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 SS_LISTEN state. When a connection
40 * request is received (the second kind of socket mentioned above), we create a
41 * new socket and refer to it as a pending socket. These pending sockets are
42 * placed on the pending connection list of the listener socket. When future
43 * packets are received for the address the listener socket is bound to, we
44 * check if the source of the packet is from one that has an existing pending
45 * connection. If it does, we process the packet for the pending socket. When
46 * that socket reaches the connected state, it is removed from the listener
47 * socket's pending list and enqueued in the listener socket's accept queue.
48 * Callers of accept(2) will accept connected sockets from the listener socket's
49 * accept queue. If the socket cannot be accepted for some reason then it is
50 * marked rejected. Once the connection is accepted, it is owned by the user
51 * process and the responsibility for cleanup falls with that user process.
53 * - It is possible that these pending sockets will never reach the connected
54 * state; in fact, we may never receive another packet after the connection
55 * request. Because of this, we must schedule a cleanup function to run in the
56 * future, after some amount of time passes where a connection should have been
57 * established. This function ensures that the socket is off all lists so it
58 * cannot be retrieved, then drops all references to the socket so it is cleaned
59 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this
60 * function will also cleanup rejected sockets, those that reach the connected
61 * state but leave it before they have been accepted.
63 * - Sockets created by user action will be cleaned up when the user process
64 * calls close(2), causing our release implementation to be called. Our release
65 * implementation will perform some cleanup then drop the last reference so our
66 * sk_destruct implementation is invoked. Our sk_destruct implementation will
67 * perform additional cleanup that's common for both types of sockets.
69 * - A socket's reference count is what ensures that the structure won't be
70 * freed. Each entry in a list (such as the "global" bound and connected tables
71 * and the listener socket's pending list and connected queue) ensures a
72 * reference. When we defer work until process context and pass a socket as our
73 * argument, we must ensure the reference count is increased to ensure the
74 * socket isn't freed before the function is run; the deferred function will
75 * then drop the reference.
78 #include <linux/types.h>
79 #include <linux/bitops.h>
80 #include <linux/cred.h>
81 #include <linux/init.h>
83 #include <linux/kernel.h>
84 #include <linux/kmod.h>
85 #include <linux/list.h>
86 #include <linux/miscdevice.h>
87 #include <linux/module.h>
88 #include <linux/mutex.h>
89 #include <linux/net.h>
90 #include <linux/poll.h>
91 #include <linux/skbuff.h>
92 #include <linux/smp.h>
93 #include <linux/socket.h>
94 #include <linux/stddef.h>
95 #include <linux/unistd.h>
96 #include <linux/wait.h>
97 #include <linux/workqueue.h>
100 #include "af_vsock.h"
101 #include "vsock_version.h"
103 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
104 static void vsock_sk_destruct(struct sock *sk);
105 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
107 /* Protocol family. */
108 static struct proto vsock_proto = {
110 .owner = THIS_MODULE,
111 .obj_size = sizeof(struct vsock_sock),
114 /* The default peer timeout indicates how long we will wait for a peer response
115 * to a control message.
117 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
119 #define SS_LISTEN 255
121 static const struct vsock_transport *transport;
122 static DEFINE_MUTEX(vsock_register_mutex);
126 /* Get the ID of the local context. This is transport dependent. */
128 int vm_sockets_get_local_cid(void)
130 return transport->get_local_cid();
132 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
136 /* Each bound VSocket is stored in the bind hash table and each connected
137 * VSocket is stored in the connected hash table.
139 * Unbound sockets are all put on the same list attached to the end of the hash
140 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
141 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
142 * represents the list that addr hashes to).
144 * Specifically, we initialize the vsock_bind_table array to a size of
145 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
146 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
147 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
148 * mods with VSOCK_HASH_SIZE - 1 to ensure this.
