1 // SPDX-License-Identifier: GPL-2.0-only
3 * VMware vSockets Driver
5 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
8 /* Implementation notes:
10 * - There are two kinds of sockets: those created by user action (such as
11 * calling socket(2)) and those created by incoming connection request packets.
13 * - There are two "global" tables, one for bound sockets (sockets that have
14 * specified an address that they are responsible for) and one for connected
15 * sockets (sockets that have established a connection with another socket).
16 * These tables are "global" in that all sockets on the system are placed
17 * within them. - Note, though, that the bound table contains an extra entry
18 * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
19 * that list. The bound table is used solely for lookup of sockets when packets
20 * are received and that's not necessary for SOCK_DGRAM sockets since we create
21 * a datagram handle for each and need not perform a lookup. Keeping SOCK_DGRAM
22 * sockets out of the bound hash buckets will reduce the chance of collisions
23 * when looking for SOCK_STREAM sockets and prevents us from having to check the
24 * socket type in the hash table lookups.
26 * - Sockets created by user action will either be "client" sockets that
27 * initiate a connection or "server" sockets that listen for connections; we do
28 * not support simultaneous connects (two "client" sockets connecting).
30 * - "Server" sockets are referred to as listener sockets throughout this
31 * implementation because they are in the TCP_LISTEN state. When a
32 * connection request is received (the second kind of socket mentioned above),
33 * we create a new socket and refer to it as a pending socket. These pending
34 * sockets are placed on the pending connection list of the listener socket.
35 * When future packets are received for the address the listener socket is
36 * bound to, we check if the source of the packet is from one that has an
37 * existing pending connection. If it does, we process the packet for the
38 * pending socket. When that socket reaches the connected state, it is removed
39 * from the listener socket's pending list and enqueued in the listener
40 * socket's accept queue. Callers of accept(2) will accept connected sockets
41 * from the listener socket's accept queue. If the socket cannot be accepted
42 * for some reason then it is marked rejected. Once the connection is
43 * accepted, it is owned by the user process and the responsibility for cleanup
44 * falls with that user process.
46 * - It is possible that these pending sockets will never reach the connected
47 * state; in fact, we may never receive another packet after the connection
48 * request. Because of this, we must schedule a cleanup function to run in the
49 * future, after some amount of time passes where a connection should have been
50 * established. This function ensures that the socket is off all lists so it
51 * cannot be retrieved, then drops all references to the socket so it is cleaned
52 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this
53 * function will also cleanup rejected sockets, those that reach the connected
54 * state but leave it before they have been accepted.
56 * - Lock ordering for pending or accept queue sockets is:
58 * lock_sock(listener);
59 * lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
61 * Using explicit nested locking keeps lockdep happy since normally only one
62 * lock of a given class may be taken at a time.
64 * - Sockets created by user action will be cleaned up when the user process
65 * calls close(2), causing our release implementation to be called. Our release
66 * implementation will perform some cleanup then drop the last reference so our
67 * sk_destruct implementation is invoked. Our sk_destruct implementation will
68 * perform additional cleanup that's common for both types of sockets.
70 * - A socket's reference count is what ensures that the structure won't be
71 * freed. Each entry in a list (such as the "global" bound and connected tables
72 * and the listener socket's pending list and connected queue) ensures a
73 * reference. When we defer work until process context and pass a socket as our
74 * argument, we must ensure the reference count is increased to ensure the
75 * socket isn't freed before the function is run; the deferred function will
76 * then drop the reference.
78 * - sk->sk_state uses the TCP state constants because they are widely used by
79 * other address families and exposed to userspace tools like ss(8):
81 * TCP_CLOSE - unconnected
82 * TCP_SYN_SENT - connecting
83 * TCP_ESTABLISHED - connected
84 * TCP_CLOSING - disconnecting
85 * TCP_LISTEN - listening
88 #include <linux/compat.h>
89 #include <linux/types.h>
90 #include <linux/bitops.h>
91 #include <linux/cred.h>
92 #include <linux/errqueue.h>
93 #include <linux/init.h>
95 #include <linux/kernel.h>
96 #include <linux/sched/signal.h>
97 #include <linux/kmod.h>
98 #include <linux/list.h>
99 #include <linux/miscdevice.h>
100 #include <linux/module.h>
101 #include <linux/mutex.h>
102 #include <linux/net.h>
103 #include <linux/poll.h>
104 #include <linux/random.h>
105 #include <linux/skbuff.h>
106 #include <linux/smp.h>
107 #include <linux/socket.h>
108 #include <linux/stddef.h>
109 #include <linux/unistd.h>
110 #include <linux/wait.h>
111 #include <linux/workqueue.h>
112 #include <net/sock.h>
113 #include <net/af_vsock.h>
114 #include <uapi/linux/vm_sockets.h>
116 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
117 static void vsock_sk_destruct(struct sock *sk);
118 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
120 /* Protocol family. */
121 struct proto vsock_proto = {
123 .owner = THIS_MODULE,
124 .obj_size = sizeof(struct vsock_sock),
125 #ifdef CONFIG_BPF_SYSCALL
126 .psock_update_sk_prot = vsock_bpf_update_proto,
130 /* The default peer timeout indicates how long we will wait for a peer response
131 * to a control message.
133 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
135 #define VSOCK_DEFAULT_BUFFER_SIZE (1024 * 256)
136 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
137 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
139 /* Transport used for host->guest communication */
140 static const struct vsock_transport *transport_h2g;
141 /* Transport used for guest->host communication */
142 static const struct vsock_transport *transport_g2h;
143 /* Transport used for DGRAM communication */
144 static const struct vsock_transport *transport_dgram;
145 /* Transport used for local communication */
146 static const struct vsock_transport *transport_local;
147 static DEFINE_MUTEX(vsock_register_mutex);
151 /* Each bound VSocket is stored in the bind hash table and each connected
152 * VSocket is stored in the connected hash table.
154 * Unbound sockets are all put on the same list attached to the end of the hash
155 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in
156 * the bucket that their local address hashes to (vsock_bound_sockets(addr)
157 * represents the list that addr hashes to).
159 * Specifically, we initialize the vsock_bind_table array to a size of
160 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
161 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
162 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function
163 * mods with VSOCK_HASH_SIZE to ensure this.
