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>
99 #include <net/af_vsock.h>
101 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
102 static void vsock_sk_destruct(struct sock *sk);
103 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
105 /* Protocol family. */
106 static struct proto vsock_proto = {
108 .owner = THIS_MODULE,
109 .obj_size = sizeof(struct vsock_sock),
112 /* The default peer timeout indicates how long we will wait for a peer response
113 * to a control message.
115 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
117 #define SS_LISTEN 255
119 static const struct vsock_transport *transport;
120 static DEFINE_MUTEX(vsock_register_mutex);
124 /* Get the ID of the local context. This is transport dependent. */
126 int vm_sockets_get_local_cid(void)
128 return transport->get_local_cid();
130 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
134 /* Each bound VSocket is stored in the bind hash table and each connected
135 * VSocket is stored in the connected hash table.
137 * Unbound sockets are all put on the same list attached to the end of the hash
138 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
139 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
140 * represents the list that addr hashes to).
142 * Specifically, we initialize the vsock_bind_table array to a size of
143 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
144 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
145 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
146 * mods with VSOCK_HASH_SIZE to ensure this.
148 #define VSOCK_HASH_SIZE 251
149 #define MAX_PORT_RETRIES 24
151 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE)
152 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
153 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
155 /* XXX This can probably be implemented in a better way. */
156 #define VSOCK_CONN_HASH(src, dst) \
157 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
158 #define vsock_connected_sockets(src, dst) \
159 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
160 #define vsock_connected_sockets_vsk(vsk) \
161 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
163 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
164 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
165 static DEFINE_SPINLOCK(vsock_table_lock);
167 /* Autobind this socket to the local address if necessary. */
168 static int vsock_auto_bind(struct vsock_sock *vsk)
170 struct sock *sk = sk_vsock(vsk);
171 struct sockaddr_vm local_addr;
173 if (vsock_addr_bound(&vsk->local_addr))
175 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
176 return __vsock_bind(sk, &local_addr);
179 static void vsock_init_tables(void)
183 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
184 INIT_LIST_HEAD(&vsock_bind_table[i]);
186 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
187 INIT_LIST_HEAD(&vsock_connected_table[i]);
190 static void __vsock_insert_bound(struct list_head *list,
191 struct vsock_sock *vsk)
194 list_add(&vsk->bound_table, list);
197 static void __vsock_insert_connected(struct list_head *list,
198 struct vsock_sock *vsk)
201 list_add(&vsk->connected_table, list);
204 static void __vsock_remove_bound(struct vsock_sock *vsk)
206 list_del_init(&vsk->bound_table);
210 static void __vsock_remove_connected(struct vsock_sock *vsk)
212 list_del_init(&vsk->connected_table);
216 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
218 struct vsock_sock *vsk;
220 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
221 if (addr->svm_port == vsk->local_addr.svm_port)
222 return sk_vsock(vsk);
227 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
228 struct sockaddr_vm *dst)
230 struct vsock_sock *vsk;
232 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
234 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
235 dst->svm_port == vsk->local_addr.svm_port) {
236 return sk_vsock(vsk);
243 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
245 return !list_empty(&vsk->bound_table);
248 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
250 return !list_empty(&vsk->connected_table);
253 static void vsock_insert_unbound(struct vsock_sock *vsk)
255 spin_lock_bh(&vsock_table_lock);
256 __vsock_insert_bound(vsock_unbound_sockets, vsk);
257 spin_unlock_bh(&vsock_table_lock);
260 void vsock_insert_connected(struct vsock_sock *vsk)
262 struct list_head *list = vsock_connected_sockets(
263 &vsk->remote_addr, &vsk->local_addr);
265 spin_lock_bh(&vsock_table_lock);
266 __vsock_insert_connected(list, vsk);
267 spin_unlock_bh(&vsock_table_lock);
269 EXPORT_SYMBOL_GPL(vsock_insert_connected);
271 void vsock_remove_bound(struct vsock_sock *vsk)
273 spin_lock_bh(&vsock_table_lock);