150 #define VSOCK_HASH_SIZE 251
151 #define MAX_PORT_RETRIES 24
153 #define VSOCK_HASH(addr) ((addr)->svm_port % (VSOCK_HASH_SIZE - 1))
154 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
155 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
157 /* XXX This can probably be implemented in a better way. */
158 #define VSOCK_CONN_HASH(src, dst) \
159 (((src)->svm_cid ^ (dst)->svm_port) % (VSOCK_HASH_SIZE - 1))
160 #define vsock_connected_sockets(src, dst) \
161 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
162 #define vsock_connected_sockets_vsk(vsk) \
163 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
165 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
166 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
167 static DEFINE_SPINLOCK(vsock_table_lock);
169 static __init void vsock_init_tables(void)
173 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
174 INIT_LIST_HEAD(&vsock_bind_table[i]);
176 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
177 INIT_LIST_HEAD(&vsock_connected_table[i]);
180 static void __vsock_insert_bound(struct list_head *list,
181 struct vsock_sock *vsk)
184 list_add(&vsk->bound_table, list);
187 static void __vsock_insert_connected(struct list_head *list,
188 struct vsock_sock *vsk)
191 list_add(&vsk->connected_table, list);
194 static void __vsock_remove_bound(struct vsock_sock *vsk)
196 list_del_init(&vsk->bound_table);
200 static void __vsock_remove_connected(struct vsock_sock *vsk)
202 list_del_init(&vsk->connected_table);
206 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
208 struct vsock_sock *vsk;
210 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
211 if (vsock_addr_equals_addr_any(addr, &vsk->local_addr))
212 return sk_vsock(vsk);
217 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
218 struct sockaddr_vm *dst)
220 struct vsock_sock *vsk;
222 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
224 if (vsock_addr_equals_addr(src, &vsk->remote_addr)
225 && vsock_addr_equals_addr(dst, &vsk->local_addr)) {
226 return sk_vsock(vsk);
233 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
235 return !list_empty(&vsk->bound_table);
238 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
240 return !list_empty(&vsk->connected_table);
243 static void vsock_insert_unbound(struct vsock_sock *vsk)
245 spin_lock_bh(&vsock_table_lock);
246 __vsock_insert_bound(vsock_unbound_sockets, vsk);
247 spin_unlock_bh(&vsock_table_lock);
250 void vsock_insert_connected(struct vsock_sock *vsk)
252 struct list_head *list = vsock_connected_sockets(
253 &vsk->remote_addr, &vsk->local_addr);
255 spin_lock_bh(&vsock_table_lock);
256 __vsock_insert_connected(list, vsk);
257 spin_unlock_bh(&vsock_table_lock);
259 EXPORT_SYMBOL_GPL(vsock_insert_connected);
261 void vsock_remove_bound(struct vsock_sock *vsk)
263 spin_lock_bh(&vsock_table_lock);
264 __vsock_remove_bound(vsk);
265 spin_unlock_bh(&vsock_table_lock);
267 EXPORT_SYMBOL_GPL(vsock_remove_bound);
269 void vsock_remove_connected(struct vsock_sock *vsk)
271 spin_lock_bh(&vsock_table_lock);
272 __vsock_remove_connected(vsk);
273 spin_unlock_bh(&vsock_table_lock);
275 EXPORT_SYMBOL_GPL(vsock_remove_connected);
277 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
281 spin_lock_bh(&vsock_table_lock);
282 sk = __vsock_find_bound_socket(addr);
286 spin_unlock_bh(&vsock_table_lock);
290 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
292 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
293 struct sockaddr_vm *dst)
297 spin_lock_bh(&vsock_table_lock);
298 sk = __vsock_find_connected_socket(src, dst);
302 spin_unlock_bh(&vsock_table_lock);
306 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
308 static bool vsock_in_bound_table(struct vsock_sock *vsk)
312 spin_lock_bh(&vsock_table_lock);
313 ret = __vsock_in_bound_table(vsk);
314 spin_unlock_bh(&vsock_table_lock);
319 static bool vsock_in_connected_table(struct vsock_sock *vsk)
323 spin_lock_bh(&vsock_table_lock);
324 ret = __vsock_in_connected_table(vsk);
325 spin_unlock_bh(&vsock_table_lock);
330 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
334 spin_lock_bh(&vsock_table_lock);
336 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
337 struct vsock_sock *vsk;
338 list_for_each_entry(vsk, &vsock_connected_table[i],
343 spin_unlock_bh(&vsock_table_lock);
345 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
347 void vsock_add_pending(struct sock *listener, struct sock *pending)
349 struct vsock_sock *vlistener;
350 struct vsock_sock *vpending;
352 vlistener = vsock_sk(listener);
353 vpending = vsock_sk(pending);
357 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
359 EXPORT_SYMBOL_GPL(vsock_add_pending);
361 void vsock_remove_pending(struct sock *listener, struct sock *pending)
363 struct vsock_sock *vpending = vsock_sk(pending);
365 list_del_init(&vpending->pending_links);
369 EXPORT_SYMBOL_GPL(vsock_remove_pending);
371 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
373 struct vsock_sock *vlistener;
374 struct vsock_sock *vconnected;
376 vlistener = vsock_sk(listener);
377 vconnected = vsock_sk(connected);
379 sock_hold(connected);
381 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
383 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
385 static struct sock *vsock_dequeue_accept(struct sock *listener)
387 struct vsock_sock *vlistener;
388 struct vsock_sock *vconnected;
390 vlistener = vsock_sk(listener);
392 if (list_empty(&vlistener->accept_queue))
395 vconnected = list_entry(vlistener->accept_queue.next,
396 struct vsock_sock, accept_queue);
398 list_del_init(&vconnected->accept_queue);
400 /* The caller will need a reference on the connected socket so we let
401 * it call sock_put().
