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  * - 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
  79 #include <linux/types.h>
  80 #include <linux/bitops.h>
  81 #include <linux/cred.h>
  82 #include <linux/init.h>
  83 #include <linux/io.h>
  84 #include <linux/kernel.h>
  85 #include <linux/kmod.h>
  86 #include <linux/list.h>
  87 #include <linux/miscdevice.h>
  88 #include <linux/module.h>
  89 #include <linux/mutex.h>
  90 #include <linux/net.h>
  91 #include <linux/poll.h>
  92 #include <linux/skbuff.h>
  93 #include <linux/smp.h>
  94 #include <linux/socket.h>
  95 #include <linux/stddef.h>
  96 #include <linux/unistd.h>
  97 #include <linux/wait.h>
  98 #include <linux/workqueue.h>
  99 #include <net/sock.h>
 100 #include <net/af_vsock.h>
 101
 102 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 103 static void vsock_sk_destruct(struct sock *sk);
 104 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 105
 106 /* Protocol family. */
 107 static struct proto vsock_proto = {
 108         .name = "AF_VSOCK",
 109         .owner = THIS_MODULE,
 110         .obj_size = sizeof(struct vsock_sock),
 111 };
 112
 113 /* The default peer timeout indicates how long we will wait for a peer response
 114  * to a control message.
 115  */
 116 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 117
 118 static const struct vsock_transport *transport;
 119 static DEFINE_MUTEX(vsock_register_mutex);
 120
 121 /**** EXPORTS ****/
 122
 123 /* Get the ID of the local context.  This is transport dependent. */
 124
 125 int vm_sockets_get_local_cid(void)
 126 {
 127         return transport->get_local_cid();
 128 }
 129 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 130
 131 /**** UTILS ****/
 132
 133 /* Each bound VSocket is stored in the bind hash table and each connected
 134  * VSocket is stored in the connected hash table.
 135  *
 136  * Unbound sockets are all put on the same list attached to the end of the hash
 137  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 138  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 139  * represents the list that addr hashes to).
 140  *
 141  * Specifically, we initialize the vsock_bind_table array to a size of
 142  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 143  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 144  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 145  * mods with VSOCK_HASH_SIZE to ensure this.
 146  */
 147 #define VSOCK_HASH_SIZE         251
 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 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 163 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 164 static DEFINE_SPINLOCK(vsock_table_lock);
 165
 166 /* Autobind this socket to the local address if necessary. */
 167 static int vsock_auto_bind(struct vsock_sock *vsk)
 168 {
 169         struct sock *sk = sk_vsock(vsk);
 170         struct sockaddr_vm local_addr;
 171
 172         if (vsock_addr_bound(&vsk->local_addr))
 173                 return 0;
 174         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 175         return __vsock_bind(sk, &local_addr);
 176 }
 177
 178 static void vsock_init_tables(void)
 179 {
 180         int i;
 181
 182         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 183                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 184
 185         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 186                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 187 }
 188
 189 static void __vsock_insert_bound(struct list_head *list,
 190                                  struct vsock_sock *vsk)
 191 {
 192         sock_hold(&vsk->sk);
 193         list_add(&vsk->bound_table, list);
 194 }
 195
 196 static void __vsock_insert_connected(struct list_head *list,
 197                                      struct vsock_sock *vsk)
 198 {
 199         sock_hold(&vsk->sk);
 200         list_add(&vsk->connected_table, list);
 201 }
 202
 203 static void __vsock_remove_bound(struct vsock_sock *vsk)
 204 {
 205         list_del_init(&vsk->bound_table);
 206         sock_put(&vsk->sk);
 207 }
 208
 209 static void __vsock_remove_connected(struct vsock_sock *vsk)
 210 {
 211         list_del_init(&vsk->connected_table);
 212         sock_put(&vsk->sk);
 213 }
 214
 215 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 216 {
 217         struct vsock_sock *vsk;
 218
 219         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 220                 if (addr->svm_port == vsk->local_addr.svm_port)
 221                         return sk_vsock(vsk);
 222
 223         return NULL;
 224 }
 225
 226 static struct sock *__vsock_find_unbound_socket(struct sockaddr_vm *addr)
 227 {
 228         struct vsock_sock *vsk;
 229
 230         list_for_each_entry(vsk, vsock_unbound_sockets, bound_table)
 231                 if (addr->svm_port == vsk->local_addr.svm_port)
 232                         return sk_vsock(vsk);
 233
 234         return NULL;
 235 }
 236
 237 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 238                                                   struct sockaddr_vm *dst)
 239 {
 240         struct vsock_sock *vsk;
 241
 242         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 243                             connected_table) {
 244                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 245                     dst->svm_port == vsk->local_addr.svm_port) {
 246                         return sk_vsock(vsk);
 247                 }
 248         }
 249
 250         return NULL;
 251 }
 252
 253 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 254 {
 255         return !list_empty(&vsk->bound_table);
 256 }
 257
 258 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 259 {
 260         return !list_empty(&vsk->connected_table);
 261 }
 262
 263 static void vsock_insert_unbound(struct vsock_sock *vsk)
 264 {
 265         spin_lock_bh(&vsock_table_lock);
 266         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 267         spin_unlock_bh(&vsock_table_lock);
 268 }
 269
 270 void vsock_insert_connected(struct vsock_sock *vsk)