274 __vsock_remove_bound(vsk);
275 spin_unlock_bh(&vsock_table_lock);
277 EXPORT_SYMBOL_GPL(vsock_remove_bound);
279 void vsock_remove_connected(struct vsock_sock *vsk)
281 spin_lock_bh(&vsock_table_lock);
282 __vsock_remove_connected(vsk);
283 spin_unlock_bh(&vsock_table_lock);
285 EXPORT_SYMBOL_GPL(vsock_remove_connected);
287 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
291 spin_lock_bh(&vsock_table_lock);
292 sk = __vsock_find_bound_socket(addr);
296 spin_unlock_bh(&vsock_table_lock);
300 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
302 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
303 struct sockaddr_vm *dst)
307 spin_lock_bh(&vsock_table_lock);
308 sk = __vsock_find_connected_socket(src, dst);
312 spin_unlock_bh(&vsock_table_lock);
316 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
318 static bool vsock_in_bound_table(struct vsock_sock *vsk)
322 spin_lock_bh(&vsock_table_lock);
323 ret = __vsock_in_bound_table(vsk);
324 spin_unlock_bh(&vsock_table_lock);
329 static bool vsock_in_connected_table(struct vsock_sock *vsk)
333 spin_lock_bh(&vsock_table_lock);
334 ret = __vsock_in_connected_table(vsk);
335 spin_unlock_bh(&vsock_table_lock);
340 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
344 spin_lock_bh(&vsock_table_lock);
346 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
347 struct vsock_sock *vsk;
348 list_for_each_entry(vsk, &vsock_connected_table[i],
353 spin_unlock_bh(&vsock_table_lock);
355 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
357 void vsock_add_pending(struct sock *listener, struct sock *pending)
359 struct vsock_sock *vlistener;
360 struct vsock_sock *vpending;
362 vlistener = vsock_sk(listener);
363 vpending = vsock_sk(pending);
367 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
369 EXPORT_SYMBOL_GPL(vsock_add_pending);
371 void vsock_remove_pending(struct sock *listener, struct sock *pending)
373 struct vsock_sock *vpending = vsock_sk(pending);
375 list_del_init(&vpending->pending_links);
379 EXPORT_SYMBOL_GPL(vsock_remove_pending);
381 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
383 struct vsock_sock *vlistener;
384 struct vsock_sock *vconnected;
386 vlistener = vsock_sk(listener);
387 vconnected = vsock_sk(connected);
389 sock_hold(connected);
391 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
393 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
395 static struct sock *vsock_dequeue_accept(struct sock *listener)
397 struct vsock_sock *vlistener;
398 struct vsock_sock *vconnected;
400 vlistener = vsock_sk(listener);
402 if (list_empty(&vlistener->accept_queue))
405 vconnected = list_entry(vlistener->accept_queue.next,
406 struct vsock_sock, accept_queue);
408 list_del_init(&vconnected->accept_queue);
410 /* The caller will need a reference on the connected socket so we let
411 * it call sock_put().
414 return sk_vsock(vconnected);
417 static bool vsock_is_accept_queue_empty(struct sock *sk)
419 struct vsock_sock *vsk = vsock_sk(sk);
420 return list_empty(&vsk->accept_queue);
423 static bool vsock_is_pending(struct sock *sk)
425 struct vsock_sock *vsk = vsock_sk(sk);
426 return !list_empty(&vsk->pending_links);
429 static int vsock_send_shutdown(struct sock *sk, int mode)
431 return transport->shutdown(vsock_sk(sk), mode);
434 void vsock_pending_work(struct work_struct *work)
437 struct sock *listener;
438 struct vsock_sock *vsk;
441 vsk = container_of(work, struct vsock_sock, dwork.work);
443 listener = vsk->listener;
449 if (vsock_is_pending(sk)) {
450 vsock_remove_pending(listener, sk);
451 } else if (!vsk->rejected) {
452 /* We are not on the pending list and accept() did not reject
453 * us, so we must have been accepted by our user process. We
454 * just need to drop our references to the sockets and be on
461 listener->sk_ack_backlog--;
463 /* We need to remove ourself from the global connected sockets list so
464 * incoming packets can't find this socket, and to reduce the reference
467 if (vsock_in_connected_table(vsk))
468 vsock_remove_connected(vsk);
470 sk->sk_state = SS_FREE;
474 release_sock(listener);
481 EXPORT_SYMBOL_GPL(vsock_pending_work);
483 /**** SOCKET OPERATIONS ****/
485 static int __vsock_bind_stream(struct vsock_sock *vsk,
486 struct sockaddr_vm *addr)
488 static u32 port = LAST_RESERVED_PORT + 1;
489 struct sockaddr_vm new_addr;
491 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
493 if (addr->svm_port == VMADDR_PORT_ANY) {
497 for (i = 0; i < MAX_PORT_RETRIES; i++) {
498 if (port <= LAST_RESERVED_PORT)
499 port = LAST_RESERVED_PORT + 1;
501 new_addr.svm_port = port++;
503 if (!__vsock_find_bound_socket(&new_addr)) {
510 return -EADDRNOTAVAIL;
512 /* If port is in reserved range, ensure caller
513 * has necessary privileges.