404 return sk_vsock(vconnected);
407 static bool vsock_is_accept_queue_empty(struct sock *sk)
409 struct vsock_sock *vsk = vsock_sk(sk);
410 return list_empty(&vsk->accept_queue);
413 static bool vsock_is_pending(struct sock *sk)
415 struct vsock_sock *vsk = vsock_sk(sk);
416 return !list_empty(&vsk->pending_links);
419 static int vsock_send_shutdown(struct sock *sk, int mode)
421 return transport->shutdown(vsock_sk(sk), mode);
424 void vsock_pending_work(struct work_struct *work)
427 struct sock *listener;
428 struct vsock_sock *vsk;
431 vsk = container_of(work, struct vsock_sock, dwork.work);
433 listener = vsk->listener;
439 if (vsock_is_pending(sk)) {
440 vsock_remove_pending(listener, sk);
441 } else if (!vsk->rejected) {
442 /* We are not on the pending list and accept() did not reject
443 * us, so we must have been accepted by our user process. We
444 * just need to drop our references to the sockets and be on
451 listener->sk_ack_backlog--;
453 /* We need to remove ourself from the global connected sockets list so
454 * incoming packets can't find this socket, and to reduce the reference
457 if (vsock_in_connected_table(vsk))
458 vsock_remove_connected(vsk);
460 sk->sk_state = SS_FREE;
464 release_sock(listener);
471 EXPORT_SYMBOL_GPL(vsock_pending_work);
473 /**** SOCKET OPERATIONS ****/
475 static int __vsock_bind_stream(struct vsock_sock *vsk,
476 struct sockaddr_vm *addr)
478 static u32 port = LAST_RESERVED_PORT + 1;
479 struct sockaddr_vm new_addr;
481 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
483 if (addr->svm_port == VMADDR_PORT_ANY) {
487 for (i = 0; i < MAX_PORT_RETRIES; i++) {
488 if (port <= LAST_RESERVED_PORT)
489 port = LAST_RESERVED_PORT + 1;
491 new_addr.svm_port = port++;
493 if (!__vsock_find_bound_socket(&new_addr)) {
500 return -EADDRNOTAVAIL;
502 /* If port is in reserved range, ensure caller
503 * has necessary privileges.
505 if (addr->svm_port <= LAST_RESERVED_PORT &&
506 !capable(CAP_NET_BIND_SERVICE)) {
510 if (__vsock_find_bound_socket(&new_addr))
514 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
516 /* Remove stream sockets from the unbound list and add them to the hash
517 * table for easy lookup by its address. The unbound list is simply an
518 * extra entry at the end of the hash table, a trick used by AF_UNIX.
520 __vsock_remove_bound(vsk);
521 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
526 static int __vsock_bind_dgram(struct vsock_sock *vsk,
527 struct sockaddr_vm *addr)
529 return transport->dgram_bind(vsk, addr);
532 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
534 struct vsock_sock *vsk = vsock_sk(sk);
538 /* First ensure this socket isn't already bound. */
539 if (vsock_addr_bound(&vsk->local_addr))
542 /* Now bind to the provided address or select appropriate values if
543 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
544 * like AF_INET prevents binding to a non-local IP address (in most
545 * cases), we only allow binding to the local CID.
547 cid = transport->get_local_cid();
548 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
549 return -EADDRNOTAVAIL;
551 switch (sk->sk_socket->type) {
553 spin_lock_bh(&vsock_table_lock);
554 retval = __vsock_bind_stream(vsk, addr);
555 spin_unlock_bh(&vsock_table_lock);
559 retval = __vsock_bind_dgram(vsk, addr);
570 struct sock *__vsock_create(struct net *net,
577 struct vsock_sock *psk;
578 struct vsock_sock *vsk;
580 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto);
584 sock_init_data(sock, sk);
586 /* sk->sk_type is normally set in sock_init_data, but only if sock is
587 * non-NULL. We make sure that our sockets always have a type by
588 * setting it here if needed.
594 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
595 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
597 sk->sk_destruct = vsock_sk_destruct;
598 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
600 sock_reset_flag(sk, SOCK_DONE);
602 INIT_LIST_HEAD(&vsk->bound_table);
603 INIT_LIST_HEAD(&vsk->connected_table);
604 vsk->listener = NULL;
605 INIT_LIST_HEAD(&vsk->pending_links);
606 INIT_LIST_HEAD(&vsk->accept_queue);
607 vsk->rejected = false;
608 vsk->sent_request = false;
609 vsk->ignore_connecting_rst = false;
610 vsk->peer_shutdown = 0;
612 psk = parent ? vsock_sk(parent) : NULL;
614 vsk->trusted = psk->trusted;
615 vsk->owner = get_cred(psk->owner);
616 vsk->connect_timeout = psk->connect_timeout;
618 vsk->trusted = capable(CAP_NET_ADMIN);
619 vsk->owner = get_current_cred();
620 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
623 if (transport->init(vsk, psk) < 0) {
629 vsock_insert_unbound(vsk);
633 EXPORT_SYMBOL_GPL(__vsock_create);
635 static void __vsock_release(struct sock *sk)
639 struct sock *pending;
640 struct vsock_sock *vsk;
643 pending = NULL; /* Compiler warning. */
645 if (vsock_in_bound_table(vsk))
646 vsock_remove_bound(vsk);
648 if (vsock_in_connected_table(vsk))
649 vsock_remove_connected(vsk);
651 transport->release(vsk);
655 sk->sk_shutdown = SHUTDOWN_MASK;
657 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
660 /* Clean up any sockets that never were accepted. */
661 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
662 __vsock_release(pending);
671 static void vsock_sk_destruct(struct sock *sk)
673 struct vsock_sock *vsk = vsock_sk(sk);
675 transport->destruct(vsk);
677 /* When clearing these addresses, there's no need to set the family and
678 * possibly register the address family with the kernel.
