net/vmw_vsock/af_vsock.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VMware vSockets Driver
   4  *
   5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   6  */
   7
   8 /* Implementation notes:
   9  *
  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.
  12  *
  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.
  25  *
  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).
  29  *
  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.
  45  *
  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.
  55  *
  56  * - Lock ordering for pending or accept queue sockets is:
  57  *
  58  *     lock_sock(listener);
  59  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  60  *
  61  * Using explicit nested locking keeps lockdep happy since normally only one
  62  * lock of a given class may be taken at a time.
  63  *
  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.
  69  *
  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.
  77  *
  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):
  80  *
  81  *   TCP_CLOSE - unconnected
  82  *   TCP_SYN_SENT - connecting
  83  *   TCP_ESTABLISHED - connected
  84  *   TCP_CLOSING - disconnecting
  85  *   TCP_LISTEN - listening
  86  */
  87
  88 #include <linux/types.h>
  89 #include <linux/bitops.h>
  90 #include <linux/cred.h>
  91 #include <linux/init.h>
  92 #include <linux/io.h>
  93 #include <linux/kernel.h>
  94 #include <linux/sched/signal.h>
  95 #include <linux/kmod.h>
  96 #include <linux/list.h>
  97 #include <linux/miscdevice.h>
  98 #include <linux/module.h>
  99 #include <linux/mutex.h>
 100 #include <linux/net.h>
 101 #include <linux/poll.h>
 102 #include <linux/random.h>
 103 #include <linux/skbuff.h>
 104 #include <linux/smp.h>
 105 #include <linux/socket.h>
 106 #include <linux/stddef.h>
 107 #include <linux/unistd.h>
 108 #include <linux/wait.h>
 109 #include <linux/workqueue.h>
 110 #include <net/sock.h>
 111 #include <net/af_vsock.h>
 112
 113 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 114 static void vsock_sk_destruct(struct sock *sk);
 115 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 116
 117 /* Protocol family. */
 118 static struct proto vsock_proto = {
 119         .name = "AF_VSOCK",
 120         .owner = THIS_MODULE,
 121         .obj_size = sizeof(struct vsock_sock),
 122 };
 123
 124 /* The default peer timeout indicates how long we will wait for a peer response
 125  * to a control message.
 126  */
 127 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 128
 129 #define VSOCK_DEFAULT_BUFFER_SIZE     (1024 * 256)
 130 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
 131 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
 132
 133 /* Transport used for host->guest communication */
 134 static const struct vsock_transport *transport_h2g;
 135 /* Transport used for guest->host communication */
 136 static const struct vsock_transport *transport_g2h;
 137 /* Transport used for DGRAM communication */
 138 static const struct vsock_transport *transport_dgram;
 139 /* Transport used for local communication */
 140 static const struct vsock_transport *transport_local;
 141 static DEFINE_MUTEX(vsock_register_mutex);
 142
 143 /**** UTILS ****/
 144
 145 /* Each bound VSocket is stored in the bind hash table and each connected
 146  * VSocket is stored in the connected hash table.
 147  *
 148  * Unbound sockets are all put on the same list attached to the end of the hash
 149  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 150  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 151  * represents the list that addr hashes to).
 152  *
 153  * Specifically, we initialize the vsock_bind_table array to a size of
 154  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 155  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 156  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 157  * mods with VSOCK_HASH_SIZE to ensure this.
 158  */
 159 #define MAX_PORT_RETRIES        24
 160
 161 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 162 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 163 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 164
 165 /* XXX This can probably be implemented in a better way. */
 166 #define VSOCK_CONN_HASH(src, dst)                               \
 167         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 168 #define vsock_connected_sockets(src, dst)               \
 169         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 170 #define vsock_connected_sockets_vsk(vsk)                                \
 171         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 172
 173 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 174 EXPORT_SYMBOL_GPL(vsock_bind_table);
 175 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 176 EXPORT_SYMBOL_GPL(vsock_connected_table);
 177 DEFINE_SPINLOCK(vsock_table_lock);
 178 EXPORT_SYMBOL_GPL(vsock_table_lock);
 179
 180 /* Autobind this socket to the local address if necessary. */
 181 static int vsock_auto_bind(struct vsock_sock *vsk)
 182 {
 183         struct sock *sk = sk_vsock(vsk);
 184         struct sockaddr_vm local_addr;
 185
 186         if (vsock_addr_bound(&vsk->local_addr))
 187                 return 0;
 188         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 189         return __vsock_bind(sk, &local_addr);
 190 }
 191
 192 static void vsock_init_tables(void)
 193 {
 194         int i;
 195
 196         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 197                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 198
 199         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 200                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 201 }
 202
 203 static void __vsock_insert_bound(struct list_head *list,
 204                                  struct vsock_sock *vsk)
 205 {
 206         sock_hold(&vsk->sk);
 207         list_add(&vsk->bound_table, list);
 208 }
 209
 210 static void __vsock_insert_connected(struct list_head *list,
 211                                      struct vsock_sock *vsk)
 212 {
 213         sock_hold(&vsk->sk);
 214         list_add(&vsk->connected_table, list);
 215 }
 216
 217 static void __vsock_remove_bound(struct vsock_sock *vsk)
 218 {
 219         list_del_init(&vsk->bound_table);
 220         sock_put(&vsk->sk);
 221 }
 222
 223 static void __vsock_remove_connected(struct vsock_sock *vsk)
 224 {
 225         list_del_init(&vsk->connected_table);
 226         sock_put(&vsk->sk);
 227 }
 228
 229 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 230 {
 231         struct vsock_sock *vsk;
 232
 233         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
 234                 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
 235                         return sk_vsock(vsk);
 236
 237                 if (addr->svm_port == vsk->local_addr.svm_port &&
 238                     (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
 239                      addr->svm_cid == VMADDR_CID_ANY))
 240                         return sk_vsock(vsk);
 241         }
 242
 243         return NULL;
 244 }
 245
 246 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 247                                                   struct sockaddr_vm *dst)
 248 {
 249         struct vsock_sock *vsk;
 250
 251         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 252                             connected_table) {
 253                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 254                     dst->svm_port == vsk->local_addr.svm_port) {
 255                         return sk_vsock(vsk);
 256                 }
 257         }
 258
 259         return NULL;
 260 }
 261
 262 static void vsock_insert_unbound(struct vsock_sock *vsk)
 263 {
 264         spin_lock_bh(&vsock_table_lock);
 265         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 266         spin_unlock_bh(&vsock_table_lock);
 267 }
 268
 269 void vsock_insert_connected(struct vsock_sock *vsk)
 270 {
 271         struct list_head *list = vsock_connected_sockets(
 272                 &vsk->remote_addr, &vsk->local_addr);
 273
 274         spin_lock_bh(&vsock_table_lock);
 275         __vsock_insert_connected(list, vsk);
 276         spin_unlock_bh(&vsock_table_lock);
 277 }
 278 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 279
 280 void vsock_remove_bound(struct vsock_sock *vsk)
 281 {
 282         spin_lock_bh(&vsock_table_lock);
 283         if (__vsock_in_bound_table(vsk))
 284                 __vsock_remove_bound(vsk);
 285         spin_unlock_bh(&vsock_table_lock);
 286 }
 287 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 288
 289 void vsock_remove_connected(struct vsock_sock *vsk)
 290 {
 291         spin_lock_bh(&vsock_table_lock);
 292         if (__vsock_in_connected_table(vsk))
 293                 __vsock_remove_connected(vsk);
 294         spin_unlock_bh(&vsock_table_lock);
 295 }
 296 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 297
 298 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 299 {
 300         struct sock *sk;
 301
 302         spin_lock_bh(&vsock_table_lock);
 303         sk = __vsock_find_bound_socket(addr);
 304         if (sk)
 305                 sock_hold(sk);
 306
 307         spin_unlock_bh(&vsock_table_lock);
 308
 309         return sk;
 310 }
 311 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 312
 313 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 314                                          struct sockaddr_vm *dst)
 315 {
