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