515 if (addr->svm_port <= LAST_RESERVED_PORT &&
516 !capable(CAP_NET_BIND_SERVICE)) {
520 if (__vsock_find_bound_socket(&new_addr))
524 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
526 /* Remove stream sockets from the unbound list and add them to the hash
527 * table for easy lookup by its address. The unbound list is simply an
528 * extra entry at the end of the hash table, a trick used by AF_UNIX.
530 __vsock_remove_bound(vsk);
531 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
536 static int __vsock_bind_dgram(struct vsock_sock *vsk,
537 struct sockaddr_vm *addr)
539 return transport->dgram_bind(vsk, addr);
542 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
544 struct vsock_sock *vsk = vsock_sk(sk);
548 /* First ensure this socket isn't already bound. */
549 if (vsock_addr_bound(&vsk->local_addr))
552 /* Now bind to the provided address or select appropriate values if
553 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
554 * like AF_INET prevents binding to a non-local IP address (in most
555 * cases), we only allow binding to the local CID.
557 cid = transport->get_local_cid();
558 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
559 return -EADDRNOTAVAIL;
561 switch (sk->sk_socket->type) {
563 spin_lock_bh(&vsock_table_lock);
564 retval = __vsock_bind_stream(vsk, addr);
565 spin_unlock_bh(&vsock_table_lock);
569 retval = __vsock_bind_dgram(vsk, addr);
580 struct sock *__vsock_create(struct net *net,
588 struct vsock_sock *psk;
589 struct vsock_sock *vsk;
591 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
595 sock_init_data(sock, sk);
597 /* sk->sk_type is normally set in sock_init_data, but only if sock is
598 * non-NULL. We make sure that our sockets always have a type by
599 * setting it here if needed.
605 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
606 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
608 sk->sk_destruct = vsock_sk_destruct;
609 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
611 sock_reset_flag(sk, SOCK_DONE);
613 INIT_LIST_HEAD(&vsk->bound_table);
614 INIT_LIST_HEAD(&vsk->connected_table);
615 vsk->listener = NULL;
616 INIT_LIST_HEAD(&vsk->pending_links);
617 INIT_LIST_HEAD(&vsk->accept_queue);
618 vsk->rejected = false;
619 vsk->sent_request = false;
620 vsk->ignore_connecting_rst = false;
621 vsk->peer_shutdown = 0;
623 psk = parent ? vsock_sk(parent) : NULL;
625 vsk->trusted = psk->trusted;
626 vsk->owner = get_cred(psk->owner);
627 vsk->connect_timeout = psk->connect_timeout;
629 vsk->trusted = capable(CAP_NET_ADMIN);
630 vsk->owner = get_current_cred();
631 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
634 if (transport->init(vsk, psk) < 0) {
640 vsock_insert_unbound(vsk);
644 EXPORT_SYMBOL_GPL(__vsock_create);
646 static void __vsock_release(struct sock *sk)
650 struct sock *pending;
651 struct vsock_sock *vsk;
654 pending = NULL; /* Compiler warning. */
656 if (vsock_in_bound_table(vsk))
657 vsock_remove_bound(vsk);
659 if (vsock_in_connected_table(vsk))
660 vsock_remove_connected(vsk);
662 transport->release(vsk);
666 sk->sk_shutdown = SHUTDOWN_MASK;
668 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
671 /* Clean up any sockets that never were accepted. */
672 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
673 __vsock_release(pending);
682 static void vsock_sk_destruct(struct sock *sk)
684 struct vsock_sock *vsk = vsock_sk(sk);
686 transport->destruct(vsk);
688 /* When clearing these addresses, there's no need to set the family and
689 * possibly register the address family with the kernel.