680 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
681 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
683 put_cred(vsk->owner);
686 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
690 err = sock_queue_rcv_skb(sk, skb);
697 s64 vsock_stream_has_data(struct vsock_sock *vsk)
699 return transport->stream_has_data(vsk);
701 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
703 s64 vsock_stream_has_space(struct vsock_sock *vsk)
705 return transport->stream_has_space(vsk);
707 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
709 static int vsock_release(struct socket *sock)
711 __vsock_release(sock->sk);
713 sock->state = SS_FREE;
719 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
723 struct sockaddr_vm *vm_addr;
727 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
731 err = __vsock_bind(sk, vm_addr);
737 static int vsock_getname(struct socket *sock,
738 struct sockaddr *addr, int *addr_len, int peer)
742 struct vsock_sock *vsk;
743 struct sockaddr_vm *vm_addr;
752 if (sock->state != SS_CONNECTED) {
756 vm_addr = &vsk->remote_addr;
758 vm_addr = &vsk->local_addr;
766 /* sys_getsockname() and sys_getpeername() pass us a
767 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
768 * that macro is defined in socket.c instead of .h, so we hardcode its
771 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
772 memcpy(addr, vm_addr, sizeof(*vm_addr));
773 *addr_len = sizeof(*vm_addr);
780 static int vsock_shutdown(struct socket *sock, int mode)
785 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
786 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
787 * here like the other address families do. Note also that the
788 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
789 * which is what we want.
793 if ((mode & ~SHUTDOWN_MASK) || !mode)
796 /* If this is a STREAM socket and it is not connected then bail out
797 * immediately. If it is a DGRAM socket then we must first kick the
798 * socket so that it wakes up from any sleeping calls, for example
799 * recv(), and then afterwards return the error.
803 if (sock->state == SS_UNCONNECTED) {
805 if (sk->sk_type == SOCK_STREAM)
808 sock->state = SS_DISCONNECTING;
812 /* Receive and send shutdowns are treated alike. */
813 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
816 sk->sk_shutdown |= mode;
817 sk->sk_state_change(sk);
820 if (sk->sk_type == SOCK_STREAM) {
821 sock_reset_flag(sk, SOCK_DONE);
822 vsock_send_shutdown(sk, mode);
829 static unsigned int vsock_poll(struct file *file, struct socket *sock,
834 struct vsock_sock *vsk;
839 poll_wait(file, sk_sleep(sk), wait);
843 /* Signify that there has been an error on this socket. */
846 /* INET sockets treat local write shutdown and peer write shutdown as a
847 * case of POLLHUP set.
849 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
850 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
851 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
855 if (sk->sk_shutdown & RCV_SHUTDOWN ||
856 vsk->peer_shutdown & SEND_SHUTDOWN) {
860 if (sock->type == SOCK_DGRAM) {
861 /* For datagram sockets we can read if there is something in
862 * the queue and write as long as the socket isn't shutdown for
865 if (!skb_queue_empty(&sk->sk_receive_queue) ||
866 (sk->sk_shutdown & RCV_SHUTDOWN)) {
867 mask |= POLLIN | POLLRDNORM;
870 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
871 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
873 } else if (sock->type == SOCK_STREAM) {
876 /* Listening sockets that have connections in their accept
879 if (sk->sk_state == SS_LISTEN
880 && !vsock_is_accept_queue_empty(sk))
881 mask |= POLLIN | POLLRDNORM;
883 /* If there is something in the queue then we can read. */
884 if (transport->stream_is_active(vsk) &&
885 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
886 bool data_ready_now = false;
887 int ret = transport->notify_poll_in(
888 vsk, 1, &data_ready_now);
893 mask |= POLLIN | POLLRDNORM;
898 /* Sockets whose connections have been closed, reset, or
899 * terminated should also be considered read, and we check the
900 * shutdown flag for that.
902 if (sk->sk_shutdown & RCV_SHUTDOWN ||
903 vsk->peer_shutdown & SEND_SHUTDOWN) {
904 mask |= POLLIN | POLLRDNORM;
907 /* Connected sockets that can produce data can be written. */
908 if (sk->sk_state == SS_CONNECTED) {
909 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
910 bool space_avail_now = false;
911 int ret = transport->notify_poll_out(
912 vsk, 1, &space_avail_now);
917 /* Remove POLLWRBAND since INET
918 * sockets are not setting it.
