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(struct vsock_transport *transport,
 337                                      void (*fn)(struct sock *sk))
 338 {
 339         int i;
 340
 341         spin_lock_bh(&vsock_table_lock);
 342
 343         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 344                 struct vsock_sock *vsk;
 345                 list_for_each_entry(vsk, &vsock_connected_table[i],
 346                                     connected_table) {
 347                         if (vsk->transport != transport)
 348                                 continue;
 349
 350                         fn(sk_vsock(vsk));
 351                 }
 352         }
 353
 354         spin_unlock_bh(&vsock_table_lock);
 355 }
 356 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 357
 358 void vsock_add_pending(struct sock *listener, struct sock *pending)
 359 {
 360         struct vsock_sock *vlistener;
 361         struct vsock_sock *vpending;
 362
 363         vlistener = vsock_sk(listener);
 364         vpending = vsock_sk(pending);
 365
 366         sock_hold(pending);
 367         sock_hold(listener);
 368         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 369 }
 370 EXPORT_SYMBOL_GPL(vsock_add_pending);
 371
 372 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 373 {
 374         struct vsock_sock *vpending = vsock_sk(pending);
 375
 376         list_del_init(&vpending->pending_links);
 377         sock_put(listener);
 378         sock_put(pending);
 379 }
 380 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 381
 382 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 383 {
 384         struct vsock_sock *vlistener;
 385         struct vsock_sock *vconnected;
 386
 387         vlistener = vsock_sk(listener);
 388         vconnected = vsock_sk(connected);
 389
 390         sock_hold(connected);
 391         sock_hold(listener);
 392         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 393 }
 394 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 395
 396 static bool vsock_use_local_transport(unsigned int remote_cid)
 397 {
 398         if (!transport_local)
 399                 return false;
 400
 401         if (remote_cid == VMADDR_CID_LOCAL)
 402                 return true;
 403
 404         if (transport_g2h) {
 405                 return remote_cid == transport_g2h->get_local_cid();
 406         } else {
 407                 return remote_cid == VMADDR_CID_HOST;
 408         }
 409 }
 410
 411 static void vsock_deassign_transport(struct vsock_sock *vsk)
 412 {
 413         if (!vsk->transport)
 414                 return;
 415
 416         vsk->transport->destruct(vsk);
 417         module_put(vsk->transport->module);
 418         vsk->transport = NULL;
 419 }
 420
 421 /* Assign a transport to a socket and call the .init transport callback.
 422  *
 423  * Note: for connection oriented socket this must be called when vsk->remote_addr
 424  * is set (e.g. during the connect() or when a connection request on a listener
 425  * socket is received).
 426  * The vsk->remote_addr is used to decide which transport to use:
 427  *  - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
 428  *    g2h is not loaded, will use local transport;
 429  *  - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
 430  *    includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
 431  *  - remote CID > VMADDR_CID_HOST will use host->guest transport;
 432  */
 433 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
 434 {
 435         const struct vsock_transport *new_transport;
 436         struct sock *sk = sk_vsock(vsk);
 437         unsigned int remote_cid = vsk->remote_addr.svm_cid;
 438         __u8 remote_flags;
 439         int ret;
 440
 441         /* If the packet is coming with the source and destination CIDs higher
 442          * than VMADDR_CID_HOST, then a vsock channel where all the packets are
 443          * forwarded to the host should be established. Then the host will
 444          * need to forward the packets to the guest.
 445          *
 446          * The flag is set on the (listen) receive path (psk is not NULL). On
 447          * the connect path the flag can be set by the user space application.
 448          */
 449         if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
 450             vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
 451                 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
 452
 453         remote_flags = vsk->remote_addr.svm_flags;
 454
 455         switch (sk->sk_type) {
 456         case SOCK_DGRAM:
 457                 new_transport = transport_dgram;
 458                 break;
 459         case SOCK_STREAM:
 460         case SOCK_SEQPACKET:
 461                 if (vsock_use_local_transport(remote_cid))
 462                         new_transport = transport_local;
 463                 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
 464                          (remote_flags & VMADDR_FLAG_TO_HOST))
 465                         new_transport = transport_g2h;
 466                 else
 467                         new_transport = transport_h2g;
 468                 break;
 469         default:
 470                 return -ESOCKTNOSUPPORT;
 471         }
 472
 473         if (vsk->transport) {
 474                 if (vsk->transport == new_transport)
 475                         return 0;
 476
 477                 /* transport->release() must be called with sock lock acquired.
 478                  * This path can only be taken during vsock_connect(), where we
 479                  * have already held the sock lock. In the other cases, this
 480                  * function is called on a new socket which is not assigned to
 481                  * any transport.
 482                  */
 483                 vsk->transport->release(vsk);
 484                 vsock_deassign_transport(vsk);
 485         }
 486
 487         /* We increase the module refcnt to prevent the transport unloading
 488          * while there are open sockets assigned to it.
 489          */
 490         if (!new_transport || !try_module_get(new_transport->module))
 491                 return -ENODEV;
 492
 493         if (sk->sk_type == SOCK_SEQPACKET) {
 494                 if (!new_transport->seqpacket_allow ||
 495                     !new_transport->seqpacket_allow(remote_cid)) {
 496                         module_put(new_transport->module);
 497                         return -ESOCKTNOSUPPORT;
 498                 }
 499         }
 500
 501         ret = new_transport->init(vsk, psk);
 502         if (ret) {
 503                 module_put(new_transport->module);
 504                 return ret;
 505         }
 506
 507         vsk->transport = new_transport;
 508
 509         return 0;
 510 }
 511 EXPORT_SYMBOL_GPL(vsock_assign_transport);
 512
 513 bool vsock_find_cid(unsigned int cid)
 514 {
 515         if (transport_g2h && cid == transport_g2h->get_local_cid())
 516                 return true;
 517
 518         if (transport_h2g && cid == VMADDR_CID_HOST)
 519                 return true;
 520
 521         if (transport_local && cid == VMADDR_CID_LOCAL)
 522                 return true;
 523
 524         return false;
 525 }
 526 EXPORT_SYMBOL_GPL(vsock_find_cid);
 527
 528 static struct sock *vsock_dequeue_accept(struct sock *listener)
 529 {
 530         struct vsock_sock *vlistener;
 531         struct vsock_sock *vconnected;
 532
 533         vlistener = vsock_sk(listener);
 534
 535         if (list_empty(&vlistener->accept_queue))
 536                 return NULL;
 537
 538         vconnected = list_entry(vlistener->accept_queue.next,
 539                                 struct vsock_sock, accept_queue);
 540
 541         list_del_init(&vconnected->accept_queue);
 542         sock_put(listener);
 543         /* The caller will need a reference on the connected socket so we let
 544          * it call sock_put().
 545          */
 546
 547         return sk_vsock(vconnected);
 548 }
 549
 550 static bool vsock_is_accept_queue_empty(struct sock *sk)
 551 {
 552         struct vsock_sock *vsk = vsock_sk(sk);
 553         return list_empty(&vsk->accept_queue);
 554 }
 555
 556 static bool vsock_is_pending(struct sock *sk)
 557 {
 558         struct vsock_sock *vsk = vsock_sk(sk);
 559         return !list_empty(&vsk->pending_links);
 560 }
 561
 562 static int vsock_send_shutdown(struct sock *sk, int mode)
 563 {
 564         struct vsock_sock *vsk = vsock_sk(sk);
 565
 566         if (!vsk->transport)
 567                 return -ENODEV;
 568
 569         return vsk->transport->shutdown(vsk, mode);
 570 }
 571
 572 static void vsock_pending_work(struct work_struct *work)
 573 {
 574         struct sock *sk;
 575         struct sock *listener;
 576         struct vsock_sock *vsk;
 577         bool cleanup;
 578
 579         vsk = container_of(work, struct vsock_sock, pending_work.work);
 580         sk = sk_vsock(vsk);
 581         listener = vsk->listener;
 582         cleanup = true;
 583
 584         lock_sock(listener);
 585         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 586
 587         if (vsock_is_pending(sk)) {
 588                 vsock_remove_pending(listener, sk);
 589
 590                 sk_acceptq_removed(listener);
 591         } else if (!vsk->rejected) {
 592                 /* We are not on the pending list and accept() did not reject
 593                  * us, so we must have been accepted by our user process.  We
 594                  * just need to drop our references to the sockets and be on
 595                  * our way.