165 #define MAX_PORT_RETRIES 24
167 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE)
168 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
169 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE])
171 /* XXX This can probably be implemented in a better way. */
172 #define VSOCK_CONN_HASH(src, dst) \
173 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
174 #define vsock_connected_sockets(src, dst) \
175 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
176 #define vsock_connected_sockets_vsk(vsk) \
177 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
179 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
180 EXPORT_SYMBOL_GPL(vsock_bind_table);
181 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
182 EXPORT_SYMBOL_GPL(vsock_connected_table);
183 DEFINE_SPINLOCK(vsock_table_lock);
184 EXPORT_SYMBOL_GPL(vsock_table_lock);
186 /* Autobind this socket to the local address if necessary. */
187 static int vsock_auto_bind(struct vsock_sock *vsk)
189 struct sock *sk = sk_vsock(vsk);
190 struct sockaddr_vm local_addr;
192 if (vsock_addr_bound(&vsk->local_addr))
194 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
195 return __vsock_bind(sk, &local_addr);
198 static void vsock_init_tables(void)
202 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
203 INIT_LIST_HEAD(&vsock_bind_table[i]);
205 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
206 INIT_LIST_HEAD(&vsock_connected_table[i]);
209 static void __vsock_insert_bound(struct list_head *list,
210 struct vsock_sock *vsk)
213 list_add(&vsk->bound_table, list);
216 static void __vsock_insert_connected(struct list_head *list,
217 struct vsock_sock *vsk)
220 list_add(&vsk->connected_table, list);
223 static void __vsock_remove_bound(struct vsock_sock *vsk)
225 list_del_init(&vsk->bound_table);
229 static void __vsock_remove_connected(struct vsock_sock *vsk)
231 list_del_init(&vsk->connected_table);
235 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
237 struct vsock_sock *vsk;
239 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
240 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
241 return sk_vsock(vsk);
243 if (addr->svm_port == vsk->local_addr.svm_port &&
244 (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
245 addr->svm_cid == VMADDR_CID_ANY))
246 return sk_vsock(vsk);
252 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
253 struct sockaddr_vm *dst)
255 struct vsock_sock *vsk;
257 list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
259 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
260 dst->svm_port == vsk->local_addr.svm_port) {
261 return sk_vsock(vsk);
268 static void vsock_insert_unbound(struct vsock_sock *vsk)
270 spin_lock_bh(&vsock_table_lock);
271 __vsock_insert_bound(vsock_unbound_sockets, vsk);
272 spin_unlock_bh(&vsock_table_lock);
275 void vsock_insert_connected(struct vsock_sock *vsk)
277 struct list_head *list = vsock_connected_sockets(
278 &vsk->remote_addr, &vsk->local_addr);
280 spin_lock_bh(&vsock_table_lock);
281 __vsock_insert_connected(list, vsk);
282 spin_unlock_bh(&vsock_table_lock);
284 EXPORT_SYMBOL_GPL(vsock_insert_connected);
286 void vsock_remove_bound(struct vsock_sock *vsk)
288 spin_lock_bh(&vsock_table_lock);
289 if (__vsock_in_bound_table(vsk))
290 __vsock_remove_bound(vsk);
291 spin_unlock_bh(&vsock_table_lock);
293 EXPORT_SYMBOL_GPL(vsock_remove_bound);
295 void vsock_remove_connected(struct vsock_sock *vsk)
297 spin_lock_bh(&vsock_table_lock);
298 if (__vsock_in_connected_table(vsk))
299 __vsock_remove_connected(vsk);
300 spin_unlock_bh(&vsock_table_lock);
302 EXPORT_SYMBOL_GPL(vsock_remove_connected);
304 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
308 spin_lock_bh(&vsock_table_lock);
309 sk = __vsock_find_bound_socket(addr);
313 spin_unlock_bh(&vsock_table_lock);
317 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
319 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
320 struct sockaddr_vm *dst)
324 spin_lock_bh(&vsock_table_lock);
325 sk = __vsock_find_connected_socket(src, dst);
329 spin_unlock_bh(&vsock_table_lock);
333 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
335 void vsock_remove_sock(struct vsock_sock *vsk)
337 vsock_remove_bound(vsk);
338 vsock_remove_connected(vsk);
340 EXPORT_SYMBOL_GPL(vsock_remove_sock);
342 void vsock_for_each_connected_socket(struct vsock_transport *transport,
343 void (*fn)(struct sock *sk))
347 spin_lock_bh(&vsock_table_lock);
349 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
350 struct vsock_sock *vsk;
351 list_for_each_entry(vsk, &vsock_connected_table[i],
353 if (vsk->transport != transport)
360 spin_unlock_bh(&vsock_table_lock);
362 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
364 void vsock_add_pending(struct sock *listener, struct sock *pending)
366 struct vsock_sock *vlistener;
367 struct vsock_sock *vpending;
369 vlistener = vsock_sk(listener);
370 vpending = vsock_sk(pending);
374 list_add_tail(&vpending->pending_links, &vlistener->pending_links);
376 EXPORT_SYMBOL_GPL(vsock_add_pending);
378 void vsock_remove_pending(struct sock *listener, struct sock *pending)
380 struct vsock_sock *vpending = vsock_sk(pending);
382 list_del_init(&vpending->pending_links);
386 EXPORT_SYMBOL_GPL(vsock_remove_pending);
388 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
390 struct vsock_sock *vlistener;
391 struct vsock_sock *vconnected;
393 vlistener = vsock_sk(listener);
394 vconnected = vsock_sk(connected);
396 sock_hold(connected);
398 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
400 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
402 static bool vsock_use_local_transport(unsigned int remote_cid)
404 if (!transport_local)
407 if (remote_cid == VMADDR_CID_LOCAL)
411 return remote_cid == transport_g2h->get_local_cid();
413 return remote_cid == VMADDR_CID_HOST;
417 static void vsock_deassign_transport(struct vsock_sock *vsk)
422 vsk->transport->destruct(vsk);
423 module_put(vsk->transport->module);
424 vsk->transport = NULL;
427 /* Assign a transport to a socket and call the .init transport callback.
429 * Note: for connection oriented socket this must be called when vsk->remote_addr
430 * is set (e.g. during the connect() or when a connection request on a listener
431 * socket is received).
432 * The vsk->remote_addr is used to decide which transport to use:
433 * - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
434 * g2h is not loaded, will use local transport;
435 * - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
436 * includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
437 * - remote CID > VMADDR_CID_HOST will use host->guest transport;
439 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
441 const struct vsock_transport *new_transport;
442 struct sock *sk = sk_vsock(vsk);
443 unsigned int remote_cid = vsk->remote_addr.svm_cid;
447 /* If the packet is coming with the source and destination CIDs higher
448 * than VMADDR_CID_HOST, then a vsock channel where all the packets are
449 * forwarded to the host should be established. Then the host will
450 * need to forward the packets to the guest.
452 * The flag is set on the (listen) receive path (psk is not NULL). On
453 * the connect path the flag can be set by the user space application.
455 if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
456 vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
457 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
459 remote_flags = vsk->remote_addr.svm_flags;
461 switch (sk->sk_type) {
463 new_transport = transport_dgram;
467 if (vsock_use_local_transport(remote_cid))
468 new_transport = transport_local;
469 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
470 (remote_flags & VMADDR_FLAG_TO_HOST))
471 new_transport = transport_g2h;
473 new_transport = transport_h2g;
476 return -ESOCKTNOSUPPORT;
479 if (vsk->transport) {
480 if (vsk->transport == new_transport)
483 /* transport->release() must be called with sock lock acquired.