920 mask |= POLLOUT | POLLWRNORM;
926 /* Simulate INET socket poll behaviors, which sets
927 * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
928 * but local send is not shutdown.
930 if (sk->sk_state == SS_UNCONNECTED) {
931 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
932 mask |= POLLOUT | POLLWRNORM;
942 static int vsock_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
943 struct msghdr *msg, size_t len)
947 struct vsock_sock *vsk;
948 struct sockaddr_vm *remote_addr;
950 if (msg->msg_flags & MSG_OOB)
953 /* For now, MSG_DONTWAIT is always assumed... */
960 if (!vsock_addr_bound(&vsk->local_addr)) {
961 struct sockaddr_vm local_addr;
963 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
964 err = __vsock_bind(sk, &local_addr);
970 /* If the provided message contains an address, use that. Otherwise
971 * fall back on the socket's remote handle (if it has been connected).
974 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
975 &remote_addr) == 0) {
976 /* Ensure this address is of the right type and is a valid
980 if (remote_addr->svm_cid == VMADDR_CID_ANY)
981 remote_addr->svm_cid = transport->get_local_cid();
983 if (!vsock_addr_bound(remote_addr)) {
987 } else if (sock->state == SS_CONNECTED) {
988 remote_addr = &vsk->remote_addr;
990 if (remote_addr->svm_cid == VMADDR_CID_ANY)
991 remote_addr->svm_cid = transport->get_local_cid();
993 /* XXX Should connect() or this function ensure remote_addr is
996 if (!vsock_addr_bound(&vsk->remote_addr)) {
1005 if (!transport->dgram_allow(remote_addr->svm_cid,
1006 remote_addr->svm_port)) {
1011 err = transport->dgram_enqueue(vsk, remote_addr, msg->msg_iov, len);
1018 static int vsock_dgram_connect(struct socket *sock,
1019 struct sockaddr *addr, int addr_len, int flags)
1023 struct vsock_sock *vsk;
1024 struct sockaddr_vm *remote_addr;
1029 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1030 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1032 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1034 sock->state = SS_UNCONNECTED;
1037 } else if (err != 0)
1042 if (!vsock_addr_bound(&vsk->local_addr)) {
1043 struct sockaddr_vm local_addr;
1045 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
1046 err = __vsock_bind(sk, &local_addr);
1052 if (!transport->dgram_allow(remote_addr->svm_cid,
1053 remote_addr->svm_port)) {
1058 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1059 sock->state = SS_CONNECTED;
1066 static int vsock_dgram_recvmsg(struct kiocb *kiocb, struct socket *sock,
1067 struct msghdr *msg, size_t len, int flags)
1069 return transport->dgram_dequeue(kiocb, vsock_sk(sock->sk), msg, len,
1073 static const struct proto_ops vsock_dgram_ops = {
1075 .owner = THIS_MODULE,
1076 .release = vsock_release,
1078 .connect = vsock_dgram_connect,
1079 .socketpair = sock_no_socketpair,
1080 .accept = sock_no_accept,
1081 .getname = vsock_getname,
1083 .ioctl = sock_no_ioctl,
1084 .listen = sock_no_listen,
1085 .shutdown = vsock_shutdown,
1086 .setsockopt = sock_no_setsockopt,
1087 .getsockopt = sock_no_getsockopt,
1088 .sendmsg = vsock_dgram_sendmsg,
1089 .recvmsg = vsock_dgram_recvmsg,
1090 .mmap = sock_no_mmap,
1091 .sendpage = sock_no_sendpage,
1094 static void vsock_connect_timeout(struct work_struct *work)
1097 struct vsock_sock *vsk;
1099 vsk = container_of(work, struct vsock_sock, dwork.work);
1103 if (sk->sk_state == SS_CONNECTING &&
1104 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1105 sk->sk_state = SS_UNCONNECTED;
1106 sk->sk_err = ETIMEDOUT;
1107 sk->sk_error_report(sk);
1114 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1115 int addr_len, int flags)
1119 struct vsock_sock *vsk;
1120 struct sockaddr_vm *remote_addr;
1130 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1131 switch (sock->state) {
1135 case SS_DISCONNECTING:
1139 /* This continues on so we can move sock into the SS_CONNECTED
1140 * state once the connection has completed (at which point err
1141 * will be set to zero also). Otherwise, we will either wait
1142 * for the connection or return -EALREADY should this be a
1143 * non-blocking call.