 271 {
 272         struct list_head *list = vsock_connected_sockets(
 273                 &vsk->remote_addr, &vsk->local_addr);
 274
 275         spin_lock_bh(&vsock_table_lock);
 276         __vsock_insert_connected(list, vsk);
 277         spin_unlock_bh(&vsock_table_lock);
 278 }
 279 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 280
 281 void vsock_remove_bound(struct vsock_sock *vsk)
 282 {
 283         spin_lock_bh(&vsock_table_lock);
 284         __vsock_remove_bound(vsk);
 285         spin_unlock_bh(&vsock_table_lock);
 286 }
 287 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 288
 289 void vsock_remove_connected(struct vsock_sock *vsk)
 290 {
 291         spin_lock_bh(&vsock_table_lock);
 292         __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_unbound_socket(struct sockaddr_vm *addr)
 313 {
 314         struct sock *sk;
 315
 316         spin_lock_bh(&vsock_table_lock);
 317         sk = __vsock_find_unbound_socket(addr);
 318         if (sk)
 319                 sock_hold(sk);
 320
 321         spin_unlock_bh(&vsock_table_lock);
 322
 323         return sk;
 324 }
 325 EXPORT_SYMBOL_GPL(vsock_find_unbound_socket);
 326
 327 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 328                                          struct sockaddr_vm *dst)
 329 {
 330         struct sock *sk;
 331
 332         spin_lock_bh(&vsock_table_lock);
 333         sk = __vsock_find_connected_socket(src, dst);
 334         if (sk)
 335                 sock_hold(sk);
 336
 337         spin_unlock_bh(&vsock_table_lock);
 338
 339         return sk;
 340 }
 341 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 342
 343 static bool vsock_in_bound_table(struct vsock_sock *vsk)
 344 {
 345         bool ret;
 346
 347         spin_lock_bh(&vsock_table_lock);
 348         ret = __vsock_in_bound_table(vsk);
 349         spin_unlock_bh(&vsock_table_lock);
 350
 351         return ret;
 352 }
 353
 354 static bool vsock_in_connected_table(struct vsock_sock *vsk)
 355 {
 356         bool ret;
 357
 358         spin_lock_bh(&vsock_table_lock);
 359         ret = __vsock_in_connected_table(vsk);
 360         spin_unlock_bh(&vsock_table_lock);
 361
 362         return ret;
 363 }
 364
 365 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 366 {
 367         int i;
 368
 369         spin_lock_bh(&vsock_table_lock);
 370
 371         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 372                 struct vsock_sock *vsk;
 373                 list_for_each_entry(vsk, &vsock_connected_table[i],
 374                                     connected_table)
 375                         fn(sk_vsock(vsk));
 376         }
 377
 378         spin_unlock_bh(&vsock_table_lock);
 379 }
 380 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 381
 382 void vsock_add_pending(struct sock *listener, struct sock *pending)
 383 {
 384         struct vsock_sock *vlistener;
 385         struct vsock_sock *vpending;
 386
 387         vlistener = vsock_sk(listener);
 388         vpending = vsock_sk(pending);
 389
 390         sock_hold(pending);
 391         sock_hold(listener);
 392         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 393 }
 394 EXPORT_SYMBOL_GPL(vsock_add_pending);
 395
 396 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 397 {
 398         struct vsock_sock *vpending = vsock_sk(pending);
 399
 400         list_del_init(&vpending->pending_links);
 401         sock_put(listener);
 402         sock_put(pending);
 403 }
 404 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 405
 406 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 407 {
 408         struct vsock_sock *vlistener;
 409         struct vsock_sock *vconnected;
 410
 411         vlistener = vsock_sk(listener);
 412         vconnected = vsock_sk(connected);
 413
 414         sock_hold(connected);
 415         sock_hold(listener);
 416         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 417 }
 418 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 419
 420 static struct sock *vsock_dequeue_accept(struct sock *listener)
 421 {
 422         struct vsock_sock *vlistener;
 423         struct vsock_sock *vconnected;
 424
 425         vlistener = vsock_sk(listener);
 426
 427         if (list_empty(&vlistener->accept_queue))
 428                 return NULL;
 429
 430         vconnected = list_entry(vlistener->accept_queue.next,
 431                                 struct vsock_sock, accept_queue);
 432
 433         list_del_init(&vconnected->accept_queue);
 434         sock_put(listener);
 435         /* The caller will need a reference on the connected socket so we let
 436          * it call sock_put().
 437          */
 438
 439         return sk_vsock(vconnected);
 440 }
 441
 442 static bool vsock_is_accept_queue_empty(struct sock *sk)
 443 {
 444         struct vsock_sock *vsk = vsock_sk(sk);
 445         return list_empty(&vsk->accept_queue);
 446 }
 447
 448 static bool vsock_is_pending(struct sock *sk)
 449 {
 450         struct vsock_sock *vsk = vsock_sk(sk);
 451         return !list_empty(&vsk->pending_links);
 452 }
 453
 454 static int vsock_send_shutdown(struct sock *sk, int mode)
 455 {
 456         return transport->shutdown(vsock_sk(sk), mode);
 457 }
 458
 459 void vsock_pending_work(struct work_struct *work)
 460 {
 461         struct sock *sk;
 462         struct sock *listener;
 463         struct vsock_sock *vsk;
 464         bool cleanup;
 465
 466         vsk = container_of(work, struct vsock_sock, dwork.work);
 467         sk = sk_vsock(vsk);
 468         listener = vsk->listener;
 469         cleanup = true;
 470
 471         lock_sock(listener);
 472         lock_sock(sk);
 473
 474         if (vsock_is_pending(sk)) {
 475                 vsock_remove_pending(listener, sk);
 476         } else if (!vsk->rejected) {
 477                 /* We are not on the pending list and accept() did not reject
 478                  * us, so we must have been accepted by our user process.  We
 479                  * just need to drop our references to the sockets and be on
 480                  * our way.