484 * This path can only be taken during vsock_connect(), where we
485 * have already held the sock lock. In the other cases, this
486 * function is called on a new socket which is not assigned to
489 vsk->transport->release(vsk);
490 vsock_deassign_transport(vsk);
493 /* We increase the module refcnt to prevent the transport unloading
494 * while there are open sockets assigned to it.
496 if (!new_transport || !try_module_get(new_transport->module))
499 if (sk->sk_type == SOCK_SEQPACKET) {
500 if (!new_transport->seqpacket_allow ||
501 !new_transport->seqpacket_allow(remote_cid)) {
502 module_put(new_transport->module);
503 return -ESOCKTNOSUPPORT;
507 ret = new_transport->init(vsk, psk);
509 module_put(new_transport->module);
513 vsk->transport = new_transport;
517 EXPORT_SYMBOL_GPL(vsock_assign_transport);
519 bool vsock_find_cid(unsigned int cid)
521 if (transport_g2h && cid == transport_g2h->get_local_cid())
524 if (transport_h2g && cid == VMADDR_CID_HOST)
527 if (transport_local && cid == VMADDR_CID_LOCAL)
532 EXPORT_SYMBOL_GPL(vsock_find_cid);
534 static struct sock *vsock_dequeue_accept(struct sock *listener)
536 struct vsock_sock *vlistener;
537 struct vsock_sock *vconnected;
539 vlistener = vsock_sk(listener);
541 if (list_empty(&vlistener->accept_queue))
544 vconnected = list_entry(vlistener->accept_queue.next,
545 struct vsock_sock, accept_queue);
547 list_del_init(&vconnected->accept_queue);
549 /* The caller will need a reference on the connected socket so we let
550 * it call sock_put().
553 return sk_vsock(vconnected);
556 static bool vsock_is_accept_queue_empty(struct sock *sk)
558 struct vsock_sock *vsk = vsock_sk(sk);
559 return list_empty(&vsk->accept_queue);
562 static bool vsock_is_pending(struct sock *sk)
564 struct vsock_sock *vsk = vsock_sk(sk);
565 return !list_empty(&vsk->pending_links);
568 static int vsock_send_shutdown(struct sock *sk, int mode)
570 struct vsock_sock *vsk = vsock_sk(sk);
575 return vsk->transport->shutdown(vsk, mode);
578 static void vsock_pending_work(struct work_struct *work)
581 struct sock *listener;
582 struct vsock_sock *vsk;
585 vsk = container_of(work, struct vsock_sock, pending_work.work);
587 listener = vsk->listener;
591 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
593 if (vsock_is_pending(sk)) {
594 vsock_remove_pending(listener, sk);
596 sk_acceptq_removed(listener);
597 } else if (!vsk->rejected) {
598 /* We are not on the pending list and accept() did not reject
599 * us, so we must have been accepted by our user process. We
600 * just need to drop our references to the sockets and be on
607 /* We need to remove ourself from the global connected sockets list so
608 * incoming packets can't find this socket, and to reduce the reference
611 vsock_remove_connected(vsk);
613 sk->sk_state = TCP_CLOSE;
617 release_sock(listener);
625 /**** SOCKET OPERATIONS ****/
627 static int __vsock_bind_connectible(struct vsock_sock *vsk,
628 struct sockaddr_vm *addr)
631 struct sockaddr_vm new_addr;
634 port = get_random_u32_above(LAST_RESERVED_PORT);
636 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
638 if (addr->svm_port == VMADDR_PORT_ANY) {
642 for (i = 0; i < MAX_PORT_RETRIES; i++) {
643 if (port <= LAST_RESERVED_PORT)
644 port = LAST_RESERVED_PORT + 1;
646 new_addr.svm_port = port++;
648 if (!__vsock_find_bound_socket(&new_addr)) {
655 return -EADDRNOTAVAIL;
657 /* If port is in reserved range, ensure caller
658 * has necessary privileges.
660 if (addr->svm_port <= LAST_RESERVED_PORT &&
661 !capable(CAP_NET_BIND_SERVICE)) {
665 if (__vsock_find_bound_socket(&new_addr))
669 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
671 /* Remove connection oriented sockets from the unbound list and add them
672 * to the hash table for easy lookup by its address. The unbound list
673 * is simply an extra entry at the end of the hash table, a trick used
676 __vsock_remove_bound(vsk);
677 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
682 static int __vsock_bind_dgram(struct vsock_sock *vsk,
683 struct sockaddr_vm *addr)
685 return vsk->transport->dgram_bind(vsk, addr);
688 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
690 struct vsock_sock *vsk = vsock_sk(sk);
693 /* First ensure this socket isn't already bound. */
694 if (vsock_addr_bound(&vsk->local_addr))
697 /* Now bind to the provided address or select appropriate values if
698 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that
699 * like AF_INET prevents binding to a non-local IP address (in most
700 * cases), we only allow binding to a local CID.
702 if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
703 return -EADDRNOTAVAIL;
705 switch (sk->sk_socket->type) {
708 spin_lock_bh(&vsock_table_lock);
709 retval = __vsock_bind_connectible(vsk, addr);
710 spin_unlock_bh(&vsock_table_lock);
714 retval = __vsock_bind_dgram(vsk, addr);
725 static void vsock_connect_timeout(struct work_struct *work);
727 static struct sock *__vsock_create(struct net *net,
735 struct vsock_sock *psk;
736 struct vsock_sock *vsk;
738 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
742 sock_init_data(sock, sk);
744 /* sk->sk_type is normally set in sock_init_data, but only if sock is
745 * non-NULL. We make sure that our sockets always have a type by
746 * setting it here if needed.
752 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
753 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
755 sk->sk_destruct = vsock_sk_destruct;
756 sk->sk_backlog_rcv = vsock_queue_rcv_skb;
757 sock_reset_flag(sk, SOCK_DONE);
759 INIT_LIST_HEAD(&vsk->bound_table);
760 INIT_LIST_HEAD(&vsk->connected_table);
761 vsk->listener = NULL;
762 INIT_LIST_HEAD(&vsk->pending_links);
763 INIT_LIST_HEAD(&vsk->accept_queue);
764 vsk->rejected = false;
765 vsk->sent_request = false;
766 vsk->ignore_connecting_rst = false;
767 vsk->peer_shutdown = 0;
768 INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
769 INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
771 psk = parent ? vsock_sk(parent) : NULL;
773 vsk->trusted = psk->trusted;
774 vsk->owner = get_cred(psk->owner);
775 vsk->connect_timeout = psk->connect_timeout;
776 vsk->buffer_size = psk->buffer_size;
777 vsk->buffer_min_size = psk->buffer_min_size;
778 vsk->buffer_max_size = psk->buffer_max_size;
779 security_sk_clone(parent, sk);
781 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
782 vsk->owner = get_current_cred();
783 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
784 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
785 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
786 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
792 static bool sock_type_connectible(u16 type)
794 return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
797 static void __vsock_release(struct sock *sk, int level)
800 struct sock *pending;
801 struct vsock_sock *vsk;
804 pending = NULL; /* Compiler warning. */
806 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
807 * version to avoid the warning "possible recursive locking
808 * detected". When "level" is 0, lock_sock_nested(sk, level)
809 * is the same as lock_sock(sk).
