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>
81 #include <linux/bitops.h>
82 #include <linux/cred.h>
83 #include <linux/init.h>
85 #include <linux/kernel.h>
86 #include <linux/kmod.h>
87 #include <linux/list.h>
88 #include <linux/miscdevice.h>
89 #include <linux/module.h>
90 #include <linux/mutex.h>
91 #include <linux/net.h>
92 #include <linux/poll.h>
93 #include <linux/skbuff.h>
94 #include <linux/smp.h>
95 #include <linux/socket.h>
96 #include <linux/stddef.h>
97 #include <linux/unistd.h>
98 #include <linux/wait.h>
99 #include <linux/workqueue.h>
100 #include <net/sock.h>
102 #include "af_vsock.h"
103 #include "vsock_version.h"
105 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
106 static void vsock_sk_destruct(struct sock *sk);
107 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
109 /* Protocol family. */
110 static struct proto vsock_proto = {
112 .owner = THIS_MODULE,
113 .obj_size = sizeof(struct vsock_sock),
116 /* The default peer timeout indicates how long we will wait for a peer response
117 * to a control message.
119 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
121 #define SS_LISTEN 255
123 static const struct vsock_transport *transport;
124 static DEFINE_MUTEX(vsock_register_mutex);
128 /* Get the ID of the local context. This is transport dependent. */
130 int vm_sockets_get_local_cid(void)
132 return transport->get_local_cid();
134 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
138 /* Each bound VSocket is stored in the bind hash table and each connected
139 * VSocket is stored in the connected hash table.
141 * Unbound sockets are all put on the same list attached to the end of the hash
142 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
143 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
144 * represents the list that addr hashes to).
146 * Specifically, we initialize the vsock_bind_table array to a size of
147 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
148 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
149 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
150 * mods with VSOCK_HASH_SIZE - 1 to ensure this.
152 #define VSOCK_HASH_SIZE 251
153 #define MAX_PORT_RETRIES 24
155 #define VSOCK_HASH(addr) ((addr)->svm_port % (VSOCK_HASH_SIZE - 1))
156 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
157 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
159 /* XXX This can probably be implemented in a better way. */
160 #define VSOCK_CONN_HASH(src, dst) \
161 (((src)->svm_cid ^ (dst)->svm_port) % (VSOCK_HASH_SIZE - 1))
162 #define vsock_connected_sockets(src, dst) \
163 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
164 #define vsock_connected_sockets_vsk(vsk) \
165 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
167 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
168 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
169 static DEFINE_SPINLOCK(vsock_table_lock);
171 static __init void vsock_init_tables(void)
175 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
176 INIT_LIST_HEAD(&vsock_bind_table[i]);
178 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
179 INIT_LIST_HEAD(&vsock_connected_table[i]);
182 static void __vsock_insert_bound(struct list_head *list,
183 struct vsock_sock *vsk)
186 list_add(&vsk->bound_table, list);
189 static void __vsock_insert_connected(struct list_head *list,
190 struct vsock_sock *vsk)
193 list_add(&vsk->connected_table, list);
196 static void __vsock_remove_bound(struct vsock_sock *vsk)
198 list_del_init(&vsk->bound_table);
202 static void __vsock_remove_connected(struct vsock_sock *vsk)
204 list_del_init(&vsk->connected_table);
208 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
210 struct vsock_sock *vsk;
212 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
213 if (vsock_addr_equals_addr_any(addr, &vsk->local_addr))
214 return sk_vsock(vsk);
219 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
220 struct sockaddr_vm *dst)
222 struct vsock_sock *vsk;
224 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
226 if (vsock_addr_equals_addr(src, &vsk->remote_addr)
227 && vsock_addr_equals_addr(dst, &vsk->local_addr)) {
228 return sk_vsock(vsk);
235 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
237 return !list_empty(&vsk->bound_table);
240 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
242 return !list_empty(&vsk->connected_table);
245 static void vsock_insert_unbound(struct vsock_sock *vsk)
247 spin_lock_bh(&vsock_table_lock);
248 __vsock_insert_bound(vsock_unbound_sockets, vsk);
249 spin_unlock_bh(&vsock_table_lock);
252 void vsock_insert_connected(struct vsock_sock *vsk)
254 struct list_head *list = vsock_connected_sockets(
255 &vsk->remote_addr, &vsk->local_addr);
257 spin_lock_bh(&vsock_table_lock);
258 __vsock_insert_connected(list, vsk);
259 spin_unlock_bh(&vsock_table_lock);
