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_connectible(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_connectible(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                 security_sk_clone(parent, sk);
 759         } else {
 760                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 761                 vsk->owner = get_current_cred();
 762                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 763                 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
 764                 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
 765                 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
 766         }
 767
 768         return sk;
 769 }
 770
 771 static bool sock_type_connectible(u16 type)
 772 {
 773         return type == SOCK_STREAM;
 774 }
 775
 776 static void __vsock_release(struct sock *sk, int level)
 777 {
 778         if (sk) {
 779                 struct sock *pending;
 780                 struct vsock_sock *vsk;
 781
 782                 vsk = vsock_sk(sk);
 783                 pending = NULL; /* Compiler warning. */
 784
 785                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 786                  * version to avoid the warning "possible recursive locking
 787                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 788                  * is the same as lock_sock(sk).
 789                  */
 790                 lock_sock_nested(sk, level);
 791
 792                 if (vsk->transport)
 793                         vsk->transport->release(vsk);
 794                 else if (sock_type_connectible(sk->sk_type))
 795                         vsock_remove_sock(vsk);
 796
 797                 sock_orphan(sk);
 798                 sk->sk_shutdown = SHUTDOWN_MASK;
 799
 800                 skb_queue_purge(&sk->sk_receive_queue);
 801
 802                 /* Clean up any sockets that never were accepted. */
 803                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 804                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 805                         sock_put(pending);
 806                 }
 807
 808                 release_sock(sk);
 809                 sock_put(sk);
 810         }
 811 }
 812
 813 static void vsock_sk_destruct(struct sock *sk)
 814 {
 815         struct vsock_sock *vsk = vsock_sk(sk);
 816
 817         vsock_deassign_transport(vsk);
 818
 819         /* When clearing these addresses, there's no need to set the family and
 820          * possibly register the address family with the kernel.
 821          */
 822         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 823         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 824
 825         put_cred(vsk->owner);
 826 }
 827
 828 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 829 {
 830         int err;
 831
 832         err = sock_queue_rcv_skb(sk, skb);
 833         if (err)
 834                 kfree_skb(skb);
 835
 836         return err;
 837 }
 838
 839 struct sock *vsock_create_connected(struct sock *parent)
 840 {
 841         return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
 842                               parent->sk_type, 0);
 843 }
 844 EXPORT_SYMBOL_GPL(vsock_create_connected);
 845
 846 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 847 {
 848         return vsk->transport->stream_has_data(vsk);
 849 }
 850 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 851
 852 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 853 {
 854         return vsk->transport->stream_has_space(vsk);
 855 }
 856 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 857
 858 static int vsock_release(struct socket *sock)
 859 {
 860         __vsock_release(sock->sk, 0);
 861         sock->sk = NULL;
 862         sock->state = SS_FREE;
 863
 864         return 0;
 865 }
 866
 867 static int
 868 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 869 {
 870         int err;
 871         struct sock *sk;
 872         struct sockaddr_vm *vm_addr;
 873
 874         sk = sock->sk;
 875
 876         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 877                 return -EINVAL;
 878
 879         lock_sock(sk);
 880         err = __vsock_bind(sk, vm_addr);
 881         release_sock(sk);
 882
 883         return err;
 884 }
 885
 886 static int vsock_getname(struct socket *sock,
 887                          struct sockaddr *addr, int peer)
 888 {
 889         int err;
 890         struct sock *sk;
 891         struct vsock_sock *vsk;
 892         struct sockaddr_vm *vm_addr;
 893
 894         sk = sock->sk;
 895         vsk = vsock_sk(sk);
 896         err = 0;
 897
 898         lock_sock(sk);
 899
 900         if (peer) {
 901                 if (sock->state != SS_CONNECTED) {
 902                         err = -ENOTCONN;
 903                         goto out;
 904                 }
 905                 vm_addr = &vsk->remote_addr;
 906         } else {
 907                 vm_addr = &vsk->local_addr;
 908         }
 909
 910         if (!vm_addr) {
 911                 err = -EINVAL;
 912                 goto out;
 913         }
 914
 915         /* sys_getsockname() and sys_getpeername() pass us a
 916          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 917          * that macro is defined in socket.c instead of .h, so we hardcode its
 918          * value here.
 919          */
 920         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 921         memcpy(addr, vm_addr, sizeof(*vm_addr));
 922         err = sizeof(*vm_addr);
 923
 924 out:
 925         release_sock(sk);
 926         return err;
 927 }
 928
 929 static int vsock_shutdown(struct socket *sock, int mode)
 930 {
 931         int err;
 932         struct sock *sk;
 933
 934         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 935          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 936          * here like the other address families do.  Note also that the
 937          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 938          * which is what we want.
 939          */
 940         mode++;
 941
 942         if ((mode & ~SHUTDOWN_MASK) || !mode)
 943                 return -EINVAL;
 944
 945         /* If this is a STREAM socket and it is not connected then bail out
 946          * immediately.  If it is a DGRAM socket then we must first kick the
 947          * socket so that it wakes up from any sleeping calls, for example
 948          * recv(), and then afterwards return the error.
 949          */
 950
 951         sk = sock->sk;
 952
 953         lock_sock(sk);
 954         if (sock->state == SS_UNCONNECTED) {
 955                 err = -ENOTCONN;
 956                 if (sock_type_connectible(sk->sk_type))
 957                         goto out;
 958         } else {
 959                 sock->state = SS_DISCONNECTING;
 960                 err = 0;
 961         }
 962
 963         /* Receive and send shutdowns are treated alike. */
 964         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 965         if (mode) {
 966                 sk->sk_shutdown |= mode;
 967                 sk->sk_state_change(sk);
 968
 969                 if (sock_type_connectible(sk->sk_type)) {
 970                         sock_reset_flag(sk, SOCK_DONE);
 971                         vsock_send_shutdown(sk, mode);
 972                 }
 973         }
 974
 975 out:
 976         release_sock(sk);
 977         return err;
 978 }
 979
 980 static __poll_t vsock_poll(struct file *file, struct socket *sock,
 981                                poll_table *wait)
 982 {
 983         struct sock *sk;
 984         __poll_t mask;
 985         struct vsock_sock *vsk;
 986
 987         sk = sock->sk;
 988         vsk = vsock_sk(sk);
 989
 990         poll_wait(file, sk_sleep(sk), wait);
 991         mask = 0;
 992
 993         if (sk->sk_err)
 994                 /* Signify that there has been an error on this socket. */
 995                 mask |= EPOLLERR;
 996
 997         /* INET sockets treat local write shutdown and peer write shutdown as a
 998          * case of EPOLLHUP set.