 316         struct sock *sk;
 317
 318         spin_lock_bh(&vsock_table_lock);
 319         sk = __vsock_find_connected_socket(src, dst);
 320         if (sk)
 321                 sock_hold(sk);
 322
 323         spin_unlock_bh(&vsock_table_lock);
 324
 325         return sk;
 326 }
 327 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 328
 329 void vsock_remove_sock(struct vsock_sock *vsk)
 330 {
 331         vsock_remove_bound(vsk);
 332         vsock_remove_connected(vsk);
 333 }
 334 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 335
 336 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 337 {
 338         int i;
 339
 340         spin_lock_bh(&vsock_table_lock);
 341
 342         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 343                 struct vsock_sock *vsk;
 344                 list_for_each_entry(vsk, &vsock_connected_table[i],
 345                                     connected_table)
 346                         fn(sk_vsock(vsk));
 347         }
 348
 349         spin_unlock_bh(&vsock_table_lock);
 350 }
 351 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 352
 353 void vsock_add_pending(struct sock *listener, struct sock *pending)
 354 {
 355         struct vsock_sock *vlistener;
 356         struct vsock_sock *vpending;
 357
 358         vlistener = vsock_sk(listener);
 359         vpending = vsock_sk(pending);
 360
 361         sock_hold(pending);
 362         sock_hold(listener);
 363         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 364 }
 365 EXPORT_SYMBOL_GPL(vsock_add_pending);
 366
 367 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 368 {
 369         struct vsock_sock *vpending = vsock_sk(pending);
 370
 371         list_del_init(&vpending->pending_links);
 372         sock_put(listener);
 373         sock_put(pending);
 374 }
 375 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 376
 377 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 378 {
 379         struct vsock_sock *vlistener;
 380         struct vsock_sock *vconnected;
 381
 382         vlistener = vsock_sk(listener);
 383         vconnected = vsock_sk(connected);
 384
 385         sock_hold(connected);
 386         sock_hold(listener);
 387         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 388 }
 389 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 390
 391 static bool vsock_use_local_transport(unsigned int remote_cid)
 392 {
 393         if (!transport_local)
 394                 return false;
 395
 396         if (remote_cid == VMADDR_CID_LOCAL)
 397                 return true;
 398
 399         if (transport_g2h) {
 400                 return remote_cid == transport_g2h->get_local_cid();
 401         } else {
 402                 return remote_cid == VMADDR_CID_HOST;
 403         }
 404 }
 405
 406 static void vsock_deassign_transport(struct vsock_sock *vsk)
 407 {
 408         if (!vsk->transport)
 409                 return;
 410
 411         vsk->transport->destruct(vsk);
 412         module_put(vsk->transport->module);
 413         vsk->transport = NULL;
 414 }
 415
 416 /* Assign a transport to a socket and call the .init transport callback.
 417  *
 418  * Note: for stream socket this must be called when vsk->remote_addr is set
 419  * (e.g. during the connect() or when a connection request on a listener
 420  * socket is received).
 421  * The vsk->remote_addr is used to decide which transport to use:
 422  *  - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
 423  *    g2h is not loaded, will use local transport;
 424  *  - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
 425  *    includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
 426  *  - remote CID > VMADDR_CID_HOST will use host->guest transport;
 427  */
 428 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
 429 {
 430         const struct vsock_transport *new_transport;
 431         struct sock *sk = sk_vsock(vsk);
 432         unsigned int remote_cid = vsk->remote_addr.svm_cid;
 433         __u8 remote_flags;
 434         int ret;
 435
 436         /* If the packet is coming with the source and destination CIDs higher
 437          * than VMADDR_CID_HOST, then a vsock channel where all the packets are
 438          * forwarded to the host should be established. Then the host will
 439          * need to forward the packets to the guest.
 440          *
 441          * The flag is set on the (listen) receive path (psk is not NULL). On
 442          * the connect path the flag can be set by the user space application.
 443          */
 444         if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
 445             vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
 446                 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
 447
 448         remote_flags = vsk->remote_addr.svm_flags;
 449
 450         switch (sk->sk_type) {
 451         case SOCK_DGRAM:
 452                 new_transport = transport_dgram;
 453                 break;
 454         case SOCK_STREAM:
 455                 if (vsock_use_local_transport(remote_cid))
 456                         new_transport = transport_local;
 457                 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
 458                          (remote_flags & VMADDR_FLAG_TO_HOST))
 459                         new_transport = transport_g2h;
 460                 else
 461                         new_transport = transport_h2g;
 462                 break;
 463         default:
 464                 return -ESOCKTNOSUPPORT;
 465         }
 466
 467         if (vsk->transport) {
 468                 if (vsk->transport == new_transport)
 469                         return 0;
 470
 471                 /* transport->release() must be called with sock lock acquired.
 472                  * This path can only be taken during vsock_stream_connect(),
 473                  * where we have already held the sock lock.
 474                  * In the other cases, this function is called on a new socket
 475                  * which is not assigned to any transport.
 476                  */
 477                 vsk->transport->release(vsk);
 478                 vsock_deassign_transport(vsk);
 479         }
 480
 481         /* We increase the module refcnt to prevent the transport unloading
 482          * while there are open sockets assigned to it.
 483          */
 484         if (!new_transport || !try_module_get(new_transport->module))
 485                 return -ENODEV;
 486
 487         ret = new_transport->init(vsk, psk);
 488         if (ret) {
 489                 module_put(new_transport->module);
 490                 return ret;
 491         }
 492
 493         vsk->transport = new_transport;
 494
 495         return 0;
 496 }
 497 EXPORT_SYMBOL_GPL(vsock_assign_transport);
 498
 499 bool vsock_find_cid(unsigned int cid)
 500 {
 501         if (transport_g2h && cid == transport_g2h->get_local_cid())
 502                 return true;
 503
 504         if (transport_h2g && cid == VMADDR_CID_HOST)
 505                 return true;
 506
 507         if (transport_local && cid == VMADDR_CID_LOCAL)
 508                 return true;
 509
 510         return false;
 511 }
 512 EXPORT_SYMBOL_GPL(vsock_find_cid);
 513
 514 static struct sock *vsock_dequeue_accept(struct sock *listener)
 515 {
 516         struct vsock_sock *vlistener;
 517         struct vsock_sock *vconnected;
 518
 519         vlistener = vsock_sk(listener);
 520
 521         if (list_empty(&vlistener->accept_queue))
 522                 return NULL;
 523
 524         vconnected = list_entry(vlistener->accept_queue.next,
 525                                 struct vsock_sock, accept_queue);
 526
 527         list_del_init(&vconnected->accept_queue);
 528         sock_put(listener);
 529         /* The caller will need a reference on the connected socket so we let
 530          * it call sock_put().
 531          */
 532
 533         return sk_vsock(vconnected);
 534 }
 535
 536 static bool vsock_is_accept_queue_empty(struct sock *sk)
 537 {
 538         struct vsock_sock *vsk = vsock_sk(sk);
 539         return list_empty(&vsk->accept_queue);
 540 }
 541
 542 static bool vsock_is_pending(struct sock *sk)
 543 {
 544         struct vsock_sock *vsk = vsock_sk(sk);
 545         return !list_empty(&vsk->pending_links);
 546 }
 547
 548 static int vsock_send_shutdown(struct sock *sk, int mode)
 549 {
 550         struct vsock_sock *vsk = vsock_sk(sk);
 551
 552         if (!vsk->transport)
 553                 return -ENODEV;
 554
 555         return vsk->transport->shutdown(vsk, mode);
 556 }
 557
 558 static void vsock_pending_work(struct work_struct *work)
 559 {
 560         struct sock *sk;
 561         struct sock *listener;
 562         struct vsock_sock *vsk;
 563         bool cleanup;
 564
 565         vsk = container_of(work, struct vsock_sock, pending_work.work);
 566         sk = sk_vsock(vsk);
 567         listener = vsk->listener;
 568         cleanup = true;
 569
 570         lock_sock(listener);
 571         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 572
 573         if (vsock_is_pending(sk)) {
 574                 vsock_remove_pending(listener, sk);
 575
 576                 sk_acceptq_removed(listener);
 577         } else if (!vsk->rejected) {
 578                 /* We are not on the pending list and accept() did not reject
 579                  * us, so we must have been accepted by our user process.  We
 580                  * just need to drop our references to the sockets and be on
 581                  * our way.