261 EXPORT_SYMBOL_GPL(vsock_insert_connected);
263 void vsock_remove_bound(struct vsock_sock *vsk)
265 spin_lock_bh(&vsock_table_lock);
266 __vsock_remove_bound(vsk);
267 spin_unlock_bh(&vsock_table_lock);
269 EXPORT_SYMBOL_GPL(vsock_remove_bound);
271 void vsock_remove_connected(struct vsock_sock *vsk)
273 spin_lock_bh(&vsock_table_lock);
274 __vsock_remove_connected(vsk);
275 spin_unlock_bh(&vsock_table_lock);
277 EXPORT_SYMBOL_GPL(vsock_remove_connected);
279 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
283 spin_lock_bh(&vsock_table_lock);
284 sk = __vsock_find_bound_socket(addr);
288 spin_unlock_bh(&vsock_table_lock);
292 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
294 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
295 struct sockaddr_vm *dst)
299 spin_lock_bh(&vsock_table_lock);
300 sk = __vsock_find_connected_socket(src, dst);
304 spin_unlock_bh(&vsock_table_lock);
308 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
310 static bool vsock_in_bound_table(struct vsock_sock *vsk)
314 spin_lock_bh(&vsock_table_lock);
315 ret = __vsock_in_bound_table(vsk);
316 spin_unlock_bh(&vsock_table_lock);
321 static bool vsock_in_connected_table(struct vsock_sock *vsk)
325 spin_lock_bh(&vsock_table_lock);
326 ret = __vsock_in_connected_table(vsk);
327 spin_unlock_bh(&vsock_table_lock);
332 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
336 spin_lock_bh(&vsock_table_lock);
338 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
339 struct vsock_sock *vsk;
340 list_for_each_entry(vsk, &vsock_connected_table[i],
345 spin_unlock_bh(&vsock_table_lock);
347 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
349 void vsock_add_pending(struct sock *listener, struct sock *pending)
351 struct vsock_sock *vlistener;
352 struct vsock_sock *vpending;
354 vlistener = vsock_sk(listener);
355 vpending = vsock_sk(pending);
359 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
361 EXPORT_SYMBOL_GPL(vsock_add_pending);
363 void vsock_remove_pending(struct sock *listener, struct sock *pending)
365 struct vsock_sock *vpending = vsock_sk(pending);
367 list_del_init(&vpending->pending_links);
371 EXPORT_SYMBOL_GPL(vsock_remove_pending);
373 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
375 struct vsock_sock *vlistener;
376 struct vsock_sock *vconnected;
378 vlistener = vsock_sk(listener);
379 vconnected = vsock_sk(connected);
381 sock_hold(connected);
383 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
385 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
387 static struct sock *vsock_dequeue_accept(struct sock *listener)
389 struct vsock_sock *vlistener;
390 struct vsock_sock *vconnected;
392 vlistener = vsock_sk(listener);
394 if (list_empty(&vlistener->accept_queue))
397 vconnected = list_entry(vlistener->accept_queue.next,
398 struct vsock_sock, accept_queue);
400 list_del_init(&vconnected->accept_queue);
402 /* The caller will need a reference on the connected socket so we let
403 * it call sock_put().
406 return sk_vsock(vconnected);
409 static bool vsock_is_accept_queue_empty(struct sock *sk)
411 struct vsock_sock *vsk = vsock_sk(sk);
412 return list_empty(&vsk->accept_queue);
415 static bool vsock_is_pending(struct sock *sk)
417 struct vsock_sock *vsk = vsock_sk(sk);
418 return !list_empty(&vsk->pending_links);
421 static int vsock_send_shutdown(struct sock *sk, int mode)
423 return transport->shutdown(vsock_sk(sk), mode);
426 void vsock_pending_work(struct work_struct *work)
429 struct sock *listener;
430 struct vsock_sock *vsk;
433 vsk = container_of(work, struct vsock_sock, dwork.work);
435 listener = vsk->listener;
441 if (vsock_is_pending(sk)) {
442 vsock_remove_pending(listener, sk);
443 } else if (!vsk->rejected) {
444 /* We are not on the pending list and accept() did not reject
445 * us, so we must have been accepted by our user process. We
446 * just need to drop our references to the sockets and be on
453 listener->sk_ack_backlog--;
455 /* We need to remove ourself from the global connected sockets list so
456 * incoming packets can't find this socket, and to reduce the reference
459 if (vsock_in_connected_table(vsk))
460 vsock_remove_connected(vsk);
462 sk->sk_state = SS_FREE;
466 release_sock(listener);
473 EXPORT_SYMBOL_GPL(vsock_pending_work);
475 /**** SOCKET OPERATIONS ****/
477 static int __vsock_bind_stream(struct vsock_sock *vsk,
478 struct sockaddr_vm *addr)
480 static u32 port = LAST_RESERVED_PORT + 1;
481 struct sockaddr_vm new_addr;
483 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
485 if (addr->svm_port == VMADDR_PORT_ANY) {
489 for (i = 0; i < MAX_PORT_RETRIES; i++) {
490 if (port <= LAST_RESERVED_PORT)
491 port = LAST_RESERVED_PORT + 1;
493 new_addr.svm_port = port++;
495 if (!__vsock_find_bound_socket(&new_addr)) {
502 return -EADDRNOTAVAIL;
504 /* If port is in reserved range, ensure caller
505 * has necessary privileges.