811 lock_sock_nested(sk, level);
814 vsk->transport->release(vsk);
815 else if (sock_type_connectible(sk->sk_type))
816 vsock_remove_sock(vsk);
819 sk->sk_shutdown = SHUTDOWN_MASK;
821 skb_queue_purge(&sk->sk_receive_queue);
823 /* Clean up any sockets that never were accepted. */
824 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
825 __vsock_release(pending, SINGLE_DEPTH_NESTING);
834 static void vsock_sk_destruct(struct sock *sk)
836 struct vsock_sock *vsk = vsock_sk(sk);
838 vsock_deassign_transport(vsk);
840 /* When clearing these addresses, there's no need to set the family and
841 * possibly register the address family with the kernel.
843 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
844 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
846 put_cred(vsk->owner);
849 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
853 err = sock_queue_rcv_skb(sk, skb);
860 struct sock *vsock_create_connected(struct sock *parent)
862 return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
865 EXPORT_SYMBOL_GPL(vsock_create_connected);
867 s64 vsock_stream_has_data(struct vsock_sock *vsk)
869 return vsk->transport->stream_has_data(vsk);
871 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
873 s64 vsock_connectible_has_data(struct vsock_sock *vsk)
875 struct sock *sk = sk_vsock(vsk);
877 if (sk->sk_type == SOCK_SEQPACKET)
878 return vsk->transport->seqpacket_has_data(vsk);
880 return vsock_stream_has_data(vsk);
882 EXPORT_SYMBOL_GPL(vsock_connectible_has_data);
884 s64 vsock_stream_has_space(struct vsock_sock *vsk)
886 return vsk->transport->stream_has_space(vsk);
888 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
890 void vsock_data_ready(struct sock *sk)
892 struct vsock_sock *vsk = vsock_sk(sk);
894 if (vsock_stream_has_data(vsk) >= sk->sk_rcvlowat ||
895 sock_flag(sk, SOCK_DONE))
896 sk->sk_data_ready(sk);
898 EXPORT_SYMBOL_GPL(vsock_data_ready);
900 static int vsock_release(struct socket *sock)
902 __vsock_release(sock->sk, 0);
904 sock->state = SS_FREE;
910 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
914 struct sockaddr_vm *vm_addr;
918 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
922 err = __vsock_bind(sk, vm_addr);
928 static int vsock_getname(struct socket *sock,
929 struct sockaddr *addr, int peer)
933 struct vsock_sock *vsk;
934 struct sockaddr_vm *vm_addr;
943 if (sock->state != SS_CONNECTED) {
947 vm_addr = &vsk->remote_addr;
949 vm_addr = &vsk->local_addr;
957 /* sys_getsockname() and sys_getpeername() pass us a
958 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately
959 * that macro is defined in socket.c instead of .h, so we hardcode its
962 BUILD_BUG_ON(sizeof(*vm_addr) > 128);
963 memcpy(addr, vm_addr, sizeof(*vm_addr));
964 err = sizeof(*vm_addr);
971 static int vsock_shutdown(struct socket *sock, int mode)
976 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
977 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
978 * here like the other address families do. Note also that the
979 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
980 * which is what we want.
984 if ((mode & ~SHUTDOWN_MASK) || !mode)
987 /* If this is a connection oriented socket and it is not connected then
988 * bail out immediately. If it is a DGRAM socket then we must first
989 * kick the socket so that it wakes up from any sleeping calls, for
990 * example recv(), and then afterwards return the error.
996 if (sock->state == SS_UNCONNECTED) {
998 if (sock_type_connectible(sk->sk_type))
1001 sock->state = SS_DISCONNECTING;
1005 /* Receive and send shutdowns are treated alike. */
1006 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
1008 sk->sk_shutdown |= mode;
1009 sk->sk_state_change(sk);
1011 if (sock_type_connectible(sk->sk_type)) {
1012 sock_reset_flag(sk, SOCK_DONE);
1013 vsock_send_shutdown(sk, mode);
1022 static __poll_t vsock_poll(struct file *file, struct socket *sock,
1027 struct vsock_sock *vsk;
1032 poll_wait(file, sk_sleep(sk), wait);
1035 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
1036 /* Signify that there has been an error on this socket. */
1039 /* INET sockets treat local write shutdown and peer write shutdown as a
1040 * case of EPOLLHUP set.
1042 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1043 ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1044 (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1048 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1049 vsk->peer_shutdown & SEND_SHUTDOWN) {
1053 if (sock->type == SOCK_DGRAM) {
1054 /* For datagram sockets we can read if there is something in
1055 * the queue and write as long as the socket isn't shutdown for
1058 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1059 (sk->sk_shutdown & RCV_SHUTDOWN)) {
1060 mask |= EPOLLIN | EPOLLRDNORM;
1063 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1064 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1066 } else if (sock_type_connectible(sk->sk_type)) {
1067 const struct vsock_transport *transport;
1071 transport = vsk->transport;
1073 /* Listening sockets that have connections in their accept
1074 * queue can be read.
1076 if (sk->sk_state == TCP_LISTEN
1077 && !vsock_is_accept_queue_empty(sk))
1078 mask |= EPOLLIN | EPOLLRDNORM;
1080 /* If there is something in the queue then we can read. */
1081 if (transport && transport->stream_is_active(vsk) &&
1082 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1083 bool data_ready_now = false;
1084 int target = sock_rcvlowat(sk, 0, INT_MAX);
1085 int ret = transport->notify_poll_in(
1086 vsk, target, &data_ready_now);
1091 mask |= EPOLLIN | EPOLLRDNORM;
1096 /* Sockets whose connections have been closed, reset, or
1097 * terminated should also be considered read, and we check the
1098 * shutdown flag for that.
1100 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1101 vsk->peer_shutdown & SEND_SHUTDOWN) {
1102 mask |= EPOLLIN | EPOLLRDNORM;
1105 /* Connected sockets that can produce data can be written. */
1106 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1107 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1108 bool space_avail_now = false;
1109 int ret = transport->notify_poll_out(
1110 vsk, 1, &space_avail_now);
1114 if (space_avail_now)
1115 /* Remove EPOLLWRBAND since INET
1116 * sockets are not setting it.