 999          */
1000         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1001             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1002              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1003                 mask |= EPOLLHUP;
1004         }
1005
1006         if (sk->sk_shutdown & RCV_SHUTDOWN ||
1007             vsk->peer_shutdown & SEND_SHUTDOWN) {
1008                 mask |= EPOLLRDHUP;
1009         }
1010
1011         if (sock->type == SOCK_DGRAM) {
1012                 /* For datagram sockets we can read if there is something in
1013                  * the queue and write as long as the socket isn't shutdown for
1014                  * sending.
1015                  */
1016                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1017                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
1018                         mask |= EPOLLIN | EPOLLRDNORM;
1019                 }
1020
1021                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1022                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1023
1024         } else if (sock_type_connectible(sk->sk_type)) {
1025                 const struct vsock_transport *transport;
1026
1027                 lock_sock(sk);
1028
1029                 transport = vsk->transport;
1030
1031                 /* Listening sockets that have connections in their accept
1032                  * queue can be read.
1033                  */
1034                 if (sk->sk_state == TCP_LISTEN
1035                     && !vsock_is_accept_queue_empty(sk))
1036                         mask |= EPOLLIN | EPOLLRDNORM;
1037
1038                 /* If there is something in the queue then we can read. */
1039                 if (transport && transport->stream_is_active(vsk) &&
1040                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1041                         bool data_ready_now = false;
1042                         int ret = transport->notify_poll_in(
1043                                         vsk, 1, &data_ready_now);
1044                         if (ret < 0) {
1045                                 mask |= EPOLLERR;
1046                         } else {
1047                                 if (data_ready_now)
1048                                         mask |= EPOLLIN | EPOLLRDNORM;
1049
1050                         }
1051                 }
1052
1053                 /* Sockets whose connections have been closed, reset, or
1054                  * terminated should also be considered read, and we check the
1055                  * shutdown flag for that.
1056                  */
1057                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1058                     vsk->peer_shutdown & SEND_SHUTDOWN) {
1059                         mask |= EPOLLIN | EPOLLRDNORM;
1060                 }
1061
1062                 /* Connected sockets that can produce data can be written. */
1063                 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1064                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1065                                 bool space_avail_now = false;
1066                                 int ret = transport->notify_poll_out(
1067                                                 vsk, 1, &space_avail_now);
1068                                 if (ret < 0) {
1069                                         mask |= EPOLLERR;
1070                                 } else {
1071                                         if (space_avail_now)
1072                                                 /* Remove EPOLLWRBAND since INET
1073                                                  * sockets are not setting it.
1074                                                  */
1075                                                 mask |= EPOLLOUT | EPOLLWRNORM;
1076
1077                                 }
1078                         }
1079                 }
1080
1081                 /* Simulate INET socket poll behaviors, which sets
1082                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1083                  * but local send is not shutdown.
1084                  */
1085                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1086                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1087                                 mask |= EPOLLOUT | EPOLLWRNORM;
1088
1089                 }
1090
1091                 release_sock(sk);
1092         }
1093
1094         return mask;
1095 }
1096
1097 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1098                                size_t len)
1099 {
1100         int err;
1101         struct sock *sk;
1102         struct vsock_sock *vsk;
1103         struct sockaddr_vm *remote_addr;
1104         const struct vsock_transport *transport;
1105
1106         if (msg->msg_flags & MSG_OOB)
1107                 return -EOPNOTSUPP;
1108
1109         /* For now, MSG_DONTWAIT is always assumed... */
1110         err = 0;
1111         sk = sock->sk;
1112         vsk = vsock_sk(sk);
1113
1114         lock_sock(sk);
1115
1116         transport = vsk->transport;
1117
1118         err = vsock_auto_bind(vsk);
1119         if (err)
1120                 goto out;
1121
1122
1123         /* If the provided message contains an address, use that.  Otherwise
1124          * fall back on the socket's remote handle (if it has been connected).
1125          */
1126         if (msg->msg_name &&
1127             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1128                             &remote_addr) == 0) {
1129                 /* Ensure this address is of the right type and is a valid
1130                  * destination.
1131                  */
1132
1133                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1134                         remote_addr->svm_cid = transport->get_local_cid();
1135
1136                 if (!vsock_addr_bound(remote_addr)) {
1137                         err = -EINVAL;
1138                         goto out;
1139                 }
1140         } else if (sock->state == SS_CONNECTED) {
1141                 remote_addr = &vsk->remote_addr;
1142
1143                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1144                         remote_addr->svm_cid = transport->get_local_cid();
1145
1146                 /* XXX Should connect() or this function ensure remote_addr is
1147                  * bound?