507 if (addr->svm_port <= LAST_RESERVED_PORT &&
508 !capable(CAP_NET_BIND_SERVICE)) {
512 if (__vsock_find_bound_socket(&new_addr))
516 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
518 /* Remove stream sockets from the unbound list and add them to the hash
519 * table for easy lookup by its address. The unbound list is simply an
520 * extra entry at the end of the hash table, a trick used by AF_UNIX.
522 __vsock_remove_bound(vsk);
523 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
528 static int __vsock_bind_dgram(struct vsock_sock *vsk,
529 struct sockaddr_vm *addr)
531 return transport->dgram_bind(vsk, addr);
534 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
536 struct vsock_sock *vsk = vsock_sk(sk);
540 /* First ensure this socket isn't already bound. */
541 if (vsock_addr_bound(&vsk->local_addr))
544 /* Now bind to the provided address or select appropriate values if
545 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
546 * like AF_INET prevents binding to a non-local IP address (in most
547 * cases), we only allow binding to the local CID.
549 cid = transport->get_local_cid();
550 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
551 return -EADDRNOTAVAIL;
553 switch (sk->sk_socket->type) {
555 spin_lock_bh(&vsock_table_lock);
556 retval = __vsock_bind_stream(vsk, addr);
557 spin_unlock_bh(&vsock_table_lock);
561 retval = __vsock_bind_dgram(vsk, addr);
572 struct sock *__vsock_create(struct net *net,
579 struct vsock_sock *psk;
580 struct vsock_sock *vsk;
582 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto);
586 sock_init_data(sock, sk);
588 /* sk->sk_type is normally set in sock_init_data, but only if sock is
589 * non-NULL. We make sure that our sockets always have a type by
590 * setting it here if needed.
596 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
597 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
599 sk->sk_destruct = vsock_sk_destruct;
600 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
602 sock_reset_flag(sk, SOCK_DONE);
604 INIT_LIST_HEAD(&vsk->bound_table);
605 INIT_LIST_HEAD(&vsk->connected_table);
606 vsk->listener = NULL;
607 INIT_LIST_HEAD(&vsk->pending_links);
608 INIT_LIST_HEAD(&vsk->accept_queue);
609 vsk->rejected = false;
610 vsk->sent_request = false;
611 vsk->ignore_connecting_rst = false;
612 vsk->peer_shutdown = 0;
614 psk = parent ? vsock_sk(parent) : NULL;
616 vsk->trusted = psk->trusted;
617 vsk->owner = get_cred(psk->owner);
618 vsk->connect_timeout = psk->connect_timeout;
620 vsk->trusted = capable(CAP_NET_ADMIN);
621 vsk->owner = get_current_cred();
622 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
625 if (transport->init(vsk, psk) < 0) {
631 vsock_insert_unbound(vsk);
635 EXPORT_SYMBOL_GPL(__vsock_create);
637 static void __vsock_release(struct sock *sk)
641 struct sock *pending;
642 struct vsock_sock *vsk;
645 pending = NULL; /* Compiler warning. */
647 if (vsock_in_bound_table(vsk))
648 vsock_remove_bound(vsk);
650 if (vsock_in_connected_table(vsk))
651 vsock_remove_connected(vsk);
653 transport->release(vsk);
657 sk->sk_shutdown = SHUTDOWN_MASK;
659 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
662 /* Clean up any sockets that never were accepted. */
663 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
664 __vsock_release(pending);
673 static void vsock_sk_destruct(struct sock *sk)
675 struct vsock_sock *vsk = vsock_sk(sk);
677 transport->destruct(vsk);
679 /* When clearing these addresses, there's no need to set the family and
680 * possibly register the address family with the kernel.
