net/vmw_vsock/af_vsock.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 /* Implementation notes:
  17  *
  18  * - There are two kinds of sockets: those created by user action (such as
  19  * calling socket(2)) and those created by incoming connection request packets.
  20  *
  21  * - There are two "global" tables, one for bound sockets (sockets that have
  22  * specified an address that they are responsible for) and one for connected
  23  * sockets (sockets that have established a connection with another socket).
  24  * These tables are "global" in that all sockets on the system are placed
  25  * within them. - Note, though, that the bound table contains an extra entry
  26  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  27  * that list. The bound table is used solely for lookup of sockets when packets
  28  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  29  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  30  * sockets out of the bound hash buckets will reduce the chance of collisions
  31  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  32  * socket type in the hash table lookups.
  33  *
  34  * - Sockets created by user action will either be "client" sockets that
  35  * initiate a connection or "server" sockets that listen for connections; we do
  36  * not support simultaneous connects (two "client" sockets connecting).
  37  *
  38  * - "Server" sockets are referred to as listener sockets throughout this
  39  * implementation because they are in the VSOCK_SS_LISTEN state.  When a
  40  * connection request is received (the second kind of socket mentioned above),
  41  * we create a new socket and refer to it as a pending socket.  These pending
  42  * sockets are placed on the pending connection list of the listener socket.
  43  * When future packets are received for the address the listener socket is
  44  * bound to, we check if the source of the packet is from one that has an
  45  * existing pending connection.  If it does, we process the packet for the
  46  * pending socket.  When that socket reaches the connected state, it is removed
  47  * from the listener socket's pending list and enqueued in the listener
  48  * socket's accept queue.  Callers of accept(2) will accept connected sockets
  49  * from the listener socket's accept queue.  If the socket cannot be accepted
  50  * for some reason then it is marked rejected.  Once the connection is
  51  * accepted, it is owned by the user process and the responsibility for cleanup
  52  * falls with that user process.
  53  *
  54  * - It is possible that these pending sockets will never reach the connected
  55  * state; in fact, we may never receive another packet after the connection
  56  * request.  Because of this, we must schedule a cleanup function to run in the
  57  * future, after some amount of time passes where a connection should have been
  58  * established.  This function ensures that the socket is off all lists so it
  59  * cannot be retrieved, then drops all references to the socket so it is cleaned
  60  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  61  * function will also cleanup rejected sockets, those that reach the connected
  62  * state but leave it before they have been accepted.
  63  *
  64  * - Lock ordering for pending or accept queue sockets is:
  65  *
  66  *     lock_sock(listener);
  67  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  68  *
  69  * Using explicit nested locking keeps lockdep happy since normally only one
  70  * lock of a given class may be taken at a time.
  71  *
  72  * - Sockets created by user action will be cleaned up when the user process
  73  * calls close(2), causing our release implementation to be called. Our release
  74  * implementation will perform some cleanup then drop the last reference so our
  75  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  76  * perform additional cleanup that's common for both types of sockets.
  77  *
  78  * - A socket's reference count is what ensures that the structure won't be
  79  * freed.  Each entry in a list (such as the "global" bound and connected tables
  80  * and the listener socket's pending list and connected queue) ensures a
  81  * reference.  When we defer work until process context and pass a socket as our
  82  * argument, we must ensure the reference count is increased to ensure the
  83  * socket isn't freed before the function is run; the deferred function will
  84  * then drop the reference.
  85  */
  86
  87 #include <linux/types.h>
  88 #include <linux/bitops.h>
  89 #include <linux/cred.h>
  90 #include <linux/init.h>
  91 #include <linux/io.h>
  92 #include <linux/kernel.h>
  93 #include <linux/kmod.h>
  94 #include <linux/list.h>
  95 #include <linux/miscdevice.h>
  96 #include <linux/module.h>
  97 #include <linux/mutex.h>
  98 #include <linux/net.h>
  99 #include <linux/poll.h>
 100 #include <linux/random.h>
 101 #include <linux/skbuff.h>
 102 #include <linux/smp.h>
 103 #include <linux/socket.h>
 104 #include <linux/stddef.h>
 105 #include <linux/unistd.h>
 106 #include <linux/wait.h>
 107 #include <linux/workqueue.h>
 108 #include <net/sock.h>
 109 #include <net/af_vsock.h>
 110
 111 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 112 static void vsock_sk_destruct(struct sock *sk);
 113 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 114
 115 /* Protocol family. */
 116 static struct proto vsock_proto = {
 117         .name = "AF_VSOCK",
 118         .owner = THIS_MODULE,
 119         .obj_size = sizeof(struct vsock_sock),
 120 };
 121
 122 /* The default peer timeout indicates how long we will wait for a peer response
 123  * to a control message.
 124  */
 125 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 126
 127 static const struct vsock_transport *transport;
 128 static DEFINE_MUTEX(vsock_register_mutex);
 129
 130 /**** EXPORTS ****/
 131
 132 /* Get the ID of the local context.  This is transport dependent. */
 133
 134 int vm_sockets_get_local_cid(void)
 135 {
 136         return transport->get_local_cid();
 137 }
 138 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 139
 140 /**** UTILS ****/
 141
 142 /* Each bound VSocket is stored in the bind hash table and each connected
 143  * VSocket is stored in the connected hash table.
 144  *
 145  * Unbound sockets are all put on the same list attached to the end of the hash
 146  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 147  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 148  * represents the list that addr hashes to).
 149  *
 150  * Specifically, we initialize the vsock_bind_table array to a size of
 151  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 152  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 153  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 154  * mods with VSOCK_HASH_SIZE to ensure this.