 582                  */
 583                 cleanup = false;
 584                 goto out;
 585         }
 586
 587         /* We need to remove ourself from the global connected sockets list so
 588          * incoming packets can't find this socket, and to reduce the reference
 589          * count.
 590          */
 591         vsock_remove_connected(vsk);
 592
 593         sk->sk_state = TCP_CLOSE;
 594
 595 out:
 596         release_sock(sk);
 597         release_sock(listener);
 598         if (cleanup)
 599                 sock_put(sk);
 600
 601         sock_put(sk);
 602         sock_put(listener);
 603 }
 604
 605 /**** SOCKET OPERATIONS ****/
 606
 607 static int __vsock_bind_stream(struct vsock_sock *vsk,
 608                                struct sockaddr_vm *addr)
 609 {
 610         static u32 port;
 611         struct sockaddr_vm new_addr;
 612
 613         if (!port)
 614                 port = LAST_RESERVED_PORT + 1 +
 615                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 616
 617         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 618
 619         if (addr->svm_port == VMADDR_PORT_ANY) {
 620                 bool found = false;
 621                 unsigned int i;
 622
 623                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 624                         if (port <= LAST_RESERVED_PORT)
 625                                 port = LAST_RESERVED_PORT + 1;
 626
 627                         new_addr.svm_port = port++;
 628
 629                         if (!__vsock_find_bound_socket(&new_addr)) {
 630                                 found = true;
 631                                 break;
 632                         }
 633                 }
 634
 635                 if (!found)
 636                         return -EADDRNOTAVAIL;
 637         } else {
 638                 /* If port is in reserved range, ensure caller
 639                  * has necessary privileges.
 640                  */
 641                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 642                     !capable(CAP_NET_BIND_SERVICE)) {
 643                         return -EACCES;
 644                 }
 645
 646                 if (__vsock_find_bound_socket(&new_addr))
 647                         return -EADDRINUSE;
 648         }
 649
 650         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 651
 652         /* Remove stream sockets from the unbound list and add them to the hash
 653          * table for easy lookup by its address.  The unbound list is simply an
 654          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 655          */
 656         __vsock_remove_bound(vsk);
 657         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 658
 659         return 0;
 660 }
 661
 662 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 663                               struct sockaddr_vm *addr)
 664 {
 665         return vsk->transport->dgram_bind(vsk, addr);
 666 }
 667
 668 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 669 {
 670         struct vsock_sock *vsk = vsock_sk(sk);
 671         int retval;
 672
 673         /* First ensure this socket isn't already bound. */
 674         if (vsock_addr_bound(&vsk->local_addr))
 675                 return -EINVAL;
 676
 677         /* Now bind to the provided address or select appropriate values if
 678          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 679          * like AF_INET prevents binding to a non-local IP address (in most
 680          * cases), we only allow binding to a local CID.
 681          */
 682         if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
 683                 return -EADDRNOTAVAIL;
 684
 685         switch (sk->sk_socket->type) {
 686         case SOCK_STREAM:
 687                 spin_lock_bh(&vsock_table_lock);
 688                 retval = __vsock_bind_stream(vsk, addr);
 689                 spin_unlock_bh(&vsock_table_lock);
 690                 break;
 691
 692         case SOCK_DGRAM:
 693                 retval = __vsock_bind_dgram(vsk, addr);
 694                 break;
 695
 696         default:
 697                 retval = -EINVAL;
 698                 break;
 699         }
 700
 701         return retval;
 702 }
 703
 704 static void vsock_connect_timeout(struct work_struct *work);
 705
 706 static struct sock *__vsock_create(struct net *net,
 707                                    struct socket *sock,
 708                                    struct sock *parent,
 709                                    gfp_t priority,
 710                                    unsigned short type,
 711                                    int kern)
 712 {
 713         struct sock *sk;
 714         struct vsock_sock *psk;
 715         struct vsock_sock *vsk;
 716
 717         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 718         if (!sk)
 719                 return NULL;
 720
 721         sock_init_data(sock, sk);
 722
 723         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 724          * non-NULL. We make sure that our sockets always have a type by
 725          * setting it here if needed.
 726          */
 727         if (!sock)
 728                 sk->sk_type = type;
 729
 730         vsk = vsock_sk(sk);
 731         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 732         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 733
 734         sk->sk_destruct = vsock_sk_destruct;
 735         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 736         sock_reset_flag(sk, SOCK_DONE);
 737
 738         INIT_LIST_HEAD(&vsk->bound_table);
 739         INIT_LIST_HEAD(&vsk->connected_table);
 740         vsk->listener = NULL;
 741         INIT_LIST_HEAD(&vsk->pending_links);
 742         INIT_LIST_HEAD(&vsk->accept_queue);
 743         vsk->rejected = false;
 744         vsk->sent_request = false;
 745         vsk->ignore_connecting_rst = false;
 746         vsk->peer_shutdown = 0;
 747         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 748         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 749
 750         psk = parent ? vsock_sk(parent) : NULL;
 751         if (parent) {
 752                 vsk->trusted = psk->trusted;
 753                 vsk->owner = get_cred(psk->owner);
 754                 vsk->connect_timeout = psk->connect_timeout;
 755                 vsk->buffer_size = psk->buffer_size;
 756                 vsk->buffer_min_size = psk->buffer_min_size;
 757                 vsk->buffer_max_size = psk->buffer_max_size;
 758         } else {
 759                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 760                 vsk->owner = get_current_cred();
 761                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 762                 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
 763                 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
 764                 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
 765         }
 766
 767         return sk;
 768 }
 769
 770 static void __vsock_release(struct sock *sk, int level)
 771 {
 772         if (sk) {
 773                 struct sock *pending;
 774                 struct vsock_sock *vsk;
 775
 776                 vsk = vsock_sk(sk);
 777                 pending = NULL; /* Compiler warning. */
 778
 779                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 780                  * version to avoid the warning "possible recursive locking
 781                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 782                  * is the same as lock_sock(sk).
 783                  */
 784                 lock_sock_nested(sk, level);
 785
 786                 if (vsk->transport)
 787                         vsk->transport->release(vsk);
 788                 else if (sk->sk_type == SOCK_STREAM)
 789                         vsock_remove_sock(vsk);
 790
 791                 sock_orphan(sk);
 792                 sk->sk_shutdown = SHUTDOWN_MASK;
 793
 794                 skb_queue_purge(&sk->sk_receive_queue);
 795
 796                 /* Clean up any sockets that never were accepted. */
 797                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 798                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 799                         sock_put(pending);
 800                 }
 801
 802                 release_sock(sk);
 803                 sock_put(sk);
 804         }
 805 }
 806
 807 static void vsock_sk_destruct(struct sock *sk)
 808 {
 809         struct vsock_sock *vsk = vsock_sk(sk);
 810
 811         vsock_deassign_transport(vsk);
 812
 813         /* When clearing these addresses, there's no need to set the family and
 814          * possibly register the address family with the kernel.
 815          */
 816         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 817         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 818
 819         put_cred(vsk->owner);
 820 }
 821
 822 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 823 {
 824         int err;
 825
 826         err = sock_queue_rcv_skb(sk, skb);
 827         if (err)
 828                 kfree_skb(skb);
 829
 830         return err;
 831 }
 832
 833 struct sock *vsock_create_connected(struct sock *parent)
 834 {
 835         return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
 836                               parent->sk_type, 0);
 837 }
 838 EXPORT_SYMBOL_GPL(vsock_create_connected);
 839
 840 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 841 {
 842         return vsk->transport->stream_has_data(vsk);
 843 }
 844 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 845
 846 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 847 {
 848         return vsk->transport->stream_has_space(vsk);
 849 }
 850 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 851
 852 static int vsock_release(struct socket *sock)
 853 {
 854         __vsock_release(sock->sk, 0);
 855         sock->sk = NULL;
 856         sock->state = SS_FREE;
 857
 858         return 0;
 859 }
 860
 861 static int
 862 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 863 {
 864         int err;
 865         struct sock *sk;
 866         struct sockaddr_vm *vm_addr;
 867
 868         sk = sock->sk;
 869
 870         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 871                 return -EINVAL;
 872
 873         lock_sock(sk);
 874         err = __vsock_bind(sk, vm_addr);
 875         release_sock(sk);
 876
 877         return err;
 878 }
 879
 880 static int vsock_getname(struct socket *sock,
 881                          struct sockaddr *addr, int peer)
 882 {
 883         int err;
 884         struct sock *sk;
 885         struct vsock_sock *vsk;
 886         struct sockaddr_vm *vm_addr;
 887
 888         sk = sock->sk;
 889         vsk = vsock_sk(sk);
 890         err = 0;
 891
 892         lock_sock(sk);
 893
 894         if (peer) {
 895                 if (sock->state != SS_CONNECTED) {
 896                         err = -ENOTCONN;
 897                         goto out;
 898                 }
 899                 vm_addr = &vsk->remote_addr;
 900         } else {
 901                 vm_addr = &vsk->local_addr;
 902         }
 903
 904         if (!vm_addr) {
 905                 err = -EINVAL;
 906                 goto out;
 907         }
 908
 909         /* sys_getsockname() and sys_getpeername() pass us a
 910          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 911          * that macro is defined in socket.c instead of .h, so we hardcode its
 912          * value here.