1118 mask |= EPOLLOUT | EPOLLWRNORM;
1124 /* Simulate INET socket poll behaviors, which sets
1125 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1126 * but local send is not shutdown.
1128 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1129 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1130 mask |= EPOLLOUT | EPOLLWRNORM;
1140 static int vsock_read_skb(struct sock *sk, skb_read_actor_t read_actor)
1142 struct vsock_sock *vsk = vsock_sk(sk);
1144 return vsk->transport->read_skb(vsk, read_actor);
1147 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1152 struct vsock_sock *vsk;
1153 struct sockaddr_vm *remote_addr;
1154 const struct vsock_transport *transport;
1156 if (msg->msg_flags & MSG_OOB)
1159 /* For now, MSG_DONTWAIT is always assumed... */
1166 transport = vsk->transport;
1168 err = vsock_auto_bind(vsk);
1173 /* If the provided message contains an address, use that. Otherwise
1174 * fall back on the socket's remote handle (if it has been connected).
1176 if (msg->msg_name &&
1177 vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1178 &remote_addr) == 0) {
1179 /* Ensure this address is of the right type and is a valid
1183 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1184 remote_addr->svm_cid = transport->get_local_cid();
1186 if (!vsock_addr_bound(remote_addr)) {
1190 } else if (sock->state == SS_CONNECTED) {
1191 remote_addr = &vsk->remote_addr;
1193 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1194 remote_addr->svm_cid = transport->get_local_cid();
1196 /* XXX Should connect() or this function ensure remote_addr is
1199 if (!vsock_addr_bound(&vsk->remote_addr)) {
1208 if (!transport->dgram_allow(remote_addr->svm_cid,
1209 remote_addr->svm_port)) {
1214 err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1221 static int vsock_dgram_connect(struct socket *sock,
1222 struct sockaddr *addr, int addr_len, int flags)
1226 struct vsock_sock *vsk;
1227 struct sockaddr_vm *remote_addr;
1232 err = vsock_addr_cast(addr, addr_len, &remote_addr);
1233 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1235 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1237 sock->state = SS_UNCONNECTED;
1240 } else if (err != 0)
1245 err = vsock_auto_bind(vsk);
1249 if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1250 remote_addr->svm_port)) {
1255 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1256 sock->state = SS_CONNECTED;
1258 /* sock map disallows redirection of non-TCP sockets with sk_state !=
1259 * TCP_ESTABLISHED (see sock_map_redirect_allowed()), so we set
1260 * TCP_ESTABLISHED here to allow redirection of connected vsock dgrams.
1262 * This doesn't seem to be abnormal state for datagram sockets, as the
1263 * same approach can be see in other datagram socket types as well
1264 * (such as unix sockets).
1266 sk->sk_state = TCP_ESTABLISHED;
1273 int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1274 size_t len, int flags)
1276 #ifdef CONFIG_BPF_SYSCALL
1277 const struct proto *prot;
1279 struct vsock_sock *vsk;
1285 #ifdef CONFIG_BPF_SYSCALL
1286 prot = READ_ONCE(sk->sk_prot);
1287 if (prot != &vsock_proto)
1288 return prot->recvmsg(sk, msg, len, flags, NULL);
1291 return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1293 EXPORT_SYMBOL_GPL(vsock_dgram_recvmsg);
1295 static const struct proto_ops vsock_dgram_ops = {
1297 .owner = THIS_MODULE,
1298 .release = vsock_release,
1300 .connect = vsock_dgram_connect,
1301 .socketpair = sock_no_socketpair,
1302 .accept = sock_no_accept,
1303 .getname = vsock_getname,
1305 .ioctl = sock_no_ioctl,
1306 .listen = sock_no_listen,
1307 .shutdown = vsock_shutdown,
1308 .sendmsg = vsock_dgram_sendmsg,
1309 .recvmsg = vsock_dgram_recvmsg,
1310 .mmap = sock_no_mmap,
1311 .read_skb = vsock_read_skb,
1314 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1316 const struct vsock_transport *transport = vsk->transport;
1318 if (!transport || !transport->cancel_pkt)
1321 return transport->cancel_pkt(vsk);
1324 static void vsock_connect_timeout(struct work_struct *work)
1327 struct vsock_sock *vsk;
1329 vsk = container_of(work, struct vsock_sock, connect_work.work);
1333 if (sk->sk_state == TCP_SYN_SENT &&
1334 (sk->sk_shutdown != SHUTDOWN_MASK)) {
1335 sk->sk_state = TCP_CLOSE;
1336 sk->sk_socket->state = SS_UNCONNECTED;
1337 sk->sk_err = ETIMEDOUT;
1338 sk_error_report(sk);
1339 vsock_transport_cancel_pkt(vsk);
1346 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1347 int addr_len, int flags)
1351 struct vsock_sock *vsk;
1352 const struct vsock_transport *transport;
1353 struct sockaddr_vm *remote_addr;
1363 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1364 switch (sock->state) {
1368 case SS_DISCONNECTING:
1372 /* This continues on so we can move sock into the SS_CONNECTED
1373 * state once the connection has completed (at which point err
1374 * will be set to zero also). Otherwise, we will either wait
1375 * for the connection or return -EALREADY should this be a
1376 * non-blocking call.
1379 if (flags & O_NONBLOCK)
1383 if ((sk->sk_state == TCP_LISTEN) ||
1384 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1389 /* Set the remote address that we are connecting to. */
1390 memcpy(&vsk->remote_addr, remote_addr,
1391 sizeof(vsk->remote_addr));
1393 err = vsock_assign_transport(vsk, NULL);
1397 transport = vsk->transport;
1399 /* The hypervisor and well-known contexts do not have socket
1403 !transport->stream_allow(remote_addr->svm_cid,
1404 remote_addr->svm_port)) {
1409 if (vsock_msgzerocopy_allow(transport)) {
1410 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1411 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1412 /* If this option was set before 'connect()',
1413 * when transport was unknown, check that this
1414 * feature is supported here.
1420 err = vsock_auto_bind(vsk);
1424 sk->sk_state = TCP_SYN_SENT;
1426 err = transport->connect(vsk);
1430 /* Mark sock as connecting and set the error code to in
1431 * progress in case this is a non-blocking connect.
1433 sock->state = SS_CONNECTING;
1437 /* The receive path will handle all communication until we are able to
1438 * enter the connected state. Here we wait for the connection to be
1439 * completed or a notification of an error.