 481                  */
 482                 cleanup = false;
 483                 goto out;
 484         }
 485
 486         listener->sk_ack_backlog--;
 487
 488         /* We need to remove ourself from the global connected sockets list so
 489          * incoming packets can't find this socket, and to reduce the reference
 490          * count.
 491          */
 492         if (vsock_in_connected_table(vsk))
 493                 vsock_remove_connected(vsk);
 494
 495         sk->sk_state = SS_FREE;
 496
 497 out:
 498         release_sock(sk);
 499         release_sock(listener);
 500         if (cleanup)
 501                 sock_put(sk);
 502
 503         sock_put(sk);
 504         sock_put(listener);
 505 }
 506 EXPORT_SYMBOL_GPL(vsock_pending_work);
 507
 508 /**** SOCKET OPERATIONS ****/
 509
 510 static int __vsock_bind_stream(struct vsock_sock *vsk,
 511                                struct sockaddr_vm *addr)
 512 {
 513         static u32 port = LAST_RESERVED_PORT + 1;
 514         struct sockaddr_vm new_addr;
 515
 516         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 517
 518         if (addr->svm_port == VMADDR_PORT_ANY) {
 519                 bool found = false;
 520                 unsigned int i;
 521
 522                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 523                         if (port <= LAST_RESERVED_PORT)
 524                                 port = LAST_RESERVED_PORT + 1;
 525
 526                         new_addr.svm_port = port++;
 527
 528                         if (!__vsock_find_bound_socket(&new_addr)) {
 529                                 found = true;
 530                                 break;
 531                         }
 532                 }
 533
 534                 if (!found)
 535                         return -EADDRNOTAVAIL;
 536         } else {
 537                 /* If port is in reserved range, ensure caller
 538                  * has necessary privileges.
 539                  */
 540                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 541                     !capable(CAP_NET_BIND_SERVICE)) {
 542                         return -EACCES;
 543                 }
 544
 545                 if (__vsock_find_bound_socket(&new_addr))
 546                         return -EADDRINUSE;
 547         }
 548
 549         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 550
 551         /* Remove stream sockets from the unbound list and add them to the hash
 552          * table for easy lookup by its address.  The unbound list is simply an
 553          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 554          */
 555         __vsock_remove_bound(vsk);
 556         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 557
 558         return 0;
 559 }
 560
 561 int vsock_bind_dgram_generic(struct vsock_sock *vsk, struct sockaddr_vm *addr)
 562 {
 563         static u32 port = LAST_RESERVED_PORT + 1;
 564         struct sockaddr_vm new_addr;
 565
 566         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 567
 568         if (addr->svm_port == VMADDR_PORT_ANY) {
 569                 bool found = false;
 570                 unsigned int i;
 571
 572                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 573                         if (port <= LAST_RESERVED_PORT)
 574                                 port = LAST_RESERVED_PORT + 1;
 575
 576                         new_addr.svm_port = port++;
 577
 578                         if (!__vsock_find_unbound_socket(&new_addr)) {
 579                                 found = true;
 580                                 break;
 581                         }
 582                 }
 583
 584                 if (!found)
 585                         return -EADDRNOTAVAIL;
 586         } else {
 587                 /* If port is in reserved range, ensure caller
 588                  * has necessary privileges.
 589                  */
 590                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 591                     !capable(CAP_NET_BIND_SERVICE)) {
 592                         return -EACCES;
 593                 }
 594
 595                 if (__vsock_find_unbound_socket(&new_addr))
 596                         return -EADDRINUSE;
 597         }
 598
 599         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 600
 601         return 0;
 602 }
 603 EXPORT_SYMBOL_GPL(vsock_bind_dgram_generic);
 604
 605 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 606                               struct sockaddr_vm *addr)
 607 {
 608         return transport->dgram_bind(vsk, addr);
 609 }
 610
 611 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 612 {
 613         struct vsock_sock *vsk = vsock_sk(sk);
 614         u32 cid;
 615         int retval;
 616
 617         /* First ensure this socket isn't already bound. */
 618         if (vsock_addr_bound(&vsk->local_addr))
 619                 return -EINVAL;
 620
 621         /* Now bind to the provided address or select appropriate values if
 622          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 623          * like AF_INET prevents binding to a non-local IP address (in most
 624          * cases), we only allow binding to the local CID.