1148                  */
1149                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1150                         err = -EINVAL;
1151                         goto out;
1152                 }
1153         } else {
1154                 err = -EINVAL;
1155                 goto out;
1156         }
1157
1158         if (!transport->dgram_allow(remote_addr->svm_cid,
1159                                     remote_addr->svm_port)) {
1160                 err = -EINVAL;
1161                 goto out;
1162         }
1163
1164         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1165
1166 out:
1167         release_sock(sk);
1168         return err;
1169 }
1170
1171 static int vsock_dgram_connect(struct socket *sock,
1172                                struct sockaddr *addr, int addr_len, int flags)
1173 {
1174         int err;
1175         struct sock *sk;
1176         struct vsock_sock *vsk;
1177         struct sockaddr_vm *remote_addr;
1178
1179         sk = sock->sk;
1180         vsk = vsock_sk(sk);
1181
1182         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1183         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1184                 lock_sock(sk);
1185                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1186                                 VMADDR_PORT_ANY);
1187                 sock->state = SS_UNCONNECTED;
1188                 release_sock(sk);
1189                 return 0;
1190         } else if (err != 0)
1191                 return -EINVAL;
1192
1193         lock_sock(sk);
1194
1195         err = vsock_auto_bind(vsk);
1196         if (err)
1197                 goto out;
1198
1199         if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1200                                          remote_addr->svm_port)) {
1201                 err = -EINVAL;
1202                 goto out;
1203         }
1204
1205         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1206         sock->state = SS_CONNECTED;
1207
1208 out:
1209         release_sock(sk);
1210         return err;
1211 }
1212
1213 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1214                                size_t len, int flags)
1215 {
1216         struct vsock_sock *vsk = vsock_sk(sock->sk);
1217
1218         return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1219 }
1220
1221 static const struct proto_ops vsock_dgram_ops = {
1222         .family = PF_VSOCK,
1223         .owner = THIS_MODULE,
1224         .release = vsock_release,
1225         .bind = vsock_bind,
1226         .connect = vsock_dgram_connect,
1227         .socketpair = sock_no_socketpair,
1228         .accept = sock_no_accept,
1229         .getname = vsock_getname,
1230         .poll = vsock_poll,
1231         .ioctl = sock_no_ioctl,
1232         .listen = sock_no_listen,
1233         .shutdown = vsock_shutdown,
1234         .sendmsg = vsock_dgram_sendmsg,
1235         .recvmsg = vsock_dgram_recvmsg,
1236         .mmap = sock_no_mmap,
1237         .sendpage = sock_no_sendpage,
1238 };
1239
1240 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1241 {
1242         const struct vsock_transport *transport = vsk->transport;
1243
1244         if (!transport || !transport->cancel_pkt)
1245                 return -EOPNOTSUPP;
1246
1247         return transport->cancel_pkt(vsk);
1248 }
1249
1250 static void vsock_connect_timeout(struct work_struct *work)
1251 {
1252         struct sock *sk;
1253         struct vsock_sock *vsk;
1254
1255         vsk = container_of(work, struct vsock_sock, connect_work.work);
1256         sk = sk_vsock(vsk);
1257
1258         lock_sock(sk);
1259         if (sk->sk_state == TCP_SYN_SENT &&
1260             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1261                 sk->sk_state = TCP_CLOSE;
1262                 sk->sk_err = ETIMEDOUT;
1263                 sk->sk_error_report(sk);
1264                 vsock_transport_cancel_pkt(vsk);
1265         }
1266         release_sock(sk);
1267
1268         sock_put(sk);
1269 }
1270
1271 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1272                          int addr_len, int flags)
1273 {
1274         int err;
1275         struct sock *sk;
1276         struct vsock_sock *vsk;
1277         const struct vsock_transport *transport;
1278         struct sockaddr_vm *remote_addr;
1279         long timeout;
1280         DEFINE_WAIT(wait);
1281
1282         err = 0;
1283         sk = sock->sk;
1284         vsk = vsock_sk(sk);
1285
1286         lock_sock(sk);
1287
1288         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1289         switch (sock->state) {
1290         case SS_CONNECTED:
1291                 err = -EISCONN;
1292                 goto out;
1293         case SS_DISCONNECTING:
1294                 err = -EINVAL;
1295                 goto out;
1296         case SS_CONNECTING:
1297                 /* This continues on so we can move sock into the SS_CONNECTED
1298                  * state once the connection has completed (at which point err
1299                  * will be set to zero also).  Otherwise, we will either wait
1300                  * for the connection or return -EALREADY should this be a
1301                  * non-blocking call.
1302                  */
1303                 err = -EALREADY;
1304                 break;
1305         default:
1306                 if ((sk->sk_state == TCP_LISTEN) ||
1307                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1308                         err = -EINVAL;
1309                         goto out;
1310                 }
1311
1312                 /* Set the remote address that we are connecting to. */
1313                 memcpy(&vsk->remote_addr, remote_addr,
1314                        sizeof(vsk->remote_addr));
1315
1316                 err = vsock_assign_transport(vsk, NULL);
1317                 if (err)
1318                         goto out;
1319
1320                 transport = vsk->transport;
1321
1322                 /* The hypervisor and well-known contexts do not have socket
1323                  * endpoints.
1324                  */
1325                 if (!transport ||
1326                     !transport->stream_allow(remote_addr->svm_cid,
1327                                              remote_addr->svm_port)) {
1328                         err = -ENETUNREACH;
1329                         goto out;
1330                 }
1331
1332                 err = vsock_auto_bind(vsk);
1333                 if (err)
1334                         goto out;
1335
1336                 sk->sk_state = TCP_SYN_SENT;
1337
1338                 err = transport->connect(vsk);
1339                 if (err < 0)
1340                         goto out;
1341
1342                 /* Mark sock as connecting and set the error code to in
1343                  * progress in case this is a non-blocking connect.
1344                  */
1345                 sock->state = SS_CONNECTING;
1346                 err = -EINPROGRESS;
1347         }
1348
1349         /* The receive path will handle all communication until we are able to
1350          * enter the connected state.  Here we wait for the connection to be
1351          * completed or a notification of an error.
1352          */
1353         timeout = vsk->connect_timeout;
1354         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1355
1356         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1357                 if (flags & O_NONBLOCK) {
1358                         /* If we're not going to block, we schedule a timeout
1359                          * function to generate a timeout on the connection
1360                          * attempt, in case the peer doesn't respond in a
1361                          * timely manner. We hold on to the socket until the
1362                          * timeout fires.