 913          */
 914         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 915         memcpy(addr, vm_addr, sizeof(*vm_addr));
 916         err = sizeof(*vm_addr);
 917
 918 out:
 919         release_sock(sk);
 920         return err;
 921 }
 922
 923 static int vsock_shutdown(struct socket *sock, int mode)
 924 {
 925         int err;
 926         struct sock *sk;
 927
 928         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 929          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 930          * here like the other address families do.  Note also that the
 931          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 932          * which is what we want.
 933          */
 934         mode++;
 935
 936         if ((mode & ~SHUTDOWN_MASK) || !mode)
 937                 return -EINVAL;
 938
 939         /* If this is a STREAM socket and it is not connected then bail out
 940          * immediately.  If it is a DGRAM socket then we must first kick the
 941          * socket so that it wakes up from any sleeping calls, for example
 942          * recv(), and then afterwards return the error.
 943          */
 944
 945         sk = sock->sk;
 946
 947         lock_sock(sk);
 948         if (sock->state == SS_UNCONNECTED) {
 949                 err = -ENOTCONN;
 950                 if (sk->sk_type == SOCK_STREAM)
 951                         goto out;
 952         } else {
 953                 sock->state = SS_DISCONNECTING;
 954                 err = 0;
 955         }
 956
 957         /* Receive and send shutdowns are treated alike. */
 958         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 959         if (mode) {
 960                 sk->sk_shutdown |= mode;
 961                 sk->sk_state_change(sk);
 962
 963                 if (sk->sk_type == SOCK_STREAM) {
 964                         sock_reset_flag(sk, SOCK_DONE);
 965                         vsock_send_shutdown(sk, mode);
 966                 }
 967         }
 968
 969 out:
 970         release_sock(sk);
 971         return err;
 972 }
 973
 974 static __poll_t vsock_poll(struct file *file, struct socket *sock,
 975                                poll_table *wait)
 976 {
 977         struct sock *sk;
 978         __poll_t mask;
 979         struct vsock_sock *vsk;
 980
 981         sk = sock->sk;
 982         vsk = vsock_sk(sk);
 983
 984         poll_wait(file, sk_sleep(sk), wait);
 985         mask = 0;
 986
 987         if (sk->sk_err)
 988                 /* Signify that there has been an error on this socket. */
 989                 mask |= EPOLLERR;
 990
 991         /* INET sockets treat local write shutdown and peer write shutdown as a
 992          * case of EPOLLHUP set.
 993          */
 994         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 995             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 996              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 997                 mask |= EPOLLHUP;
 998         }
 999
1000         if (sk->sk_shutdown & RCV_SHUTDOWN ||
1001             vsk->peer_shutdown & SEND_SHUTDOWN) {
1002                 mask |= EPOLLRDHUP;
1003         }
1004
1005         if (sock->type == SOCK_DGRAM) {
1006                 /* For datagram sockets we can read if there is something in
1007                  * the queue and write as long as the socket isn't shutdown for
1008                  * sending.
1009                  */
1010                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1011                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
1012                         mask |= EPOLLIN | EPOLLRDNORM;
1013                 }
1014
1015                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1016                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1017
1018         } else if (sock->type == SOCK_STREAM) {
1019                 const struct vsock_transport *transport;
1020
1021                 lock_sock(sk);
1022
1023                 transport = vsk->transport;
1024
1025                 /* Listening sockets that have connections in their accept
1026                  * queue can be read.
1027                  */
1028                 if (sk->sk_state == TCP_LISTEN
1029                     && !vsock_is_accept_queue_empty(sk))
1030                         mask |= EPOLLIN | EPOLLRDNORM;
1031
1032                 /* If there is something in the queue then we can read. */
1033                 if (transport && transport->stream_is_active(vsk) &&
1034                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1035                         bool data_ready_now = false;
1036                         int ret = transport->notify_poll_in(
1037                                         vsk, 1, &data_ready_now);
1038                         if (ret < 0) {
1039                                 mask |= EPOLLERR;
1040                         } else {
1041                                 if (data_ready_now)
1042                                         mask |= EPOLLIN | EPOLLRDNORM;
1043
1044                         }
1045                 }
1046
1047                 /* Sockets whose connections have been closed, reset, or
1048                  * terminated should also be considered read, and we check the
1049                  * shutdown flag for that.
1050                  */
1051                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1052                     vsk->peer_shutdown & SEND_SHUTDOWN) {
1053                         mask |= EPOLLIN | EPOLLRDNORM;
1054                 }
1055
1056                 /* Connected sockets that can produce data can be written. */
1057                 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1058                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1059                                 bool space_avail_now = false;
1060                                 int ret = transport->notify_poll_out(
1061                                                 vsk, 1, &space_avail_now);
1062                                 if (ret < 0) {
1063                                         mask |= EPOLLERR;
1064                                 } else {
1065                                         if (space_avail_now)
1066                                                 /* Remove EPOLLWRBAND since INET
1067                                                  * sockets are not setting it.
1068                                                  */
1069                                                 mask |= EPOLLOUT | EPOLLWRNORM;
1070
1071                                 }
1072                         }
1073                 }
1074
1075                 /* Simulate INET socket poll behaviors, which sets
1076                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1077                  * but local send is not shutdown.
1078                  */
1079                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1080                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1081                                 mask |= EPOLLOUT | EPOLLWRNORM;
1082
1083                 }
1084
1085                 release_sock(sk);
1086         }
1087
1088         return mask;
1089 }
1090
1091 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1092                                size_t len)
1093 {
1094         int err;
1095         struct sock *sk;
1096         struct vsock_sock *vsk;
1097         struct sockaddr_vm *remote_addr;
1098         const struct vsock_transport *transport;
1099
1100         if (msg->msg_flags & MSG_OOB)
1101                 return -EOPNOTSUPP;
1102
1103         /* For now, MSG_DONTWAIT is always assumed... */
1104         err = 0;
1105         sk = sock->sk;
1106         vsk = vsock_sk(sk);
1107
1108         lock_sock(sk);
1109
1110         transport = vsk->transport;
1111
1112         err = vsock_auto_bind(vsk);
1113         if (err)
1114                 goto out;
1115
1116
1117         /* If the provided message contains an address, use that.  Otherwise
1118          * fall back on the socket's remote handle (if it has been connected).
1119          */
1120         if (msg->msg_name &&
1121             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1122                             &remote_addr) == 0) {
1123                 /* Ensure this address is of the right type and is a valid
1124                  * destination.
1125                  */
1126
1127                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1128                         remote_addr->svm_cid = transport->get_local_cid();
1129
1130                 if (!vsock_addr_bound(remote_addr)) {
1131                         err = -EINVAL;
1132                         goto out;
1133                 }
1134         } else if (sock->state == SS_CONNECTED) {
1135                 remote_addr = &vsk->remote_addr;
1136
1137                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1138                         remote_addr->svm_cid = transport->get_local_cid();
1139
1140                 /* XXX Should connect() or this function ensure remote_addr is
1141                  * bound?