 625          */
 626         cid = transport->get_local_cid();
 627         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 628                 return -EADDRNOTAVAIL;
 629
 630         switch (sk->sk_socket->type) {
 631         case SOCK_STREAM:
 632                 spin_lock_bh(&vsock_table_lock);
 633                 retval = __vsock_bind_stream(vsk, addr);
 634                 spin_unlock_bh(&vsock_table_lock);
 635                 break;
 636
 637         case SOCK_DGRAM:
 638                 retval = __vsock_bind_dgram(vsk, addr);
 639                 break;
 640
 641         default:
 642                 retval = -EINVAL;
 643                 break;
 644         }
 645
 646         return retval;
 647 }
 648
 649 struct sock *__vsock_create(struct net *net,
 650                             struct socket *sock,
 651                             struct sock *parent,
 652                             gfp_t priority,
 653                             unsigned short type,
 654                             int kern)
 655 {
 656         struct sock *sk;
 657         struct vsock_sock *psk;
 658         struct vsock_sock *vsk;
 659
 660         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 661         if (!sk)
 662                 return NULL;
 663
 664         sock_init_data(sock, sk);
 665
 666         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 667          * non-NULL. We make sure that our sockets always have a type by
 668          * setting it here if needed.
 669          */
 670         if (!sock)
 671                 sk->sk_type = type;
 672
 673         vsk = vsock_sk(sk);
 674         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 675         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 676
 677         sk->sk_destruct = vsock_sk_destruct;
 678         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 679         sk->sk_state = 0;
 680         sock_reset_flag(sk, SOCK_DONE);
 681
 682         INIT_LIST_HEAD(&vsk->bound_table);
 683         INIT_LIST_HEAD(&vsk->connected_table);
 684         vsk->listener = NULL;
 685         INIT_LIST_HEAD(&vsk->pending_links);
 686         INIT_LIST_HEAD(&vsk->accept_queue);
 687         vsk->rejected = false;
 688         vsk->sent_request = false;
 689         vsk->ignore_connecting_rst = false;
 690         vsk->peer_shutdown = 0;
 691
 692         psk = parent ? vsock_sk(parent) : NULL;
 693         if (parent) {
 694                 vsk->trusted = psk->trusted;
 695                 vsk->owner = get_cred(psk->owner);
 696                 vsk->connect_timeout = psk->connect_timeout;
 697         } else {
 698                 vsk->trusted = capable(CAP_NET_ADMIN);
 699                 vsk->owner = get_current_cred();
 700                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 701         }
 702
 703         if (transport->init(vsk, psk) < 0) {
 704                 sk_free(sk);
 705                 return NULL;
 706         }
 707
 708         if (sock)
 709                 vsock_insert_unbound(vsk);
 710
 711         return sk;
 712 }
 713 EXPORT_SYMBOL_GPL(__vsock_create);
 714
 715 static void __vsock_release(struct sock *sk)
 716 {
 717         if (sk) {
 718                 struct sk_buff *skb;
 719                 struct sock *pending;
 720                 struct vsock_sock *vsk;
 721
 722                 vsk = vsock_sk(sk);
 723                 pending = NULL; /* Compiler warning. */
 724
 725                 if (vsock_in_bound_table(vsk))
 726                         vsock_remove_bound(vsk);
 727
 728                 if (vsock_in_connected_table(vsk))
 729                         vsock_remove_connected(vsk);
 730
 731                 transport->release(vsk);
 732
 733                 lock_sock(sk);
 734                 sock_orphan(sk);
 735                 sk->sk_shutdown = SHUTDOWN_MASK;
 736
 737                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 738                         kfree_skb(skb);
 739
 740                 /* Clean up any sockets that never were accepted. */
 741                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 742                         __vsock_release(pending);
 743                         sock_put(pending);
 744                 }
 745
 746                 release_sock(sk);
 747                 sock_put(sk);
 748         }
 749 }
 750
 751 static void vsock_sk_destruct(struct sock *sk)
 752 {
 753         struct vsock_sock *vsk = vsock_sk(sk);
 754
 755         transport->destruct(vsk);
 756
 757         /* When clearing these addresses, there's no need to set the family and
 758          * possibly register the address family with the kernel.
 759          */
 760         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 761         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 762
 763         put_cred(vsk->owner);
 764 }
 765
 766 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 767 {
 768         int err;
 769
 770         err = sock_queue_rcv_skb(sk, skb);
 771         if (err)
 772                 kfree_skb(skb);
 773
 774         return err;
 775 }
 776
 777 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 778 {
 779         return transport->stream_has_data(vsk);
 780 }
 781 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 782
 783 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 784 {
 785         return transport->stream_has_space(vsk);
 786 }
 787 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 788
 789 static int vsock_release(struct socket *sock)
 790 {
 791         __vsock_release(sock->sk);
 792         sock->sk = NULL;
 793         sock->state = SS_FREE;
 794
 795         return 0;
 796 }
 797
 798 static int
 799 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 800 {
 801         int err;
 802         struct sock *sk;
 803         struct sockaddr_vm *vm_addr;
 804
 805         sk = sock->sk;
 806
 807         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 808                 return -EINVAL;
 809
 810         lock_sock(sk);
 811         err = __vsock_bind(sk, vm_addr);
 812         release_sock(sk);
 813
 814         return err;
 815 }
 816
 817 static int vsock_getname(struct socket *sock,
 818                          struct sockaddr *addr, int *addr_len, int peer)
 819 {
 820         int err;
 821         struct sock *sk;
 822         struct vsock_sock *vsk;
 823         struct sockaddr_vm *vm_addr;
 824
 825         sk = sock->sk;
 826         vsk = vsock_sk(sk);
 827         err = 0;
 828
 829         lock_sock(sk);
 830
 831         if (peer) {
 832                 if (sock->state != SS_CONNECTED) {
 833                         err = -ENOTCONN;
 834                         goto out;
 835                 }
 836                 vm_addr = &vsk->remote_addr;
 837         } else {
 838                 vm_addr = &vsk->local_addr;
 839         }
 840
 841         if (!vm_addr) {
 842                 err = -EINVAL;
 843                 goto out;
 844         }
 845
 846         /* sys_getsockname() and sys_getpeername() pass us a
 847          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 848          * that macro is defined in socket.c instead of .h, so we hardcode its
 849          * value here.