1363                          */
1364                         sock_hold(sk);
1365                         schedule_delayed_work(&vsk->connect_work, timeout);
1366
1367                         /* Skip ahead to preserve error code set above. */
1368                         goto out_wait;
1369                 }
1370
1371                 release_sock(sk);
1372                 timeout = schedule_timeout(timeout);
1373                 lock_sock(sk);
1374
1375                 if (signal_pending(current)) {
1376                         err = sock_intr_errno(timeout);
1377                         sk->sk_state = TCP_CLOSE;
1378                         sock->state = SS_UNCONNECTED;
1379                         vsock_transport_cancel_pkt(vsk);
1380                         goto out_wait;
1381                 } else if (timeout == 0) {
1382                         err = -ETIMEDOUT;
1383                         sk->sk_state = TCP_CLOSE;
1384                         sock->state = SS_UNCONNECTED;
1385                         vsock_transport_cancel_pkt(vsk);
1386                         goto out_wait;
1387                 }
1388
1389                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1390         }
1391
1392         if (sk->sk_err) {
1393                 err = -sk->sk_err;
1394                 sk->sk_state = TCP_CLOSE;
1395                 sock->state = SS_UNCONNECTED;
1396         } else {
1397                 err = 0;
1398         }
1399
1400 out_wait:
1401         finish_wait(sk_sleep(sk), &wait);
1402 out:
1403         release_sock(sk);
1404         return err;
1405 }
1406
1407 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1408                         bool kern)
1409 {
1410         struct sock *listener;
1411         int err;
1412         struct sock *connected;
1413         struct vsock_sock *vconnected;
1414         long timeout;
1415         DEFINE_WAIT(wait);
1416
1417         err = 0;
1418         listener = sock->sk;
1419
1420         lock_sock(listener);
1421
1422         if (!sock_type_connectible(sock->type)) {
1423                 err = -EOPNOTSUPP;
1424                 goto out;
1425         }
1426
1427         if (listener->sk_state != TCP_LISTEN) {
1428                 err = -EINVAL;
1429                 goto out;
1430         }
1431
1432         /* Wait for children sockets to appear; these are the new sockets
1433          * created upon connection establishment.
1434          */
1435         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1436         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1437
1438         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1439                listener->sk_err == 0) {
1440                 release_sock(listener);
1441                 timeout = schedule_timeout(timeout);
1442                 finish_wait(sk_sleep(listener), &wait);
1443                 lock_sock(listener);
1444
1445                 if (signal_pending(current)) {
1446                         err = sock_intr_errno(timeout);
1447                         goto out;
1448                 } else if (timeout == 0) {
1449                         err = -EAGAIN;
1450                         goto out;
1451                 }
1452
1453                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1454         }
1455         finish_wait(sk_sleep(listener), &wait);
1456
1457         if (listener->sk_err)
1458                 err = -listener->sk_err;
1459
1460         if (connected) {
1461                 sk_acceptq_removed(listener);
1462
1463                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1464                 vconnected = vsock_sk(connected);
1465
1466                 /* If the listener socket has received an error, then we should
1467                  * reject this socket and return.  Note that we simply mark the
1468                  * socket rejected, drop our reference, and let the cleanup
1469                  * function handle the cleanup; the fact that we found it in
1470                  * the listener's accept queue guarantees that the cleanup
1471                  * function hasn't run yet.
1472                  */
1473                 if (err) {
1474                         vconnected->rejected = true;
1475                 } else {
1476                         newsock->state = SS_CONNECTED;
1477                         sock_graft(connected, newsock);
1478                 }
1479
1480                 release_sock(connected);
1481                 sock_put(connected);
1482         }
1483
1484 out:
1485         release_sock(listener);
1486         return err;
1487 }
1488
1489 static int vsock_listen(struct socket *sock, int backlog)
1490 {
1491         int err;
1492         struct sock *sk;
1493         struct vsock_sock *vsk;
1494
1495         sk = sock->sk;
1496
1497         lock_sock(sk);
1498
1499         if (!sock_type_connectible(sk->sk_type)) {
1500                 err = -EOPNOTSUPP;
1501                 goto out;
1502         }
1503
1504         if (sock->state != SS_UNCONNECTED) {
1505                 err = -EINVAL;
1506                 goto out;
1507         }
1508
1509         vsk = vsock_sk(sk);
1510
1511         if (!vsock_addr_bound(&vsk->local_addr)) {
1512                 err = -EINVAL;
1513                 goto out;
1514         }
1515
1516         sk->sk_max_ack_backlog = backlog;
1517         sk->sk_state = TCP_LISTEN;
1518
1519         err = 0;
1520
1521 out:
1522         release_sock(sk);
1523         return err;
1524 }
1525
1526 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1527                                      const struct vsock_transport *transport,
1528                                      u64 val)
1529 {
1530         if (val > vsk->buffer_max_size)
1531                 val = vsk->buffer_max_size;
1532
1533         if (val < vsk->buffer_min_size)
1534                 val = vsk->buffer_min_size;
1535
1536         if (val != vsk->buffer_size &&
1537             transport && transport->notify_buffer_size)
1538                 transport->notify_buffer_size(vsk, &val);
1539
1540         vsk->buffer_size = val;
1541 }
1542
1543 static int vsock_connectible_setsockopt(struct socket *sock,
1544                                         int level,
1545                                         int optname,
1546                                         sockptr_t optval,
1547                                         unsigned int optlen)
1548 {
1549         int err;
1550         struct sock *sk;
1551         struct vsock_sock *vsk;
1552         const struct vsock_transport *transport;
1553         u64 val;
1554
1555         if (level != AF_VSOCK)
1556                 return -ENOPROTOOPT;
1557
1558 #define COPY_IN(_v)                                       \
1559         do {                                              \
1560                 if (optlen < sizeof(_v)) {                \
1561                         err = -EINVAL;                    \
1562                         goto exit;                        \
1563                 }                                         \
1564                 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) {  \
1565                         err = -EFAULT;                                  \
1566                         goto exit;                                      \
1567                 }                                                       \
1568         } while (0)
1569
1570         err = 0;
1571         sk = sock->sk;
1572         vsk = vsock_sk(sk);
1573
1574         lock_sock(sk);
1575
1576         transport = vsk->transport;
1577
1578         switch (optname) {
1579         case SO_VM_SOCKETS_BUFFER_SIZE:
1580                 COPY_IN(val);
1581                 vsock_update_buffer_size(vsk, transport, val);
1582                 break;
1583