 596                  */
 597                 cleanup = false;
 598                 goto out;
 599         }
 600
 601         /* We need to remove ourself from the global connected sockets list so
 602          * incoming packets can't find this socket, and to reduce the reference
 603          * count.
 604          */
 605         vsock_remove_connected(vsk);
 606
 607         sk->sk_state = TCP_CLOSE;
 608
 609 out:
 610         release_sock(sk);
 611         release_sock(listener);
 612         if (cleanup)
 613                 sock_put(sk);
 614
 615         sock_put(sk);
 616         sock_put(listener);
 617 }
 618
 619 /**** SOCKET OPERATIONS ****/
 620
 621 static int __vsock_bind_connectible(struct vsock_sock *vsk,
 622                                     struct sockaddr_vm *addr)
 623 {
 624         static u32 port;
 625         struct sockaddr_vm new_addr;
 626
 627         if (!port)
 628                 port = LAST_RESERVED_PORT + 1 +
 629                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 630
 631         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 632
 633         if (addr->svm_port == VMADDR_PORT_ANY) {
 634                 bool found = false;
 635                 unsigned int i;
 636
 637                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 638                         if (port <= LAST_RESERVED_PORT)
 639                                 port = LAST_RESERVED_PORT + 1;
 640
 641                         new_addr.svm_port = port++;
 642
 643                         if (!__vsock_find_bound_socket(&new_addr)) {
 644                                 found = true;
 645                                 break;
 646                         }
 647                 }
 648
 649                 if (!found)
 650                         return -EADDRNOTAVAIL;
 651         } else {
 652                 /* If port is in reserved range, ensure caller
 653                  * has necessary privileges.
 654                  */
 655                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 656                     !capable(CAP_NET_BIND_SERVICE)) {
 657                         return -EACCES;
 658                 }
 659
 660                 if (__vsock_find_bound_socket(&new_addr))
 661                         return -EADDRINUSE;
 662         }
 663
 664         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 665
 666         /* Remove connection oriented sockets from the unbound list and add them
 667          * to the hash table for easy lookup by its address.  The unbound list
 668          * is simply an extra entry at the end of the hash table, a trick used
 669          * by AF_UNIX.
 670          */
 671         __vsock_remove_bound(vsk);
 672         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 673
 674         return 0;
 675 }
 676
 677 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 678                               struct sockaddr_vm *addr)
 679 {
 680         return vsk->transport->dgram_bind(vsk, addr);
 681 }
 682
 683 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 684 {
 685         struct vsock_sock *vsk = vsock_sk(sk);
 686         int retval;
 687
 688         /* First ensure this socket isn't already bound. */
 689         if (vsock_addr_bound(&vsk->local_addr))
 690                 return -EINVAL;
 691
 692         /* Now bind to the provided address or select appropriate values if
 693          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 694          * like AF_INET prevents binding to a non-local IP address (in most
 695          * cases), we only allow binding to a local CID.
 696          */
 697         if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
 698                 return -EADDRNOTAVAIL;
 699
 700         switch (sk->sk_socket->type) {
 701         case SOCK_STREAM:
 702         case SOCK_SEQPACKET:
 703                 spin_lock_bh(&vsock_table_lock);
 704                 retval = __vsock_bind_connectible(vsk, addr);
 705                 spin_unlock_bh(&vsock_table_lock);
 706                 break;
 707
 708         case SOCK_DGRAM:
 709                 retval = __vsock_bind_dgram(vsk, addr);
 710                 break;
 711
 712         default:
 713                 retval = -EINVAL;
 714                 break;
 715         }
 716
 717         return retval;
 718 }
 719
 720 static void vsock_connect_timeout(struct work_struct *work);
 721
 722 static struct sock *__vsock_create(struct net *net,
 723                                    struct socket *sock,
 724                                    struct sock *parent,
 725                                    gfp_t priority,
 726                                    unsigned short type,
 727                                    int kern)
 728 {
 729         struct sock *sk;
 730         struct vsock_sock *psk;
 731         struct vsock_sock *vsk;
 732
 733         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 734         if (!sk)
 735                 return NULL;
 736
 737         sock_init_data(sock, sk);
 738
 739         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 740          * non-NULL. We make sure that our sockets always have a type by
 741          * setting it here if needed.
 742          */
 743         if (!sock)
 744                 sk->sk_type = type;
 745
 746         vsk = vsock_sk(sk);
 747         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 748         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 749
 750         sk->sk_destruct = vsock_sk_destruct;
 751         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 752         sock_reset_flag(sk, SOCK_DONE);
 753
 754         INIT_LIST_HEAD(&vsk->bound_table);
 755         INIT_LIST_HEAD(&vsk->connected_table);
 756         vsk->listener = NULL;
 757         INIT_LIST_HEAD(&vsk->pending_links);
 758         INIT_LIST_HEAD(&vsk->accept_queue);
 759         vsk->rejected = false;
 760         vsk->sent_request = false;
 761         vsk->ignore_connecting_rst = false;
 762         vsk->peer_shutdown = 0;
 763         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 764         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 765
 766         psk = parent ? vsock_sk(parent) : NULL;
 767         if (parent) {
 768                 vsk->trusted = psk->trusted;
 769                 vsk->owner = get_cred(psk->owner);
 770                 vsk->connect_timeout = psk->connect_timeout;
 771                 vsk->buffer_size = psk->buffer_size;
 772                 vsk->buffer_min_size = psk->buffer_min_size;
 773                 vsk->buffer_max_size = psk->buffer_max_size;
 774                 security_sk_clone(parent, sk);
 775         } else {
 776                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 777                 vsk->owner = get_current_cred();
 778                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 779                 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
 780                 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
 781                 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
 782         }
 783
 784         return sk;
 785 }
 786
 787 static bool sock_type_connectible(u16 type)
 788 {
 789         return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
 790 }
 791
 792 static void __vsock_release(struct sock *sk, int level)
 793 {
 794         if (sk) {
 795                 struct sock *pending;
 796                 struct vsock_sock *vsk;
 797
 798                 vsk = vsock_sk(sk);
 799                 pending = NULL; /* Compiler warning. */
 800
 801                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 802                  * version to avoid the warning "possible recursive locking
 803                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 804                  * is the same as lock_sock(sk).
 805                  */
 806                 lock_sock_nested(sk, level);
 807
 808                 if (vsk->transport)
 809                         vsk->transport->release(vsk);
 810                 else if (sock_type_connectible(sk->sk_type))
 811                         vsock_remove_sock(vsk);
 812
 813                 sock_orphan(sk);
 814                 sk->sk_shutdown = SHUTDOWN_MASK;
 815
 816                 skb_queue_purge(&sk->sk_receive_queue);
 817
 818                 /* Clean up any sockets that never were accepted. */
 819                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 820                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 821                         sock_put(pending);
 822                 }
 823
 824                 release_sock(sk);
 825                 sock_put(sk);
 826         }
 827 }
 828
 829 static void vsock_sk_destruct(struct sock *sk)
 830 {
 831         struct vsock_sock *vsk = vsock_sk(sk);
 832
 833         vsock_deassign_transport(vsk);
 834
 835         /* When clearing these addresses, there's no need to set the family and
 836          * possibly register the address family with the kernel.