682 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
683 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
685 put_cred(vsk->owner);
688 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
692 err = sock_queue_rcv_skb(sk, skb);
699 s64 vsock_stream_has_data(struct vsock_sock *vsk)
701 return transport->stream_has_data(vsk);
703 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
705 s64 vsock_stream_has_space(struct vsock_sock *vsk)
707 return transport->stream_has_space(vsk);
709 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
711 static int vsock_release(struct socket *sock)
713 __vsock_release(sock->sk);
715 sock->state = SS_FREE;
721 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
725 struct sockaddr_vm *vm_addr;
729 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
733 err = __vsock_bind(sk, vm_addr);
739 static int vsock_getname(struct socket *sock,
740 struct sockaddr *addr, int *addr_len, int peer)
744 struct vsock_sock *vsk;
745 struct sockaddr_vm *vm_addr;
754 if (sock->state != SS_CONNECTED) {
758 vm_addr = &vsk->remote_addr;
760 vm_addr = &vsk->local_addr;
768 /* sys_getsockname() and sys_getpeername() pass us a
769 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
770 * that macro is defined in socket.c instead of .h, so we hardcode its
773 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
774 memcpy(addr, vm_addr, sizeof(*vm_addr));
775 *addr_len = sizeof(*vm_addr);
782 static int vsock_shutdown(struct socket *sock, int mode)
787 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
788 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
789 * here like the other address families do. Note also that the
790 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
791 * which is what we want.
795 if ((mode & ~SHUTDOWN_MASK) || !mode)
798 /* If this is a STREAM socket and it is not connected then bail out
799 * immediately. If it is a DGRAM socket then we must first kick the
800 * socket so that it wakes up from any sleeping calls, for example
801 * recv(), and then afterwards return the error.
805 if (sock->state == SS_UNCONNECTED) {
807 if (sk->sk_type == SOCK_STREAM)
810 sock->state = SS_DISCONNECTING;
814 /* Receive and send shutdowns are treated alike. */
815 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
818 sk->sk_shutdown |= mode;
819 sk->sk_state_change(sk);
822 if (sk->sk_type == SOCK_STREAM) {
823 sock_reset_flag(sk, SOCK_DONE);
824 vsock_send_shutdown(sk, mode);
831 static unsigned int vsock_poll(struct file *file, struct socket *sock,
836 struct vsock_sock *vsk;
841 poll_wait(file, sk_sleep(sk), wait);
845 /* Signify that there has been an error on this socket. */
848 /* INET sockets treat local write shutdown and peer write shutdown as a
849 * case of POLLHUP set.
851 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
852 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
853 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
857 if (sk->sk_shutdown & RCV_SHUTDOWN ||
858 vsk->peer_shutdown & SEND_SHUTDOWN) {
862 if (sock->type == SOCK_DGRAM) {
863 /* For datagram sockets we can read if there is something in
864 * the queue and write as long as the socket isn't shutdown for
867 if (!skb_queue_empty(&sk->sk_receive_queue) ||
868 (sk->sk_shutdown & RCV_SHUTDOWN)) {
869 mask |= POLLIN | POLLRDNORM;
872 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
873 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
875 } else if (sock->type == SOCK_STREAM) {
878 /* Listening sockets that have connections in their accept
881 if (sk->sk_state == SS_LISTEN
882 && !vsock_is_accept_queue_empty(sk))
883 mask |= POLLIN | POLLRDNORM;
885 /* If there is something in the queue then we can read. */
886 if (transport->stream_is_active(vsk) &&
887 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
888 bool data_ready_now = false;
889 int ret = transport->notify_poll_in(
890 vsk, 1, &data_ready_now);
895 mask |= POLLIN | POLLRDNORM;
900 /* Sockets whose connections have been closed, reset, or
901 * terminated should also be considered read, and we check the
902 * shutdown flag for that.
904 if (sk->sk_shutdown & RCV_SHUTDOWN ||
905 vsk->peer_shutdown & SEND_SHUTDOWN) {
906 mask |= POLLIN | POLLRDNORM;
909 /* Connected sockets that can produce data can be written. */
910 if (sk->sk_state == SS_CONNECTED) {
911 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
912 bool space_avail_now = false;
913 int ret = transport->notify_poll_out(
914 vsk, 1, &space_avail_now);
919 /* Remove POLLWRBAND since INET
920 * sockets are not setting it.
922 mask |= POLLOUT | POLLWRNORM;
928 /* Simulate INET socket poll behaviors, which sets
929 * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
930 * but local send is not shutdown.
932 if (sk->sk_state == SS_UNCONNECTED) {
933 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
934 mask |= POLLOUT | POLLWRNORM;
944 static int vsock_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
945 struct msghdr *msg, size_t len)
949 struct vsock_sock *vsk;
950 struct sockaddr_vm *remote_addr;
952 if (msg->msg_flags & MSG_OOB)
955 /* For now, MSG_DONTWAIT is always assumed... */
962 if (!vsock_addr_bound(&vsk->local_addr)) {
963 struct sockaddr_vm local_addr;
965 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
966 err = __vsock_bind(sk, &local_addr);
972 /* If the provided message contains an address, use that. Otherwise
973 * fall back on the socket's remote handle (if it has been connected).