1584         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1585                 COPY_IN(val);
1586                 vsk->buffer_max_size = val;
1587                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1588                 break;
1589
1590         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1591                 COPY_IN(val);
1592                 vsk->buffer_min_size = val;
1593                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1594                 break;
1595
1596         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1597                 struct __kernel_old_timeval tv;
1598                 COPY_IN(tv);
1599                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1600                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1601                         vsk->connect_timeout = tv.tv_sec * HZ +
1602                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1603                         if (vsk->connect_timeout == 0)
1604                                 vsk->connect_timeout =
1605                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1606
1607                 } else {
1608                         err = -ERANGE;
1609                 }
1610                 break;
1611         }
1612
1613         default:
1614                 err = -ENOPROTOOPT;
1615                 break;
1616         }
1617
1618 #undef COPY_IN
1619
1620 exit:
1621         release_sock(sk);
1622         return err;
1623 }
1624
1625 static int vsock_connectible_getsockopt(struct socket *sock,
1626                                         int level, int optname,
1627                                         char __user *optval,
1628                                         int __user *optlen)
1629 {
1630         int err;
1631         int len;
1632         struct sock *sk;
1633         struct vsock_sock *vsk;
1634         u64 val;
1635
1636         if (level != AF_VSOCK)
1637                 return -ENOPROTOOPT;
1638
1639         err = get_user(len, optlen);
1640         if (err != 0)
1641                 return err;
1642
1643 #define COPY_OUT(_v)                            \
1644         do {                                    \
1645                 if (len < sizeof(_v))           \
1646                         return -EINVAL;         \
1647                                                 \
1648                 len = sizeof(_v);               \
1649                 if (copy_to_user(optval, &_v, len) != 0)        \
1650                         return -EFAULT;                         \
1651                                                                 \
1652         } while (0)
1653
1654         err = 0;
1655         sk = sock->sk;
1656         vsk = vsock_sk(sk);
1657
1658         switch (optname) {
1659         case SO_VM_SOCKETS_BUFFER_SIZE:
1660                 val = vsk->buffer_size;
1661                 COPY_OUT(val);
1662                 break;
1663
1664         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1665                 val = vsk->buffer_max_size;
1666                 COPY_OUT(val);
1667                 break;
1668
1669         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1670                 val = vsk->buffer_min_size;
1671                 COPY_OUT(val);
1672                 break;
1673
1674         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1675                 struct __kernel_old_timeval tv;
1676                 tv.tv_sec = vsk->connect_timeout / HZ;
1677                 tv.tv_usec =
1678                     (vsk->connect_timeout -
1679                      tv.tv_sec * HZ) * (1000000 / HZ);
1680                 COPY_OUT(tv);
1681                 break;
1682         }
1683         default:
1684                 return -ENOPROTOOPT;
1685         }
1686
1687         err = put_user(len, optlen);
1688         if (err != 0)
1689                 return -EFAULT;
1690
1691 #undef COPY_OUT
1692
1693         return 0;
1694 }
1695
1696 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1697                                      size_t len)
1698 {
1699         struct sock *sk;
1700         struct vsock_sock *vsk;
1701         const struct vsock_transport *transport;
1702         ssize_t total_written;
1703         long timeout;
1704         int err;
1705         struct vsock_transport_send_notify_data send_data;
1706         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1707
1708         sk = sock->sk;
1709         vsk = vsock_sk(sk);
1710         total_written = 0;
1711         err = 0;
1712
1713         if (msg->msg_flags & MSG_OOB)
1714                 return -EOPNOTSUPP;
1715
1716         lock_sock(sk);
1717
1718         transport = vsk->transport;
1719
1720         /* Callers should not provide a destination with stream sockets. */
1721         if (msg->msg_namelen) {
1722                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1723                 goto out;
1724         }
1725
1726         /* Send data only if both sides are not shutdown in the direction. */
1727         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1728             vsk->peer_shutdown & RCV_SHUTDOWN) {
1729                 err = -EPIPE;
1730                 goto out;
1731         }
1732
1733         if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1734             !vsock_addr_bound(&vsk->local_addr)) {
1735                 err = -ENOTCONN;
1736                 goto out;
1737         }
1738
1739         if (!vsock_addr_bound(&vsk->remote_addr)) {
1740                 err = -EDESTADDRREQ;
1741                 goto out;
1742         }
1743
1744         /* Wait for room in the produce queue to enqueue our user's data. */
1745         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1746
1747         err = transport->notify_send_init(vsk, &send_data);
1748         if (err < 0)
1749                 goto out;
1750
1751         while (total_written < len) {
1752                 ssize_t written;
1753
1754                 add_wait_queue(sk_sleep(sk), &wait);
1755                 while (vsock_stream_has_space(vsk) == 0 &&
1756                        sk->sk_err == 0 &&
1757                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1758                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1759
1760                         /* Don't wait for non-blocking sockets. */
1761                         if (timeout == 0) {
1762                                 err = -EAGAIN;
1763                                 remove_wait_queue(sk_sleep(sk), &wait);
1764                                 goto out_err;
1765                         }
1766
1767                         err = transport->notify_send_pre_block(vsk, &send_data);
1768                         if (err < 0) {
1769                                 remove_wait_queue(sk_sleep(sk), &wait);
1770                                 goto out_err;
1771                         }
1772
1773                         release_sock(sk);
1774                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1775                         lock_sock(sk);
1776                         if (signal_pending(current)) {
1777                                 err = sock_intr_errno(timeout);
1778                                 remove_wait_queue(sk_sleep(sk), &wait);
1779                                 goto out_err;
1780                         } else if (timeout == 0) {
1781                                 err = -EAGAIN;
1782                                 remove_wait_queue(sk_sleep(sk), &wait);
1783                                 goto out_err;
1784                         }
1785                 }
1786                 remove_wait_queue(sk_sleep(sk), &wait);
1787
1788                 /* These checks occur both as part of and after the loop
1789                  * conditional since we need to check before and after
1790                  * sleeping.