 155  */
 156 #define VSOCK_HASH_SIZE         251
 157 #define MAX_PORT_RETRIES        24
 158
 159 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 160 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 161 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 162
 163 /* XXX This can probably be implemented in a better way. */
 164 #define VSOCK_CONN_HASH(src, dst)                               \
 165         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 166 #define vsock_connected_sockets(src, dst)               \
 167         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 168 #define vsock_connected_sockets_vsk(vsk)                                \
 169         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 170
 171 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 172 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 173 static DEFINE_SPINLOCK(vsock_table_lock);
 174
 175 /* Autobind this socket to the local address if necessary. */
 176 static int vsock_auto_bind(struct vsock_sock *vsk)
 177 {
 178         struct sock *sk = sk_vsock(vsk);
 179         struct sockaddr_vm local_addr;
 180
 181         if (vsock_addr_bound(&vsk->local_addr))
 182                 return 0;
 183         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 184         return __vsock_bind(sk, &local_addr);
 185 }
 186
 187 static void vsock_init_tables(void)
 188 {
 189         int i;
 190
 191         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 192                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 193
 194         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 195                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 196 }
 197
 198 static void __vsock_insert_bound(struct list_head *list,
 199                                  struct vsock_sock *vsk)
 200 {
 201         sock_hold(&vsk->sk);
 202         list_add(&vsk->bound_table, list);
 203 }
 204
 205 static void __vsock_insert_connected(struct list_head *list,
 206                                      struct vsock_sock *vsk)
 207 {
 208         sock_hold(&vsk->sk);
 209         list_add(&vsk->connected_table, list);
 210 }
 211
 212 static void __vsock_remove_bound(struct vsock_sock *vsk)
 213 {
 214         list_del_init(&vsk->bound_table);
 215         sock_put(&vsk->sk);
 216 }
 217
 218 static void __vsock_remove_connected(struct vsock_sock *vsk)
 219 {
 220         list_del_init(&vsk->connected_table);
 221         sock_put(&vsk->sk);
 222 }
 223
 224 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 225 {
 226         struct vsock_sock *vsk;
 227
 228         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 229                 if (addr->svm_port == vsk->local_addr.svm_port)
 230                         return sk_vsock(vsk);
 231
 232         return NULL;
 233 }
 234
 235 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 236                                                   struct sockaddr_vm *dst)
 237 {
 238         struct vsock_sock *vsk;
 239
 240         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 241                             connected_table) {
 242                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 243                     dst->svm_port == vsk->local_addr.svm_port) {
 244                         return sk_vsock(vsk);
 245                 }
 246         }
 247
 248         return NULL;
 249 }
 250
 251 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 252 {
 253         return !list_empty(&vsk->bound_table);
 254 }
 255
 256 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 257 {
 258         return !list_empty(&vsk->connected_table);
 259 }
 260
 261 static void vsock_insert_unbound(struct vsock_sock *vsk)
 262 {
 263         spin_lock_bh(&vsock_table_lock);
 264         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 265         spin_unlock_bh(&vsock_table_lock);
 266 }
 267
 268 void vsock_insert_connected(struct vsock_sock *vsk)
 269 {
 270         struct list_head *list = vsock_connected_sockets(
 271                 &vsk->remote_addr, &vsk->local_addr);
 272
 273         spin_lock_bh(&vsock_table_lock);
 274         __vsock_insert_connected(list, vsk);
 275         spin_unlock_bh(&vsock_table_lock);
 276 }
 277 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 278
 279 void vsock_remove_bound(struct vsock_sock *vsk)
 280 {
 281         spin_lock_bh(&vsock_table_lock);
 282         if (__vsock_in_bound_table(vsk))
 283                 __vsock_remove_bound(vsk);
 284         spin_unlock_bh(&vsock_table_lock);
 285 }
 286 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 287
 288 void vsock_remove_connected(struct vsock_sock *vsk)
 289 {
 290         spin_lock_bh(&vsock_table_lock);
 291         if (__vsock_in_connected_table(vsk))
 292                 __vsock_remove_connected(vsk);
 293         spin_unlock_bh(&vsock_table_lock);
 294 }
 295 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 296
 297 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 298 {
 299         struct sock *sk;
 300
 301         spin_lock_bh(&vsock_table_lock);
 302         sk = __vsock_find_bound_socket(addr);
 303         if (sk)
 304                 sock_hold(sk);
 305
 306         spin_unlock_bh(&vsock_table_lock);
 307
 308         return sk;
 309 }
 310 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 311
 312 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 313                                          struct sockaddr_vm *dst)
 314 {
 315         struct sock *sk;
 316
 317         spin_lock_bh(&vsock_table_lock);
 318         sk = __vsock_find_connected_socket(src, dst);
 319         if (sk)
 320                 sock_hold(sk);
 321
 322         spin_unlock_bh(&vsock_table_lock);
 323
 324         return sk;
 325 }
 326 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 327
 328 void vsock_remove_sock(struct vsock_sock *vsk)
 329 {
 330         vsock_remove_bound(vsk);
 331         vsock_remove_connected(vsk);
 332 }
 333 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 334
 335 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 336 {
 337         int i;
 338
 339         spin_lock_bh(&vsock_table_lock);
 340
 341         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 342                 struct vsock_sock *vsk;
 343                 list_for_each_entry(vsk, &vsock_connected_table[i],
 344                                     connected_table)
 345                         fn(sk_vsock(vsk));
 346         }
 347
 348         spin_unlock_bh(&vsock_table_lock);
 349 }
 350 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 351
 352 void vsock_add_pending(struct sock *listener, struct sock *pending)
 353 {
 354         struct vsock_sock *vlistener;
 355         struct vsock_sock *vpending;
 356
 357         vlistener = vsock_sk(listener);
 358         vpending = vsock_sk(pending);
 359
 360         sock_hold(pending);
 361         sock_hold(listener);
 362         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 363 }
 364 EXPORT_SYMBOL_GPL(vsock_add_pending);
 365
 366 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 367 {
 368         struct vsock_sock *vpending = vsock_sk(pending);
 369
 370         list_del_init(&vpending->pending_links);
 371         sock_put(listener);
 372         sock_put(pending);
 373 }
 374 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 375
 376 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 377 {
 378         struct vsock_sock *vlistener;
 379         struct vsock_sock *vconnected;
 380
 381         vlistener = vsock_sk(listener);
 382         vconnected = vsock_sk(connected);
 383
 384         sock_hold(connected);
 385         sock_hold(listener);
 386         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 387 }
 388 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 389
 390 static struct sock *vsock_dequeue_accept(struct sock *listener)
 391 {
 392         struct vsock_sock *vlistener;
 393         struct vsock_sock *vconnected;
 394
 395         vlistener = vsock_sk(listener);
 396
 397         if (list_empty(&vlistener->accept_queue))
 398                 return NULL;
 399
 400         vconnected = list_entry(vlistener->accept_queue.next,
 401                                 struct vsock_sock, accept_queue);
 402
 403         list_del_init(&vconnected->accept_queue);
 404         sock_put(listener);
 405         /* The caller will need a reference on the connected socket so we let
 406          * it call sock_put().
 407          */
 408
 409         return sk_vsock(vconnected);
 410 }
 411
 412 static bool vsock_is_accept_queue_empty(struct sock *sk)
 413 {
 414         struct vsock_sock *vsk = vsock_sk(sk);
 415         return list_empty(&vsk->accept_queue);
 416 }
 417
 418 static bool vsock_is_pending(struct sock *sk)
 419 {
 420         struct vsock_sock *vsk = vsock_sk(sk);
 421         return !list_empty(&vsk->pending_links);
 422 }
 423
 424 static int vsock_send_shutdown(struct sock *sk, int mode)
 425 {
 426         return transport->shutdown(vsock_sk(sk), mode);
 427 }
 428
 429 static void vsock_pending_work(struct work_struct *work)
 430 {
 431         struct sock *sk;
 432         struct sock *listener;
 433         struct vsock_sock *vsk;
 434         bool cleanup;
 435
 436         vsk = container_of(work, struct vsock_sock, pending_work.work);
 437         sk = sk_vsock(vsk);
 438         listener = vsk->listener;
 439         cleanup = true;
 440
 441         lock_sock(listener);
 442         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 443
 444         if (vsock_is_pending(sk)) {
 445                 vsock_remove_pending(listener, sk);
 446
 447                 listener->sk_ack_backlog--;
 448         } else if (!vsk->rejected) {
 449                 /* We are not on the pending list and accept() did not reject
 450                  * us, so we must have been accepted by our user process.  We
 451                  * just need to drop our references to the sockets and be on
 452                  * our way.
 453                  */
 454                 cleanup = false;
 455                 goto out;
 456         }
 457
 458         /* We need to remove ourself from the global connected sockets list so
 459          * incoming packets can't find this socket, and to reduce the reference
 460          * count.