976 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
977 &remote_addr) == 0) {
978 /* Ensure this address is of the right type and is a valid
982 if (remote_addr->svm_cid == VMADDR_CID_ANY)
983 remote_addr->svm_cid = transport->get_local_cid();
985 if (!vsock_addr_bound(remote_addr)) {
989 } else if (sock->state == SS_CONNECTED) {
990 remote_addr = &vsk->remote_addr;
992 if (remote_addr->svm_cid == VMADDR_CID_ANY)
993 remote_addr->svm_cid = transport->get_local_cid();
995 /* XXX Should connect() or this function ensure remote_addr is
998 if (!vsock_addr_bound(&vsk->remote_addr)) {
1007 if (!transport->dgram_allow(remote_addr->svm_cid,
1008 remote_addr->svm_port)) {
1013 err = transport->dgram_enqueue(vsk, remote_addr, msg->msg_iov, len);
1020 static int vsock_dgram_connect(struct socket *sock,
1021 struct sockaddr *addr, int addr_len, int flags)
1025 struct vsock_sock *vsk;
1026 struct sockaddr_vm *remote_addr;
1031 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1032 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1034 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1036 sock->state = SS_UNCONNECTED;
1039 } else if (err != 0)
1044 if (!vsock_addr_bound(&vsk->local_addr)) {
1045 struct sockaddr_vm local_addr;
1047 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
1048 err = __vsock_bind(sk, &local_addr);
1054 if (!transport->dgram_allow(remote_addr->svm_cid,
1055 remote_addr->svm_port)) {
1060 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1061 sock->state = SS_CONNECTED;
1068 static int vsock_dgram_recvmsg(struct kiocb *kiocb, struct socket *sock,
1069 struct msghdr *msg, size_t len, int flags)
1071 return transport->dgram_dequeue(kiocb, vsock_sk(sock->sk), msg, len,
1075 static const struct proto_ops vsock_dgram_ops = {
1077 .owner = THIS_MODULE,
1078 .release = vsock_release,
1080 .connect = vsock_dgram_connect,
1081 .socketpair = sock_no_socketpair,
1082 .accept = sock_no_accept,
1083 .getname = vsock_getname,
1085 .ioctl = sock_no_ioctl,
1086 .listen = sock_no_listen,
1087 .shutdown = vsock_shutdown,
1088 .setsockopt = sock_no_setsockopt,
1089 .getsockopt = sock_no_getsockopt,
1090 .sendmsg = vsock_dgram_sendmsg,
1091 .recvmsg = vsock_dgram_recvmsg,
1092 .mmap = sock_no_mmap,
1093 .sendpage = sock_no_sendpage,
1096 static void vsock_connect_timeout(struct work_struct *work)
1099 struct vsock_sock *vsk;
1101 vsk = container_of(work, struct vsock_sock, dwork.work);
1105 if (sk->sk_state == SS_CONNECTING &&
1106 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1107 sk->sk_state = SS_UNCONNECTED;
1108 sk->sk_err = ETIMEDOUT;
1109 sk->sk_error_report(sk);
1116 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1117 int addr_len, int flags)
1121 struct vsock_sock *vsk;
1122 struct sockaddr_vm *remote_addr;
1132 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1133 switch (sock->state) {
1137 case SS_DISCONNECTING:
1141 /* This continues on so we can move sock into the SS_CONNECTED
1142 * state once the connection has completed (at which point err
1143 * will be set to zero also). Otherwise, we will either wait
1144 * for the connection or return -EALREADY should this be a
1145 * non-blocking call.
1150 if ((sk->sk_state == SS_LISTEN) ||
1151 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1156 /* The hypervisor and well-known contexts do not have socket
1159 if (!transport->stream_allow(remote_addr->svm_cid,
1160 remote_addr->svm_port)) {
1165 /* Set the remote address that we are connecting to. */
1166 memcpy(&vsk->remote_addr, remote_addr,
1167 sizeof(vsk->remote_addr));
1169 /* Autobind this socket to the local address if necessary. */
1170 if (!vsock_addr_bound(&vsk->local_addr)) {
1171 struct sockaddr_vm local_addr;
1173 vsock_addr_init(&local_addr, VMADDR_CID_ANY,
1175 err = __vsock_bind(sk, &local_addr);
1181 sk->sk_state = SS_CONNECTING;
1183 err = transport->connect(vsk);
1187 /* Mark sock as connecting and set the error code to in
1188 * progress in case this is a non-blocking connect.
1190 sock->state = SS_CONNECTING;
1194 /* The receive path will handle all communication until we are able to
1195 * enter the connected state. Here we wait for the connection to be
1196 * completed or a notification of an error.