1791                  */
1792                 if (sk->sk_err) {
1793                         err = -sk->sk_err;
1794                         goto out_err;
1795                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1796                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1797                         err = -EPIPE;
1798                         goto out_err;
1799                 }
1800
1801                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1802                 if (err < 0)
1803                         goto out_err;
1804
1805                 /* Note that enqueue will only write as many bytes as are free
1806                  * in the produce queue, so we don't need to ensure len is
1807                  * smaller than the queue size.  It is the caller's
1808                  * responsibility to check how many bytes we were able to send.
1809                  */
1810
1811                 if (sk->sk_type == SOCK_SEQPACKET) {
1812                         written = transport->seqpacket_enqueue(vsk,
1813                                                 msg, len - total_written);
1814                 } else {
1815                         written = transport->stream_enqueue(vsk,
1816                                         msg, len - total_written);
1817                 }
1818                 if (written < 0) {
1819                         err = -ENOMEM;
1820                         goto out_err;
1821                 }
1822
1823                 total_written += written;
1824
1825                 err = transport->notify_send_post_enqueue(
1826                                 vsk, written, &send_data);
1827                 if (err < 0)
1828                         goto out_err;
1829
1830         }
1831
1832 out_err:
1833         if (total_written > 0) {
1834                 /* Return number of written bytes only if:
1835                  * 1) SOCK_STREAM socket.
1836                  * 2) SOCK_SEQPACKET socket when whole buffer is sent.
1837                  */
1838                 if (sk->sk_type == SOCK_STREAM || total_written == len)
1839                         err = total_written;
1840         }
1841 out:
1842         release_sock(sk);
1843         return err;
1844 }
1845
1846 static int vsock_wait_data(struct sock *sk, struct wait_queue_entry *wait,
1847                            long timeout,
1848                            struct vsock_transport_recv_notify_data *recv_data,
1849                            size_t target)
1850 {
1851         const struct vsock_transport *transport;
1852         struct vsock_sock *vsk;
1853         s64 data;
1854         int err;
1855
1856         vsk = vsock_sk(sk);
1857         err = 0;
1858         transport = vsk->transport;
1859
1860         while ((data = vsock_stream_has_data(vsk)) == 0) {
1861                 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
1862
1863                 if (sk->sk_err != 0 ||
1864                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1865                     (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1866                         break;
1867                 }
1868
1869                 /* Don't wait for non-blocking sockets. */
1870                 if (timeout == 0) {
1871                         err = -EAGAIN;
1872                         break;
1873                 }
1874
1875                 if (recv_data) {
1876                         err = transport->notify_recv_pre_block(vsk, target, recv_data);
1877                         if (err < 0)
1878                                 break;
1879                 }
1880
1881                 release_sock(sk);
1882                 timeout = schedule_timeout(timeout);
1883                 lock_sock(sk);
1884
1885                 if (signal_pending(current)) {
1886                         err = sock_intr_errno(timeout);
1887                         break;
1888                 } else if (timeout == 0) {
1889                         err = -EAGAIN;
1890                         break;
1891                 }
1892         }
1893
1894         finish_wait(sk_sleep(sk), wait);
1895
1896         if (err)
1897                 return err;
1898
1899         /* Internal transport error when checking for available
1900          * data. XXX This should be changed to a connection
1901          * reset in a later change.
1902          */
1903         if (data < 0)
1904                 return -ENOMEM;
1905
1906         return data;
1907 }
1908
1909 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
1910                                   size_t len, int flags)
1911 {
1912         struct vsock_transport_recv_notify_data recv_data;
1913         const struct vsock_transport *transport;
1914         struct vsock_sock *vsk;
1915         ssize_t copied;
1916         size_t target;
1917         long timeout;
1918         int err;
1919
1920         DEFINE_WAIT(wait);
1921
1922         vsk = vsock_sk(sk);
1923         transport = vsk->transport;
1924
1925         /* We must not copy less than target bytes into the user's buffer
1926          * before returning successfully, so we wait for the consume queue to
1927          * have that much data to consume before dequeueing.  Note that this
1928          * makes it impossible to handle cases where target is greater than the
1929          * queue size.