691 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
692 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
694 put_cred(vsk->owner);
697 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
701 err = sock_queue_rcv_skb(sk, skb);
708 s64 vsock_stream_has_data(struct vsock_sock *vsk)
710 return transport->stream_has_data(vsk);
712 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
714 s64 vsock_stream_has_space(struct vsock_sock *vsk)
716 return transport->stream_has_space(vsk);
718 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
720 static int vsock_release(struct socket *sock)
722 __vsock_release(sock->sk);
724 sock->state = SS_FREE;
730 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
734 struct sockaddr_vm *vm_addr;
738 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
742 err = __vsock_bind(sk, vm_addr);
748 static int vsock_getname(struct socket *sock,
749 struct sockaddr *addr, int *addr_len, int peer)
753 struct vsock_sock *vsk;
754 struct sockaddr_vm *vm_addr;
763 if (sock->state != SS_CONNECTED) {
767 vm_addr = &vsk->remote_addr;
769 vm_addr = &vsk->local_addr;
777 /* sys_getsockname() and sys_getpeername() pass us a
778 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
779 * that macro is defined in socket.c instead of .h, so we hardcode its
782 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
783 memcpy(addr, vm_addr, sizeof(*vm_addr));
784 *addr_len = sizeof(*vm_addr);
791 static int vsock_shutdown(struct socket *sock, int mode)
796 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
797 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
798 * here like the other address families do. Note also that the
799 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
800 * which is what we want.
804 if ((mode & ~SHUTDOWN_MASK) || !mode)
807 /* If this is a STREAM socket and it is not connected then bail out
808 * immediately. If it is a DGRAM socket then we must first kick the
809 * socket so that it wakes up from any sleeping calls, for example
810 * recv(), and then afterwards return the error.
814 if (sock->state == SS_UNCONNECTED) {
816 if (sk->sk_type == SOCK_STREAM)
819 sock->state = SS_DISCONNECTING;
823 /* Receive and send shutdowns are treated alike. */
824 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
827 sk->sk_shutdown |= mode;
828 sk->sk_state_change(sk);
831 if (sk->sk_type == SOCK_STREAM) {
832 sock_reset_flag(sk, SOCK_DONE);
833 vsock_send_shutdown(sk, mode);
840 static unsigned int vsock_poll(struct file *file, struct socket *sock,
845 struct vsock_sock *vsk;
850 poll_wait(file, sk_sleep(sk), wait);
854 /* Signify that there has been an error on this socket. */
857 /* INET sockets treat local write shutdown and peer write shutdown as a
858 * case of POLLHUP set.
860 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
861 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
862 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
866 if (sk->sk_shutdown & RCV_SHUTDOWN ||
867 vsk->peer_shutdown & SEND_SHUTDOWN) {
871 if (sock->type == SOCK_DGRAM) {
872 /* For datagram sockets we can read if there is something in
873 * the queue and write as long as the socket isn't shutdown for
876 if (!skb_queue_empty(&sk->sk_receive_queue) ||
877 (sk->sk_shutdown & RCV_SHUTDOWN)) {
878 mask |= POLLIN | POLLRDNORM;
881 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
882 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
884 } else if (sock->type == SOCK_STREAM) {
887 /* Listening sockets that have connections in their accept
890 if (sk->sk_state == SS_LISTEN
891 && !vsock_is_accept_queue_empty(sk))
892 mask |= POLLIN | POLLRDNORM;
894 /* If there is something in the queue then we can read. */
895 if (transport->stream_is_active(vsk) &&
896 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
897 bool data_ready_now = false;
898 int ret = transport->notify_poll_in(
899 vsk, 1, &data_ready_now);
904 mask |= POLLIN | POLLRDNORM;
909 /* Sockets whose connections have been closed, reset, or
910 * terminated should also be considered read, and we check the
911 * shutdown flag for that.
913 if (sk->sk_shutdown & RCV_SHUTDOWN ||
914 vsk->peer_shutdown & SEND_SHUTDOWN) {
915 mask |= POLLIN | POLLRDNORM;
918 /* Connected sockets that can produce data can be written. */
919 if (sk->sk_state == SS_CONNECTED) {
920 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
921 bool space_avail_now = false;
922 int ret = transport->notify_poll_out(
923 vsk, 1, &space_avail_now);
928 /* Remove POLLWRBAND since INET
929 * sockets are not setting it.
931 mask |= POLLOUT | POLLWRNORM;
937 /* Simulate INET socket poll behaviors, which sets
938 * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
939 * but local send is not shutdown.