1142                  */
1143                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1144                         err = -EINVAL;
1145                         goto out;
1146                 }
1147         } else {
1148                 err = -EINVAL;
1149                 goto out;
1150         }
1151
1152         if (!transport->dgram_allow(remote_addr->svm_cid,
1153                                     remote_addr->svm_port)) {
1154                 err = -EINVAL;
1155                 goto out;
1156         }
1157
1158         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1159
1160 out:
1161         release_sock(sk);
1162         return err;
1163 }
1164
1165 static int vsock_dgram_connect(struct socket *sock,
1166                                struct sockaddr *addr, int addr_len, int flags)
1167 {
1168         int err;
1169         struct sock *sk;
1170         struct vsock_sock *vsk;
1171         struct sockaddr_vm *remote_addr;
1172
1173         sk = sock->sk;
1174         vsk = vsock_sk(sk);
1175
1176         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1177         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1178                 lock_sock(sk);
1179                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1180                                 VMADDR_PORT_ANY);
1181                 sock->state = SS_UNCONNECTED;
1182                 release_sock(sk);
1183                 return 0;
1184         } else if (err != 0)
1185                 return -EINVAL;
1186
1187         lock_sock(sk);
1188
1189         err = vsock_auto_bind(vsk);
1190         if (err)
1191                 goto out;
1192
1193         if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1194                                          remote_addr->svm_port)) {
1195                 err = -EINVAL;
1196                 goto out;
1197         }
1198
1199         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1200         sock->state = SS_CONNECTED;
1201
1202 out:
1203         release_sock(sk);
1204         return err;
1205 }
1206
1207 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1208                                size_t len, int flags)
1209 {
1210         struct vsock_sock *vsk = vsock_sk(sock->sk);
1211
1212         return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1213 }
1214
1215 static const struct proto_ops vsock_dgram_ops = {
1216         .family = PF_VSOCK,
1217         .owner = THIS_MODULE,
1218         .release = vsock_release,
1219         .bind = vsock_bind,
1220         .connect = vsock_dgram_connect,
1221         .socketpair = sock_no_socketpair,
1222         .accept = sock_no_accept,
1223         .getname = vsock_getname,
1224         .poll = vsock_poll,
1225         .ioctl = sock_no_ioctl,
1226         .listen = sock_no_listen,
1227         .shutdown = vsock_shutdown,
1228         .sendmsg = vsock_dgram_sendmsg,
1229         .recvmsg = vsock_dgram_recvmsg,
1230         .mmap = sock_no_mmap,
1231         .sendpage = sock_no_sendpage,
1232 };
1233
1234 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1235 {
1236         const struct vsock_transport *transport = vsk->transport;
1237
1238         if (!transport || !transport->cancel_pkt)
1239                 return -EOPNOTSUPP;
1240
1241         return transport->cancel_pkt(vsk);
1242 }
1243
1244 static void vsock_connect_timeout(struct work_struct *work)
1245 {
1246         struct sock *sk;
1247         struct vsock_sock *vsk;
1248
1249         vsk = container_of(work, struct vsock_sock, connect_work.work);
1250         sk = sk_vsock(vsk);
1251
1252         lock_sock(sk);
1253         if (sk->sk_state == TCP_SYN_SENT &&
1254             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1255                 sk->sk_state = TCP_CLOSE;
1256                 sk->sk_err = ETIMEDOUT;
1257                 sk->sk_error_report(sk);
1258                 vsock_transport_cancel_pkt(vsk);
1259         }
1260         release_sock(sk);
1261
1262         sock_put(sk);
1263 }
1264
1265 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1266                                 int addr_len, int flags)
1267 {
1268         int err;
1269         struct sock *sk;
1270         struct vsock_sock *vsk;
1271         const struct vsock_transport *transport;
1272         struct sockaddr_vm *remote_addr;
1273         long timeout;
1274         DEFINE_WAIT(wait);
1275
1276         err = 0;
1277         sk = sock->sk;
1278         vsk = vsock_sk(sk);
1279
1280         lock_sock(sk);
1281
1282         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1283         switch (sock->state) {
1284         case SS_CONNECTED:
1285                 err = -EISCONN;
1286                 goto out;
1287         case SS_DISCONNECTING:
1288                 err = -EINVAL;
1289                 goto out;
1290         case SS_CONNECTING:
1291                 /* This continues on so we can move sock into the SS_CONNECTED
1292                  * state once the connection has completed (at which point err
1293                  * will be set to zero also).  Otherwise, we will either wait
1294                  * for the connection or return -EALREADY should this be a
1295                  * non-blocking call.
1296                  */
1297                 err = -EALREADY;
1298                 break;
1299         default:
1300                 if ((sk->sk_state == TCP_LISTEN) ||
1301                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1302                         err = -EINVAL;
1303                         goto out;
1304                 }
1305
1306                 /* Set the remote address that we are connecting to. */
1307                 memcpy(&vsk->remote_addr, remote_addr,
1308                        sizeof(vsk->remote_addr));
1309
1310                 err = vsock_assign_transport(vsk, NULL);
1311                 if (err)
1312                         goto out;
1313
1314                 transport = vsk->transport;
1315
1316                 /* The hypervisor and well-known contexts do not have socket
1317                  * endpoints.
1318                  */
1319                 if (!transport ||
1320                     !transport->stream_allow(remote_addr->svm_cid,
1321                                              remote_addr->svm_port)) {
1322                         err = -ENETUNREACH;
1323                         goto out;
1324                 }
1325
1326                 err = vsock_auto_bind(vsk);
1327                 if (err)
1328                         goto out;
1329
1330                 sk->sk_state = TCP_SYN_SENT;
1331
1332                 err = transport->connect(vsk);
1333                 if (err < 0)
1334                         goto out;
1335
1336                 /* Mark sock as connecting and set the error code to in
1337                  * progress in case this is a non-blocking connect.
1338                  */
1339                 sock->state = SS_CONNECTING;
1340                 err = -EINPROGRESS;
1341         }
1342
1343         /* The receive path will handle all communication until we are able to
1344          * enter the connected state.  Here we wait for the connection to be
1345          * completed or a notification of an error.
1346          */
1347         timeout = vsk->connect_timeout;
1348         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1349
1350         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1351                 if (flags & O_NONBLOCK) {
1352                         /* If we're not going to block, we schedule a timeout
1353                          * function to generate a timeout on the connection
1354                          * attempt, in case the peer doesn't respond in a
1355                          * timely manner. We hold on to the socket until the
1356                          * timeout fires.
1357                          */
1358                         sock_hold(sk);
1359                         schedule_delayed_work(&vsk->connect_work, timeout);
1360
1361                         /* Skip ahead to preserve error code set above. */
1362                         goto out_wait;
1363                 }
1364
1365                 release_sock(sk);
1366                 timeout = schedule_timeout(timeout);
1367                 lock_sock(sk);
1368
1369                 if (signal_pending(current)) {
1370                         err = sock_intr_errno(timeout);
1371                         sk->sk_state = TCP_CLOSE;
1372                         sock->state = SS_UNCONNECTED;
1373                         vsock_transport_cancel_pkt(vsk);
1374                         goto out_wait;
1375                 } else if (timeout == 0) {
1376                         err = -ETIMEDOUT;
1377                         sk->sk_state = TCP_CLOSE;
1378                         sock->state = SS_UNCONNECTED;
1379                         vsock_transport_cancel_pkt(vsk);
1380                         goto out_wait;
1381                 }
1382
1383                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1384         }
1385
1386         if (sk->sk_err) {
1387                 err = -sk->sk_err;
1388                 sk->sk_state = TCP_CLOSE;
1389                 sock->state = SS_UNCONNECTED;
1390         } else {
1391                 err = 0;
1392         }
1393
1394 out_wait:
1395         finish_wait(sk_sleep(sk), &wait);
1396 out:
1397         release_sock(sk);
1398         return err;
1399 }
1400
1401 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1402                         bool kern)
1403 {
1404         struct sock *listener;
1405         int err;
1406         struct sock *connected;
1407         struct vsock_sock *vconnected;
1408         long timeout;
1409         DEFINE_WAIT(wait);
1410
1411         err = 0;
1412         listener = sock->sk;
1413
1414         lock_sock(listener);
1415
1416         if (sock->type != SOCK_STREAM) {
1417                 err = -EOPNOTSUPP;
1418                 goto out;
1419         }
1420
1421         if (listener->sk_state != TCP_LISTEN) {
1422                 err = -EINVAL;
1423                 goto out;
1424         }
1425
1426         /* Wait for children sockets to appear; these are the new sockets
1427          * created upon connection establishment.