1930          */
1931         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1932         if (target >= transport->stream_rcvhiwat(vsk)) {
1933                 err = -ENOMEM;
1934                 goto out;
1935         }
1936         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1937         copied = 0;
1938
1939         err = transport->notify_recv_init(vsk, target, &recv_data);
1940         if (err < 0)
1941                 goto out;
1942
1943
1944         while (1) {
1945                 ssize_t read;
1946
1947                 err = vsock_wait_data(sk, &wait, timeout, &recv_data, target);
1948                 if (err <= 0)
1949                         break;
1950
1951                 err = transport->notify_recv_pre_dequeue(vsk, target,
1952                                                          &recv_data);
1953                 if (err < 0)
1954                         break;
1955
1956                 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
1957                 if (read < 0) {
1958                         err = -ENOMEM;
1959                         break;
1960                 }
1961
1962                 copied += read;
1963
1964                 err = transport->notify_recv_post_dequeue(vsk, target, read,
1965                                                 !(flags & MSG_PEEK), &recv_data);
1966                 if (err < 0)
1967                         goto out;
1968
1969                 if (read >= target || flags & MSG_PEEK)
1970                         break;
1971
1972                 target -= read;
1973         }
1974
1975         if (sk->sk_err)
1976                 err = -sk->sk_err;
1977         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1978                 err = 0;
1979
1980         if (copied > 0)
1981                 err = copied;
1982
1983 out:
1984         return err;
1985 }
1986
1987 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
1988                                      size_t len, int flags)
1989 {
1990         const struct vsock_transport *transport;
1991         struct vsock_sock *vsk;
1992         ssize_t record_len;
1993         long timeout;
1994         int err = 0;
1995         DEFINE_WAIT(wait);
1996
1997         vsk = vsock_sk(sk);
1998         transport = vsk->transport;
1999
2000         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2001
2002         err = vsock_wait_data(sk, &wait, timeout, NULL, 0);
2003         if (err <= 0)
2004                 goto out;
2005
2006         record_len = transport->seqpacket_dequeue(vsk, msg, flags);
2007
2008         if (record_len < 0) {
2009                 err = -ENOMEM;
2010                 goto out;
2011         }
2012
2013         if (sk->sk_err) {
2014                 err = -sk->sk_err;
2015         } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2016                 err = 0;
2017         } else {
2018                 /* User sets MSG_TRUNC, so return real length of
2019                  * packet.
2020                  */
2021                 if (flags & MSG_TRUNC)
2022                         err = record_len;
2023                 else
2024                         err = len - msg_data_left(msg);
2025
2026                 /* Always set MSG_TRUNC if real length of packet is
2027                  * bigger than user's buffer.
2028                  */
2029                 if (record_len > len)
2030                         msg->msg_flags |= MSG_TRUNC;
2031         }
2032
2033 out:
2034         return err;
2035 }
2036
2037 static int
2038 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2039                           int flags)
2040 {
2041         struct sock *sk;
2042         struct vsock_sock *vsk;
2043         const struct vsock_transport *transport;
2044         int err;
2045
2046         DEFINE_WAIT(wait);
2047
2048         sk = sock->sk;
2049         vsk = vsock_sk(sk);
2050         err = 0;
2051
2052         lock_sock(sk);
2053
2054         transport = vsk->transport;
2055
2056         if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2057                 /* Recvmsg is supposed to return 0 if a peer performs an
2058                  * orderly shutdown. Differentiate between that case and when a
2059                  * peer has not connected or a local shutdown occurred with the
2060                  * SOCK_DONE flag.
2061                  */
2062                 if (sock_flag(sk, SOCK_DONE))
2063                         err = 0;
2064                 else
2065                         err = -ENOTCONN;
2066
2067                 goto out;
2068         }
2069
2070         if (flags & MSG_OOB) {
2071                 err = -EOPNOTSUPP;
2072                 goto out;
2073         }
2074
2075         /* We don't check peer_shutdown flag here since peer may actually shut
2076          * down, but there can be data in the queue that a local socket can
2077          * receive.
2078          */
2079         if (sk->sk_shutdown & RCV_SHUTDOWN) {
2080                 err = 0;
2081                 goto out;
2082         }
2083
2084         /* It is valid on Linux to pass in a zero-length receive buffer.  This
2085          * is not an error.  We may as well bail out now.
2086          */
2087         if (!len) {
2088                 err = 0;
2089                 goto out;
2090         }
2091
2092         if (sk->sk_type == SOCK_STREAM)
2093                 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2094         else
2095                 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2096
2097 out:
2098         release_sock(sk);
2099         return err;
2100 }
2101
2102 static const struct proto_ops vsock_stream_ops = {
2103         .family = PF_VSOCK,
2104         .owner = THIS_MODULE,
2105         .release = vsock_release,
2106         .bind = vsock_bind,
2107         .connect = vsock_connect,
2108         .socketpair = sock_no_socketpair,
2109         .accept = vsock_accept,
2110         .getname = vsock_getname,
2111         .poll = vsock_poll,
2112         .ioctl = sock_no_ioctl,
2113         .listen = vsock_listen,
2114         .shutdown = vsock_shutdown,
2115         .setsockopt = vsock_connectible_setsockopt,
2116         .getsockopt = vsock_connectible_getsockopt,
2117         .sendmsg = vsock_connectible_sendmsg,
2118         .recvmsg = vsock_connectible_recvmsg,
2119         .mmap = sock_no_mmap,
2120         .sendpage = sock_no_sendpage,
2121 };
2122
2123 static int vsock_create(struct net *net, struct socket *sock,
2124                         int protocol, int kern)
2125 {
2126         struct vsock_sock *vsk;
2127         struct sock *sk;
2128         int ret;
2129
2130         if (!sock)
2131                 return -EINVAL;
2132
2133         if (protocol && protocol != PF_VSOCK)
2134                 return -EPROTONOSUPPORT;
2135
2136         switch (sock->type) {
2137         case SOCK_DGRAM:
2138                 sock->ops = &vsock_dgram_ops;
2139                 break;