 461          */
 462         vsock_remove_connected(vsk);
 463
 464         sk->sk_state = SS_FREE;
 465
 466 out:
 467         release_sock(sk);
 468         release_sock(listener);
 469         if (cleanup)
 470                 sock_put(sk);
 471
 472         sock_put(sk);
 473         sock_put(listener);
 474 }
 475
 476 /**** SOCKET OPERATIONS ****/
 477
 478 static int __vsock_bind_stream(struct vsock_sock *vsk,
 479                                struct sockaddr_vm *addr)
 480 {
 481         static u32 port = 0;
 482         struct sockaddr_vm new_addr;
 483
 484         if (!port)
 485                 port = LAST_RESERVED_PORT + 1 +
 486                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 487
 488         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 489
 490         if (addr->svm_port == VMADDR_PORT_ANY) {
 491                 bool found = false;
 492                 unsigned int i;
 493
 494                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 495                         if (port <= LAST_RESERVED_PORT)
 496                                 port = LAST_RESERVED_PORT + 1;
 497
 498                         new_addr.svm_port = port++;
 499
 500                         if (!__vsock_find_bound_socket(&new_addr)) {
 501                                 found = true;
 502                                 break;
 503                         }
 504                 }
 505
 506                 if (!found)
 507                         return -EADDRNOTAVAIL;
 508         } else {
 509                 /* If port is in reserved range, ensure caller
 510                  * has necessary privileges.
 511                  */
 512                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 513                     !capable(CAP_NET_BIND_SERVICE)) {
 514                         return -EACCES;
 515                 }
 516
 517                 if (__vsock_find_bound_socket(&new_addr))
 518                         return -EADDRINUSE;
 519         }
 520
 521         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 522
 523         /* Remove stream sockets from the unbound list and add them to the hash
 524          * table for easy lookup by its address.  The unbound list is simply an
 525          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 526          */
 527         __vsock_remove_bound(vsk);
 528         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 529
 530         return 0;
 531 }
 532
 533 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 534                               struct sockaddr_vm *addr)
 535 {
 536         return transport->dgram_bind(vsk, addr);
 537 }
 538
 539 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 540 {
 541         struct vsock_sock *vsk = vsock_sk(sk);
 542         u32 cid;
 543         int retval;
 544
 545         /* First ensure this socket isn't already bound. */
 546         if (vsock_addr_bound(&vsk->local_addr))
 547                 return -EINVAL;
 548
 549         /* Now bind to the provided address or select appropriate values if
 550          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 551          * like AF_INET prevents binding to a non-local IP address (in most
 552          * cases), we only allow binding to the local CID.
 553          */
 554         cid = transport->get_local_cid();
 555         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 556                 return -EADDRNOTAVAIL;
 557
 558         switch (sk->sk_socket->type) {
 559         case SOCK_STREAM:
 560                 spin_lock_bh(&vsock_table_lock);
 561                 retval = __vsock_bind_stream(vsk, addr);
 562                 spin_unlock_bh(&vsock_table_lock);
 563                 break;
 564
 565         case SOCK_DGRAM:
 566                 retval = __vsock_bind_dgram(vsk, addr);
 567                 break;
 568
 569         default:
 570                 retval = -EINVAL;
 571                 break;
 572         }
 573
 574         return retval;
 575 }
 576
 577 static void vsock_connect_timeout(struct work_struct *work);
 578
 579 struct sock *__vsock_create(struct net *net,
 580                             struct socket *sock,
 581                             struct sock *parent,
 582                             gfp_t priority,
 583                             unsigned short type,
 584                             int kern)
 585 {
 586         struct sock *sk;
 587         struct vsock_sock *psk;
 588         struct vsock_sock *vsk;
 589
 590         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 591         if (!sk)
 592                 return NULL;
 593
 594         sock_init_data(sock, sk);
 595
 596         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 597          * non-NULL. We make sure that our sockets always have a type by
 598          * setting it here if needed.
 599          */
 600         if (!sock)
 601                 sk->sk_type = type;
 602
 603         vsk = vsock_sk(sk);
 604         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 605         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 606
 607         sk->sk_destruct = vsock_sk_destruct;
 608         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 609         sk->sk_state = 0;
 610         sock_reset_flag(sk, SOCK_DONE);
 611
 612         INIT_LIST_HEAD(&vsk->bound_table);
 613         INIT_LIST_HEAD(&vsk->connected_table);
 614         vsk->listener = NULL;
 615         INIT_LIST_HEAD(&vsk->pending_links);
 616         INIT_LIST_HEAD(&vsk->accept_queue);
 617         vsk->rejected = false;
 618         vsk->sent_request = false;
 619         vsk->ignore_connecting_rst = false;
 620         vsk->peer_shutdown = 0;
 621         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 622         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 623
 624         psk = parent ? vsock_sk(parent) : NULL;
 625         if (parent) {
 626                 vsk->trusted = psk->trusted;
 627                 vsk->owner = get_cred(psk->owner);
 628                 vsk->connect_timeout = psk->connect_timeout;
 629                 security_sk_clone(parent, sk);
 630         } else {
 631                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 632                 vsk->owner = get_current_cred();
 633                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 634         }
 635
 636         if (transport->init(vsk, psk) < 0) {
 637                 sk_free(sk);
 638                 return NULL;
 639         }
 640
 641         if (sock)
 642                 vsock_insert_unbound(vsk);
 643
 644         return sk;
 645 }
 646 EXPORT_SYMBOL_GPL(__vsock_create);
 647
 648 static void __vsock_release(struct sock *sk)
 649 {
 650         if (sk) {
 651                 struct sk_buff *skb;
 652                 struct sock *pending;
 653                 struct vsock_sock *vsk;
 654
 655                 vsk = vsock_sk(sk);
 656                 pending = NULL; /* Compiler warning. */
 657
 658                 transport->release(vsk);
 659
 660                 lock_sock(sk);
 661                 sock_orphan(sk);
 662                 sk->sk_shutdown = SHUTDOWN_MASK;
 663
 664                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 665                         kfree_skb(skb);
 666
 667                 /* Clean up any sockets that never were accepted. */
 668                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 669                         __vsock_release(pending);
 670                         sock_put(pending);
 671                 }
 672
 673                 release_sock(sk);
 674                 sock_put(sk);
 675         }
 676 }
 677
 678 static void vsock_sk_destruct(struct sock *sk)
 679 {
 680         struct vsock_sock *vsk = vsock_sk(sk);
 681
 682         transport->destruct(vsk);
 683
 684         /* When clearing these addresses, there's no need to set the family and
 685          * possibly register the address family with the kernel.
 686          */
 687         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 688         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 689
 690         put_cred(vsk->owner);
 691 }
 692
 693 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 694 {
 695         int err;
 696
 697         err = sock_queue_rcv_skb(sk, skb);
 698         if (err)
 699                 kfree_skb(skb);
 700
 701         return err;
 702 }
 703
 704 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 705 {
 706         return transport->stream_has_data(vsk);
 707 }
 708 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 709
 710 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 711 {
 712         return transport->stream_has_space(vsk);
 713 }
 714 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 715
 716 static int vsock_release(struct socket *sock)
 717 {
 718         __vsock_release(sock->sk);
 719         sock->sk = NULL;
 720         sock->state = SS_FREE;
 721
 722         return 0;
 723 }
 724
 725 static int
 726 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 727 {
 728         int err;
 729         struct sock *sk;
 730         struct sockaddr_vm *vm_addr;
 731
 732         sk = sock->sk;
 733
 734         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 735                 return -EINVAL;
 736
 737         lock_sock(sk);
 738         err = __vsock_bind(sk, vm_addr);
 739         release_sock(sk);
 740
 741         return err;
 742 }
 743
 744 static int vsock_getname(struct socket *sock,
 745                          struct sockaddr *addr, int *addr_len, int peer)
 746 {
 747         int err;
 748         struct sock *sk;
 749         struct vsock_sock *vsk;
 750         struct sockaddr_vm *vm_addr;
 751
 752         sk = sock->sk;
 753         vsk = vsock_sk(sk);
 754         err = 0;
 755
 756         lock_sock(sk);
 757
 758         if (peer) {
 759                 if (sock->state != SS_CONNECTED) {
 760                         err = -ENOTCONN;
 761                         goto out;
 762                 }
 763                 vm_addr = &vsk->remote_addr;
 764         } else {
 765                 vm_addr = &vsk->local_addr;
 766         }
 767
 768         if (!vm_addr) {
 769                 err = -EINVAL;
 770                 goto out;
 771         }
 772
 773         /* sys_getsockname() and sys_getpeername() pass us a
 774          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 775          * that macro is defined in socket.c instead of .h, so we hardcode its
 776          * value here.
