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