941 if (sk->sk_state == SS_UNCONNECTED) {
942 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
943 mask |= POLLOUT | POLLWRNORM;
953 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
958 struct vsock_sock *vsk;
959 struct sockaddr_vm *remote_addr;
961 if (msg->msg_flags & MSG_OOB)
964 /* For now, MSG_DONTWAIT is always assumed... */
971 err = vsock_auto_bind(vsk);
976 /* If the provided message contains an address, use that. Otherwise
977 * fall back on the socket's remote handle (if it has been connected).
980 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
981 &remote_addr) == 0) {
982 /* Ensure this address is of the right type and is a valid
986 if (remote_addr->svm_cid == VMADDR_CID_ANY)
987 remote_addr->svm_cid = transport->get_local_cid();
989 if (!vsock_addr_bound(remote_addr)) {
993 } else if (sock->state == SS_CONNECTED) {
994 remote_addr = &vsk->remote_addr;
996 if (remote_addr->svm_cid == VMADDR_CID_ANY)
997 remote_addr->svm_cid = transport->get_local_cid();
999 /* XXX Should connect() or this function ensure remote_addr is
1002 if (!vsock_addr_bound(&vsk->remote_addr)) {
1011 if (!transport->dgram_allow(remote_addr->svm_cid,
1012 remote_addr->svm_port)) {
1017 err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1024 static int vsock_dgram_connect(struct socket *sock,
1025 struct sockaddr *addr, int addr_len, int flags)
1029 struct vsock_sock *vsk;
1030 struct sockaddr_vm *remote_addr;
1035 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1036 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1038 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1040 sock->state = SS_UNCONNECTED;
1043 } else if (err != 0)
1048 err = vsock_auto_bind(vsk);
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 socket *sock, struct msghdr *msg,
1067 size_t len, int flags)
1069 return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1072 static const struct proto_ops vsock_dgram_ops = {
1074 .owner = THIS_MODULE,
1075 .release = vsock_release,
1077 .connect = vsock_dgram_connect,
1078 .socketpair = sock_no_socketpair,
1079 .accept = sock_no_accept,
1080 .getname = vsock_getname,
1082 .ioctl = sock_no_ioctl,
1083 .listen = sock_no_listen,
1084 .shutdown = vsock_shutdown,
1085 .setsockopt = sock_no_setsockopt,
1086 .getsockopt = sock_no_getsockopt,
1087 .sendmsg = vsock_dgram_sendmsg,
1088 .recvmsg = vsock_dgram_recvmsg,
1089 .mmap = sock_no_mmap,
1090 .sendpage = sock_no_sendpage,
1093 static void vsock_connect_timeout(struct work_struct *work)
1096 struct vsock_sock *vsk;
1098 vsk = container_of(work, struct vsock_sock, dwork.work);
1102 if (sk->sk_state == SS_CONNECTING &&
1103 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1104 sk->sk_state = SS_UNCONNECTED;
1105 sk->sk_err = ETIMEDOUT;
1106 sk->sk_error_report(sk);
1113 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1114 int addr_len, int flags)
1118 struct vsock_sock *vsk;
1119 struct sockaddr_vm *remote_addr;
1129 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1130 switch (sock->state) {
1134 case SS_DISCONNECTING:
1138 /* This continues on so we can move sock into the SS_CONNECTED
1139 * state once the connection has completed (at which point err
1140 * will be set to zero also). Otherwise, we will either wait
1141 * for the connection or return -EALREADY should this be a
1142 * non-blocking call.
1147 if ((sk->sk_state == SS_LISTEN) ||
1148 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1153 /* The hypervisor and well-known contexts do not have socket
1156 if (!transport->stream_allow(remote_addr->svm_cid,
1157 remote_addr->svm_port)) {
1162 /* Set the remote address that we are connecting to. */
1163 memcpy(&vsk->remote_addr, remote_addr,
1164 sizeof(vsk->remote_addr));
1166 err = vsock_auto_bind(vsk);
1170 sk->sk_state = SS_CONNECTING;
1172 err = transport->connect(vsk);
1176 /* Mark sock as connecting and set the error code to in
1177 * progress in case this is a non-blocking connect.
1179 sock->state = SS_CONNECTING;
1183 /* The receive path will handle all communication until we are able to
1184 * enter the connected state. Here we wait for the connection to be
1185 * completed or a notification of an error.