 777          */
 778         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 779         memcpy(addr, vm_addr, sizeof(*vm_addr));
 780         *addr_len = sizeof(*vm_addr);
 781
 782 out:
 783         release_sock(sk);
 784         return err;
 785 }
 786
 787 static int vsock_shutdown(struct socket *sock, int mode)
 788 {
 789         int err;
 790         struct sock *sk;
 791
 792         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 793          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 794          * here like the other address families do.  Note also that the
 795          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 796          * which is what we want.
 797          */
 798         mode++;
 799
 800         if ((mode & ~SHUTDOWN_MASK) || !mode)
 801                 return -EINVAL;
 802
 803         /* If this is a STREAM socket and it is not connected then bail out
 804          * immediately.  If it is a DGRAM socket then we must first kick the
 805          * socket so that it wakes up from any sleeping calls, for example
 806          * recv(), and then afterwards return the error.
 807          */
 808
 809         sk = sock->sk;
 810
 811         lock_sock(sk);
 812         if (sock->state == SS_UNCONNECTED) {
 813                 err = -ENOTCONN;
 814                 if (sk->sk_type == SOCK_STREAM)
 815                         goto out;
 816         } else {
 817                 sock->state = SS_DISCONNECTING;
 818                 err = 0;
 819         }
 820
 821         /* Receive and send shutdowns are treated alike. */
 822         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 823         if (mode) {
 824                 sk->sk_shutdown |= mode;
 825                 sk->sk_state_change(sk);
 826
 827                 if (sk->sk_type == SOCK_STREAM) {
 828                         sock_reset_flag(sk, SOCK_DONE);
 829                         vsock_send_shutdown(sk, mode);
 830                 }
 831         }
 832
 833 out:
 834         release_sock(sk);
 835         return err;
 836 }
 837
 838 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 839                                poll_table *wait)
 840 {
 841         struct sock *sk;
 842         unsigned int mask;
 843         struct vsock_sock *vsk;
 844
 845         sk = sock->sk;
 846         vsk = vsock_sk(sk);
 847
 848         poll_wait(file, sk_sleep(sk), wait);
 849         mask = 0;
 850
 851         if (sk->sk_err)
 852                 /* Signify that there has been an error on this socket. */
 853                 mask |= POLLERR;
 854
 855         /* INET sockets treat local write shutdown and peer write shutdown as a
 856          * case of POLLHUP set.
 857          */
 858         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 859             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 860              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 861                 mask |= POLLHUP;
 862         }
 863
 864         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 865             vsk->peer_shutdown & SEND_SHUTDOWN) {
 866                 mask |= POLLRDHUP;
 867         }
 868
 869         if (sock->type == SOCK_DGRAM) {
 870                 /* For datagram sockets we can read if there is something in
 871                  * the queue and write as long as the socket isn't shutdown for
 872                  * sending.
 873                  */
 874                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 875                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 876                         mask |= POLLIN | POLLRDNORM;
 877                 }
 878
 879                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 880                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 881
 882         } else if (sock->type == SOCK_STREAM) {
 883                 lock_sock(sk);
 884
 885                 /* Listening sockets that have connections in their accept
 886                  * queue can be read.
 887                  */
 888                 if (sk->sk_state == VSOCK_SS_LISTEN
 889                     && !vsock_is_accept_queue_empty(sk))
 890                         mask |= POLLIN | POLLRDNORM;
 891
 892                 /* If there is something in the queue then we can read. */
 893                 if (transport->stream_is_active(vsk) &&
 894                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 895                         bool data_ready_now = false;
 896                         int ret = transport->notify_poll_in(
 897                                         vsk, 1, &data_ready_now);
 898                         if (ret < 0) {
 899                                 mask |= POLLERR;
 900                         } else {
 901                                 if (data_ready_now)
 902                                         mask |= POLLIN | POLLRDNORM;
 903
 904                         }
 905                 }
 906
 907                 /* Sockets whose connections have been closed, reset, or
 908                  * terminated should also be considered read, and we check the
 909                  * shutdown flag for that.
 910                  */
 911                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 912                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 913                         mask |= POLLIN | POLLRDNORM;
 914                 }
 915
 916                 /* Connected sockets that can produce data can be written. */
 917                 if (sk->sk_state == SS_CONNECTED) {
 918                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 919                                 bool space_avail_now = false;
 920                                 int ret = transport->notify_poll_out(
 921                                                 vsk, 1, &space_avail_now);
 922                                 if (ret < 0) {
 923                                         mask |= POLLERR;
 924                                 } else {
 925                                         if (space_avail_now)
 926                                                 /* Remove POLLWRBAND since INET
 927                                                  * sockets are not setting it.
 928                                                  */
 929                                                 mask |= POLLOUT | POLLWRNORM;
 930
 931                                 }
 932                         }
 933                 }
 934
 935                 /* Simulate INET socket poll behaviors, which sets
 936                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 937                  * but local send is not shutdown.
 938                  */
 939                 if (sk->sk_state == SS_UNCONNECTED) {
 940                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 941                                 mask |= POLLOUT | POLLWRNORM;
 942
 943                 }
 944
 945                 release_sock(sk);
 946         }
 947
 948         return mask;
 949 }
 950
 951 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 952                                size_t len)
 953 {
 954         int err;
 955         struct sock *sk;
 956         struct vsock_sock *vsk;
 957         struct sockaddr_vm *remote_addr;
 958
 959         if (msg->msg_flags & MSG_OOB)
 960                 return -EOPNOTSUPP;
 961
 962         /* For now, MSG_DONTWAIT is always assumed... */
 963         err = 0;
 964         sk = sock->sk;
 965         vsk = vsock_sk(sk);
 966
 967         lock_sock(sk);
 968
 969         err = vsock_auto_bind(vsk);
 970         if (err)
 971                 goto out;
 972
 973
 974         /* If the provided message contains an address, use that.  Otherwise
 975          * fall back on the socket's remote handle (if it has been connected).
 976          */
 977         if (msg->msg_name &&
 978             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 979                             &remote_addr) == 0) {
 980                 /* Ensure this address is of the right type and is a valid
 981                  * destination.
 982                  */
 983
 984                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 985                         remote_addr->svm_cid = transport->get_local_cid();
 986
 987                 if (!vsock_addr_bound(remote_addr)) {
 988                         err = -EINVAL;
 989                         goto out;
 990                 }
 991         } else if (sock->state == SS_CONNECTED) {
 992                 remote_addr = &vsk->remote_addr;
 993
 994                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 995                         remote_addr->svm_cid = transport->get_local_cid();
 996
 997                 /* XXX Should connect() or this function ensure remote_addr is
 998                  * bound?