1428          */
1429         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1430         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1431
1432         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1433                listener->sk_err == 0) {
1434                 release_sock(listener);
1435                 timeout = schedule_timeout(timeout);
1436                 finish_wait(sk_sleep(listener), &wait);
1437                 lock_sock(listener);
1438
1439                 if (signal_pending(current)) {
1440                         err = sock_intr_errno(timeout);
1441                         goto out;
1442                 } else if (timeout == 0) {
1443                         err = -EAGAIN;
1444                         goto out;
1445                 }
1446
1447                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1448         }
1449         finish_wait(sk_sleep(listener), &wait);
1450
1451         if (listener->sk_err)
1452                 err = -listener->sk_err;
1453
1454         if (connected) {
1455                 sk_acceptq_removed(listener);
1456
1457                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1458                 vconnected = vsock_sk(connected);
1459
1460                 /* If the listener socket has received an error, then we should
1461                  * reject this socket and return.  Note that we simply mark the
1462                  * socket rejected, drop our reference, and let the cleanup
1463                  * function handle the cleanup; the fact that we found it in
1464                  * the listener's accept queue guarantees that the cleanup
1465                  * function hasn't run yet.
1466                  */
1467                 if (err) {
1468                         vconnected->rejected = true;
1469                 } else {
1470                         newsock->state = SS_CONNECTED;
1471                         sock_graft(connected, newsock);
1472                 }
1473
1474                 release_sock(connected);
1475                 sock_put(connected);
1476         }
1477
1478 out:
1479         release_sock(listener);
1480         return err;
1481 }
1482
1483 static int vsock_listen(struct socket *sock, int backlog)
1484 {
1485         int err;
1486         struct sock *sk;
1487         struct vsock_sock *vsk;
1488
1489         sk = sock->sk;
1490
1491         lock_sock(sk);
1492
1493         if (sock->type != SOCK_STREAM) {
1494                 err = -EOPNOTSUPP;
1495                 goto out;
1496         }
1497
1498         if (sock->state != SS_UNCONNECTED) {
1499                 err = -EINVAL;
1500                 goto out;
1501         }
1502
1503         vsk = vsock_sk(sk);
1504
1505         if (!vsock_addr_bound(&vsk->local_addr)) {
1506                 err = -EINVAL;
1507                 goto out;
1508         }
1509
1510         sk->sk_max_ack_backlog = backlog;
1511         sk->sk_state = TCP_LISTEN;
1512
1513         err = 0;
1514
1515 out:
1516         release_sock(sk);
1517         return err;
1518 }
1519
1520 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1521                                      const struct vsock_transport *transport,
1522                                      u64 val)
1523 {
1524         if (val > vsk->buffer_max_size)
1525                 val = vsk->buffer_max_size;
1526
1527         if (val < vsk->buffer_min_size)
1528                 val = vsk->buffer_min_size;
1529
1530         if (val != vsk->buffer_size &&
1531             transport && transport->notify_buffer_size)
1532                 transport->notify_buffer_size(vsk, &val);
1533
1534         vsk->buffer_size = val;
1535 }
1536
1537 static int vsock_stream_setsockopt(struct socket *sock,
1538                                    int level,
1539                                    int optname,
1540                                    sockptr_t optval,
1541                                    unsigned int optlen)
1542 {
1543         int err;
1544         struct sock *sk;
1545         struct vsock_sock *vsk;
1546         const struct vsock_transport *transport;
1547         u64 val;
1548
1549         if (level != AF_VSOCK)
1550                 return -ENOPROTOOPT;
1551
1552 #define COPY_IN(_v)                                       \
1553         do {                                              \
1554                 if (optlen < sizeof(_v)) {                \
1555                         err = -EINVAL;                    \
1556                         goto exit;                        \
1557                 }                                         \
1558                 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) {  \
1559                         err = -EFAULT;                                  \
1560                         goto exit;                                      \
1561                 }                                                       \
1562         } while (0)
1563
1564         err = 0;
1565         sk = sock->sk;
1566         vsk = vsock_sk(sk);
1567
1568         lock_sock(sk);
1569
1570         transport = vsk->transport;
1571
1572         switch (optname) {
1573         case SO_VM_SOCKETS_BUFFER_SIZE:
1574                 COPY_IN(val);
1575                 vsock_update_buffer_size(vsk, transport, val);
1576                 break;
1577
1578         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1579                 COPY_IN(val);
1580                 vsk->buffer_max_size = val;
1581                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1582                 break;
1583
1584         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1585                 COPY_IN(val);
1586                 vsk->buffer_min_size = val;
1587                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1588                 break;
1589
1590         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1591                 struct __kernel_old_timeval tv;
1592                 COPY_IN(tv);
1593                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1594                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1595                         vsk->connect_timeout = tv.tv_sec * HZ +
1596                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1597                         if (vsk->connect_timeout == 0)
1598                                 vsk->connect_timeout =
1599                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1600
1601                 } else {
1602                         err = -ERANGE;
1603                 }
1604                 break;
1605         }
1606
1607         default:
1608                 err = -ENOPROTOOPT;
1609                 break;
1610         }
1611
1612 #undef COPY_IN
1613
1614 exit:
1615         release_sock(sk);
1616         return err;
1617 }
1618
1619 static int vsock_stream_getsockopt(struct socket *sock,
1620                                    int level, int optname,
1621                                    char __user *optval,
1622                                    int __user *optlen)
1623 {
1624         int err;
1625         int len;
1626         struct sock *sk;
1627         struct vsock_sock *vsk;
1628         u64 val;
1629
1630         if (level != AF_VSOCK)
1631                 return -ENOPROTOOPT;
1632
1633         err = get_user(len, optlen);
1634         if (err != 0)
1635                 return err;
1636
1637 #define COPY_OUT(_v)                            \
1638         do {                                    \
1639                 if (len < sizeof(_v))           \
1640                         return -EINVAL;         \
1641                                                 \
1642                 len = sizeof(_v);               \
1643                 if (copy_to_user(optval, &_v, len) != 0)        \
1644                         return -EFAULT;                         \
1645                                                                 \
1646         } while (0)
1647
1648         err = 0;
1649         sk = sock->sk;
1650         vsk = vsock_sk(sk);
1651
1652         switch (optname) {
1653         case SO_VM_SOCKETS_BUFFER_SIZE:
1654                 val = vsk->buffer_size;
1655                 COPY_OUT(val);
1656                 break;
1657
1658         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1659                 val = vsk->buffer_max_size;
1660                 COPY_OUT(val);
1661                 break;
1662
1663         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1664                 val = vsk->buffer_min_size;
1665                 COPY_OUT(val);
1666                 break;
1667
1668         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1669                 struct __kernel_old_timeval tv;
1670                 tv.tv_sec = vsk->connect_timeout / HZ;
1671                 tv.tv_usec =
1672                     (vsk->connect_timeout -
1673                      tv.tv_sec * HZ) * (1000000 / HZ);
1674                 COPY_OUT(tv);
1675                 break;
1676         }
1677         default:
1678                 return -ENOPROTOOPT;
1679         }
1680
1681         err = put_user(len, optlen);
1682         if (err != 0)
1683                 return -EFAULT;
1684
1685 #undef COPY_OUT
1686
1687         return 0;
1688 }
1689
1690 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1691                                 size_t len)
1692 {
1693         struct sock *sk;
1694         struct vsock_sock *vsk;
1695         const struct vsock_transport *transport;
1696         ssize_t total_written;
1697         long timeout;
1698         int err;
1699         struct vsock_transport_send_notify_data send_data;
1700         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1701
1702         sk = sock->sk;
1703         vsk = vsock_sk(sk);
1704         total_written = 0;
1705         err = 0;
1706
1707         if (msg->msg_flags & MSG_OOB)
1708                 return -EOPNOTSUPP;
1709
1710         lock_sock(sk);
1711
1712         transport = vsk->transport;
1713
1714         /* Callers should not provide a destination with stream sockets. */
1715         if (msg->msg_namelen) {
1716                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1717                 goto out;
1718         }
1719
1720         /* Send data only if both sides are not shutdown in the direction. */
1721         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1722             vsk->peer_shutdown & RCV_SHUTDOWN) {
1723                 err = -EPIPE;
1724                 goto out;
1725         }
1726
1727         if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1728             !vsock_addr_bound(&vsk->local_addr)) {
1729                 err = -ENOTCONN;
1730                 goto out;
1731         }
1732
1733         if (!vsock_addr_bound(&vsk->remote_addr)) {
1734                 err = -EDESTADDRREQ;
1735                 goto out;
1736         }
1737
1738         /* Wait for room in the produce queue to enqueue our user's data. */
1739         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1740
1741         err = transport->notify_send_init(vsk, &send_data);
1742         if (err < 0)
1743                 goto out;
1744
1745         while (total_written < len) {
1746                 ssize_t written;
1747
1748                 add_wait_queue(sk_sleep(sk), &wait);
1749                 while (vsock_stream_has_space(vsk) == 0 &&
1750                        sk->sk_err == 0 &&
1751                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1752                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1753
1754                         /* Don't wait for non-blocking sockets. */
1755                         if (timeout == 0) {
1756                                 err = -EAGAIN;
1757                                 remove_wait_queue(sk_sleep(sk), &wait);
1758                                 goto out_err;
1759                         }
1760
1761                         err = transport->notify_send_pre_block(vsk, &send_data);
1762                         if (err < 0) {
1763                                 remove_wait_queue(sk_sleep(sk), &wait);
1764                                 goto out_err;
1765                         }
1766
1767                         release_sock(sk);
1768                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1769                         lock_sock(sk);
1770                         if (signal_pending(current)) {
1771                                 err = sock_intr_errno(timeout);
1772                                 remove_wait_queue(sk_sleep(sk), &wait);
1773                                 goto out_err;
1774                         } else if (timeout == 0) {
1775                                 err = -EAGAIN;
1776                                 remove_wait_queue(sk_sleep(sk), &wait);
1777                                 goto out_err;
1778                         }
1779                 }
1780                 remove_wait_queue(sk_sleep(sk), &wait);
1781
1782                 /* These checks occur both as part of and after the loop
1783                  * conditional since we need to check before and after
1784                  * sleeping.