1187 timeout = vsk->connect_timeout;
1188 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1190 while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1191 if (flags & O_NONBLOCK) {
1192 /* If we're not going to block, we schedule a timeout
1193 * function to generate a timeout on the connection
1194 * attempt, in case the peer doesn't respond in a
1195 * timely manner. We hold on to the socket until the
1199 INIT_DELAYED_WORK(&vsk->dwork,
1200 vsock_connect_timeout);
1201 schedule_delayed_work(&vsk->dwork, timeout);
1203 /* Skip ahead to preserve error code set above. */
1208 timeout = schedule_timeout(timeout);
1211 if (signal_pending(current)) {
1212 err = sock_intr_errno(timeout);
1213 goto out_wait_error;
1214 } else if (timeout == 0) {
1216 goto out_wait_error;
1219 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1224 goto out_wait_error;
1229 finish_wait(sk_sleep(sk), &wait);
1235 sk->sk_state = SS_UNCONNECTED;
1236 sock->state = SS_UNCONNECTED;
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 != 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 lock_sock(listener);
1276 if (signal_pending(current)) {
1277 err = sock_intr_errno(timeout);
1279 } else if (timeout == 0) {
1284 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1287 if (listener->sk_err)
1288 err = -listener->sk_err;
1291 listener->sk_ack_backlog--;
1293 lock_sock(connected);
1294 vconnected = vsock_sk(connected);
1296 /* If the listener socket has received an error, then we should
1297 * reject this socket and return. Note that we simply mark the
1298 * socket rejected, drop our reference, and let the cleanup
1299 * function handle the cleanup; the fact that we found it in
1300 * the listener's accept queue guarantees that the cleanup
1301 * function hasn't run yet.
1304 vconnected->rejected = true;
1305 release_sock(connected);
1306 sock_put(connected);
1310 newsock->state = SS_CONNECTED;
1311 sock_graft(connected, newsock);
1312 release_sock(connected);
1313 sock_put(connected);
1317 finish_wait(sk_sleep(listener), &wait);
1319 release_sock(listener);
1323 static int vsock_listen(struct socket *sock, int backlog)
1327 struct vsock_sock *vsk;
1333 if (sock->type != SOCK_STREAM) {
1338 if (sock->state != SS_UNCONNECTED) {
1345 if (!vsock_addr_bound(&vsk->local_addr)) {
1350 sk->sk_max_ack_backlog = backlog;
1351 sk->sk_state = SS_LISTEN;
1360 static int vsock_stream_setsockopt(struct socket *sock,
1363 char __user *optval,
1364 unsigned int optlen)
1368 struct vsock_sock *vsk;
1371 if (level != AF_VSOCK)
1372 return -ENOPROTOOPT;
1374 #define COPY_IN(_v) \
1376 if (optlen < sizeof(_v)) { \
1380 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \
1393 case SO_VM_SOCKETS_BUFFER_SIZE:
1395 transport->set_buffer_size(vsk, val);
1398 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1400 transport->set_max_buffer_size(vsk, val);
1403 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1405 transport->set_min_buffer_size(vsk, val);
1408 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1411 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1412 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1413 vsk->connect_timeout = tv.tv_sec * HZ +
1414 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1415 if (vsk->connect_timeout == 0)
1416 vsk->connect_timeout =
1417 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1437 static int vsock_stream_getsockopt(struct socket *sock,
1438 int level, int optname,
1439 char __user *optval,
1445 struct vsock_sock *vsk;
1448 if (level != AF_VSOCK)
1449 return -ENOPROTOOPT;
1451 err = get_user(len, optlen);
1455 #define COPY_OUT(_v) \
1457 if (len < sizeof(_v)) \
1461 if (copy_to_user(optval, &_v, len) != 0) \
1471 case SO_VM_SOCKETS_BUFFER_SIZE:
1472 val = transport->get_buffer_size(vsk);
1476 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1477 val = transport->get_max_buffer_size(vsk);
1481 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1482 val = transport->get_min_buffer_size(vsk);
1486 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1488 tv.tv_sec = vsk->connect_timeout / HZ;
1490 (vsk->connect_timeout -
1491 tv.tv_sec * HZ) * (1000000 / HZ);
1496 return -ENOPROTOOPT;
1499 err = put_user(len, optlen);
1508 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1512 struct vsock_sock *vsk;
1513 ssize_t total_written;
1516 struct vsock_transport_send_notify_data send_data;
1525 if (msg->msg_flags & MSG_OOB)
1530 /* Callers should not provide a destination with stream sockets. */