 999                  */
1000                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1001                         err = -EINVAL;
1002                         goto out;
1003                 }
1004         } else {
1005                 err = -EINVAL;
1006                 goto out;
1007         }
1008
1009         if (!transport->dgram_allow(remote_addr->svm_cid,
1010                                     remote_addr->svm_port)) {
1011                 err = -EINVAL;
1012                 goto out;
1013         }
1014
1015         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1016
1017 out:
1018         release_sock(sk);
1019         return err;
1020 }
1021
1022 static int vsock_dgram_connect(struct socket *sock,
1023                                struct sockaddr *addr, int addr_len, int flags)
1024 {
1025         int err;
1026         struct sock *sk;
1027         struct vsock_sock *vsk;
1028         struct sockaddr_vm *remote_addr;
1029
1030         sk = sock->sk;
1031         vsk = vsock_sk(sk);
1032
1033         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1034         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1035                 lock_sock(sk);
1036                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1037                                 VMADDR_PORT_ANY);
1038                 sock->state = SS_UNCONNECTED;
1039                 release_sock(sk);
1040                 return 0;
1041         } else if (err != 0)
1042                 return -EINVAL;
1043
1044         lock_sock(sk);
1045
1046         err = vsock_auto_bind(vsk);
1047         if (err)
1048                 goto out;
1049
1050         if (!transport->dgram_allow(remote_addr->svm_cid,
1051                                     remote_addr->svm_port)) {
1052                 err = -EINVAL;
1053                 goto out;
1054         }
1055
1056         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1057         sock->state = SS_CONNECTED;
1058
1059 out:
1060         release_sock(sk);
1061         return err;
1062 }
1063
1064 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1065                                size_t len, int flags)
1066 {
1067         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1068 }
1069
1070 static const struct proto_ops vsock_dgram_ops = {
1071         .family = PF_VSOCK,
1072         .owner = THIS_MODULE,
1073         .release = vsock_release,
1074         .bind = vsock_bind,
1075         .connect = vsock_dgram_connect,
1076         .socketpair = sock_no_socketpair,
1077         .accept = sock_no_accept,
1078         .getname = vsock_getname,
1079         .poll = vsock_poll,
1080         .ioctl = sock_no_ioctl,
1081         .listen = sock_no_listen,
1082         .shutdown = vsock_shutdown,
1083         .setsockopt = sock_no_setsockopt,
1084         .getsockopt = sock_no_getsockopt,
1085         .sendmsg = vsock_dgram_sendmsg,
1086         .recvmsg = vsock_dgram_recvmsg,
1087         .mmap = sock_no_mmap,
1088         .sendpage = sock_no_sendpage,
1089 };
1090
1091 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1092 {
1093         if (!transport->cancel_pkt)
1094                 return -EOPNOTSUPP;
1095
1096         return transport->cancel_pkt(vsk);
1097 }
1098
1099 static void vsock_connect_timeout(struct work_struct *work)
1100 {
1101         struct sock *sk;
1102         struct vsock_sock *vsk;
1103
1104         vsk = container_of(work, struct vsock_sock, connect_work.work);
1105         sk = sk_vsock(vsk);
1106
1107         lock_sock(sk);
1108         if (sk->sk_state == SS_CONNECTING &&
1109             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1110                 sk->sk_state = SS_UNCONNECTED;
1111                 sk->sk_err = ETIMEDOUT;
1112                 sk->sk_error_report(sk);
1113                 vsock_transport_cancel_pkt(vsk);
1114         }
1115         release_sock(sk);
1116
1117         sock_put(sk);
1118 }
1119
1120 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1121                                 int addr_len, int flags)
1122 {
1123         int err;
1124         struct sock *sk;
1125         struct vsock_sock *vsk;
1126         struct sockaddr_vm *remote_addr;
1127         long timeout;
1128         DEFINE_WAIT(wait);
1129
1130         err = 0;
1131         sk = sock->sk;
1132         vsk = vsock_sk(sk);
1133
1134         lock_sock(sk);
1135
1136         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1137         switch (sock->state) {
1138         case SS_CONNECTED:
1139                 err = -EISCONN;
1140                 goto out;
1141         case SS_DISCONNECTING:
1142                 err = -EINVAL;
1143                 goto out;
1144         case SS_CONNECTING:
1145                 /* This continues on so we can move sock into the SS_CONNECTED
1146                  * state once the connection has completed (at which point err
1147                  * will be set to zero also).  Otherwise, we will either wait
1148                  * for the connection or return -EALREADY should this be a
1149                  * non-blocking call.
1150                  */
1151                 err = -EALREADY;
1152                 if (flags & O_NONBLOCK)
1153                         goto out;
1154                 break;
1155         default:
1156                 if ((sk->sk_state == VSOCK_SS_LISTEN) ||
1157                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1158                         err = -EINVAL;
1159                         goto out;
1160                 }
1161
1162                 /* The hypervisor and well-known contexts do not have socket
1163                  * endpoints.
1164                  */
1165                 if (!transport->stream_allow(remote_addr->svm_cid,
1166                                              remote_addr->svm_port)) {
1167                         err = -ENETUNREACH;
1168                         goto out;
1169                 }
1170
1171                 /* Set the remote address that we are connecting to. */
1172                 memcpy(&vsk->remote_addr, remote_addr,
1173                        sizeof(vsk->remote_addr));
1174
1175                 err = vsock_auto_bind(vsk);
1176                 if (err)
1177                         goto out;
1178
1179                 sk->sk_state = SS_CONNECTING;
1180
1181                 err = transport->connect(vsk);
1182                 if (err < 0)
1183                         goto out;
1184
1185                 /* Mark sock as connecting and set the error code to in
1186                  * progress in case this is a non-blocking connect.
1187                  */
1188                 sock->state = SS_CONNECTING;
1189                 err = -EINPROGRESS;
1190         }
1191
1192         /* The receive path will handle all communication until we are able to
1193          * enter the connected state.  Here we wait for the connection to be
1194          * completed or a notification of an error.
1195          */
1196         timeout = vsk->connect_timeout;
1197         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1198
1199         while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1200                 if (flags & O_NONBLOCK) {
1201                         /* If we're not going to block, we schedule a timeout
1202                          * function to generate a timeout on the connection
1203                          * attempt, in case the peer doesn't respond in a
1204                          * timely manner. We hold on to the socket until the
1205                          * timeout fires.
1206                          */
1207                         sock_hold(sk);
1208
1209                         /* If the timeout function is already scheduled,
1210                          * reschedule it, then ungrab the socket refcount to
1211                          * keep it balanced.
1212                          */
1213                         if (mod_delayed_work(system_wq, &vsk->connect_work,
1214                                              timeout))
1215                                 sock_put(sk);
1216
1217                         /* Skip ahead to preserve error code set above. */
1218                         goto out_wait;
1219                 }
1220
1221                 release_sock(sk);
1222                 timeout = schedule_timeout(timeout);
1223                 lock_sock(sk);
1224
1225                 if (signal_pending(current)) {
1226                         err = sock_intr_errno(timeout);
1227                         sk->sk_state = SS_UNCONNECTED;
1228                         sock->state = SS_UNCONNECTED;
1229                         vsock_transport_cancel_pkt(vsk);
1230                         vsock_remove_connected(vsk);
1231                         goto out_wait;
1232                 } else if (timeout == 0) {
1233                         err = -ETIMEDOUT;
1234                         sk->sk_state = SS_UNCONNECTED;
1235                         sock->state = SS_UNCONNECTED;
1236                         vsock_transport_cancel_pkt(vsk);
1237                         goto out_wait;
1238                 }
1239
1240                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1241         }
1242
1243         if (sk->sk_err) {
1244                 err = -sk->sk_err;
1245                 sk->sk_state = SS_UNCONNECTED;
1246                 sock->state = SS_UNCONNECTED;
1247         } else {
1248                 err = 0;
1249         }
1250
1251 out_wait:
1252         finish_wait(sk_sleep(sk), &wait);
1253 out:
1254         release_sock(sk);
1255         return err;
1256 }
1257
1258 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1259 {
1260         struct sock *listener;
1261         int err;
1262         struct sock *connected;
1263         struct vsock_sock *vconnected;
1264         long timeout;
1265         DEFINE_WAIT(wait);
1266
1267         err = 0;
1268         listener = sock->sk;
1269
1270         lock_sock(listener);
1271
1272         if (sock->type != SOCK_STREAM) {
1273                 err = -EOPNOTSUPP;
1274                 goto out;
1275         }
1276
1277         if (listener->sk_state != VSOCK_SS_LISTEN) {
1278                 err = -EINVAL;
1279                 goto out;
1280         }
1281
1282         /* Wait for children sockets to appear; these are the new sockets
1283          * created upon connection establishment.