 837          */
 838         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 839         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 840
 841         put_cred(vsk->owner);
 842 }
 843
 844 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 845 {
 846         int err;
 847
 848         err = sock_queue_rcv_skb(sk, skb);
 849         if (err)
 850                 kfree_skb(skb);
 851
 852         return err;
 853 }
 854
 855 struct sock *vsock_create_connected(struct sock *parent)
 856 {
 857         return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
 858                               parent->sk_type, 0);
 859 }
 860 EXPORT_SYMBOL_GPL(vsock_create_connected);
 861
 862 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 863 {
 864         return vsk->transport->stream_has_data(vsk);
 865 }
 866 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 867
 868 static s64 vsock_connectible_has_data(struct vsock_sock *vsk)
 869 {
 870         struct sock *sk = sk_vsock(vsk);
 871
 872         if (sk->sk_type == SOCK_SEQPACKET)
 873                 return vsk->transport->seqpacket_has_data(vsk);
 874         else
 875                 return vsock_stream_has_data(vsk);
 876 }
 877
 878 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 879 {
 880         return vsk->transport->stream_has_space(vsk);
 881 }
 882 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 883
 884 static int vsock_release(struct socket *sock)
 885 {
 886         __vsock_release(sock->sk, 0);
 887         sock->sk = NULL;
 888         sock->state = SS_FREE;
 889
 890         return 0;
 891 }
 892
 893 static int
 894 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 895 {
 896         int err;
 897         struct sock *sk;
 898         struct sockaddr_vm *vm_addr;
 899
 900         sk = sock->sk;
 901
 902         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 903                 return -EINVAL;
 904
 905         lock_sock(sk);
 906         err = __vsock_bind(sk, vm_addr);
 907         release_sock(sk);
 908
 909         return err;
 910 }
 911
 912 static int vsock_getname(struct socket *sock,
 913                          struct sockaddr *addr, int peer)
 914 {
 915         int err;
 916         struct sock *sk;
 917         struct vsock_sock *vsk;
 918         struct sockaddr_vm *vm_addr;
 919
 920         sk = sock->sk;
 921         vsk = vsock_sk(sk);
 922         err = 0;
 923
 924         lock_sock(sk);
 925
 926         if (peer) {
 927                 if (sock->state != SS_CONNECTED) {
 928                         err = -ENOTCONN;
 929                         goto out;
 930                 }
 931                 vm_addr = &vsk->remote_addr;
 932         } else {
 933                 vm_addr = &vsk->local_addr;
 934         }
 935
 936         if (!vm_addr) {
 937                 err = -EINVAL;
 938                 goto out;
 939         }
 940
 941         /* sys_getsockname() and sys_getpeername() pass us a
 942          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 943          * that macro is defined in socket.c instead of .h, so we hardcode its
 944          * value here.
 945          */
 946         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 947         memcpy(addr, vm_addr, sizeof(*vm_addr));
 948         err = sizeof(*vm_addr);
 949
 950 out:
 951         release_sock(sk);
 952         return err;
 953 }
 954
 955 static int vsock_shutdown(struct socket *sock, int mode)
 956 {
 957         int err;
 958         struct sock *sk;
 959
 960         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 961          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 962          * here like the other address families do.  Note also that the
 963          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 964          * which is what we want.
 965          */
 966         mode++;
 967
 968         if ((mode & ~SHUTDOWN_MASK) || !mode)
 969                 return -EINVAL;
 970
 971         /* If this is a connection oriented socket and it is not connected then
 972          * bail out immediately.  If it is a DGRAM socket then we must first
 973          * kick the socket so that it wakes up from any sleeping calls, for
 974          * example recv(), and then afterwards return the error.
 975          */
 976
 977         sk = sock->sk;
 978
 979         lock_sock(sk);
 980         if (sock->state == SS_UNCONNECTED) {
 981                 err = -ENOTCONN;
 982                 if (sock_type_connectible(sk->sk_type))
 983                         goto out;
 984         } else {
 985                 sock->state = SS_DISCONNECTING;
 986                 err = 0;
 987         }
 988
 989         /* Receive and send shutdowns are treated alike. */
 990         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 991         if (mode) {
 992                 sk->sk_shutdown |= mode;
 993                 sk->sk_state_change(sk);
 994
 995                 if (sock_type_connectible(sk->sk_type)) {
 996                         sock_reset_flag(sk, SOCK_DONE);
 997                         vsock_send_shutdown(sk, mode);
 998                 }
 999         }
1000
1001 out:
1002         release_sock(sk);
1003         return err;
1004 }
1005
1006 static __poll_t vsock_poll(struct file *file, struct socket *sock,
1007                                poll_table *wait)
1008 {
1009         struct sock *sk;
1010         __poll_t mask;
1011         struct vsock_sock *vsk;
1012
1013         sk = sock->sk;
1014         vsk = vsock_sk(sk);
1015
1016         poll_wait(file, sk_sleep(sk), wait);
1017         mask = 0;
1018
1019         if (sk->sk_err)
1020                 /* Signify that there has been an error on this socket. */
1021                 mask |= EPOLLERR;
1022
1023         /* INET sockets treat local write shutdown and peer write shutdown as a
1024          * case of EPOLLHUP set.
1025          */
1026         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1027             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1028              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1029                 mask |= EPOLLHUP;
1030         }
1031
1032         if (sk->sk_shutdown & RCV_SHUTDOWN ||
1033             vsk->peer_shutdown & SEND_SHUTDOWN) {
1034                 mask |= EPOLLRDHUP;
1035         }
1036
1037         if (sock->type == SOCK_DGRAM) {
1038                 /* For datagram sockets we can read if there is something in
1039                  * the queue and write as long as the socket isn't shutdown for
1040                  * sending.
1041                  */
1042                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1043                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
1044                         mask |= EPOLLIN | EPOLLRDNORM;
1045                 }
1046
1047                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1048                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1049
1050         } else if (sock_type_connectible(sk->sk_type)) {
1051                 const struct vsock_transport *transport;
1052
1053                 lock_sock(sk);
1054
1055                 transport = vsk->transport;
1056
1057                 /* Listening sockets that have connections in their accept
1058                  * queue can be read.
1059                  */
1060                 if (sk->sk_state == TCP_LISTEN
1061                     && !vsock_is_accept_queue_empty(sk))
1062                         mask |= EPOLLIN | EPOLLRDNORM;
1063
1064                 /* If there is something in the queue then we can read. */
1065                 if (transport && transport->stream_is_active(vsk) &&
1066                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1067                         bool data_ready_now = false;
1068                         int ret = transport->notify_poll_in(
1069                                         vsk, 1, &data_ready_now);
1070                         if (ret < 0) {
1071                                 mask |= EPOLLERR;
1072                         } else {
1073                                 if (data_ready_now)
1074                                         mask |= EPOLLIN | EPOLLRDNORM;
1075
1076                         }
1077                 }
1078
1079                 /* Sockets whose connections have been closed, reset, or
1080                  * terminated should also be considered read, and we check the
1081                  * shutdown flag for that.
1082                  */
1083                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1084                     vsk->peer_shutdown & SEND_SHUTDOWN) {
1085                         mask |= EPOLLIN | EPOLLRDNORM;
1086                 }
1087
1088                 /* Connected sockets that can produce data can be written. */
1089                 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1090                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1091                                 bool space_avail_now = false;
1092                                 int ret = transport->notify_poll_out(
1093                                                 vsk, 1, &space_avail_now);
1094                                 if (ret < 0) {
1095                                         mask |= EPOLLERR;
1096                                 } else {
1097                                         if (space_avail_now)
1098                                                 /* Remove EPOLLWRBAND since INET
1099                                                  * sockets are not setting it.
