNOSUPPORT;
1860 sock->state = SS_UNCONNECTED;
1862 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1865 static const struct net_proto_family vsock_family_ops = {
1867 .create = vsock_create,
1868 .owner = THIS_MODULE,
1871 static long vsock_dev_do_ioctl(struct file *filp,
1872 unsigned int cmd, void __user *ptr)
1874 u32 __user *p = ptr;
1878 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1879 if (put_user(transport->get_local_cid(), p) != 0)
1884 pr_err("Unknown ioctl %d\n", cmd);
1891 static long vsock_dev_ioctl(struct file *filp,
1892 unsigned int cmd, unsigned long arg)
1894 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1897 #ifdef CONFIG_COMPAT
1898 static long vsock_dev_compat_ioctl(struct file *filp,
1899 unsigned int cmd, unsigned long arg)
1901 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1905 static const struct file_operations vsock_device_ops = {
1906 .owner = THIS_MODULE,
1907 .unlocked_ioctl = vsock_dev_ioctl,
1908 #ifdef CONFIG_COMPAT
1909 .compat_ioctl = vsock_dev_compat_ioctl,
1911 .open = nonseekable_open,
1914 static struct miscdevice vsock_device = {
1916 .fops = &vsock_device_ops,
1919 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1921 int err = mutex_lock_interruptible(&vsock_register_mutex);
1931 /* Transport must be the owner of the protocol so that it can't
1932 * unload while there are open sockets.
1934 vsock_proto.owner = owner;
1937 vsock_init_tables();
1939 vsock_device.minor = MISC_DYNAMIC_MINOR;
1940 err = misc_register(&vsock_device);
1942 pr_err("Failed to register misc device\n");
1943 goto err_reset_transport;
1946 err = proto_register(&vsock_proto, 1); /* we want our slab */
1948 pr_err("Cannot register vsock protocol\n");
1949 goto err_deregister_misc;
1952 err = sock_register(&vsock_family_ops);
1954 pr_err("could not register af_vsock (%d) address family: %d\n",
1956 goto err_unregister_proto;
1959 mutex_unlock(&vsock_register_mutex);
1962 err_unregister_proto:
1963 proto_unregister(&vsock_proto);
1964 err_deregister_misc:
1965 misc_deregister(&vsock_device);
1966 err_reset_transport:
1969 mutex_unlock(&vsock_register_mutex);
1972 EXPORT_SYMBOL_GPL(__vsock_core_init);
1974 void vsock_core_exit(void)
1976 mutex_lock(&vsock_register_mutex);
1978 misc_deregister(&vsock_device);
1979 sock_unregister(AF_VSOCK);
1980 proto_unregister(&vsock_proto);
1982 /* We do not want the assignment below re-ordered. */
1986 mutex_unlock(&vsock_register_mutex);
1988 EXPORT_SYMBOL_GPL(vsock_core_exit);
1990 MODULE_AUTHOR("VMware, Inc.");
1991 MODULE_DESCRIPTION("VMware Virtual Socket Family");
1992 MODULE_VERSION("1.0.1.0-k");
1993 MODULE_LICENSE("GPL v2");