1785                  */
1786                 if (sk->sk_err) {
1787                         err = -sk->sk_err;
1788                         goto out_err;
1789                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1790                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1791                         err = -EPIPE;
1792                         goto out_err;
1793                 }
1794
1795                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1796                 if (err < 0)
1797                         goto out_err;
1798
1799                 /* Note that enqueue will only write as many bytes as are free
1800                  * in the produce queue, so we don't need to ensure len is
1801                  * smaller than the queue size.  It is the caller's
1802                  * responsibility to check how many bytes we were able to send.
1803                  */
1804
1805                 written = transport->stream_enqueue(
1806                                 vsk, msg,
1807                                 len - total_written);
1808                 if (written < 0) {
1809                         err = -ENOMEM;
1810                         goto out_err;
1811                 }
1812
1813                 total_written += written;
1814
1815                 err = transport->notify_send_post_enqueue(
1816                                 vsk, written, &send_data);
1817                 if (err < 0)
1818                         goto out_err;
1819
1820         }
1821
1822 out_err:
1823         if (total_written > 0)
1824                 err = total_written;
1825 out:
1826         release_sock(sk);
1827         return err;
1828 }
1829
1830
1831 static int
1832 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1833                      int flags)
1834 {
1835         struct sock *sk;
1836         struct vsock_sock *vsk;
1837         const struct vsock_transport *transport;
1838         int err;
1839         size_t target;
1840         ssize_t copied;
1841         long timeout;
1842         struct vsock_transport_recv_notify_data recv_data;
1843
1844         DEFINE_WAIT(wait);
1845
1846         sk = sock->sk;
1847         vsk = vsock_sk(sk);
1848         err = 0;
1849
1850         lock_sock(sk);
1851
1852         transport = vsk->transport;
1853
1854         if (!transport || sk->sk_state != TCP_ESTABLISHED) {
1855                 /* Recvmsg is supposed to return 0 if a peer performs an
1856                  * orderly shutdown. Differentiate between that case and when a
1857                  * peer has not connected or a local shutdown occured with the
1858                  * SOCK_DONE flag.
1859                  */
1860                 if (sock_flag(sk, SOCK_DONE))
1861                         err = 0;
1862                 else
1863                         err = -ENOTCONN;
1864
1865                 goto out;
1866         }
1867
1868         if (flags & MSG_OOB) {
1869                 err = -EOPNOTSUPP;
1870                 goto out;
1871         }
1872
1873         /* We don't check peer_shutdown flag here since peer may actually shut
1874          * down, but there can be data in the queue that a local socket can
1875          * receive.
1876          */
1877         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1878                 err = 0;
1879                 goto out;
1880         }
1881
1882         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1883          * is not an error.  We may as well bail out now.
1884          */
1885         if (!len) {
1886                 err = 0;
1887                 goto out;
1888         }
1889
1890         /* We must not copy less than target bytes into the user's buffer
1891          * before returning successfully, so we wait for the consume queue to
1892          * have that much data to consume before dequeueing.  Note that this
1893          * makes it impossible to handle cases where target is greater than the
1894          * queue size.
1895          */
1896         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1897         if (target >= transport->stream_rcvhiwat(vsk)) {
1898                 err = -ENOMEM;
1899                 goto out;
1900         }
1901         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1902         copied = 0;
1903
1904         err = transport->notify_recv_init(vsk, target, &recv_data);
1905         if (err < 0)
1906                 goto out;
1907
1908
1909         while (1) {
1910                 s64 ready;
1911
1912                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1913                 ready = vsock_stream_has_data(vsk);
1914
1915                 if (ready == 0) {
1916                         if (sk->sk_err != 0 ||
1917                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1918                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1919                                 finish_wait(sk_sleep(sk), &wait);
1920                                 break;
1921                         }
1922                         /* Don't wait for non-blocking sockets. */
1923                         if (timeout == 0) {
1924                                 err = -EAGAIN;
1925                                 finish_wait(sk_sleep(sk), &wait);
1926                                 break;
1927                         }
1928
1929                         err = transport->notify_recv_pre_block(
1930                                         vsk, target, &recv_data);
1931                         if (err < 0) {
1932                                 finish_wait(sk_sleep(sk), &wait);
1933                                 break;
1934                         }
1935                         release_sock(sk);
1936                         timeout = schedule_timeout(timeout);
1937                         lock_sock(sk);
1938
1939                         if (signal_pending(current)) {
1940                                 err = sock_intr_errno(timeout);
1941                                 finish_wait(sk_sleep(sk), &wait);
1942                                 break;
1943                         } else if (timeout == 0) {
1944                                 err = -EAGAIN;
1945                                 finish_wait(sk_sleep(sk), &wait);
1946                                 break;
1947                         }
1948                 } else {
1949                         ssize_t read;
1950
1951                         finish_wait(sk_sleep(sk), &wait);
1952
1953                         if (ready < 0) {
1954                                 /* Invalid queue pair content. XXX This should
1955                                 * be changed to a connection reset in a later
1956                                 * change.