1100                                                  */
1101                                                 mask |= EPOLLOUT | EPOLLWRNORM;
1102
1103                                 }
1104                         }
1105                 }
1106
1107                 /* Simulate INET socket poll behaviors, which sets
1108                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1109                  * but local send is not shutdown.
1110                  */
1111                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1112                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1113                                 mask |= EPOLLOUT | EPOLLWRNORM;
1114
1115                 }
1116
1117                 release_sock(sk);
1118         }
1119
1120         return mask;
1121 }
1122
1123 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1124                                size_t len)
1125 {
1126         int err;
1127         struct sock *sk;
1128         struct vsock_sock *vsk;
1129         struct sockaddr_vm *remote_addr;
1130         const struct vsock_transport *transport;
1131
1132         if (msg->msg_flags & MSG_OOB)
1133                 return -EOPNOTSUPP;
1134
1135         /* For now, MSG_DONTWAIT is always assumed... */
1136         err = 0;
1137         sk = sock->sk;
1138         vsk = vsock_sk(sk);
1139
1140         lock_sock(sk);
1141
1142         transport = vsk->transport;
1143
1144         err = vsock_auto_bind(vsk);
1145         if (err)
1146                 goto out;
1147
1148
1149         /* If the provided message contains an address, use that.  Otherwise
1150          * fall back on the socket's remote handle (if it has been connected).
1151          */
1152         if (msg->msg_name &&
1153             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1154                             &remote_addr) == 0) {
1155                 /* Ensure this address is of the right type and is a valid
1156                  * destination.
1157                  */
1158
1159                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1160                         remote_addr->svm_cid = transport->get_local_cid();
1161
1162                 if (!vsock_addr_bound(remote_addr)) {
1163                         err = -EINVAL;
1164                         goto out;
1165                 }
1166         } else if (sock->state == SS_CONNECTED) {
1167                 remote_addr = &vsk->remote_addr;
1168
1169                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1170                         remote_addr->svm_cid = transport->get_local_cid();
1171
1172                 /* XXX Should connect() or this function ensure remote_addr is
1173                  * bound?
1174                  */
1175                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1176                         err = -EINVAL;
1177                         goto out;
1178                 }
1179         } else {
1180                 err = -EINVAL;
1181                 goto out;
1182         }
1183
1184         if (!transport->dgram_allow(remote_addr->svm_cid,
1185                                     remote_addr->svm_port)) {
1186                 err = -EINVAL;
1187                 goto out;
1188         }
1189
1190         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1191
1192 out:
1193         release_sock(sk);
1194         return err;
1195 }
1196
1197 static int vsock_dgram_connect(struct socket *sock,
1198                                struct sockaddr *addr, int addr_len, int flags)
1199 {
1200         int err;
1201         struct sock *sk;
1202         struct vsock_sock *vsk;
1203         struct sockaddr_vm *remote_addr;
1204
1205         sk = sock->sk;
1206         vsk = vsock_sk(sk);
1207
1208         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1209         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1210                 lock_sock(sk);
1211                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1212                                 VMADDR_PORT_ANY);
1213                 sock->state = SS_UNCONNECTED;
1214                 release_sock(sk);
1215                 return 0;
1216         } else if (err != 0)
1217                 return -EINVAL;
1218
1219         lock_sock(sk);
1220
1221         err = vsock_auto_bind(vsk);
1222         if (err)
1223                 goto out;
1224
1225         if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1226                                          remote_addr->svm_port)) {
1227                 err = -EINVAL;
1228                 goto out;
1229         }
1230
1231         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1232         sock->state = SS_CONNECTED;
1233
1234 out:
1235         release_sock(sk);
1236         return err;
1237 }
1238
1239 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1240                                size_t len, int flags)
1241 {
1242         struct vsock_sock *vsk = vsock_sk(sock->sk);
1243
1244         return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1245 }
1246
1247 static const struct proto_ops vsock_dgram_ops = {
1248         .family = PF_VSOCK,
1249         .owner = THIS_MODULE,
1250         .release = vsock_release,
1251         .bind = vsock_bind,
1252         .connect = vsock_dgram_connect,
1253         .socketpair = sock_no_socketpair,
1254         .accept = sock_no_accept,
1255         .getname = vsock_getname,
1256         .poll = vsock_poll,
1257         .ioctl = sock_no_ioctl,
1258         .listen = sock_no_listen,
1259         .shutdown = vsock_shutdown,
1260         .sendmsg = vsock_dgram_sendmsg,
1261         .recvmsg = vsock_dgram_recvmsg,
1262         .mmap = sock_no_mmap,
1263         .sendpage = sock_no_sendpage,
1264 };
1265
1266 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1267 {
1268         const struct vsock_transport *transport = vsk->transport;
1269
1270         if (!transport || !transport->cancel_pkt)
1271                 return -EOPNOTSUPP;
1272
1273         return transport->cancel_pkt(vsk);
1274 }
1275
1276 static void vsock_connect_timeout(struct work_struct *work)
1277 {
1278         struct sock *sk;
1279         struct vsock_sock *vsk;
1280
1281         vsk = container_of(work, struct vsock_sock, connect_work.work);
1282         sk = sk_vsock(vsk);
1283
1284         lock_sock(sk);
1285         if (sk->sk_state == TCP_SYN_SENT &&
1286             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1287                 sk->sk_state = TCP_CLOSE;
1288                 sk->sk_socket->state = SS_UNCONNECTED;
1289                 sk->sk_err = ETIMEDOUT;
1290                 sk_error_report(sk);
1291                 vsock_transport_cancel_pkt(vsk);
1292         }
1293         release_sock(sk);
1294
1295         sock_put(sk);
1296 }
1297
1298 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1299                          int addr_len, int flags)
1300 {
1301         int err;
1302         struct sock *sk;
1303         struct vsock_sock *vsk;
1304         const struct vsock_transport *transport;
1305         struct sockaddr_vm *remote_addr;
1306         long timeout;
1307         DEFINE_WAIT(wait);
1308
1309         err = 0;
1310         sk = sock->sk;
1311         vsk = vsock_sk(sk);
1312
1313         lock_sock(sk);
1314
1315         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1316         switch (sock->state) {
1317         case SS_CONNECTED:
1318                 err = -EISCONN;
1319                 goto out;
1320         case SS_DISCONNECTING:
1321                 err = -EINVAL;
1322                 goto out;
1323         case SS_CONNECTING:
1324                 /* This continues on so we can move sock into the SS_CONNECTED
1325                  * state once the connection has completed (at which point err
1326                  * will be set to zero also).  Otherwise, we will either wait
1327                  * for the connection or return -EALREADY should this be a
1328                  * non-blocking call.
1329                  */
1330                 err = -EALREADY;
1331                 if (flags & O_NONBLOCK)
1332                         goto out;
1333                 break;
1334         default:
1335                 if ((sk->sk_state == TCP_LISTEN) ||
1336                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1337                         err = -EINVAL;
1338                         goto out;
1339                 }
1340
1341                 /* Set the remote address that we are connecting to. */
1342                 memcpy(&vsk->remote_addr, remote_addr,
1343                        sizeof(vsk->remote_addr));
1344
1345                 err = vsock_assign_transport(vsk, NULL);
1346                 if (err)
1347                         goto out;
1348
1349                 transport = vsk->transport;
1350
1351                 /* The hypervisor and well-known contexts do not have socket
1352                  * endpoints.
1353                  */
1354                 if (!transport ||
1355                     !transport->stream_allow(remote_addr->svm_cid,
1356                                              remote_addr->svm_port)) {
1357                         err = -ENETUNREACH;
1358                         goto out;
1359                 }
1360
1361                 err = vsock_auto_bind(vsk);
1362                 if (err)
1363                         goto out;
1364
1365                 sk->sk_state = TCP_SYN_SENT;
1366
1367                 err = transport->connect(vsk);
1368                 if (err < 0)
1369                         goto out;
1370
1371                 /* Mark sock as connecting and set the error code to in
1372                  * progress in case this is a non-blocking connect.