 850          */
 851         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 852         memcpy(addr, vm_addr, sizeof(*vm_addr));
 853         *addr_len = sizeof(*vm_addr);
 854
 855 out:
 856         release_sock(sk);
 857         return err;
 858 }
 859
 860 static int vsock_shutdown(struct socket *sock, int mode)
 861 {
 862         int err;
 863         struct sock *sk;
 864
 865         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 866          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 867          * here like the other address families do.  Note also that the
 868          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 869          * which is what we want.
 870          */
 871         mode++;
 872
 873         if ((mode & ~SHUTDOWN_MASK) || !mode)
 874                 return -EINVAL;
 875
 876         /* If this is a STREAM socket and it is not connected then bail out
 877          * immediately.  If it is a DGRAM socket then we must first kick the
 878          * socket so that it wakes up from any sleeping calls, for example
 879          * recv(), and then afterwards return the error.
 880          */
 881
 882         sk = sock->sk;
 883         if (sock->state == SS_UNCONNECTED) {
 884                 err = -ENOTCONN;
 885                 if (sk->sk_type == SOCK_STREAM)
 886                         return err;
 887         } else {
 888                 sock->state = SS_DISCONNECTING;
 889                 err = 0;
 890         }
 891
 892         /* Receive and send shutdowns are treated alike. */
 893         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 894         if (mode) {
 895                 lock_sock(sk);
 896                 sk->sk_shutdown |= mode;
 897                 sk->sk_state_change(sk);
 898                 release_sock(sk);
 899
 900                 if (sk->sk_type == SOCK_STREAM) {
 901                         sock_reset_flag(sk, SOCK_DONE);
 902                         vsock_send_shutdown(sk, mode);
 903                 }
 904         }
 905
 906         return err;
 907 }
 908
 909 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 910                                poll_table *wait)
 911 {
 912         struct sock *sk;
 913         unsigned int mask;
 914         struct vsock_sock *vsk;
 915
 916         sk = sock->sk;
 917         vsk = vsock_sk(sk);
 918
 919         poll_wait(file, sk_sleep(sk), wait);
 920         mask = 0;
 921
 922         if (sk->sk_err)
 923                 /* Signify that there has been an error on this socket. */
 924                 mask |= POLLERR;
 925
 926         /* INET sockets treat local write shutdown and peer write shutdown as a
 927          * case of POLLHUP set.
 928          */
 929         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 930             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 931              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 932                 mask |= POLLHUP;
 933         }
 934
 935         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 936             vsk->peer_shutdown & SEND_SHUTDOWN) {
 937                 mask |= POLLRDHUP;
 938         }
 939
 940         if (sock->type == SOCK_DGRAM) {
 941                 /* For datagram sockets we can read if there is something in
 942                  * the queue and write as long as the socket isn't shutdown for
 943                  * sending.
 944                  */
 945                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 946                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 947                         mask |= POLLIN | POLLRDNORM;
 948                 }
 949
 950                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 951                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 952
 953         } else if (sock->type == SOCK_STREAM) {
 954                 lock_sock(sk);
 955
 956                 /* Listening sockets that have connections in their accept
 957                  * queue can be read.
 958                  */
 959                 if (sk->sk_state == VSOCK_SS_LISTEN
 960                     && !vsock_is_accept_queue_empty(sk))
 961                         mask |= POLLIN | POLLRDNORM;
 962
 963                 /* If there is something in the queue then we can read. */
 964                 if (transport->stream_is_active(vsk) &&
 965                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 966                         bool data_ready_now = false;
 967                         int ret = transport->notify_poll_in(
 968                                         vsk, 1, &data_ready_now);
 969                         if (ret < 0) {
 970                                 mask |= POLLERR;
 971                         } else {
 972                                 if (data_ready_now)
 973                                         mask |= POLLIN | POLLRDNORM;
 974
 975                         }
 976                 }
 977
 978                 /* Sockets whose connections have been closed, reset, or
 979                  * terminated should also be considered read, and we check the
 980                  * shutdown flag for that.
 981                  */
 982                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 983                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 984                         mask |= POLLIN | POLLRDNORM;
 985                 }
 986
 987                 /* Connected sockets that can produce data can be written. */
 988                 if (sk->sk_state == SS_CONNECTED) {
 989                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 990                                 bool space_avail_now = false;
 991                                 int ret = transport->notify_poll_out(
 992                                                 vsk, 1, &space_avail_now);
 993                                 if (ret < 0) {
 994                                         mask |= POLLERR;
 995                                 } else {
 996                                         if (space_avail_now)
 997                                                 /* Remove POLLWRBAND since INET
 998                                                  * sockets are not setting it.