1198 timeout = vsk->connect_timeout;
1199 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1201 while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1202 if (flags & O_NONBLOCK) {
1203 /* If we're not going to block, we schedule a timeout
1204 * function to generate a timeout on the connection
1205 * attempt, in case the peer doesn't respond in a
1206 * timely manner. We hold on to the socket until the
1210 INIT_DELAYED_WORK(&vsk->dwork,
1211 vsock_connect_timeout);
1212 schedule_delayed_work(&vsk->dwork, timeout);
1214 /* Skip ahead to preserve error code set above. */
1219 timeout = schedule_timeout(timeout);
1222 if (signal_pending(current)) {
1223 err = sock_intr_errno(timeout);
1224 goto out_wait_error;
1225 } else if (timeout == 0) {
1227 goto out_wait_error;
1230 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1235 goto out_wait_error;
1240 finish_wait(sk_sleep(sk), &wait);
1246 sk->sk_state = SS_UNCONNECTED;
1247 sock->state = SS_UNCONNECTED;
1251 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1253 struct sock *listener;
1255 struct sock *connected;
1256 struct vsock_sock *vconnected;
1261 listener = sock->sk;
1263 lock_sock(listener);
1265 if (sock->type != SOCK_STREAM) {
1270 if (listener->sk_state != SS_LISTEN) {
1275 /* Wait for children sockets to appear; these are the new sockets
1276 * created upon connection establishment.
1278 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1279 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1281 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1282 listener->sk_err == 0) {
1283 release_sock(listener);
1284 timeout = schedule_timeout(timeout);
1285 lock_sock(listener);
1287 if (signal_pending(current)) {
1288 err = sock_intr_errno(timeout);
1290 } else if (timeout == 0) {
1295 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1298 if (listener->sk_err)
1299 err = -listener->sk_err;
1302 listener->sk_ack_backlog--;
1304 lock_sock(connected);
1305 vconnected = vsock_sk(connected);
1307 /* If the listener socket has received an error, then we should
1308 * reject this socket and return. Note that we simply mark the
1309 * socket rejected, drop our reference, and let the cleanup
1310 * function handle the cleanup; the fact that we found it in
1311 * the listener's accept queue guarantees that the cleanup
1312 * function hasn't run yet.
1315 vconnected->rejected = true;
1316 release_sock(connected);
1317 sock_put(connected);
1321 newsock->state = SS_CONNECTED;
1322 sock_graft(connected, newsock);
1323 release_sock(connected);
1324 sock_put(connected);
1328 finish_wait(sk_sleep(listener), &wait);
1330 release_sock(listener);
1334 static int vsock_listen(struct socket *sock, int backlog)
1338 struct vsock_sock *vsk;
1344 if (sock->type != SOCK_STREAM) {
1349 if (sock->state != SS_UNCONNECTED) {
1356 if (!vsock_addr_bound(&vsk->local_addr)) {
1361 sk->sk_max_ack_backlog = backlog;
1362 sk->sk_state = SS_LISTEN;
1371 static int vsock_stream_setsockopt(struct socket *sock,
1374 char __user *optval,
1375 unsigned int optlen)
1379 struct vsock_sock *vsk;
1382 if (level != AF_VSOCK)
1383 return -ENOPROTOOPT;
1385 #define COPY_IN(_v) \
1387 if (optlen < sizeof(_v)) { \
1391 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \
1404 case SO_VM_SOCKETS_BUFFER_SIZE:
1406 transport->set_buffer_size(vsk, val);
1409 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1411 transport->set_max_buffer_size(vsk, val);
1414 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1416 transport->set_min_buffer_size(vsk, val);
1419 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1422 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1423 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1424 vsk->connect_timeout = tv.tv_sec * HZ +
1425 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1426 if (vsk->connect_timeout == 0)
1427 vsk->connect_timeout =
1428 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1448 static int vsock_stream_getsockopt(struct socket *sock,
1449 int level, int optname,
1450 char __user *optval,
1456 struct vsock_sock *vsk;
1459 if (level != AF_VSOCK)
1460 return -ENOPROTOOPT;
1462 err = get_user(len, optlen);
1466 #define COPY_OUT(_v) \
1468 if (len < sizeof(_v)) \
1472 if (copy_to_user(optval, &_v, len) != 0) \
1482 case SO_VM_SOCKETS_BUFFER_SIZE:
1483 val = transport->get_buffer_size(vsk);
1487 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1488 val = transport->get_max_buffer_size(vsk);
1492 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1493 val = transport->get_min_buffer_size(vsk);
1497 case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1499 tv.tv_sec = vsk->connect_timeout / HZ;
1501 (vsk->connect_timeout -
1502 tv.tv_sec * HZ) * (1000000 / HZ);
1507 return -ENOPROTOOPT;
1510 err = put_user(len, optlen);