1373                  */
1374                 sock->state = SS_CONNECTING;
1375                 err = -EINPROGRESS;
1376         }
1377
1378         /* The receive path will handle all communication until we are able to
1379          * enter the connected state.  Here we wait for the connection to be
1380          * completed or a notification of an error.
1381          */
1382         timeout = vsk->connect_timeout;
1383         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1384
1385         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1386                 if (flags & O_NONBLOCK) {
1387                         /* If we're not going to block, we schedule a timeout
1388                          * function to generate a timeout on the connection
1389                          * attempt, in case the peer doesn't respond in a
1390                          * timely manner. We hold on to the socket until the
1391                          * timeout fires.
1392                          */
1393                         sock_hold(sk);
1394
1395                         /* If the timeout function is already scheduled,
1396                          * reschedule it, then ungrab the socket refcount to
1397                          * keep it balanced.
1398                          */
1399                         if (mod_delayed_work(system_wq, &vsk->connect_work,
1400                                              timeout))
1401                                 sock_put(sk);
1402
1403                         /* Skip ahead to preserve error code set above. */
1404                         goto out_wait;
1405                 }
1406
1407                 release_sock(sk);
1408                 timeout = schedule_timeout(timeout);
1409                 lock_sock(sk);
1410
1411                 if (signal_pending(current)) {
1412                         err = sock_intr_errno(timeout);
1413                         sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1414                         sock->state = SS_UNCONNECTED;
1415                         vsock_transport_cancel_pkt(vsk);
1416                         vsock_remove_connected(vsk);
1417                         goto out_wait;
1418                 } else if (timeout == 0) {
1419                         err = -ETIMEDOUT;
1420                         sk->sk_state = TCP_CLOSE;
1421                         sock->state = SS_UNCONNECTED;
1422                         vsock_transport_cancel_pkt(vsk);
1423                         goto out_wait;
1424                 }
1425
1426                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1427         }
1428
1429         if (sk->sk_err) {
1430                 err = -sk->sk_err;
1431                 sk->sk_state = TCP_CLOSE;
1432                 sock->state = SS_UNCONNECTED;
1433         } else {
1434                 err = 0;
1435         }
1436
1437 out_wait:
1438         finish_wait(sk_sleep(sk), &wait);
1439 out:
1440         release_sock(sk);
1441         return err;
1442 }
1443
1444 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1445                         bool kern)
1446 {
1447         struct sock *listener;
1448         int err;
1449         struct sock *connected;
1450         struct vsock_sock *vconnected;
1451         long timeout;
1452         DEFINE_WAIT(wait);
1453
1454         err = 0;
1455         listener = sock->sk;
1456
1457         lock_sock(listener);
1458
1459         if (!sock_type_connectible(sock->type)) {
1460                 err = -EOPNOTSUPP;
1461                 goto out;
1462         }
1463
1464         if (listener->sk_state != TCP_LISTEN) {
1465                 err = -EINVAL;
1466                 goto out;
1467         }
1468
1469         /* Wait for children sockets to appear; these are the new sockets
1470          * created upon connection establishment.
1471          */
1472         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1473         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1474
1475         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1476                listener->sk_err == 0) {
1477                 release_sock(listener);
1478                 timeout = schedule_timeout(timeout);
1479                 finish_wait(sk_sleep(listener), &wait);
1480                 lock_sock(listener);
1481
1482                 if (signal_pending(current)) {
1483                         err = sock_intr_errno(timeout);
1484                         goto out;
1485                 } else if (timeout == 0) {
1486                         err = -EAGAIN;
1487                         goto out;
1488                 }
1489
1490                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1491         }
1492         finish_wait(sk_sleep(listener), &wait);
1493
1494         if (listener->sk_err)
1495                 err = -listener->sk_err;
1496
1497         if (connected) {
1498                 sk_acceptq_removed(listener);
1499
1500                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1501                 vconnected = vsock_sk(connected);
1502
1503                 /* If the listener socket has received an error, then we should
1504                  * reject this socket and return.  Note that we simply mark the
1505                  * socket rejected, drop our reference, and let the cleanup
1506                  * function handle the cleanup; the fact that we found it in
1507                  * the listener's accept queue guarantees that the cleanup
1508                  * function hasn't run yet.
1509                  */
1510                 if (err) {
1511                         vconnected->rejected = true;
1512                 } else {
1513                         newsock->state = SS_CONNECTED;
1514                         sock_graft(connected, newsock);
1515                 }
1516
1517                 release_sock(connected);
1518                 sock_put(connected);
1519         }
1520
1521 out:
1522         release_sock(listener);
1523         return err;
1524 }
1525
1526 static int vsock_listen(struct socket *sock, int backlog)
1527 {
1528         int err;
1529         struct sock *sk;
1530         struct vsock_sock *vsk;
1531
1532         sk = sock->sk;
1533
1534         lock_sock(sk);
1535
1536         if (!sock_type_connectible(sk->sk_type)) {
1537                 err = -EOPNOTSUPP;
1538                 goto out;
1539         }
1540
1541         if (sock->state != SS_UNCONNECTED) {
1542                 err = -EINVAL;
1543                 goto out;
1544         }
1545
1546         vsk = vsock_sk(sk);
1547
1548         if (!vsock_addr_bound(&vsk->local_addr)) {
1549                 err = -EINVAL;
1550                 goto out;
1551         }
1552
1553         sk->sk_max_ack_backlog = backlog;
1554         sk->sk_state = TCP_LISTEN;
1555
1556         err = 0;
1557
1558 out:
1559         release_sock(sk);
1560         return err;
1561 }
1562
1563 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1564                                      const struct vsock_transport *transport,
1565                                      u64 val)
1566 {
1567         if (val > vsk->buffer_max_size)
1568                 val = vsk->buffer_max_size;
1569
1570         if (val < vsk->buffer_min_size)
1571                 val = vsk->buffer_min_size;
1572
1573         if (val != vsk->buffer_size &&
1574             transport && transport->notify_buffer_size)
1575                 transport->notify_buffer_size(vsk, &val);
1576
1577         vsk->buffer_size = val;
1578 }
1579
1580 static int vsock_connectible_setsockopt(struct socket *sock,
1581                                         int level,
1582                                         int optname,
1583                                         sockptr_t optval,
1584                                         unsigned int optlen)
1585 {
1586         int err;
1587         struct sock *sk;
1588         struct vsock_sock *vsk;
1589         const struct vsock_transport *transport;
1590         u64 val;
1591
1592         if (level != AF_VSOCK)
1593                 return -ENOPROTOOPT;
1594
1595 #define COPY_IN(_v)                                       \
1596         do {                                              \
1597                 if (optlen < sizeof(_v)) {                \
1598                         err = -EINVAL;                    \
1599                         goto exit;                        \
1600                 }                                         \
1601                 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) {  \
1602                         err = -EFAULT;                                  \
1603                         goto exit;                                      \
1604                 }                                                       \
1605         } while (0)
1606
1607         err = 0;
1608         sk = sock->sk;
1609         vsk = vsock_sk(sk);
1610
1611         lock_sock(sk);
1612
1613         transport = vsk->transport;
1614
1615         switch (optname) {
1616         case SO_VM_SOCKETS_BUFFER_SIZE:
1617                 COPY_IN(val);
1618                 vsock_update_buffer_size(vsk, transport, val);
1619                 break;
1620
1621         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1622                 COPY_IN(val);
1623                 vsk->buffer_max_size = val;
1624                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1625                 break;
1626
1627         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1628                 COPY_IN(val);