 999                                                  */
1000                                                 mask |= POLLOUT | POLLWRNORM;
1001
1002                                 }
1003                         }
1004                 }
1005
1006                 /* Simulate INET socket poll behaviors, which sets
1007                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
1008                  * but local send is not shutdown.
1009                  */
1010                 if (sk->sk_state == SS_UNCONNECTED) {
1011                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1012                                 mask |= POLLOUT | POLLWRNORM;
1013
1014                 }
1015
1016                 release_sock(sk);
1017         }
1018
1019         return mask;
1020 }
1021
1022 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1023                                size_t len)
1024 {
1025         int err;
1026         struct sock *sk;
1027         struct vsock_sock *vsk;
1028         struct sockaddr_vm *remote_addr;
1029
1030         if (msg->msg_flags & MSG_OOB)
1031                 return -EOPNOTSUPP;
1032
1033         /* For now, MSG_DONTWAIT is always assumed... */
1034         err = 0;
1035         sk = sock->sk;
1036         vsk = vsock_sk(sk);
1037
1038         lock_sock(sk);
1039
1040         err = vsock_auto_bind(vsk);
1041         if (err)
1042                 goto out;
1043
1044
1045         /* If the provided message contains an address, use that.  Otherwise
1046          * fall back on the socket's remote handle (if it has been connected).
1047          */
1048         if (msg->msg_name &&
1049             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1050                             &remote_addr) == 0) {
1051                 /* Ensure this address is of the right type and is a valid
1052                  * destination.
1053                  */
1054
1055                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1056                         remote_addr->svm_cid = transport->get_local_cid();
1057
1058                 if (!vsock_addr_bound(remote_addr)) {
1059                         err = -EINVAL;
1060                         goto out;
1061                 }
1062         } else if (sock->state == SS_CONNECTED) {
1063                 remote_addr = &vsk->remote_addr;
1064
1065                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1066                         remote_addr->svm_cid = transport->get_local_cid();
1067
1068                 /* XXX Should connect() or this function ensure remote_addr is
1069                  * bound?
1070                  */
1071                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1072                         err = -EINVAL;
1073                         goto out;
1074                 }
1075         } else {
1076                 err = -EINVAL;
1077                 goto out;
1078         }
1079
1080         if (!transport->dgram_allow(remote_addr->svm_cid,
1081                                     remote_addr->svm_port)) {
1082                 err = -EINVAL;
1083                 goto out;
1084         }
1085
1086         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1087
1088 out:
1089         release_sock(sk);
1090         return err;
1091 }
1092
1093 static int vsock_dgram_connect(struct socket *sock,
1094                                struct sockaddr *addr, int addr_len, int flags)
1095 {
1096         int err;
1097         struct sock *sk;
1098         struct vsock_sock *vsk;
1099         struct sockaddr_vm *remote_addr;
1100
1101         sk = sock->sk;
1102         vsk = vsock_sk(sk);
1103
1104         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1105         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1106                 lock_sock(sk);
1107                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1108                                 VMADDR_PORT_ANY);
1109                 sock->state = SS_UNCONNECTED;
1110                 release_sock(sk);
1111                 return 0;
1112         } else if (err != 0)
1113                 return -EINVAL;
1114
1115         lock_sock(sk);
1116
1117         err = vsock_auto_bind(vsk);
1118         if (err)
1119                 goto out;
1120
1121         if (!transport->dgram_allow(remote_addr->svm_cid,
1122                                     remote_addr->svm_port)) {
1123                 err = -EINVAL;
1124                 goto out;
1125         }
1126
1127         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1128         sock->state = SS_CONNECTED;
1129
1130 out:
1131         release_sock(sk);
1132         return err;
1133 }
1134
1135 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1136                                size_t len, int flags)
1137 {
1138         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1139 }
1140
1141 static const struct proto_ops vsock_dgram_ops = {
1142         .family = PF_VSOCK,
1143         .owner = THIS_MODULE,
1144         .release = vsock_release,
1145         .bind = vsock_bind,
1146         .connect = vsock_dgram_connect,
1147         .socketpair = sock_no_socketpair,
1148         .accept = sock_no_accept,
1149         .getname = vsock_getname,
1150         .poll = vsock_poll,
1151         .ioctl = sock_no_ioctl,
1152         .listen = sock_no_listen,
1153         .shutdown = vsock_shutdown,
1154         .setsockopt = sock_no_setsockopt,
1155         .getsockopt = sock_no_getsockopt,
1156         .sendmsg = vsock_dgram_sendmsg,
1157         .recvmsg = vsock_dgram_recvmsg,
1158         .mmap = sock_no_mmap,
1159         .sendpage = sock_no_sendpage,
1160 };
1161
1162 static void vsock_connect_timeout(struct work_struct *work)
1163 {
1164         struct sock *sk;
1165         struct vsock_sock *vsk;
1166
1167         vsk = container_of(work, struct vsock_sock, dwork.work);
1168         sk = sk_vsock(vsk);
1169
1170         lock_sock(sk);
1171         if (sk->sk_state == SS_CONNECTING &&
1172             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1173                 sk->sk_state = SS_UNCONNECTED;
1174                 sk->sk_err = ETIMEDOUT;
1175                 sk->sk_error_report(sk);
1176         }
1177         release_sock(sk);
1178
1179         sock_put(sk);
1180 }
1181
1182 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1183                                 int addr_len, int flags)
1184 {
1185         int err;
1186         struct sock *sk;
1187         struct vsock_sock *vsk;
1188         struct sockaddr_vm *remote_addr;
1189         long timeout;
1190         DEFINE_WAIT(wait);
1191
1192         err = 0;
1193         sk = sock->sk;
1194         vsk = vsock_sk(sk);
1195
1196         lock_sock(sk);
1197
1198         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1199         switch (sock->state) {
1200         case SS_CONNECTED:
1201                 err = -EISCONN;
1202                 goto out;
1203         case SS_DISCONNECTING:
1204                 err = -EINVAL;
1205                 goto out;
1206         case SS_CONNECTING:
1207                 /* This continues on so we can move sock into the SS_CONNECTED
1208                  * state once the connection has completed (at which point err
1209                  * will be set to zero also).  Otherwise, we will either wait
1210                  * for the connection or return -EALREADY should this be a
1211                  * non-blocking call.