1441 timeout = vsk->connect_timeout;
1442 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1444 while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1445 if (flags & O_NONBLOCK) {
1446 /* If we're not going to block, we schedule a timeout
1447 * function to generate a timeout on the connection
1448 * attempt, in case the peer doesn't respond in a
1449 * timely manner. We hold on to the socket until the
1454 /* If the timeout function is already scheduled,
1455 * reschedule it, then ungrab the socket refcount to
1458 if (mod_delayed_work(system_wq, &vsk->connect_work,
1462 /* Skip ahead to preserve error code set above. */
1467 timeout = schedule_timeout(timeout);
1470 if (signal_pending(current)) {
1471 err = sock_intr_errno(timeout);
1472 sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1473 sock->state = SS_UNCONNECTED;
1474 vsock_transport_cancel_pkt(vsk);
1475 vsock_remove_connected(vsk);
1477 } else if ((sk->sk_state != TCP_ESTABLISHED) && (timeout == 0)) {
1479 sk->sk_state = TCP_CLOSE;
1480 sock->state = SS_UNCONNECTED;
1481 vsock_transport_cancel_pkt(vsk);
1485 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1490 sk->sk_state = TCP_CLOSE;
1491 sock->state = SS_UNCONNECTED;
1497 finish_wait(sk_sleep(sk), &wait);
1503 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1506 struct sock *listener;
1508 struct sock *connected;
1509 struct vsock_sock *vconnected;
1514 listener = sock->sk;
1516 lock_sock(listener);
1518 if (!sock_type_connectible(sock->type)) {
1523 if (listener->sk_state != TCP_LISTEN) {
1528 /* Wait for children sockets to appear; these are the new sockets
1529 * created upon connection establishment.
1531 timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1532 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1534 while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1535 listener->sk_err == 0) {
1536 release_sock(listener);
1537 timeout = schedule_timeout(timeout);
1538 finish_wait(sk_sleep(listener), &wait);
1539 lock_sock(listener);
1541 if (signal_pending(current)) {
1542 err = sock_intr_errno(timeout);
1544 } else if (timeout == 0) {
1549 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1551 finish_wait(sk_sleep(listener), &wait);
1553 if (listener->sk_err)
1554 err = -listener->sk_err;
1557 sk_acceptq_removed(listener);
1559 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1560 vconnected = vsock_sk(connected);
1562 /* If the listener socket has received an error, then we should
1563 * reject this socket and return. Note that we simply mark the
1564 * socket rejected, drop our reference, and let the cleanup
1565 * function handle the cleanup; the fact that we found it in
1566 * the listener's accept queue guarantees that the cleanup
1567 * function hasn't run yet.
1570 vconnected->rejected = true;
1572 newsock->state = SS_CONNECTED;
1573 sock_graft(connected, newsock);
1574 if (vsock_msgzerocopy_allow(vconnected->transport))
1575 set_bit(SOCK_SUPPORT_ZC,
1576 &connected->sk_socket->flags);
1579 release_sock(connected);
1580 sock_put(connected);
1584 release_sock(listener);
1588 static int vsock_listen(struct socket *sock, int backlog)
1592 struct vsock_sock *vsk;
1598 if (!sock_type_connectible(sk->sk_type)) {
1603 if (sock->state != SS_UNCONNECTED) {
1610 if (!vsock_addr_bound(&vsk->local_addr)) {
1615 sk->sk_max_ack_backlog = backlog;
1616 sk->sk_state = TCP_LISTEN;
1625 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1626 const struct vsock_transport *transport,
1629 if (val > vsk->buffer_max_size)
1630 val = vsk->buffer_max_size;
1632 if (val < vsk->buffer_min_size)
1633 val = vsk->buffer_min_size;
1635 if (val != vsk->buffer_size &&
1636 transport && transport->notify_buffer_size)
1637 transport->notify_buffer_size(vsk, &val);
1639 vsk->buffer_size = val;
1642 static int vsock_connectible_setsockopt(struct socket *sock,
1646 unsigned int optlen)
1650 struct vsock_sock *vsk;
1651 const struct vsock_transport *transport;
1654 if (level != AF_VSOCK && level != SOL_SOCKET)
1655 return -ENOPROTOOPT;
1657 #define COPY_IN(_v) \
1659 if (optlen < sizeof(_v)) { \
1663 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) { \
1675 transport = vsk->transport;
1677 if (level == SOL_SOCKET) {
1680 if (optname != SO_ZEROCOPY) {
1682 return sock_setsockopt(sock, level, optname, optval, optlen);
1685 /* Use 'int' type here, because variable to
1686 * set this option usually has this type.
1690 if (zerocopy < 0 || zerocopy > 1) {
1695 if (transport && !vsock_msgzerocopy_allow(transport)) {
1700 sock_valbool_flag(sk, SOCK_ZEROCOPY, zerocopy);
1705 case SO_VM_SOCKETS_BUFFER_SIZE:
1707 vsock_update_buffer_size(vsk, transport, val);
1710 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1712 vsk->buffer_max_size = val;
1713 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1716 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1718 vsk->buffer_min_size = val;
1719 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1722 case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1723 case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: {
1724 struct __kernel_sock_timeval tv;
1726 err = sock_copy_user_timeval(&tv, optval, optlen,
1727 optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1730 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1731 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1732 vsk->connect_timeout = tv.tv_sec * HZ +
1733 DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ));
1734 if (vsk->connect_timeout == 0)
1735 vsk->connect_timeout =
1736 VSOCK_DEFAULT_CONNECT_TIMEOUT;
1756 static int vsock_connectible_getsockopt(struct socket *sock,
1757 int level, int optname,
1758 char __user *optval,
1761 struct sock *sk = sock->sk;
1762 struct vsock_sock *vsk = vsock_sk(sk);
1766 struct old_timeval32 tm32;
1767 struct __kernel_old_timeval tm;
1768 struct __kernel_sock_timeval stm;
1771 int lv = sizeof(v.val64);
1774 if (level != AF_VSOCK)
1775 return -ENOPROTOOPT;
1777 if (get_user(len, optlen))
1780 memset(&v, 0, sizeof(v));
1783 case SO_VM_SOCKETS_BUFFER_SIZE:
1784 v.val64 = vsk->buffer_size;
1787 case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1788 v.val64 = vsk->buffer_max_size;
1791 case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1792 v.val64 = vsk->buffer_min_size;
1795 case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1796 case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD:
1797 lv = sock_get_timeout(vsk->connect_timeout, &v,
1798 optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1802 return -ENOPROTOOPT;
1809 if (copy_to_user(optval, &v, len))
1812 if (put_user(len, optlen))
1818 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1822 struct vsock_sock *vsk;
1823 const struct vsock_transport *transport;
1824 ssize_t total_written;
1827 struct vsock_transport_send_notify_data send_data;
1828 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1835 if (msg->msg_flags & MSG_OOB)
1840 transport = vsk->transport;
1842 /* Callers should not provide a destination with connection oriented
1845 if (msg->msg_namelen) {
1846 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1850 /* Send data only if both sides are not shutdown in the direction. */
1851 if (sk->sk_shutdown & SEND_SHUTDOWN ||
1852 vsk->peer_shutdown & RCV_SHUTDOWN) {
1857 if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1858 !vsock_addr_bound(&vsk->local_addr)) {
1863 if (!vsock_addr_bound(&vsk->remote_addr)) {
1864 err = -EDESTADDRREQ;
1868 if (msg->msg_flags & MSG_ZEROCOPY &&
1869 !vsock_msgzerocopy_allow(transport)) {
1874 /* Wait for room in the produce queue to enqueue our user's data. */
1875 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1877 err = transport->notify_send_init(vsk, &send_data);
1881 while (total_written < len) {
1884 add_wait_queue(sk_sleep(sk), &wait);
1885 while (vsock_stream_has_space(vsk) == 0 &&
1887 !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1888 !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1890 /* Don't wait for non-blocking sockets. */
1893 remove_wait_queue(sk_sleep(sk), &wait);
1897 err = transport->notify_send_pre_block(vsk, &send_data);
1899 remove_wait_queue(sk_sleep(sk), &wait);
1904 timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1906 if (signal_pending(current)) {
1907 err = sock_intr_errno(timeout);
1908 remove_wait_queue(sk_sleep(sk), &wait);
1910 } else if (timeout == 0) {
1912 remove_wait_queue(sk_sleep(sk), &wait);
1916 remove_wait_queue(sk_sleep(sk), &wait);
1918 /* These checks occur both as part of and after the loop
1919 * conditional since we need to check before and after
1925 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1926 (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1931 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1935 /* Note that enqueue will only write as many bytes as are free
1936 * in the produce queue, so we don't need to ensure len is
1937 * smaller than the queue size. It is the caller's
1938 * responsibility to check how many bytes we were able to send.