1629                 vsk->buffer_min_size = val;
1630                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1631                 break;
1632
1633         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1634                 struct __kernel_old_timeval tv;
1635                 COPY_IN(tv);
1636                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1637                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1638                         vsk->connect_timeout = tv.tv_sec * HZ +
1639                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1640                         if (vsk->connect_timeout == 0)
1641                                 vsk->connect_timeout =
1642                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1643
1644                 } else {
1645                         err = -ERANGE;
1646                 }
1647                 break;
1648         }
1649
1650         default:
1651                 err = -ENOPROTOOPT;
1652                 break;
1653         }
1654
1655 #undef COPY_IN
1656
1657 exit:
1658         release_sock(sk);
1659         return err;
1660 }
1661
1662 static int vsock_connectible_getsockopt(struct socket *sock,
1663                                         int level, int optname,
1664                                         char __user *optval,
1665                                         int __user *optlen)
1666 {
1667         int err;
1668         int len;
1669         struct sock *sk;
1670         struct vsock_sock *vsk;
1671         u64 val;
1672
1673         if (level != AF_VSOCK)
1674                 return -ENOPROTOOPT;
1675
1676         err = get_user(len, optlen);
1677         if (err != 0)
1678                 return err;
1679
1680 #define COPY_OUT(_v)                            \
1681         do {                                    \
1682                 if (len < sizeof(_v))           \
1683                         return -EINVAL;         \
1684                                                 \
1685                 len = sizeof(_v);               \
1686                 if (copy_to_user(optval, &_v, len) != 0)        \
1687                         return -EFAULT;                         \
1688                                                                 \
1689         } while (0)
1690
1691         err = 0;
1692         sk = sock->sk;
1693         vsk = vsock_sk(sk);
1694
1695         switch (optname) {
1696         case SO_VM_SOCKETS_BUFFER_SIZE:
1697                 val = vsk->buffer_size;
1698                 COPY_OUT(val);
1699                 break;
1700
1701         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1702                 val = vsk->buffer_max_size;
1703                 COPY_OUT(val);
1704                 break;
1705
1706         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1707                 val = vsk->buffer_min_size;
1708                 COPY_OUT(val);
1709                 break;
1710
1711         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1712                 struct __kernel_old_timeval tv;
1713                 tv.tv_sec = vsk->connect_timeout / HZ;
1714                 tv.tv_usec =
1715                     (vsk->connect_timeout -
1716                      tv.tv_sec * HZ) * (1000000 / HZ);
1717                 COPY_OUT(tv);
1718                 break;
1719         }
1720         default:
1721                 return -ENOPROTOOPT;
1722         }
1723
1724         err = put_user(len, optlen);
1725         if (err != 0)
1726                 return -EFAULT;
1727
1728 #undef COPY_OUT
1729
1730         return 0;
1731 }
1732
1733 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1734                                      size_t len)
1735 {
1736         struct sock *sk;
1737         struct vsock_sock *vsk;
1738         const struct vsock_transport *transport;
1739         ssize_t total_written;
1740         long timeout;
1741         int err;
1742         struct vsock_transport_send_notify_data send_data;
1743         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1744
1745         sk = sock->sk;
1746         vsk = vsock_sk(sk);
1747         total_written = 0;
1748         err = 0;
1749
1750         if (msg->msg_flags & MSG_OOB)
1751                 return -EOPNOTSUPP;
1752
1753         lock_sock(sk);
1754
1755         transport = vsk->transport;
1756
1757         /* Callers should not provide a destination with connection oriented
1758          * sockets.
1759          */
1760         if (msg->msg_namelen) {
1761                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1762                 goto out;
1763         }
1764
1765         /* Send data only if both sides are not shutdown in the direction. */
1766         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1767             vsk->peer_shutdown & RCV_SHUTDOWN) {
1768                 err = -EPIPE;
1769                 goto out;
1770         }
1771
1772         if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1773             !vsock_addr_bound(&vsk->local_addr)) {
1774                 err = -ENOTCONN;
1775                 goto out;
1776         }
1777
1778         if (!vsock_addr_bound(&vsk->remote_addr)) {
1779                 err = -EDESTADDRREQ;
1780                 goto out;
1781         }
1782
1783         /* Wait for room in the produce queue to enqueue our user's data. */
1784         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1785
1786         err = transport->notify_send_init(vsk, &send_data);
1787         if (err < 0)
1788                 goto out;
1789
1790         while (total_written < len) {
1791                 ssize_t written;
1792
1793                 add_wait_queue(sk_sleep(sk), &wait);
1794                 while (vsock_stream_has_space(vsk) == 0 &&
1795                        sk->sk_err == 0 &&
1796                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1797                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1798
1799                         /* Don't wait for non-blocking sockets. */
1800                         if (timeout == 0) {
1801                                 err = -EAGAIN;
1802                                 remove_wait_queue(sk_sleep(sk), &wait);
1803                                 goto out_err;
1804                         }
1805
1806                         err = transport->notify_send_pre_block(vsk, &send_data);
1807                         if (err < 0) {
1808                                 remove_wait_queue(sk_sleep(sk), &wait);
1809                                 goto out_err;
1810                         }
1811
1812                         release_sock(sk);
1813                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1814                         lock_sock(sk);
1815                         if (signal_pending(current)) {
1816                                 err = sock_intr_errno(timeout);
1817                                 remove_wait_queue(sk_sleep(sk), &wait);
1818                                 goto out_err;
1819                         } else if (timeout == 0) {
1820                                 err = -EAGAIN;
1821                                 remove_wait_queue(sk_sleep(sk), &wait);
1822                                 goto out_err;
1823                         }
1824                 }
1825                 remove_wait_queue(sk_sleep(sk), &wait);
1826
1827                 /* These checks occur both as part of and after the loop
1828                  * conditional since we need to check before and after
1829                  * sleeping.
1830                  */
1831                 if (sk->sk_err) {
1832                         err = -sk->sk_err;
1833                         goto out_err;
1834                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1835                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1836                         err = -EPIPE;
1837                         goto out_err;
1838                 }
1839
1840                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1841                 if (err < 0)
1842                         goto out_err;
1843
1844                 /* Note that enqueue will only write as many bytes as are free
1845                  * in the produce queue, so we don't need to ensure len is
1846                  * smaller than the queue size.  It is the caller's
1847                  * responsibility to check how many bytes we were able to send.
1848                  */
1849
1850                 if (sk->sk_type == SOCK_SEQPACKET) {
1851                         written = transport->seqpacket_enqueue(vsk,
1852                                                 msg, len - total_written);
1853                 } else {
1854                         written = transport->stream_enqueue(vsk,
1855                                         msg, len - total_written);
1856                 }
1857                 if (written < 0) {
1858                         err = -ENOMEM;
1859                         goto out_err;
1860                 }
1861
1862                 total_written += written;
1863
1864                 err = transport->notify_send_post_enqueue(
1865                                 vsk, written, &send_data);
1866                 if (err < 0)
1867                         goto out_err;
1868
1869         }
1870
1871 out_err:
1872         if (total_written > 0) {
1873                 /* Return number of written bytes only if:
1874                  * 1) SOCK_STREAM socket.
1875                  * 2) SOCK_SEQPACKET socket when whole buffer is sent.