1519 static int vsock_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1520 struct msghdr *msg, size_t len)
1523 struct vsock_sock *vsk;
1524 ssize_t total_written;
1527 struct vsock_transport_send_notify_data send_data;
1536 if (msg->msg_flags & MSG_OOB)
1541 /* Callers should not provide a destination with stream sockets. */
1542 if (msg->msg_namelen) {
1543 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1547 /* Send data only if both sides are not shutdown in the direction. */
1548 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1549 vsk->peer_shutdown & RCV_SHUTDOWN) {
1554 if (sk->sk_state != SS_CONNECTED ||
1555 !vsock_addr_bound(&vsk->local_addr)) {
1560 if (!vsock_addr_bound(&vsk->remote_addr)) {
1561 err = -EDESTADDRREQ;
1565 /* Wait for room in the produce queue to enqueue our user's data. */
1566 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1568 err = transport->notify_send_init(vsk, &send_data);
1572 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1574 while (total_written < len) {
1577 while (vsock_stream_has_space(vsk) == 0 &&
1579 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1580 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1582 /* Don't wait for non-blocking sockets. */
1588 err = transport->notify_send_pre_block(vsk, &send_data);
1593 timeout = schedule_timeout(timeout);
1595 if (signal_pending(current)) {
1596 err = sock_intr_errno(timeout);
1598 } else if (timeout == 0) {
1603 prepare_to_wait(sk_sleep(sk), &wait,
1604 TASK_INTERRUPTIBLE);
1607 /* These checks occur both as part of and after the loop
1608 * conditional since we need to check before and after
1614 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1615 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1620 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1624 /* Note that enqueue will only write as many bytes as are free
1625 * in the produce queue, so we don't need to ensure len is
1626 * smaller than the queue size. It is the caller's
1627 * responsibility to check how many bytes we were able to send.
1630 written = transport->stream_enqueue(
1632 len - total_written);
1638 total_written += written;
1640 err = transport->notify_send_post_enqueue(
1641 vsk, written, &send_data);
1648 if (total_written > 0)
1649 err = total_written;
1650 finish_wait(sk_sleep(sk), &wait);
1658 vsock_stream_recvmsg(struct kiocb *kiocb,
1659 struct socket *sock,
1660 struct msghdr *msg, size_t len, int flags)
1663 struct vsock_sock *vsk;
1668 struct vsock_transport_recv_notify_data recv_data;
1678 if (sk->sk_state != SS_CONNECTED) {
1679 /* Recvmsg is supposed to return 0 if a peer performs an
1680 * orderly shutdown. Differentiate between that case and when a
1681 * peer has not connected or a local shutdown occured with the
1684 if (sock_flag(sk, SOCK_DONE))
1692 if (flags & MSG_OOB) {
1697 /* We don't check peer_shutdown flag here since peer may actually shut
1698 * down, but there can be data in the queue that a local socket can
1701 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1706 /* It is valid on Linux to pass in a zero-length receive buffer. This
1707 * is not an error. We may as well bail out now.
1714 /* We must not copy less than target bytes into the user's buffer
1715 * before returning successfully, so we wait for the consume queue to
1716 * have that much data to consume before dequeueing. Note that this
1717 * makes it impossible to handle cases where target is greater than the
1720 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1721 if (target >= transport->stream_rcvhiwat(vsk)) {
1725 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1728 err = transport->notify_recv_init(vsk, target, &recv_data);
1732 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1735 s64 ready = vsock_stream_has_data(vsk);
1738 /* Invalid queue pair content. XXX This should be
1739 * changed to a connection reset in a later change.
1744 } else if (ready > 0) {
1747 err = transport->notify_recv_pre_dequeue(
1748 vsk, target, &recv_data);
1752 read = transport->stream_dequeue(
1754 len - copied, flags);
1762 err = transport->notify_recv_post_dequeue(
1764 !(flags & MSG_PEEK), &recv_data);
1768 if (read >= target || flags & MSG_PEEK)
1773 if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1774 || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1777 /* Don't wait for non-blocking sockets. */
1783 err = transport->notify_recv_pre_block(
1784 vsk, target, &recv_data);
1789 timeout = schedule_timeout(timeout);
1792 if (signal_pending(current)) {
1793 err = sock_intr_errno(timeout);
1795 } else if (timeout == 0) {
1800 prepare_to_wait(sk_sleep(sk), &wait,
1801 TASK_INTERRUPTIBLE);
1807 else if (sk->sk_shutdown & RCV_SHUTDOWN)
1811 /* We only do these additional bookkeeping/notification steps
1812 * if we actually copied something out of the queue pair
1813 * instead of just peeking ahead.