1284          */
1285         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1286         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1287
1288         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1289                listener->sk_err == 0) {
1290                 release_sock(listener);
1291                 timeout = schedule_timeout(timeout);
1292                 finish_wait(sk_sleep(listener), &wait);
1293                 lock_sock(listener);
1294
1295                 if (signal_pending(current)) {
1296                         err = sock_intr_errno(timeout);
1297                         goto out;
1298                 } else if (timeout == 0) {
1299                         err = -EAGAIN;
1300                         goto out;
1301                 }
1302
1303                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1304         }
1305         finish_wait(sk_sleep(listener), &wait);
1306
1307         if (listener->sk_err)
1308                 err = -listener->sk_err;
1309
1310         if (connected) {
1311                 listener->sk_ack_backlog--;
1312
1313                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1314                 vconnected = vsock_sk(connected);
1315
1316                 /* If the listener socket has received an error, then we should
1317                  * reject this socket and return.  Note that we simply mark the
1318                  * socket rejected, drop our reference, and let the cleanup
1319                  * function handle the cleanup; the fact that we found it in
1320                  * the listener's accept queue guarantees that the cleanup
1321                  * function hasn't run yet.
1322                  */
1323                 if (err) {
1324                         vconnected->rejected = true;
1325                 } else {
1326                         newsock->state = SS_CONNECTED;
1327                         sock_graft(connected, newsock);
1328                 }
1329
1330                 release_sock(connected);
1331                 sock_put(connected);
1332         }
1333
1334 out:
1335         release_sock(listener);
1336         return err;
1337 }
1338
1339 static int vsock_listen(struct socket *sock, int backlog)
1340 {
1341         int err;
1342         struct sock *sk;
1343         struct vsock_sock *vsk;
1344
1345         sk = sock->sk;
1346
1347         lock_sock(sk);
1348
1349         if (sock->type != SOCK_STREAM) {
1350                 err = -EOPNOTSUPP;
1351                 goto out;
1352         }
1353
1354         if (sock->state != SS_UNCONNECTED) {
1355                 err = -EINVAL;
1356                 goto out;
1357         }
1358
1359         vsk = vsock_sk(sk);
1360
1361         if (!vsock_addr_bound(&vsk->local_addr)) {
1362                 err = -EINVAL;
1363                 goto out;
1364         }
1365
1366         sk->sk_max_ack_backlog = backlog;
1367         sk->sk_state = VSOCK_SS_LISTEN;
1368
1369         err = 0;
1370
1371 out:
1372         release_sock(sk);
1373         return err;
1374 }
1375
1376 static int vsock_stream_setsockopt(struct socket *sock,
1377                                    int level,
1378                                    int optname,
1379                                    char __user *optval,
1380                                    unsigned int optlen)
1381 {
1382         int err;
1383         struct sock *sk;
1384         struct vsock_sock *vsk;
1385         u64 val;
1386
1387         if (level != AF_VSOCK)
1388                 return -ENOPROTOOPT;
1389
1390 #define COPY_IN(_v)                                       \
1391         do {                                              \
1392                 if (optlen < sizeof(_v)) {                \
1393                         err = -EINVAL;                    \
1394                         goto exit;                        \
1395                 }                                         \
1396                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1397                         err = -EFAULT;                                  \
1398                         goto exit;                                      \
1399                 }                                                       \
1400         } while (0)
1401
1402         err = 0;
1403         sk = sock->sk;
1404         vsk = vsock_sk(sk);
1405
1406         lock_sock(sk);
1407
1408         switch (optname) {
1409         case SO_VM_SOCKETS_BUFFER_SIZE:
1410                 COPY_IN(val);
1411                 transport->set_buffer_size(vsk, val);
1412                 break;
1413
1414         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1415                 COPY_IN(val);
1416                 transport->set_max_buffer_size(vsk, val);
1417                 break;
1418
1419         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1420                 COPY_IN(val);
1421                 transport->set_min_buffer_size(vsk, val);
1422                 break;
1423
1424         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1425                 struct timeval tv;
1426                 COPY_IN(tv);
1427                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1428                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1429                         vsk->connect_timeout = tv.tv_sec * HZ +
1430                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1431                         if (vsk->connect_timeout == 0)
1432                                 vsk->connect_timeout =
1433                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1434
1435                 } else {
1436                         err = -ERANGE;
1437                 }
1438                 break;
1439         }
1440
1441         default:
1442                 err = -ENOPROTOOPT;
1443                 break;
1444         }
1445
1446 #undef COPY_IN
1447
1448 exit:
1449         release_sock(sk);
1450         return err;
1451 }
1452
1453 static int vsock_stream_getsockopt(struct socket *sock,
1454                                    int level, int optname,
1455                                    char __user *optval,
1456                                    int __user *optlen)
1457 {
1458         int err;
1459         int len;
1460         struct sock *sk;
1461         struct vsock_sock *vsk;
1462         u64 val;
1463
1464         if (level != AF_VSOCK)
1465                 return -ENOPROTOOPT;
1466
1467         err = get_user(len, optlen);
1468         if (err != 0)
1469                 return err;
1470
1471 #define COPY_OUT(_v)                            \
1472         do {                                    \
1473                 if (len < sizeof(_v))           \
1474                         return -EINVAL;         \
1475                                                 \
1476                 len = sizeof(_v);               \
1477                 if (copy_to_user(optval, &_v, len) != 0)        \
1478                         return -EFAULT;                         \
1479                                                                 \
1480         } while (0)
1481
1482         err = 0;
1483         sk = sock->sk;
1484         vsk = vsock_sk(sk);
1485
1486         switch (optname) {
1487         case SO_VM_SOCKETS_BUFFER_SIZE:
1488                 val = transport->get_buffer_size(vsk);
1489                 COPY_OUT(val);
1490                 break;
1491
1492         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1493                 val = transport->get_max_buffer_size(vsk);
1494                 COPY_OUT(val);
1495                 break;
1496
1497         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1498                 val = transport->get_min_buffer_size(vsk);
1499                 COPY_OUT(val);
1500                 break;
1501
1502         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1503                 struct timeval tv;
1504                 tv.tv_sec = vsk->connect_timeout / HZ;
1505                 tv.tv_usec =
1506                     (vsk->connect_timeout -
1507                      tv.tv_sec * HZ) * (1000000 / HZ);
1508                 COPY_OUT(tv);
1509                 break;
1510         }
1511         default:
1512                 return -ENOPROTOOPT;
1513         }
1514
1515         err = put_user(len, optlen);
1516         if (err != 0)
1517                 return -EFAULT;
1518
1519 #undef COPY_OUT
1520
1521         return 0;
1522 }
1523
1524 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1525                                 size_t len)
1526 {
1527         struct sock *sk;
1528         struct vsock_sock *vsk;
1529         ssize_t total_written;
1530         long timeout;
1531         int err;
1532         struct vsock_transport_send_notify_data send_data;
1533         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1534
1535         sk = sock->sk;
1536         vsk = vsock_sk(sk);
1537         total_written = 0;
1538         err = 0;
1539
1540         if (msg->msg_flags & MSG_OOB)
1541                 return -EOPNOTSUPP;
1542
1543         lock_sock(sk);
1544
1545         /* Callers should not provide a destination with stream sockets. */
1546         if (msg->msg_namelen) {
1547                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1548                 goto out;
1549         }
1550
1551         /* Send data only if both sides are not shutdown in the direction. */