1148 if ((sk->sk_state == SS_LISTEN) ||
1149 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1154 /* The hypervisor and well-known contexts do not have socket
1157 if (!transport->stream_allow(remote_addr->svm_cid,
1158 remote_addr->svm_port)) {
1163 /* Set the remote address that we are connecting to. */
1164 memcpy(&vsk->remote_addr, remote_addr,
1165 sizeof(vsk->remote_addr));
1167 /* Autobind this socket to the local address if necessary. */
1168 if (!vsock_addr_bound(&vsk->local_addr)) {
1169 struct sockaddr_vm local_addr;
1171 vsock_addr_init(&local_addr, VMADDR_CID_ANY,
1173 err = __vsock_bind(sk, &local_addr);
1179 sk->sk_state = SS_CONNECTING;
1181 err = transport->connect(vsk);
1185 /* Mark sock as connecting and set the error code to in
1186 * progress in case this is a non-blocking connect.
1188 sock->state = SS_CONNECTING;
1192 /* The receive path will handle all communication until we are able to
1193 * enter the connected state. Here we wait for the connection to be
1194 * completed or a notification of an error.
1196 timeout = vsk->connect_timeout;
1197 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1199 while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1200 if (flags & O_NONBLOCK) {
1201 /* If we're not going to block, we schedule a timeout
1202 * function to generate a timeout on the connection
1203 * attempt, in case the peer doesn't respond in a
1204 * timely manner. We hold on to the socket until the
1208 INIT_DELAYED_WORK(&vsk->dwork,
1209 vsock_connect_timeout);
1210 schedule_delayed_work(&vsk->dwork, timeout);
1212 /* Skip ahead to preserve error code set above. */
1217 timeout = schedule_timeout(timeout);
1220 if (signal_pending(current)) {
1221 err = sock_intr_errno(timeout);
1222 goto out_wait_error;
1223 } else if (timeout == 0) {
1225 goto out_wait_error;
1228 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1233 goto out_wait_error;
1238 finish_wait(sk_sleep(sk), &wait);
1244 sk->sk_state = SS_UNCONNECTED;
1245 sock->state = SS_UNCONNECTED;
1249 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1251 struct sock *listener;
1253 struct sock *connected;
1254 struct vsock_sock *vconnected;
1259 listener = sock->sk;
1261 lock_sock(listener);
1263 if (sock->type != SOCK_STREAM) {
1268 if (listener->sk_state != SS_LISTEN) {
1273 /* Wait for children sockets to appear; these are the new sockets
1274 * created upon connection establishment.
1276 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1277 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1279 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1280 listener->sk_err == 0) {
1281 release_sock(listener);
1282 timeout = schedule_timeout(timeout);
1283 lock_sock(listener);
1285 if (signal_pending(current)) {
1286 err = sock_intr_errno(timeout);
1288 } else if (timeout == 0) {
1293 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1296 if (listener->sk_err)
1297 err = -listener->sk_err;
1300 listener->sk_ack_backlog--;
1302 lock_sock(connected);
1303 vconnected = vsock_sk(connected);
1305 /* If the listener socket has received an error, then we should
1306 * reject this socket and return. Note that we simply mark the
1307 * socket rejected, drop our reference, and let the cleanup
1308 * function handle the cleanup; the fact that we found it in
1309 * the listener's accept queue guarantees that the cleanup
1310 * function hasn't run yet.
1313 vconnected->rejected = true;
1314 release_sock(connected);
1315 sock_put(connected);
1319 newsock->state = SS_CONNECTED;
1320 sock_graft(connected, newsock);
1321 release_sock(connected);
1322 sock_put(connected);
1326 finish_wait(sk_sleep(listener), &wait);
1328 release_sock(listener);
1332 static int vsock_listen(struct socket *sock, int backlog)
1336 struct vsock_sock *vsk;
1342 if (sock->type != SOCK_STREAM) {
1347 if (sock->state != SS_UNCONNECTED) {
1354 if (!vsock_addr_bound(&vsk->local_addr)) {
1359 sk->sk_max_ack_backlog = backlog;
1360 sk->sk_state = SS_LISTEN;
1369 static int vsock_stream_setsockopt(struct socket *sock,
1372 char __user *optval,
1373 unsigned int optlen)
1377 struct vsock_sock *vsk;
1380 if (level != AF_VSOCK)
1381 return -ENOPROTOOPT;
1383 #define COPY_IN(_v) \
1385 if (optlen < sizeof(_v)) { \
1389 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \
1402 case SO_VM_SOCKETS_BUFFER_SIZE:
1404 transport->set_buffer_size(vsk, val);
1407 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1409 transport->set_max_buffer_size(vsk, val);
1412 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1414 transport->set_min_buffer_size(vsk, val);
1417 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1420 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1421 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1422 vsk->connect_timeout = tv.tv_sec * HZ +
1423 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1424 if (vsk->connect_timeout == 0)
1425 vsk->connect_timeout =
1426 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1446 static int vsock_stream_getsockopt(struct socket *sock,
1447 int level, int optname,
1448 char __user *optval,
1454 struct vsock_sock *vsk;
1457 if (level != AF_VSOCK)
1458 return -ENOPROTOOPT;
1460 err = get_user(len, optlen);
1464 #define COPY_OUT(_v) \
1466 if (len < sizeof(_v)) \
1470 if (copy_to_user(optval, &_v, len) != 0) \
1480 case SO_VM_SOCKETS_BUFFER_SIZE:
1481 val = transport->get_buffer_size(vsk);
1485 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1486 val = transport->get_max_buffer_size(vsk);
1490 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1491 val = transport->get_min_buffer_size(vsk);
1495 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1497 tv.tv_sec = vsk->connect_timeout / HZ;
1499 (vsk->connect_timeout -
1500 tv.tv_sec * HZ) * (1000000 / HZ);
1505 return -ENOPROTOOPT;
1508 err = put_user(len, optlen);
1517 static int vsock_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1518 struct msghdr *msg, size_t len)
1521 struct vsock_sock *vsk;
1522 ssize_t total_written;
1525 struct vsock_transport_send_notify_data send_data;
1534 if (msg->msg_flags & MSG_OOB)
1539 /* Callers should not provide a destination with stream sockets. */
1540 if (msg->msg_namelen) {
1541 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1545 /* Send data only if both sides are not shutdown in the direction. */
1546 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1547 vsk->peer_shutdown & RCV_SHUTDOWN) {
1552 if (sk->sk_state != SS_CONNECTED ||
1553 !vsock_addr_bound(&vsk->local_addr)) {
1558 if (!vsock_addr_bound(&vsk->remote_addr)) {
1559 err = -EDESTADDRREQ;
1563 /* Wait for room in the produce queue to enqueue our user's data. */
1564 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1566 err = transport->notify_send_init(vsk, &send_data);
1570 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1572 while (total_written < len) {
1575 while (vsock_stream_has_space(vsk) == 0 &&
1577 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1578 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1580 /* Don't wait for non-blocking sockets. */
1586 err = transport->notify_send_pre_block(vsk, &send_data);
1591 timeout = schedule_timeout(timeout);
1593 if (signal_pending(current)) {
1594 err = sock_intr_errno(timeout);
1596 } else if (timeout == 0) {
1601 prepare_to_wait(sk_sleep(sk), &wait,
1602 TASK_INTERRUPTIBLE);
1605 /* These checks occur both as part of and after the loop
1606 * conditional since we need to check before and after
1612 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1613 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1618 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1622 /* Note that enqueue will only write as many bytes as are free
1623 * in the produce queue, so we don't need to ensure len is
1624 * smaller than the queue size. It is the caller's
1625 * responsibility to check how many bytes we were able to send.
1628 written = transport->stream_enqueue(
1630 len - total_written);
1636 total_written += written;
1638 err = transport->notify_send_post_enqueue(
1639 vsk, written, &send_data);
1646 if (total_written > 0)
1647 err = total_written;
1648 finish_wait(sk_sleep(sk), &wait);
1656 vsock_stream_recvmsg(struct kiocb *kiocb,
1657 struct socket *sock,
1658 struct msghdr *msg, size_t len, int flags)
1661 struct vsock_sock *vsk;
1666 struct vsock_transport_recv_notify_data recv_data;
1676 if (sk->sk_state != SS_CONNECTED) {
1677 /* Recvmsg is supposed to return 0 if a peer performs an
1678 * orderly shutdown. Differentiate between that case and when a
1679 * peer has not connected or a local shutdown occured with the
1682 if (sock_flag(sk, SOCK_DONE))
1690 if (flags & MSG_OOB) {
1695 /* We don't check peer_shutdown flag here since peer may actually shut
1696 * down, but there can be data in the queue that a local socket can
1699 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1704 /* It is valid on Linux to pass in a zero-length receive buffer. This
1705 * is not an error. We may as well bail out now.
1712 /* We must not copy less than target bytes into the user's buffer
1713 * before returning successfully, so we wait for the consume queue to
1714 * have that much data to consume before dequeueing. Note that this
1715 * makes it impossible to handle cases where target is greater than the
1718 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1719 if (target >= transport->stream_rcvhiwat(vsk)) {
1723 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1726 err = transport->notify_recv_init(vsk, target, &recv_data);
1730 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1733 s64 ready = vsock_stream_has_data(vsk);
1736 /* Invalid queue pair content. XXX This should be
1737 * changed to a connection reset in a later change.
1742 } else if (ready > 0) {
1745 err = transport->notify_recv_pre_dequeue(
1746 vsk, target, &recv_data);
1750 read = transport->stream_dequeue(
1752 len - copied, flags);
1760 err = transport->notify_recv_post_dequeue(
1762 !(flags & MSG_PEEK), &recv_data);
1766 if (read >= target || flags & MSG_PEEK)
1771 if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1772 || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1775 /* Don't wait for non-blocking sockets. */
1781 err = transport->notify_recv_pre_block(
1782 vsk, target, &recv_data);
1787 timeout = schedule_timeout(timeout);
1790 if (signal_pending(current)) {
1791 err = sock_intr_errno(timeout);
1793 } else if (timeout == 0) {
1798 prepare_to_wait(sk_sleep(sk), &wait,
1799 TASK_INTERRUPTIBLE);
1805 else if (sk->sk_shutdown & RCV_SHUTDOWN)
1809 /* We only do these additional bookkeeping/notification steps
1810 * if we actually copied something out of the queue pair
1811 * instead of just peeking ahead.