1212                  */
1213                 err = -EALREADY;
1214                 break;
1215         default:
1216                 if ((sk->sk_state == VSOCK_SS_LISTEN) ||
1217                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1218                         err = -EINVAL;
1219                         goto out;
1220                 }
1221
1222                 /* The hypervisor and well-known contexts do not have socket
1223                  * endpoints.
1224                  */
1225                 if (!transport->stream_allow(remote_addr->svm_cid,
1226                                              remote_addr->svm_port)) {
1227                         err = -ENETUNREACH;
1228                         goto out;
1229                 }
1230
1231                 /* Set the remote address that we are connecting to. */
1232                 memcpy(&vsk->remote_addr, remote_addr,
1233                        sizeof(vsk->remote_addr));
1234
1235                 err = vsock_auto_bind(vsk);
1236                 if (err)
1237                         goto out;
1238
1239                 sk->sk_state = SS_CONNECTING;
1240
1241                 err = transport->connect(vsk);
1242                 if (err < 0)
1243                         goto out;
1244
1245                 /* Mark sock as connecting and set the error code to in
1246                  * progress in case this is a non-blocking connect.
1247                  */
1248                 sock->state = SS_CONNECTING;
1249                 err = -EINPROGRESS;
1250         }
1251
1252         /* The receive path will handle all communication until we are able to
1253          * enter the connected state.  Here we wait for the connection to be
1254          * completed or a notification of an error.
1255          */
1256         timeout = vsk->connect_timeout;
1257         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1258
1259         while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1260                 if (flags & O_NONBLOCK) {
1261                         /* If we're not going to block, we schedule a timeout
1262                          * function to generate a timeout on the connection
1263                          * attempt, in case the peer doesn't respond in a
1264                          * timely manner. We hold on to the socket until the
1265                          * timeout fires.
1266                          */
1267                         sock_hold(sk);
1268                         INIT_DELAYED_WORK(&vsk->dwork,
1269                                           vsock_connect_timeout);
1270                         schedule_delayed_work(&vsk->dwork, timeout);
1271
1272                         /* Skip ahead to preserve error code set above. */
1273                         goto out_wait;
1274                 }
1275
1276                 release_sock(sk);
1277                 timeout = schedule_timeout(timeout);
1278                 lock_sock(sk);
1279
1280                 if (signal_pending(current)) {
1281                         err = sock_intr_errno(timeout);
1282                         goto out_wait_error;
1283                 } else if (timeout == 0) {
1284                         err = -ETIMEDOUT;
1285                         goto out_wait_error;
1286                 }
1287
1288                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1289         }
1290
1291         if (sk->sk_err) {
1292                 err = -sk->sk_err;
1293                 goto out_wait_error;
1294         } else
1295                 err = 0;
1296
1297 out_wait:
1298         finish_wait(sk_sleep(sk), &wait);
1299 out:
1300         release_sock(sk);
1301         return err;
1302
1303 out_wait_error:
1304         sk->sk_state = SS_UNCONNECTED;
1305         sock->state = SS_UNCONNECTED;
1306         goto out_wait;
1307 }
1308
1309 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1310 {
1311         struct sock *listener;
1312         int err;
1313         struct sock *connected;
1314         struct vsock_sock *vconnected;
1315         long timeout;
1316         DEFINE_WAIT(wait);
1317
1318         err = 0;
1319         listener = sock->sk;
1320
1321         lock_sock(listener);
1322
1323         if (sock->type != SOCK_STREAM) {
1324                 err = -EOPNOTSUPP;
1325                 goto out;
1326         }
1327
1328         if (listener->sk_state != VSOCK_SS_LISTEN) {
1329                 err = -EINVAL;
1330                 goto out;
1331         }
1332
1333         /* Wait for children sockets to appear; these are the new sockets
1334          * created upon connection establishment.