1531 if (msg->msg_namelen) {
1532 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1536 /* Send data only if both sides are not shutdown in the direction. */
1537 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1538 vsk->peer_shutdown & RCV_SHUTDOWN) {
1543 if (sk->sk_state != SS_CONNECTED ||
1544 !vsock_addr_bound(&vsk->local_addr)) {
1549 if (!vsock_addr_bound(&vsk->remote_addr)) {
1550 err = -EDESTADDRREQ;
1554 /* Wait for room in the produce queue to enqueue our user's data. */
1555 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1557 err = transport->notify_send_init(vsk, &send_data);
1561 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1563 while (total_written < len) {
1566 while (vsock_stream_has_space(vsk) == 0 &&
1568 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1569 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1571 /* Don't wait for non-blocking sockets. */
1577 err = transport->notify_send_pre_block(vsk, &send_data);
1582 timeout = schedule_timeout(timeout);
1584 if (signal_pending(current)) {
1585 err = sock_intr_errno(timeout);
1587 } else if (timeout == 0) {
1592 prepare_to_wait(sk_sleep(sk), &wait,
1593 TASK_INTERRUPTIBLE);
1596 /* These checks occur both as part of and after the loop
1597 * conditional since we need to check before and after
1603 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1604 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1609 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1613 /* Note that enqueue will only write as many bytes as are free
1614 * in the produce queue, so we don't need to ensure len is
1615 * smaller than the queue size. It is the caller's
1616 * responsibility to check how many bytes we were able to send.
1619 written = transport->stream_enqueue(
1621 len - total_written);
1627 total_written += written;
1629 err = transport->notify_send_post_enqueue(
1630 vsk, written, &send_data);
1637 if (total_written > 0)
1638 err = total_written;
1639 finish_wait(sk_sleep(sk), &wait);
1647 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1651 struct vsock_sock *vsk;
1656 struct vsock_transport_recv_notify_data recv_data;
1666 if (sk->sk_state != SS_CONNECTED) {
1667 /* Recvmsg is supposed to return 0 if a peer performs an
1668 * orderly shutdown. Differentiate between that case and when a
1669 * peer has not connected or a local shutdown occured with the
1672 if (sock_flag(sk, SOCK_DONE))
1680 if (flags & MSG_OOB) {
1685 /* We don't check peer_shutdown flag here since peer may actually shut
1686 * down, but there can be data in the queue that a local socket can
1689 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1694 /* It is valid on Linux to pass in a zero-length receive buffer. This
1695 * is not an error. We may as well bail out now.
1702 /* We must not copy less than target bytes into the user's buffer
1703 * before returning successfully, so we wait for the consume queue to
1704 * have that much data to consume before dequeueing. Note that this
1705 * makes it impossible to handle cases where target is greater than the
1708 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1709 if (target >= transport->stream_rcvhiwat(vsk)) {
1713 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1716 err = transport->notify_recv_init(vsk, target, &recv_data);
1720 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1723 s64 ready = vsock_stream_has_data(vsk);
1726 /* Invalid queue pair content. XXX This should be
1727 * changed to a connection reset in a later change.
1732 } else if (ready > 0) {
1735 err = transport->notify_recv_pre_dequeue(
1736 vsk, target, &recv_data);
1740 read = transport->stream_dequeue(
1742 len - copied, flags);
1750 err = transport->notify_recv_post_dequeue(
1752 !(flags & MSG_PEEK), &recv_data);
1756 if (read >= target || flags & MSG_PEEK)
1761 if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1762 || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1765 /* Don't wait for non-blocking sockets. */
1771 err = transport->notify_recv_pre_block(
1772 vsk, target, &recv_data);
1777 timeout = schedule_timeout(timeout);
1780 if (signal_pending(current)) {
1781 err = sock_intr_errno(timeout);
1783 } else if (timeout == 0) {
1788 prepare_to_wait(sk_sleep(sk), &wait,
1789 TASK_INTERRUPTIBLE);
1795 else if (sk->sk_shutdown & RCV_SHUTDOWN)
1799 /* We only do these additional bookkeeping/notification steps
1800 * if we actually copied something out of the queue pair
1801 * instead of just peeking ahead.