1814 if (!(flags & MSG_PEEK)) {
1815 /* If the other side has shutdown for sending and there
1816 * is nothing more to read, then modify the socket
1819 if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1820 if (vsock_stream_has_data(vsk) <= 0) {
1821 sk->sk_state = SS_UNCONNECTED;
1822 sock_set_flag(sk, SOCK_DONE);
1823 sk->sk_state_change(sk);
1831 finish_wait(sk_sleep(sk), &wait);
1837 static const struct proto_ops vsock_stream_ops = {
1839 .owner = THIS_MODULE,
1840 .release = vsock_release,
1842 .connect = vsock_stream_connect,
1843 .socketpair = sock_no_socketpair,
1844 .accept = vsock_accept,
1845 .getname = vsock_getname,
1847 .ioctl = sock_no_ioctl,
1848 .listen = vsock_listen,
1849 .shutdown = vsock_shutdown,
1850 .setsockopt = vsock_stream_setsockopt,
1851 .getsockopt = vsock_stream_getsockopt,
1852 .sendmsg = vsock_stream_sendmsg,
1853 .recvmsg = vsock_stream_recvmsg,
1854 .mmap = sock_no_mmap,
1855 .sendpage = sock_no_sendpage,
1858 static int vsock_create(struct net *net, struct socket *sock,
1859 int protocol, int kern)
1865 return -EPROTONOSUPPORT;
1867 switch (sock->type) {
1869 sock->ops = &vsock_dgram_ops;
1872 sock->ops = &vsock_stream_ops;
1875 return -ESOCKTNOSUPPORT;
1878 sock->state = SS_UNCONNECTED;
1880 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0) ? 0 : -ENOMEM;
1883 static const struct net_proto_family vsock_family_ops = {
1885 .create = vsock_create,
1886 .owner = THIS_MODULE,
1889 static long vsock_dev_do_ioctl(struct file *filp,
1890 unsigned int cmd, void __user *ptr)
1892 u32 __user *p = ptr;
1896 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1897 if (put_user(transport->get_local_cid(), p) != 0)
1902 pr_err("Unknown ioctl %d\n", cmd);
1909 static long vsock_dev_ioctl(struct file *filp,
1910 unsigned int cmd, unsigned long arg)
1912 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1915 #ifdef CONFIG_COMPAT
1916 static long vsock_dev_compat_ioctl(struct file *filp,
1917 unsigned int cmd, unsigned long arg)
1919 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1923 static const struct file_operations vsock_device_ops = {
1924 .owner = THIS_MODULE,
1925 .unlocked_ioctl = vsock_dev_ioctl,
1926 #ifdef CONFIG_COMPAT
1927 .compat_ioctl = vsock_dev_compat_ioctl,
1929 .open = nonseekable_open,
1932 static struct miscdevice vsock_device = {
1934 .minor = MISC_DYNAMIC_MINOR,
1935 .fops = &vsock_device_ops,
1938 static int __vsock_core_init(void)
1942 vsock_init_tables();
1944 err = misc_register(&vsock_device);
1946 pr_err("Failed to register misc device\n");
1950 err = proto_register(&vsock_proto, 1); /* we want our slab */
1952 pr_err("Cannot register vsock protocol\n");
1953 goto err_misc_deregister;
1956 err = sock_register(&vsock_family_ops);
1958 pr_err("could not register af_vsock (%d) address family: %d\n",
1960 goto err_unregister_proto;
1965 err_unregister_proto:
1966 proto_unregister(&vsock_proto);
1967 err_misc_deregister:
1968 misc_deregister(&vsock_device);
1972 int vsock_core_init(const struct vsock_transport *t)
1974 int retval = mutex_lock_interruptible(&vsock_register_mutex);
1984 retval = __vsock_core_init();
1989 mutex_unlock(&vsock_register_mutex);
1992 EXPORT_SYMBOL_GPL(vsock_core_init);
1994 void vsock_core_exit(void)
1996 mutex_lock(&vsock_register_mutex);
1998 misc_deregister(&vsock_device);
1999 sock_unregister(AF_VSOCK);
2000 proto_unregister(&vsock_proto);
2002 /* We do not want the assignment below re-ordered. */
2006 mutex_unlock(&vsock_register_mutex);
2008 EXPORT_SYMBOL_GPL(vsock_core_exit);
2010 MODULE_AUTHOR("VMware, Inc.");
2011 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2012 MODULE_VERSION(VSOCK_DRIVER_VERSION_STRING);
2013 MODULE_LICENSE("GPL v2");