1552         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1553             vsk->peer_shutdown & RCV_SHUTDOWN) {
1554                 err = -EPIPE;
1555                 goto out;
1556         }
1557
1558         if (sk->sk_state != SS_CONNECTED ||
1559             !vsock_addr_bound(&vsk->local_addr)) {
1560                 err = -ENOTCONN;
1561                 goto out;
1562         }
1563
1564         if (!vsock_addr_bound(&vsk->remote_addr)) {
1565                 err = -EDESTADDRREQ;
1566                 goto out;
1567         }
1568
1569         /* Wait for room in the produce queue to enqueue our user's data. */
1570         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1571
1572         err = transport->notify_send_init(vsk, &send_data);
1573         if (err < 0)
1574                 goto out;
1575
1576         while (total_written < len) {
1577                 ssize_t written;
1578
1579                 add_wait_queue(sk_sleep(sk), &wait);
1580                 while (vsock_stream_has_space(vsk) == 0 &&
1581                        sk->sk_err == 0 &&
1582                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1583                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1584
1585                         /* Don't wait for non-blocking sockets. */
1586                         if (timeout == 0) {
1587                                 err = -EAGAIN;
1588                                 remove_wait_queue(sk_sleep(sk), &wait);
1589                                 goto out_err;
1590                         }
1591
1592                         err = transport->notify_send_pre_block(vsk, &send_data);
1593                         if (err < 0) {
1594                                 remove_wait_queue(sk_sleep(sk), &wait);
1595                                 goto out_err;
1596                         }
1597
1598                         release_sock(sk);
1599                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1600                         lock_sock(sk);
1601                         if (signal_pending(current)) {
1602                                 err = sock_intr_errno(timeout);
1603                                 remove_wait_queue(sk_sleep(sk), &wait);
1604                                 goto out_err;
1605                         } else if (timeout == 0) {
1606                                 err = -EAGAIN;
1607                                 remove_wait_queue(sk_sleep(sk), &wait);
1608                                 goto out_err;
1609                         }
1610                 }
1611                 remove_wait_queue(sk_sleep(sk), &wait);
1612
1613                 /* These checks occur both as part of and after the loop
1614                  * conditional since we need to check before and after
1615                  * sleeping.
1616                  */
1617                 if (sk->sk_err) {
1618                         err = -sk->sk_err;
1619                         goto out_err;
1620                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1621                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1622                         err = -EPIPE;
1623                         goto out_err;
1624                 }
1625
1626                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1627                 if (err < 0)
1628                         goto out_err;
1629
1630                 /* Note that enqueue will only write as many bytes as are free
1631                  * in the produce queue, so we don't need to ensure len is
1632                  * smaller than the queue size.  It is the caller's
1633                  * responsibility to check how many bytes we were able to send.
1634                  */
1635
1636                 written = transport->stream_enqueue(
1637                                 vsk, msg,
1638                                 len - total_written);
1639                 if (written < 0) {
1640                         err = -ENOMEM;
1641                         goto out_err;
1642                 }
1643
1644                 total_written += written;
1645
1646                 err = transport->notify_send_post_enqueue(
1647                                 vsk, written, &send_data);
1648                 if (err < 0)
1649                         goto out_err;
1650
1651         }
1652
1653 out_err:
1654         if (total_written > 0)
1655                 err = total_written;
1656 out:
1657         release_sock(sk);
1658         return err;
1659 }
1660
1661
1662 static int
1663 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1664                      int flags)
1665 {
1666         struct sock *sk;
1667         struct vsock_sock *vsk;
1668         int err;
1669         size_t target;
1670         ssize_t copied;
1671         long timeout;
1672         struct vsock_transport_recv_notify_data recv_data;
1673
1674         DEFINE_WAIT(wait);
1675
1676         sk = sock->sk;
1677         vsk = vsock_sk(sk);
1678         err = 0;
1679
1680         lock_sock(sk);
1681
1682         if (sk->sk_state != SS_CONNECTED) {
1683                 /* Recvmsg is supposed to return 0 if a peer performs an
1684                  * orderly shutdown. Differentiate between that case and when a
1685                  * peer has not connected or a local shutdown occured with the
1686                  * SOCK_DONE flag.
1687                  */
1688                 if (sock_flag(sk, SOCK_DONE))
1689                         err = 0;
1690                 else
1691                         err = -ENOTCONN;
1692
1693                 goto out;
1694         }
1695
1696         if (flags & MSG_OOB) {
1697                 err = -EOPNOTSUPP;
1698                 goto out;
1699         }
1700
1701         /* We don't check peer_shutdown flag here since peer may actually shut
1702          * down, but there can be data in the queue that a local socket can
1703          * receive.
1704          */
1705         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1706                 err = 0;
1707                 goto out;
1708         }
1709
1710         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1711          * is not an error.  We may as well bail out now.
1712          */
1713         if (!len) {
1714                 err = 0;
1715                 goto out;
1716         }
1717
1718         /* We must not copy less than target bytes into the user's buffer
1719          * before returning successfully, so we wait for the consume queue to
1720          * have that much data to consume before dequeueing.  Note that this
1721          * makes it impossible to handle cases where target is greater than the
1722          * queue size.
1723          */
1724         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1725         if (target >= transport->stream_rcvhiwat(vsk)) {
1726                 err = -ENOMEM;
1727                 goto out;
1728         }
1729         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1730         copied = 0;
1731
1732         err = transport->notify_recv_init(vsk, target, &recv_data);
1733         if (err < 0)
1734                 goto out;
1735
1736
1737         while (1) {
1738                 s64 ready;
1739
1740                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1741                 ready = vsock_stream_has_data(vsk);
1742
1743                 if (ready == 0) {
1744                         if (sk->sk_err != 0 ||
1745                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1746                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1747                                 finish_wait(sk_sleep(sk), &wait);
1748                                 break;
1749                         }
1750                         /* Don't wait for non-blocking sockets. */
1751                         if (timeout == 0) {
1752                                 err = -EAGAIN;
1753                                 finish_wait(sk_sleep(sk), &wait);
1754                                 break;
1755                         }
1756
1757                         err = transport->notify_recv_pre_block(
1758                                         vsk, target, &recv_data);
1759                         if (err < 0) {
1760                                 finish_wait(sk_sleep(sk), &wait);
1761                                 break;
1762                         }
1763                         release_sock(sk);
1764                         timeout = schedule_timeout(timeout);
1765                         lock_sock(sk);
1766
1767                         if (signal_pending(current)) {
1768                                 err = sock_intr_errno(timeout);
1769                                 finish_wait(sk_sleep(sk), &wait);
1770                                 break;
1771                         } else if (timeout == 0) {
1772                                 err = -EAGAIN;
1773                                 finish_wait(sk_sleep(sk), &wait);
1774                                 break;
1775                         }
1776                 } else {
1777                         ssize_t read;
1778
1779                         finish_wait(sk_sleep(sk), &wait);
1780
1781                         if (ready < 0) {
1782                                 /* Invalid queue pair content. XXX This should
1783                                 * be changed to a connection reset in a later
1784                                 * change.