1876                  */
1877                 if (sk->sk_type == SOCK_STREAM || total_written == len)
1878                         err = total_written;
1879         }
1880 out:
1881         release_sock(sk);
1882         return err;
1883 }
1884
1885 static int vsock_connectible_wait_data(struct sock *sk,
1886                                        struct wait_queue_entry *wait,
1887                                        long timeout,
1888                                        struct vsock_transport_recv_notify_data *recv_data,
1889                                        size_t target)
1890 {
1891         const struct vsock_transport *transport;
1892         struct vsock_sock *vsk;
1893         s64 data;
1894         int err;
1895
1896         vsk = vsock_sk(sk);
1897         err = 0;
1898         transport = vsk->transport;
1899
1900         while (1) {
1901                 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
1902                 data = vsock_connectible_has_data(vsk);
1903                 if (data != 0)
1904                         break;
1905
1906                 if (sk->sk_err != 0 ||
1907                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1908                     (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1909                         break;
1910                 }
1911
1912                 /* Don't wait for non-blocking sockets. */
1913                 if (timeout == 0) {
1914                         err = -EAGAIN;
1915                         break;
1916                 }
1917
1918                 if (recv_data) {
1919                         err = transport->notify_recv_pre_block(vsk, target, recv_data);
1920                         if (err < 0)
1921                                 break;
1922                 }
1923
1924                 release_sock(sk);
1925                 timeout = schedule_timeout(timeout);
1926                 lock_sock(sk);
1927
1928                 if (signal_pending(current)) {
1929                         err = sock_intr_errno(timeout);
1930                         break;
1931                 } else if (timeout == 0) {
1932                         err = -EAGAIN;
1933                         break;
1934                 }
1935         }
1936
1937         finish_wait(sk_sleep(sk), wait);
1938
1939         if (err)
1940                 return err;
1941
1942         /* Internal transport error when checking for available
1943          * data. XXX This should be changed to a connection
1944          * reset in a later change.
1945          */
1946         if (data < 0)
1947                 return -ENOMEM;
1948
1949         return data;
1950 }
1951
1952 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
1953                                   size_t len, int flags)
1954 {
1955         struct vsock_transport_recv_notify_data recv_data;
1956         const struct vsock_transport *transport;
1957         struct vsock_sock *vsk;
1958         ssize_t copied;
1959         size_t target;
1960         long timeout;
1961         int err;
1962
1963         DEFINE_WAIT(wait);
1964
1965         vsk = vsock_sk(sk);
1966         transport = vsk->transport;
1967
1968         /* We must not copy less than target bytes into the user's buffer
1969          * before returning successfully, so we wait for the consume queue to
1970          * have that much data to consume before dequeueing.  Note that this
1971          * makes it impossible to handle cases where target is greater than the
1972          * queue size.
1973          */
1974         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1975         if (target >= transport->stream_rcvhiwat(vsk)) {
1976                 err = -ENOMEM;
1977                 goto out;
1978         }
1979         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1980         copied = 0;
1981
1982         err = transport->notify_recv_init(vsk, target, &recv_data);
1983         if (err < 0)
1984                 goto out;
1985
1986
1987         while (1) {
1988                 ssize_t read;
1989
1990                 err = vsock_connectible_wait_data(sk, &wait, timeout,
1991                                                   &recv_data, target);
1992                 if (err <= 0)
1993                         break;
1994
1995                 err = transport->notify_recv_pre_dequeue(vsk, target,
1996                                                          &recv_data);
1997                 if (err < 0)
1998                         break;
1999
2000                 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
2001                 if (read < 0) {
2002                         err = -ENOMEM;
2003                         break;
2004                 }
2005
2006                 copied += read;
2007
2008                 err = transport->notify_recv_post_dequeue(vsk, target, read,
2009                                                 !(flags & MSG_PEEK), &recv_data);
2010                 if (err < 0)
2011                         goto out;
2012
2013                 if (read >= target || flags & MSG_PEEK)
2014                         break;
2015
2016                 target -= read;
2017         }
2018
2019         if (sk->sk_err)
2020                 err = -sk->sk_err;
2021         else if (sk->sk_shutdown & RCV_SHUTDOWN)
2022                 err = 0;
2023
2024         if (copied > 0)
2025                 err = copied;
2026
2027 out:
2028         return err;
2029 }
2030
2031 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2032                                      size_t len, int flags)
2033 {
2034         const struct vsock_transport *transport;
2035         struct vsock_sock *vsk;
2036         ssize_t msg_len;
2037         long timeout;
2038         int err = 0;
2039         DEFINE_WAIT(wait);
2040
2041         vsk = vsock_sk(sk);
2042         transport = vsk->transport;
2043
2044         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2045
2046         err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0);
2047         if (err <= 0)
2048                 goto out;
2049
2050         msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2051
2052         if (msg_len < 0) {
2053                 err = -ENOMEM;
2054                 goto out;
2055         }
2056
2057         if (sk->sk_err) {
2058                 err = -sk->sk_err;
2059         } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2060                 err = 0;
2061         } else {
2062                 /* User sets MSG_TRUNC, so return real length of
2063                  * packet.
2064                  */
2065                 if (flags & MSG_TRUNC)
2066                         err = msg_len;
2067                 else
2068                         err = len - msg_data_left(msg);
2069
2070                 /* Always set MSG_TRUNC if real length of packet is
2071                  * bigger than user's buffer.
2072                  */
2073                 if (msg_len > len)
2074                         msg->msg_flags |= MSG_TRUNC;
2075         }
2076
2077 out:
2078         return err;
2079 }
2080
2081 static int
2082 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2083                           int flags)
2084 {
2085         struct sock *sk;
2086         struct vsock_sock *vsk;
2087         const struct vsock_transport *transport;
2088         int err;
2089
2090         DEFINE_WAIT(wait);
2091
2092         sk = sock->sk;
2093         vsk = vsock_sk(sk);
2094         err = 0;
2095
2096         lock_sock(sk);
2097
2098         transport = vsk->transport;
2099
2100         if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2101                 /* Recvmsg is supposed to return 0 if a peer performs an
2102                  * orderly shutdown. Differentiate between that case and when a
2103                  * peer has not connected or a local shutdown occurred with the
2104                  * SOCK_DONE flag.
2105                  */
2106                 if (sock_flag(sk, SOCK_DONE))
2107                         err = 0;
2108                 else
2109                         err = -ENOTCONN;
2110
2111                 goto out;
2112         }
2113
2114         if (flags & MSG_OOB) {
2115                 err = -EOPNOTSUPP;
2116                 goto out;
2117         }
2118
2119         /* We don't check peer_shutdown flag here since peer may actually shut
2120          * down, but there can be data in the queue that a local socket can
2121          * receive.
2122          */
2123         if (sk->sk_shutdown & RCV_SHUTDOWN) {
2124                 err = 0;
2125                 goto out;
2126         }
2127
2128         /* It is valid on Linux to pass in a zero-length receive buffer.  This
2129          * is not an error.  We may as well bail out now.
