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