1804 if (!(flags & MSG_PEEK)) {
1805 /* If the other side has shutdown for sending and there
1806 * is nothing more to read, then modify the socket
1809 if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1810 if (vsock_stream_has_data(vsk) <= 0) {
1811 sk->sk_state = SS_UNCONNECTED;
1812 sock_set_flag(sk, SOCK_DONE);
1813 sk->sk_state_change(sk);
1821 finish_wait(sk_sleep(sk), &wait);
1827 static const struct proto_ops vsock_stream_ops = {
1829 .owner = THIS_MODULE,
1830 .release = vsock_release,
1832 .connect = vsock_stream_connect,
1833 .socketpair = sock_no_socketpair,
1834 .accept = vsock_accept,
1835 .getname = vsock_getname,
1837 .ioctl = sock_no_ioctl,
1838 .listen = vsock_listen,
1839 .shutdown = vsock_shutdown,
1840 .setsockopt = vsock_stream_setsockopt,
1841 .getsockopt = vsock_stream_getsockopt,
1842 .sendmsg = vsock_stream_sendmsg,
1843 .recvmsg = vsock_stream_recvmsg,
1844 .mmap = sock_no_mmap,
1845 .sendpage = sock_no_sendpage,
1848 static int vsock_create(struct net *net, struct socket *sock,
1849 int protocol, int kern)
1854 if (protocol && protocol != PF_VSOCK)
1855 return -EPROTONOSUPPORT;
1857 switch (sock->type) {
1859 sock->ops = &vsock_dgram_ops;
1862 sock->ops = &vsock_stream_ops;
1865 return -ESOCKTNOSUPPORT;
1868 sock->state = SS_UNCONNECTED;
1870 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1873 static const struct net_proto_family vsock_family_ops = {
1875 .create = vsock_create,
1876 .owner = THIS_MODULE,
1879 static long vsock_dev_do_ioctl(struct file *filp,
1880 unsigned int cmd, void __user *ptr)
1882 u32 __user *p = ptr;
1886 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1887 if (put_user(transport->get_local_cid(), p) != 0)
1892 pr_err("Unknown ioctl %d\n", cmd);
1899 static long vsock_dev_ioctl(struct file *filp,
1900 unsigned int cmd, unsigned long arg)
1902 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1905 #ifdef CONFIG_COMPAT
1906 static long vsock_dev_compat_ioctl(struct file *filp,
1907 unsigned int cmd, unsigned long arg)
1909 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1913 static const struct file_operations vsock_device_ops = {
1914 .owner = THIS_MODULE,
1915 .unlocked_ioctl = vsock_dev_ioctl,
1916 #ifdef CONFIG_COMPAT
1917 .compat_ioctl = vsock_dev_compat_ioctl,
1919 .open = nonseekable_open,
1922 static struct miscdevice vsock_device = {
1924 .fops = &vsock_device_ops,
1927 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1929 int err = mutex_lock_interruptible(&vsock_register_mutex);
1939 /* Transport must be the owner of the protocol so that it can't
1940 * unload while there are open sockets.
1942 vsock_proto.owner = owner;
1945 vsock_init_tables();
1947 vsock_device.minor = MISC_DYNAMIC_MINOR;
1948 err = misc_register(&vsock_device);
1950 pr_err("Failed to register misc device\n");
1954 err = proto_register(&vsock_proto, 1); /* we want our slab */
1956 pr_err("Cannot register vsock protocol\n");
1957 goto err_misc_deregister;
1960 err = sock_register(&vsock_family_ops);
1962 pr_err("could not register af_vsock (%d) address family: %d\n",
1964 goto err_unregister_proto;
1967 mutex_unlock(&vsock_register_mutex);
1970 err_unregister_proto:
1971 proto_unregister(&vsock_proto);
1972 err_misc_deregister:
1973 misc_deregister(&vsock_device);
1976 mutex_unlock(&vsock_register_mutex);
1979 EXPORT_SYMBOL_GPL(__vsock_core_init);
1981 void vsock_core_exit(void)
1983 mutex_lock(&vsock_register_mutex);
1985 misc_deregister(&vsock_device);
1986 sock_unregister(AF_VSOCK);
1987 proto_unregister(&vsock_proto);
1989 /* We do not want the assignment below re-ordered. */
1993 mutex_unlock(&vsock_register_mutex);
1995 EXPORT_SYMBOL_GPL(vsock_core_exit);
1997 MODULE_AUTHOR("VMware, Inc.");
1998 MODULE_DESCRIPTION("VMware Virtual Socket Family");
1999 MODULE_VERSION("1.0.1.0-k");
2000 MODULE_LICENSE("GPL v2");