1941 if (sk->sk_type == SOCK_SEQPACKET) {
1942 written = transport->seqpacket_enqueue(vsk,
1943 msg, len - total_written);
1945 written = transport->stream_enqueue(vsk,
1946 msg, len - total_written);
1954 total_written += written;
1956 err = transport->notify_send_post_enqueue(
1957 vsk, written, &send_data);
1964 if (total_written > 0) {
1965 /* Return number of written bytes only if:
1966 * 1) SOCK_STREAM socket.
1967 * 2) SOCK_SEQPACKET socket when whole buffer is sent.
1969 if (sk->sk_type == SOCK_STREAM || total_written == len)
1970 err = total_written;
1973 if (sk->sk_type == SOCK_STREAM)
1974 err = sk_stream_error(sk, msg->msg_flags, err);
1980 static int vsock_connectible_wait_data(struct sock *sk,
1981 struct wait_queue_entry *wait,
1983 struct vsock_transport_recv_notify_data *recv_data,
1986 const struct vsock_transport *transport;
1987 struct vsock_sock *vsk;
1993 transport = vsk->transport;
1996 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
1997 data = vsock_connectible_has_data(vsk);
2001 if (sk->sk_err != 0 ||
2002 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2003 (vsk->peer_shutdown & SEND_SHUTDOWN)) {
2007 /* Don't wait for non-blocking sockets. */
2014 err = transport->notify_recv_pre_block(vsk, target, recv_data);
2020 timeout = schedule_timeout(timeout);
2023 if (signal_pending(current)) {
2024 err = sock_intr_errno(timeout);
2026 } else if (timeout == 0) {
2032 finish_wait(sk_sleep(sk), wait);
2037 /* Internal transport error when checking for available
2038 * data. XXX This should be changed to a connection
2039 * reset in a later change.
2047 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2048 size_t len, int flags)
2050 struct vsock_transport_recv_notify_data recv_data;
2051 const struct vsock_transport *transport;
2052 struct vsock_sock *vsk;
2061 transport = vsk->transport;
2063 /* We must not copy less than target bytes into the user's buffer
2064 * before returning successfully, so we wait for the consume queue to
2065 * have that much data to consume before dequeueing. Note that this
2066 * makes it impossible to handle cases where target is greater than the
2069 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2070 if (target >= transport->stream_rcvhiwat(vsk)) {
2074 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2077 err = transport->notify_recv_init(vsk, target, &recv_data);
2085 err = vsock_connectible_wait_data(sk, &wait, timeout,
2086 &recv_data, target);
2090 err = transport->notify_recv_pre_dequeue(vsk, target,
2095 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
2103 err = transport->notify_recv_post_dequeue(vsk, target, read,
2104 !(flags & MSG_PEEK), &recv_data);
2108 if (read >= target || flags & MSG_PEEK)
2116 else if (sk->sk_shutdown & RCV_SHUTDOWN)
2126 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2127 size_t len, int flags)
2129 const struct vsock_transport *transport;
2130 struct vsock_sock *vsk;
2137 transport = vsk->transport;
2139 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2141 err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0);
2145 msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2154 } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2157 /* User sets MSG_TRUNC, so return real length of
2160 if (flags & MSG_TRUNC)
2163 err = len - msg_data_left(msg);
2165 /* Always set MSG_TRUNC if real length of packet is
2166 * bigger than user's buffer.
2169 msg->msg_flags |= MSG_TRUNC;
2177 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2181 struct vsock_sock *vsk;
2182 const struct vsock_transport *transport;
2183 #ifdef CONFIG_BPF_SYSCALL
2184 const struct proto *prot;
2190 if (unlikely(flags & MSG_ERRQUEUE))
2191 return sock_recv_errqueue(sk, msg, len, SOL_VSOCK, VSOCK_RECVERR);
2198 transport = vsk->transport;
2200 if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2201 /* Recvmsg is supposed to return 0 if a peer performs an
2202 * orderly shutdown. Differentiate between that case and when a
2203 * peer has not connected or a local shutdown occurred with the
2206 if (sock_flag(sk, SOCK_DONE))
2214 if (flags & MSG_OOB) {
2219 /* We don't check peer_shutdown flag here since peer may actually shut
2220 * down, but there can be data in the queue that a local socket can
2223 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2228 /* It is valid on Linux to pass in a zero-length receive buffer. This
2229 * is not an error. We may as well bail out now.