1957                                 */
1958
1959                                 err = -ENOMEM;
1960                                 goto out;
1961                         }
1962
1963                         err = transport->notify_recv_pre_dequeue(
1964                                         vsk, target, &recv_data);
1965                         if (err < 0)
1966                                 break;
1967
1968                         read = transport->stream_dequeue(
1969                                         vsk, msg,
1970                                         len - copied, flags);
1971                         if (read < 0) {
1972                                 err = -ENOMEM;
1973                                 break;
1974                         }
1975
1976                         copied += read;
1977
1978                         err = transport->notify_recv_post_dequeue(
1979                                         vsk, target, read,
1980                                         !(flags & MSG_PEEK), &recv_data);
1981                         if (err < 0)
1982                                 goto out;
1983
1984                         if (read >= target || flags & MSG_PEEK)
1985                                 break;
1986
1987                         target -= read;
1988                 }
1989         }
1990
1991         if (sk->sk_err)
1992                 err = -sk->sk_err;
1993         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1994                 err = 0;
1995
1996         if (copied > 0)
1997                 err = copied;
1998
1999 out:
2000         release_sock(sk);
2001         return err;
2002 }
2003
2004 static const struct proto_ops vsock_stream_ops = {
2005         .family = PF_VSOCK,
2006         .owner = THIS_MODULE,
2007         .release = vsock_release,
2008         .bind = vsock_bind,
2009         .connect = vsock_stream_connect,
2010         .socketpair = sock_no_socketpair,
2011         .accept = vsock_accept,
2012         .getname = vsock_getname,
2013         .poll = vsock_poll,
2014         .ioctl = sock_no_ioctl,
2015         .listen = vsock_listen,
2016         .shutdown = vsock_shutdown,
2017         .setsockopt = vsock_stream_setsockopt,
2018         .getsockopt = vsock_stream_getsockopt,
2019         .sendmsg = vsock_stream_sendmsg,
2020         .recvmsg = vsock_stream_recvmsg,
2021         .mmap = sock_no_mmap,
2022         .sendpage = sock_no_sendpage,
2023 };
2024
2025 static int vsock_create(struct net *net, struct socket *sock,
2026                         int protocol, int kern)
2027 {
2028         struct vsock_sock *vsk;
2029         struct sock *sk;
2030         int ret;
2031
2032         if (!sock)
2033                 return -EINVAL;
2034
2035         if (protocol && protocol != PF_VSOCK)
2036                 return -EPROTONOSUPPORT;
2037
2038         switch (sock->type) {
2039         case SOCK_DGRAM:
2040                 sock->ops = &vsock_dgram_ops;
2041                 break;
2042         case SOCK_STREAM:
2043                 sock->ops = &vsock_stream_ops;
2044                 break;
2045         default:
2046                 return -ESOCKTNOSUPPORT;
2047         }
2048
2049         sock->state = SS_UNCONNECTED;
2050
2051         sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2052         if (!sk)
2053                 return -ENOMEM;
2054
2055         vsk = vsock_sk(sk);
2056
2057         if (sock->type == SOCK_DGRAM) {
2058                 ret = vsock_assign_transport(vsk, NULL);
2059                 if (ret < 0) {
2060                         sock_put(sk);
2061                         return ret;
2062                 }
2063         }
2064
2065         vsock_insert_unbound(vsk);
2066
2067         return 0;
2068 }
2069
2070 static const struct net_proto_family vsock_family_ops = {
2071         .family = AF_VSOCK,
2072         .create = vsock_create,
2073         .owner = THIS_MODULE,
2074 };
2075
2076 static long vsock_dev_do_ioctl(struct file *filp,
2077                                unsigned int cmd, void __user *ptr)
2078 {
2079         u32 __user *p = ptr;
2080         u32 cid = VMADDR_CID_ANY;
2081         int retval = 0;
2082
2083         switch (cmd) {
2084         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2085                 /* To be compatible with the VMCI behavior, we prioritize the
2086                  * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2087                  */
2088                 if (transport_g2h)
2089                         cid = transport_g2h->get_local_cid();
2090                 else if (transport_h2g)
2091                         cid = transport_h2g->get_local_cid();
2092
2093                 if (put_user(cid, p) != 0)
2094                         retval = -EFAULT;
2095                 break;
2096
2097         default:
2098                 retval = -ENOIOCTLCMD;
2099         }
2100
2101         return retval;
2102 }
2103
2104 static long vsock_dev_ioctl(struct file *filp,
2105                             unsigned int cmd, unsigned long arg)
2106 {
2107         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2108 }
2109
2110 #ifdef CONFIG_COMPAT
2111 static long vsock_dev_compat_ioctl(struct file *filp,
2112                                    unsigned int cmd, unsigned long arg)
2113 {
2114         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2115 }
2116 #endif
2117
2118 static const struct file_operations vsock_device_ops = {
2119         .owner          = THIS_MODULE,
2120         .unlocked_ioctl = vsock_dev_ioctl,
2121 #ifdef CONFIG_COMPAT
2122         .compat_ioctl   = vsock_dev_compat_ioctl,
2123 #endif
2124         .open           = nonseekable_open,
2125 };
2126
2127 static struct miscdevice vsock_device = {
2128         .name           = "vsock",
2129         .fops           = &vsock_device_ops,
2130 };
2131
2132 static int __init vsock_init(void)
2133 {
2134         int err = 0;
2135
2136         vsock_init_tables();
2137
2138         vsock_proto.owner = THIS_MODULE;
2139         vsock_device.minor = MISC_DYNAMIC_MINOR;
2140         err = misc_register(&vsock_device);
2141         if (err) {
2142                 pr_err("Failed to register misc device\n");
2143                 goto err_reset_transport;
2144         }
2145
2146         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2147         if (err) {
2148                 pr_err("Cannot register vsock protocol\n");
2149                 goto err_deregister_misc;
2150         }
2151
2152         err = sock_register(&vsock_family_ops);
2153         if (err) {
2154                 pr_err("could not register af_vsock (%d) address family: %d\n",
2155                        AF_VSOCK, err);
2156                 goto err_unregister_proto;
2157         }
2158
2159         return 0;
2160
2161 err_unregister_proto:
2162         proto_unregister(&vsock_proto);
2163 err_deregister_misc:
2164         misc_deregister(&vsock_device);
2165 err_reset_transport:
2166         return err;
2167 }
2168
2169 static void __exit vsock_exit(void)
2170 {
2171         misc_deregister(&vsock_device);
2172         sock_unregister(AF_VSOCK);
2173         proto_unregister(&vsock_proto);
2174 }
2175
2176 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2177 {
2178         return vsk->transport;
2179 }
2180 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2181
2182 int vsock_core_register(const struct vsock_transport *t, int features)
2183 {
2184         const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2185         int err = mutex_lock_interruptible(&vsock_register_mutex);
2186
2187         if (err)
2188                 return err;
2189
2190         t_h2g = transport_h2g;
2191         t_g2h = transport_g2h;
2192         t_dgram = transport_dgram;
2193         t_local = transport_local;
2194
2195         if (features & VSOCK_TRANSPORT_F_H2G) {
2196                 if (t_h2g) {
2197                         err = -EBUSY;
2198                         goto err_busy;
2199                 }
2200                 t_h2g = t;
2201         }
2202
2203         if (features & VSOCK_TRANSPORT_F_G2H) {
2204                 if (t_g2h) {
2205                         err = -EBUSY;
2206                         goto err_busy;
2207                 }
2208                 t_g2h = t;
2209         }
2210
2211         if (features & VSOCK_TRANSPORT_F_DGRAM) {
2212                 if (t_dgram) {
2213                         err = -EBUSY;
2214                         goto err_busy;
2215                 }
2216                 t_dgram = t;
2217         }
2218
2219         if (features & VSOCK_TRANSPORT_F_LOCAL) {
2220                 if (t_local) {
2221                         err = -EBUSY;
2222                         goto err_busy;
2223                 }
2224                 t_local = t;
2225         }
2226
2227         transport_h2g = t_h2g;
2228         transport_g2h = t_g2h;
2229         transport_dgram = t_dgram;
2230         transport_local = t_local;
2231
2232 err_busy:
2233         mutex_unlock(&vsock_register_mutex);
2234         return err;
2235 }
2236 EXPORT_SYMBOL_GPL(vsock_core_register);
2237
2238 void vsock_core_unregister(const struct vsock_transport *t)
2239 {
2240         mutex_lock(&vsock_register_mutex);
2241
2242         if (transport_h2g == t)
2243                 transport_h2g = NULL;
2244
2245         if (transport_g2h == t)
2246                 transport_g2h = NULL;
2247
2248         if (transport_dgram == t)
2249                 transport_dgram = NULL;
2250
2251         if (transport_local == t)
2252                 transport_local = NULL;
2253
2254         mutex_unlock(&vsock_register_mutex);
2255 }
2256 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2257
2258 module_init(vsock_init);
2259 module_exit(vsock_exit);
2260
2261 MODULE_AUTHOR("VMware, Inc.");
2262 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2263 MODULE_VERSION("1.0.2.0-k");
2264 MODULE_LICENSE("GPL v2");