1785                                 */
1786
1787                                 err = -ENOMEM;
1788                                 goto out;
1789                         }
1790
1791                         err = transport->notify_recv_pre_dequeue(
1792                                         vsk, target, &recv_data);
1793                         if (err < 0)
1794                                 break;
1795
1796                         read = transport->stream_dequeue(
1797                                         vsk, msg,
1798                                         len - copied, flags);
1799                         if (read < 0) {
1800                                 err = -ENOMEM;
1801                                 break;
1802                         }
1803
1804                         copied += read;
1805
1806                         err = transport->notify_recv_post_dequeue(
1807                                         vsk, target, read,
1808                                         !(flags & MSG_PEEK), &recv_data);
1809                         if (err < 0)
1810                                 goto out;
1811
1812                         if (read >= target || flags & MSG_PEEK)
1813                                 break;
1814
1815                         target -= read;
1816                 }
1817         }
1818
1819         if (sk->sk_err)
1820                 err = -sk->sk_err;
1821         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1822                 err = 0;
1823
1824         if (copied > 0)
1825                 err = copied;
1826
1827 out:
1828         release_sock(sk);
1829         return err;
1830 }
1831
1832 static const struct proto_ops vsock_stream_ops = {
1833         .family = PF_VSOCK,
1834         .owner = THIS_MODULE,
1835         .release = vsock_release,
1836         .bind = vsock_bind,
1837         .connect = vsock_stream_connect,
1838         .socketpair = sock_no_socketpair,
1839         .accept = vsock_accept,
1840         .getname = vsock_getname,
1841         .poll = vsock_poll,
1842         .ioctl = sock_no_ioctl,
1843         .listen = vsock_listen,
1844         .shutdown = vsock_shutdown,
1845         .setsockopt = vsock_stream_setsockopt,
1846         .getsockopt = vsock_stream_getsockopt,
1847         .sendmsg = vsock_stream_sendmsg,
1848         .recvmsg = vsock_stream_recvmsg,
1849         .mmap = sock_no_mmap,
1850         .sendpage = sock_no_sendpage,
1851 };
1852
1853 static int vsock_create(struct net *net, struct socket *sock,
1854                         int protocol, int kern)
1855 {
1856         if (!sock)
1857                 return -EINVAL;
1858
1859         if (protocol && protocol != PF_VSOCK)
1860                 return -EPROTONOSUPPORT;
1861
1862         switch (sock->type) {
1863         case SOCK_DGRAM:
1864                 sock->ops = &vsock_dgram_ops;
1865                 break;
1866         case SOCK_STREAM:
1867                 sock->ops = &vsock_stream_ops;
1868                 break;
1869         default:
1870                 return -ESOCKTNOSUPPORT;
1871         }
1872
1873         sock->state = SS_UNCONNECTED;
1874
1875         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1876 }
1877
1878 static const struct net_proto_family vsock_family_ops = {
1879         .family = AF_VSOCK,
1880         .create = vsock_create,
1881         .owner = THIS_MODULE,
1882 };
1883
1884 static long vsock_dev_do_ioctl(struct file *filp,
1885                                unsigned int cmd, void __user *ptr)
1886 {
1887         u32 __user *p = ptr;
1888         int retval = 0;
1889
1890         switch (cmd) {
1891         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1892                 if (put_user(transport->get_local_cid(), p) != 0)
1893                         retval = -EFAULT;
1894                 break;
1895
1896         default:
1897                 pr_err("Unknown ioctl %d\n", cmd);
1898                 retval = -EINVAL;
1899         }
1900
1901         return retval;
1902 }
1903
1904 static long vsock_dev_ioctl(struct file *filp,
1905                             unsigned int cmd, unsigned long arg)
1906 {
1907         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1908 }
1909
1910 #ifdef CONFIG_COMPAT
1911 static long vsock_dev_compat_ioctl(struct file *filp,
1912                                    unsigned int cmd, unsigned long arg)
1913 {
1914         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1915 }
1916 #endif
1917
1918 static const struct file_operations vsock_device_ops = {
1919         .owner          = THIS_MODULE,
1920         .unlocked_ioctl = vsock_dev_ioctl,
1921 #ifdef CONFIG_COMPAT
1922         .compat_ioctl   = vsock_dev_compat_ioctl,
1923 #endif
1924         .open           = nonseekable_open,
1925 };
1926
1927 static struct miscdevice vsock_device = {
1928         .name           = "vsock",
1929         .fops           = &vsock_device_ops,
1930 };
1931
1932 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1933 {
1934         int err = mutex_lock_interruptible(&vsock_register_mutex);
1935
1936         if (err)
1937                 return err;
1938
1939         if (transport) {
1940                 err = -EBUSY;
1941                 goto err_busy;
1942         }
1943
1944         /* Transport must be the owner of the protocol so that it can't
1945          * unload while there are open sockets.
1946          */
1947         vsock_proto.owner = owner;
1948         transport = t;
1949
1950         vsock_init_tables();
1951
1952         vsock_device.minor = MISC_DYNAMIC_MINOR;
1953         err = misc_register(&vsock_device);
1954         if (err) {
1955                 pr_err("Failed to register misc device\n");
1956                 goto err_reset_transport;
1957         }
1958
1959         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1960         if (err) {
1961                 pr_err("Cannot register vsock protocol\n");
1962                 goto err_deregister_misc;
1963         }
1964
1965         err = sock_register(&vsock_family_ops);
1966         if (err) {
1967                 pr_err("could not register af_vsock (%d) address family: %d\n",
1968                        AF_VSOCK, err);
1969                 goto err_unregister_proto;
1970         }
1971
1972         mutex_unlock(&vsock_register_mutex);
1973         return 0;
1974
1975 err_unregister_proto:
1976         proto_unregister(&vsock_proto);
1977 err_deregister_misc:
1978         misc_deregister(&vsock_device);
1979 err_reset_transport:
1980         transport = NULL;
1981 err_busy:
1982         mutex_unlock(&vsock_register_mutex);
1983         return err;
1984 }
1985 EXPORT_SYMBOL_GPL(__vsock_core_init);
1986
1987 void vsock_core_exit(void)
1988 {
1989         mutex_lock(&vsock_register_mutex);
1990
1991         misc_deregister(&vsock_device);
1992         sock_unregister(AF_VSOCK);
1993         proto_unregister(&vsock_proto);
1994
1995         /* We do not want the assignment below re-ordered. */
1996         mb();
1997         transport = NULL;
1998
1999         mutex_unlock(&vsock_register_mutex);
2000 }
2001 EXPORT_SYMBOL_GPL(vsock_core_exit);
2002
2003 const struct vsock_transport *vsock_core_get_transport(void)
2004 {
2005         /* vsock_register_mutex not taken since only the transport uses this
2006          * function and only while registered.
2007          */
2008         return transport;
2009 }
2010 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2011
2012 MODULE_AUTHOR("VMware, Inc.");
2013 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2014 MODULE_VERSION("1.0.2.0-k");
2015 MODULE_LICENSE("GPL v2");