2130          */
2131         if (!len) {
2132                 err = 0;
2133                 goto out;
2134         }
2135
2136         if (sk->sk_type == SOCK_STREAM)
2137                 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2138         else
2139                 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2140
2141 out:
2142         release_sock(sk);
2143         return err;
2144 }
2145
2146 static const struct proto_ops vsock_stream_ops = {
2147         .family = PF_VSOCK,
2148         .owner = THIS_MODULE,
2149         .release = vsock_release,
2150         .bind = vsock_bind,
2151         .connect = vsock_connect,
2152         .socketpair = sock_no_socketpair,
2153         .accept = vsock_accept,
2154         .getname = vsock_getname,
2155         .poll = vsock_poll,
2156         .ioctl = sock_no_ioctl,
2157         .listen = vsock_listen,
2158         .shutdown = vsock_shutdown,
2159         .setsockopt = vsock_connectible_setsockopt,
2160         .getsockopt = vsock_connectible_getsockopt,
2161         .sendmsg = vsock_connectible_sendmsg,
2162         .recvmsg = vsock_connectible_recvmsg,
2163         .mmap = sock_no_mmap,
2164         .sendpage = sock_no_sendpage,
2165 };
2166
2167 static const struct proto_ops vsock_seqpacket_ops = {
2168         .family = PF_VSOCK,
2169         .owner = THIS_MODULE,
2170         .release = vsock_release,
2171         .bind = vsock_bind,
2172         .connect = vsock_connect,
2173         .socketpair = sock_no_socketpair,
2174         .accept = vsock_accept,
2175         .getname = vsock_getname,
2176         .poll = vsock_poll,
2177         .ioctl = sock_no_ioctl,
2178         .listen = vsock_listen,
2179         .shutdown = vsock_shutdown,
2180         .setsockopt = vsock_connectible_setsockopt,
2181         .getsockopt = vsock_connectible_getsockopt,
2182         .sendmsg = vsock_connectible_sendmsg,
2183         .recvmsg = vsock_connectible_recvmsg,
2184         .mmap = sock_no_mmap,
2185         .sendpage = sock_no_sendpage,
2186 };
2187
2188 static int vsock_create(struct net *net, struct socket *sock,
2189                         int protocol, int kern)
2190 {
2191         struct vsock_sock *vsk;
2192         struct sock *sk;
2193         int ret;
2194
2195         if (!sock)
2196                 return -EINVAL;
2197
2198         if (protocol && protocol != PF_VSOCK)
2199                 return -EPROTONOSUPPORT;
2200
2201         switch (sock->type) {
2202         case SOCK_DGRAM:
2203                 sock->ops = &vsock_dgram_ops;
2204                 break;
2205         case SOCK_STREAM:
2206                 sock->ops = &vsock_stream_ops;
2207                 break;
2208         case SOCK_SEQPACKET:
2209                 sock->ops = &vsock_seqpacket_ops;
2210                 break;
2211         default:
2212                 return -ESOCKTNOSUPPORT;
2213         }
2214
2215         sock->state = SS_UNCONNECTED;
2216
2217         sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2218         if (!sk)
2219                 return -ENOMEM;
2220
2221         vsk = vsock_sk(sk);
2222
2223         if (sock->type == SOCK_DGRAM) {
2224                 ret = vsock_assign_transport(vsk, NULL);
2225                 if (ret < 0) {
2226                         sock_put(sk);
2227                         return ret;
2228                 }
2229         }
2230
2231         vsock_insert_unbound(vsk);
2232
2233         return 0;
2234 }
2235
2236 static const struct net_proto_family vsock_family_ops = {
2237         .family = AF_VSOCK,
2238         .create = vsock_create,
2239         .owner = THIS_MODULE,
2240 };
2241
2242 static long vsock_dev_do_ioctl(struct file *filp,
2243                                unsigned int cmd, void __user *ptr)
2244 {
2245         u32 __user *p = ptr;
2246         u32 cid = VMADDR_CID_ANY;
2247         int retval = 0;
2248
2249         switch (cmd) {
2250         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2251                 /* To be compatible with the VMCI behavior, we prioritize the
2252                  * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2253                  */
2254                 if (transport_g2h)
2255                         cid = transport_g2h->get_local_cid();
2256                 else if (transport_h2g)
2257                         cid = transport_h2g->get_local_cid();
2258
2259                 if (put_user(cid, p) != 0)
2260                         retval = -EFAULT;
2261                 break;
2262
2263         default:
2264                 retval = -ENOIOCTLCMD;
2265         }
2266
2267         return retval;
2268 }
2269
2270 static long vsock_dev_ioctl(struct file *filp,
2271                             unsigned int cmd, unsigned long arg)
2272 {
2273         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2274 }
2275
2276 #ifdef CONFIG_COMPAT
2277 static long vsock_dev_compat_ioctl(struct file *filp,
2278                                    unsigned int cmd, unsigned long arg)
2279 {
2280         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2281 }
2282 #endif
2283
2284 static const struct file_operations vsock_device_ops = {
2285         .owner          = THIS_MODULE,
2286         .unlocked_ioctl = vsock_dev_ioctl,
2287 #ifdef CONFIG_COMPAT
2288         .compat_ioctl   = vsock_dev_compat_ioctl,
2289 #endif
2290         .open           = nonseekable_open,
2291 };
2292
2293 static struct miscdevice vsock_device = {
2294         .name           = "vsock",
2295         .fops           = &vsock_device_ops,
2296 };
2297
2298 static int __init vsock_init(void)
2299 {
2300         int err = 0;
2301
2302         vsock_init_tables();
2303
2304         vsock_proto.owner = THIS_MODULE;
2305         vsock_device.minor = MISC_DYNAMIC_MINOR;
2306         err = misc_register(&vsock_device);
2307         if (err) {
2308                 pr_err("Failed to register misc device\n");
2309                 goto err_reset_transport;
2310         }
2311
2312         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2313         if (err) {
2314                 pr_err("Cannot register vsock protocol\n");
2315                 goto err_deregister_misc;
2316         }
2317
2318         err = sock_register(&vsock_family_ops);
2319         if (err) {
2320                 pr_err("could not register af_vsock (%d) address family: %d\n",
2321                        AF_VSOCK, err);
2322                 goto err_unregister_proto;
2323         }
2324
2325         return 0;
2326
2327 err_unregister_proto:
2328         proto_unregister(&vsock_proto);
2329 err_deregister_misc:
2330         misc_deregister(&vsock_device);
2331 err_reset_transport:
2332         return err;
2333 }
2334
2335 static void __exit vsock_exit(void)
2336 {
2337         misc_deregister(&vsock_device);
2338         sock_unregister(AF_VSOCK);
2339         proto_unregister(&vsock_proto);
2340 }
2341
2342 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2343 {
2344         return vsk->transport;
2345 }
2346 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2347
2348 int vsock_core_register(const struct vsock_transport *t, int features)
2349 {
2350         const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2351         int err = mutex_lock_interruptible(&vsock_register_mutex);
2352
2353         if (err)
2354                 return err;
2355
2356         t_h2g = transport_h2g;
2357         t_g2h = transport_g2h;
2358         t_dgram = transport_dgram;
2359         t_local = transport_local;
2360
2361         if (features & VSOCK_TRANSPORT_F_H2G) {
2362                 if (t_h2g) {
2363                         err = -EBUSY;
2364                         goto err_busy;
2365                 }
2366                 t_h2g = t;
2367         }
2368
2369         if (features & VSOCK_TRANSPORT_F_G2H) {
2370                 if (t_g2h) {
2371                         err = -EBUSY;
2372                         goto err_busy;
2373                 }
2374                 t_g2h = t;
2375         }
2376
2377         if (features & VSOCK_TRANSPORT_F_DGRAM) {
2378                 if (t_dgram) {
2379                         err = -EBUSY;
2380                         goto err_busy;
2381                 }
2382                 t_dgram = t;
2383         }
2384
2385         if (features & VSOCK_TRANSPORT_F_LOCAL) {
2386                 if (t_local) {
2387                         err = -EBUSY;
2388                         goto err_busy;
2389                 }
2390                 t_local = t;
2391         }
2392
2393         transport_h2g = t_h2g;
2394         transport_g2h = t_g2h;
2395         transport_dgram = t_dgram;
2396         transport_local = t_local;
2397
2398 err_busy:
2399         mutex_unlock(&vsock_register_mutex);
2400         return err;
2401 }
2402 EXPORT_SYMBOL_GPL(vsock_core_register);
2403
2404 void vsock_core_unregister(const struct vsock_transport *t)
2405 {
2406         mutex_lock(&vsock_register_mutex);
2407
2408         if (transport_h2g == t)
2409                 transport_h2g = NULL;
2410
2411         if (transport_g2h == t)
2412                 transport_g2h = NULL;
2413
2414         if (transport_dgram == t)
2415                 transport_dgram = NULL;
2416
2417         if (transport_local == t)
2418                 transport_local = NULL;
2419
2420         mutex_unlock(&vsock_register_mutex);
2421 }
2422 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2423
2424 module_init(vsock_init);
2425 module_exit(vsock_exit);
2426
2427 MODULE_AUTHOR("VMware, Inc.");
2428 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2429 MODULE_VERSION("1.0.2.0-k");
2430 MODULE_LICENSE("GPL v2");