1816 if (!(flags & MSG_PEEK)) {
1817 /* If the other side has shutdown for sending and there
1818 * is nothing more to read, then modify the socket
1821 if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1822 if (vsock_stream_has_data(vsk) <= 0) {
1823 sk->sk_state = SS_UNCONNECTED;
1824 sock_set_flag(sk, SOCK_DONE);
1825 sk->sk_state_change(sk);
1833 finish_wait(sk_sleep(sk), &wait);
1839 static const struct proto_ops vsock_stream_ops = {
1841 .owner = THIS_MODULE,
1842 .release = vsock_release,
1844 .connect = vsock_stream_connect,
1845 .socketpair = sock_no_socketpair,
1846 .accept = vsock_accept,
1847 .getname = vsock_getname,
1849 .ioctl = sock_no_ioctl,
1850 .listen = vsock_listen,
1851 .shutdown = vsock_shutdown,
1852 .setsockopt = vsock_stream_setsockopt,
1853 .getsockopt = vsock_stream_getsockopt,
1854 .sendmsg = vsock_stream_sendmsg,
1855 .recvmsg = vsock_stream_recvmsg,
1856 .mmap = sock_no_mmap,
1857 .sendpage = sock_no_sendpage,
1860 static int vsock_create(struct net *net, struct socket *sock,
1861 int protocol, int kern)
1867 return -EPROTONOSUPPORT;
1869 switch (sock->type) {
1871 sock->ops = &vsock_dgram_ops;
1874 sock->ops = &vsock_stream_ops;
1877 return -ESOCKTNOSUPPORT;
1880 sock->state = SS_UNCONNECTED;
1882 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0) ? 0 : -ENOMEM;
1885 static const struct net_proto_family vsock_family_ops = {
1887 .create = vsock_create,
1888 .owner = THIS_MODULE,
1891 static long vsock_dev_do_ioctl(struct file *filp,
1892 unsigned int cmd, void __user *ptr)
1894 u32 __user *p = ptr;
1898 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1899 if (put_user(transport->get_local_cid(), p) != 0)
1904 pr_err("Unknown ioctl %d\n", cmd);
1911 static long vsock_dev_ioctl(struct file *filp,
1912 unsigned int cmd, unsigned long arg)
1914 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1917 #ifdef CONFIG_COMPAT
1918 static long vsock_dev_compat_ioctl(struct file *filp,
1919 unsigned int cmd, unsigned long arg)
1921 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1925 static const struct file_operations vsock_device_ops = {
1926 .owner = THIS_MODULE,
1927 .unlocked_ioctl = vsock_dev_ioctl,
1928 #ifdef CONFIG_COMPAT
1929 .compat_ioctl = vsock_dev_compat_ioctl,
1931 .open = nonseekable_open,
1934 static struct miscdevice vsock_device = {
1936 .minor = MISC_DYNAMIC_MINOR,
1937 .fops = &vsock_device_ops,
1940 static int __vsock_core_init(void)
1944 vsock_init_tables();
1946 err = misc_register(&vsock_device);
1948 pr_err("Failed to register misc device\n");
1952 err = proto_register(&vsock_proto, 1); /* we want our slab */
1954 pr_err("Cannot register vsock protocol\n");
1955 goto err_misc_deregister;
1958 err = sock_register(&vsock_family_ops);
1960 pr_err("could not register af_vsock (%d) address family: %d\n",
1962 goto err_unregister_proto;
1967 err_unregister_proto:
1968 proto_unregister(&vsock_proto);
1969 err_misc_deregister:
1970 misc_deregister(&vsock_device);
1974 int vsock_core_init(const struct vsock_transport *t)
1976 int retval = mutex_lock_interruptible(&vsock_register_mutex);
1986 retval = __vsock_core_init();
1991 mutex_unlock(&vsock_register_mutex);
1994 EXPORT_SYMBOL_GPL(vsock_core_init);
1996 void vsock_core_exit(void)
1998 mutex_lock(&vsock_register_mutex);
2000 misc_deregister(&vsock_device);
2001 sock_unregister(AF_VSOCK);
2002 proto_unregister(&vsock_proto);
2004 /* We do not want the assignment below re-ordered. */
2008 mutex_unlock(&vsock_register_mutex);
2010 EXPORT_SYMBOL_GPL(vsock_core_exit);
2012 MODULE_AUTHOR("VMware, Inc.");
2013 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2014 MODULE_VERSION(VSOCK_DRIVER_VERSION_STRING);
2015 MODULE_LICENSE("GPL v2");