2236 #ifdef CONFIG_BPF_SYSCALL
2237 prot = READ_ONCE(sk->sk_prot);
2238 if (prot != &vsock_proto) {
2240 return prot->recvmsg(sk, msg, len, flags, NULL);
2244 if (sk->sk_type == SOCK_STREAM)
2245 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2247 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2253 EXPORT_SYMBOL_GPL(vsock_connectible_recvmsg);
2255 static int vsock_set_rcvlowat(struct sock *sk, int val)
2257 const struct vsock_transport *transport;
2258 struct vsock_sock *vsk;
2262 if (val > vsk->buffer_size)
2265 transport = vsk->transport;
2267 if (transport && transport->set_rcvlowat)
2268 return transport->set_rcvlowat(vsk, val);
2270 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
2274 static const struct proto_ops vsock_stream_ops = {
2276 .owner = THIS_MODULE,
2277 .release = vsock_release,
2279 .connect = vsock_connect,
2280 .socketpair = sock_no_socketpair,
2281 .accept = vsock_accept,
2282 .getname = vsock_getname,
2284 .ioctl = sock_no_ioctl,
2285 .listen = vsock_listen,
2286 .shutdown = vsock_shutdown,
2287 .setsockopt = vsock_connectible_setsockopt,
2288 .getsockopt = vsock_connectible_getsockopt,
2289 .sendmsg = vsock_connectible_sendmsg,
2290 .recvmsg = vsock_connectible_recvmsg,
2291 .mmap = sock_no_mmap,
2292 .set_rcvlowat = vsock_set_rcvlowat,
2293 .read_skb = vsock_read_skb,
2296 static const struct proto_ops vsock_seqpacket_ops = {
2298 .owner = THIS_MODULE,
2299 .release = vsock_release,
2301 .connect = vsock_connect,
2302 .socketpair = sock_no_socketpair,
2303 .accept = vsock_accept,
2304 .getname = vsock_getname,
2306 .ioctl = sock_no_ioctl,
2307 .listen = vsock_listen,
2308 .shutdown = vsock_shutdown,
2309 .setsockopt = vsock_connectible_setsockopt,
2310 .getsockopt = vsock_connectible_getsockopt,
2311 .sendmsg = vsock_connectible_sendmsg,
2312 .recvmsg = vsock_connectible_recvmsg,
2313 .mmap = sock_no_mmap,
2314 .read_skb = vsock_read_skb,
2317 static int vsock_create(struct net *net, struct socket *sock,
2318 int protocol, int kern)
2320 struct vsock_sock *vsk;
2327 if (protocol && protocol != PF_VSOCK)
2328 return -EPROTONOSUPPORT;
2330 switch (sock->type) {
2332 sock->ops = &vsock_dgram_ops;
2335 sock->ops = &vsock_stream_ops;
2337 case SOCK_SEQPACKET:
2338 sock->ops = &vsock_seqpacket_ops;
2341 return -ESOCKTNOSUPPORT;
2344 sock->state = SS_UNCONNECTED;
2346 sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2352 if (sock->type == SOCK_DGRAM) {
2353 ret = vsock_assign_transport(vsk, NULL);
2360 /* SOCK_DGRAM doesn't have 'setsockopt' callback set in its
2361 * proto_ops, so there is no handler for custom logic.
2363 if (sock_type_connectible(sock->type))
2364 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2366 vsock_insert_unbound(vsk);
2371 static const struct net_proto_family vsock_family_ops = {
2373 .create = vsock_create,
2374 .owner = THIS_MODULE,
2377 static long vsock_dev_do_ioctl(struct file *filp,
2378 unsigned int cmd, void __user *ptr)
2380 u32 __user *p = ptr;
2381 u32 cid = VMADDR_CID_ANY;
2385 case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2386 /* To be compatible with the VMCI behavior, we prioritize the
2387 * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2390 cid = transport_g2h->get_local_cid();
2391 else if (transport_h2g)
2392 cid = transport_h2g->get_local_cid();
2394 if (put_user(cid, p) != 0)
2399 retval = -ENOIOCTLCMD;
2405 static long vsock_dev_ioctl(struct file *filp,
2406 unsigned int cmd, unsigned long arg)
2408 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2411 #ifdef CONFIG_COMPAT
2412 static long vsock_dev_compat_ioctl(struct file *filp,
2413 unsigned int cmd, unsigned long arg)
2415 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2419 static const struct file_operations vsock_device_ops = {
2420 .owner = THIS_MODULE,
2421 .unlocked_ioctl = vsock_dev_ioctl,
2422 #ifdef CONFIG_COMPAT
2423 .compat_ioctl = vsock_dev_compat_ioctl,
2425 .open = nonseekable_open,
2428 static struct miscdevice vsock_device = {
2430 .fops = &vsock_device_ops,
2433 static int __init vsock_init(void)
2437 vsock_init_tables();
2439 vsock_proto.owner = THIS_MODULE;
2440 vsock_device.minor = MISC_DYNAMIC_MINOR;
2441 err = misc_register(&vsock_device);
2443 pr_err("Failed to register misc device\n");
2444 goto err_reset_transport;
2447 err = proto_register(&vsock_proto, 1); /* we want our slab */
2449 pr_err("Cannot register vsock protocol\n");
2450 goto err_deregister_misc;
2453 err = sock_register(&vsock_family_ops);
2455 pr_err("could not register af_vsock (%d) address family: %d\n",
2457 goto err_unregister_proto;
2460 vsock_bpf_build_proto();
2464 err_unregister_proto:
2465 proto_unregister(&vsock_proto);
2466 err_deregister_misc:
2467 misc_deregister(&vsock_device);
2468 err_reset_transport:
2472 static void __exit vsock_exit(void)
2474 misc_deregister(&vsock_device);
2475 sock_unregister(AF_VSOCK);
2476 proto_unregister(&vsock_proto);
2479 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2481 return vsk->transport;
2483 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2485 int vsock_core_register(const struct vsock_transport *t, int features)
2487 const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2488 int err = mutex_lock_interruptible(&vsock_register_mutex);
2493 t_h2g = transport_h2g;
2494 t_g2h = transport_g2h;
2495 t_dgram = transport_dgram;
2496 t_local = transport_local;
2498 if (features & VSOCK_TRANSPORT_F_H2G) {
2506 if (features & VSOCK_TRANSPORT_F_G2H) {
2514 if (features & VSOCK_TRANSPORT_F_DGRAM) {
2522 if (features & VSOCK_TRANSPORT_F_LOCAL) {
2530 transport_h2g = t_h2g;
2531 transport_g2h = t_g2h;
2532 transport_dgram = t_dgram;
2533 transport_local = t_local;
2536 mutex_unlock(&vsock_register_mutex);
2539 EXPORT_SYMBOL_GPL(vsock_core_register);
2541 void vsock_core_unregister(const struct vsock_transport *t)
2543 mutex_lock(&vsock_register_mutex);
2545 if (transport_h2g == t)
2546 transport_h2g = NULL;
2548 if (transport_g2h == t)
2549 transport_g2h = NULL;
2551 if (transport_dgram == t)
2552 transport_dgram = NULL;
2554 if (transport_local == t)
2555 transport_local = NULL;
2557 mutex_unlock(&vsock_register_mutex);
2559 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2561 module_init(vsock_init);
2562 module_exit(vsock_exit);
2564 MODULE_AUTHOR("VMware, Inc.");
2565 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2566 MODULE_VERSION("1.0.2.0-k");
2567 MODULE_LICENSE("GPL v2");