2140         case SOCK_STREAM:
2141                 sock->ops = &vsock_stream_ops;
2142                 break;
2143         default:
2144                 return -ESOCKTNOSUPPORT;
2145         }
2146
2147         sock->state = SS_UNCONNECTED;
2148
2149         sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2150         if (!sk)
2151                 return -ENOMEM;
2152
2153         vsk = vsock_sk(sk);
2154
2155         if (sock->type == SOCK_DGRAM) {
2156                 ret = vsock_assign_transport(vsk, NULL);
2157                 if (ret < 0) {
2158                         sock_put(sk);
2159                         return ret;
2160                 }
2161         }
2162
2163         vsock_insert_unbound(vsk);
2164
2165         return 0;
2166 }
2167
2168 static const struct net_proto_family vsock_family_ops = {
2169         .family = AF_VSOCK,
2170         .create = vsock_create,
2171         .owner = THIS_MODULE,
2172 };
2173
2174 static long vsock_dev_do_ioctl(struct file *filp,
2175                                unsigned int cmd, void __user *ptr)
2176 {
2177         u32 __user *p = ptr;
2178         u32 cid = VMADDR_CID_ANY;
2179         int retval = 0;
2180
2181         switch (cmd) {
2182         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2183                 /* To be compatible with the VMCI behavior, we prioritize the
2184                  * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2185                  */
2186                 if (transport_g2h)
2187                         cid = transport_g2h->get_local_cid();
2188                 else if (transport_h2g)
2189                         cid = transport_h2g->get_local_cid();
2190
2191                 if (put_user(cid, p) != 0)
2192                         retval = -EFAULT;
2193                 break;
2194
2195         default:
2196                 retval = -ENOIOCTLCMD;
2197         }
2198
2199         return retval;
2200 }
2201
2202 static long vsock_dev_ioctl(struct file *filp,
2203                             unsigned int cmd, unsigned long arg)
2204 {
2205         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2206 }
2207
2208 #ifdef CONFIG_COMPAT
2209 static long vsock_dev_compat_ioctl(struct file *filp,
2210                                    unsigned int cmd, unsigned long arg)
2211 {
2212         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2213 }
2214 #endif
2215
2216 static const struct file_operations vsock_device_ops = {
2217         .owner          = THIS_MODULE,
2218         .unlocked_ioctl = vsock_dev_ioctl,
2219 #ifdef CONFIG_COMPAT
2220         .compat_ioctl   = vsock_dev_compat_ioctl,
2221 #endif
2222         .open           = nonseekable_open,
2223 };
2224
2225 static struct miscdevice vsock_device = {
2226         .name           = "vsock",
2227         .fops           = &vsock_device_ops,
2228 };
2229
2230 static int __init vsock_init(void)
2231 {
2232         int err = 0;
2233
2234         vsock_init_tables();
2235
2236         vsock_proto.owner = THIS_MODULE;
2237         vsock_device.minor = MISC_DYNAMIC_MINOR;
2238         err = misc_register(&vsock_device);
2239         if (err) {
2240                 pr_err("Failed to register misc device\n");
2241                 goto err_reset_transport;
2242         }
2243
2244         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2245         if (err) {
2246                 pr_err("Cannot register vsock protocol\n");
2247                 goto err_deregister_misc;
2248         }
2249
2250         err = sock_register(&vsock_family_ops);
2251         if (err) {
2252                 pr_err("could not register af_vsock (%d) address family: %d\n",
2253                        AF_VSOCK, err);
2254                 goto err_unregister_proto;
2255         }
2256
2257         return 0;
2258
2259 err_unregister_proto:
2260         proto_unregister(&vsock_proto);
2261 err_deregister_misc:
2262         misc_deregister(&vsock_device);
2263 err_reset_transport:
2264         return err;
2265 }
2266
2267 static void __exit vsock_exit(void)
2268 {
2269         misc_deregister(&vsock_device);
2270         sock_unregister(AF_VSOCK);
2271         proto_unregister(&vsock_proto);
2272 }
2273
2274 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2275 {
2276         return vsk->transport;
2277 }
2278 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2279
2280 int vsock_core_register(const struct vsock_transport *t, int features)
2281 {
2282         const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2283         int err = mutex_lock_interruptible(&vsock_register_mutex);
2284
2285         if (err)
2286                 return err;
2287
2288         t_h2g = transport_h2g;
2289         t_g2h = transport_g2h;
2290         t_dgram = transport_dgram;
2291         t_local = transport_local;
2292
2293         if (features & VSOCK_TRANSPORT_F_H2G) {
2294                 if (t_h2g) {
2295                         err = -EBUSY;
2296                         goto err_busy;
2297                 }
2298                 t_h2g = t;
2299         }
2300
2301         if (features & VSOCK_TRANSPORT_F_G2H) {
2302                 if (t_g2h) {
2303                         err = -EBUSY;
2304                         goto err_busy;
2305                 }
2306                 t_g2h = t;
2307         }
2308
2309         if (features & VSOCK_TRANSPORT_F_DGRAM) {
2310                 if (t_dgram) {
2311                         err = -EBUSY;
2312                         goto err_busy;
2313                 }
2314                 t_dgram = t;
2315         }
2316
2317         if (features & VSOCK_TRANSPORT_F_LOCAL) {
2318                 if (t_local) {
2319                         err = -EBUSY;
2320                         goto err_busy;
2321                 }
2322                 t_local = t;
2323         }
2324
2325         transport_h2g = t_h2g;
2326         transport_g2h = t_g2h;
2327         transport_dgram = t_dgram;
2328         transport_local = t_local;
2329
2330 err_busy:
2331         mutex_unlock(&vsock_register_mutex);
2332         return err;
2333 }
2334 EXPORT_SYMBOL_GPL(vsock_core_register);
2335
2336 void vsock_core_unregister(const struct vsock_transport *t)
2337 {
2338         mutex_lock(&vsock_register_mutex);
2339
2340         if (transport_h2g == t)
2341                 transport_h2g = NULL;
2342
2343         if (transport_g2h == t)
2344                 transport_g2h = NULL;
2345
2346         if (transport_dgram == t)
2347                 transport_dgram = NULL;
2348
2349         if (transport_local == t)
2350                 transport_local = NULL;
2351
2352         mutex_unlock(&vsock_register_mutex);
2353 }
2354 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2355
2356 module_init(vsock_init);
2357 module_exit(vsock_exit);
2358
2359 MODULE_AUTHOR("VMware, Inc.");
2360 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2361 MODULE_VERSION("1.0.2.0-k");
2362 MODULE_LICENSE("GPL v2");