1335          */
1336         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1337         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1338
1339         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1340                listener->sk_err == 0) {
1341                 release_sock(listener);
1342                 timeout = schedule_timeout(timeout);
1343                 lock_sock(listener);
1344
1345                 if (signal_pending(current)) {
1346                         err = sock_intr_errno(timeout);
1347                         goto out_wait;
1348                 } else if (timeout == 0) {
1349                         err = -EAGAIN;
1350                         goto out_wait;
1351                 }
1352
1353                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1354         }
1355
1356         if (listener->sk_err)
1357                 err = -listener->sk_err;
1358
1359         if (connected) {
1360                 listener->sk_ack_backlog--;
1361
1362                 lock_sock(connected);
1363                 vconnected = vsock_sk(connected);
1364
1365                 /* If the listener socket has received an error, then we should
1366                  * reject this socket and return.  Note that we simply mark the
1367                  * socket rejected, drop our reference, and let the cleanup
1368                  * function handle the cleanup; the fact that we found it in
1369                  * the listener's accept queue guarantees that the cleanup
1370                  * function hasn't run yet.
1371                  */
1372                 if (err) {
1373                         vconnected->rejected = true;
1374                         release_sock(connected);
1375                         sock_put(connected);
1376                         goto out_wait;
1377                 }
1378
1379                 newsock->state = SS_CONNECTED;
1380                 sock_graft(connected, newsock);
1381                 release_sock(connected);
1382                 sock_put(connected);
1383         }
1384
1385 out_wait:
1386         finish_wait(sk_sleep(listener), &wait);
1387 out:
1388         release_sock(listener);
1389         return err;
1390 }
1391
1392 static int vsock_listen(struct socket *sock, int backlog)
1393 {
1394         int err;
1395         struct sock *sk;
1396         struct vsock_sock *vsk;
1397
1398         sk = sock->sk;
1399
1400         lock_sock(sk);
1401
1402         if (sock->type != SOCK_STREAM) {
1403                 err = -EOPNOTSUPP;
1404                 goto out;
1405         }
1406
1407         if (sock->state != SS_UNCONNECTED) {
1408                 err = -EINVAL;
1409                 goto out;
1410         }
1411
1412         vsk = vsock_sk(sk);
1413
1414         if (!vsock_addr_bound(&vsk->local_addr)) {
1415                 err = -EINVAL;
1416                 goto out;
1417         }
1418
1419         sk->sk_max_ack_backlog = backlog;
1420         sk->sk_state = VSOCK_SS_LISTEN;
1421
1422         err = 0;
1423
1424 out:
1425         release_sock(sk);
1426         return err;
1427 }
1428
1429 static int vsock_stream_setsockopt(struct socket *sock,
1430                                    int level,
1431                                    int optname,
1432                                    char __user *optval,
1433                                    unsigned int optlen)
1434 {
1435         int err;
1436         struct sock *sk;
1437         struct vsock_sock *vsk;
1438         u64 val;
1439
1440         if (level != AF_VSOCK)
1441                 return -ENOPROTOOPT;
1442
1443 #define COPY_IN(_v)                                       \
1444         do {                                              \
1445                 if (optlen < sizeof(_v)) {                \
1446                         err = -EINVAL;                    \
1447                         goto exit;                        \
1448                 }                                         \
1449                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1450                         err = -EFAULT;                                  \
1451                         goto exit;                                      \
1452                 }                                                       \
1453         } while (0)
1454
1455         err = 0;
1456         sk = sock->sk;
1457         vsk = vsock_sk(sk);
1458
1459         lock_sock(sk);
1460
1461         switch (optname) {
1462         case SO_VM_SOCKETS_BUFFER_SIZE:
1463                 COPY_IN(val);
1464                 transport->set_buffer_size(vsk, val);
1465                 break;
1466
1467         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1468                 COPY_IN(val);
1469                 transport->set_max_buffer_size(vsk, val);
1470                 break;
1471
1472         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1473                 COPY_IN(val);
1474                 transport->set_min_buffer_size(vsk, val);
1475                 break;
1476
1477         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1478                 struct timeval tv;
1479                 COPY_IN(tv);
1480                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1481                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1482                         vsk->connect_timeout = tv.tv_sec * HZ +
1483                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1484                         if (vsk->connect_timeout == 0)
1485                                 vsk->connect_timeout =
1486                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1487
1488                 } else {
1489                         err = -ERANGE;
1490                 }
1491                 break;
1492         }
1493
1494         default:
1495                 err = -ENOPROTOOPT;
1496                 break;
1497         }
1498
1499 #undef COPY_IN
1500
1501 exit:
1502         release_sock(sk);
1503         return err;
1504 }
1505
1506 static int vsock_stream_getsockopt(struct socket *sock,
1507                                    int level, int optname,
1508                                    char __user *optval,
1509                                    int __user *optlen)
1510 {
1511         int err;
1512         int len;
1513         struct sock *sk;
1514         struct vsock_sock *vsk;
1515         u64 val;
1516
1517         if (level != AF_VSOCK)
1518                 return -ENOPROTOOPT;
1519
1520         err = get_user(len, optlen);
1521         if (err != 0)
1522                 return err;
1523
1524 #define COPY_OUT(_v)                            \
1525         do {                                    \
1526                 if (len < sizeof(_v))           \
1527                         return -EINVAL;         \
1528                                                 \
1529                 len = sizeof(_v);               \
1530                 if (copy_to_user(optval, &_v, len) != 0)        \
1531                         return -EFAULT;                         \
1532                                                                 \
1533         } while (0)
1534
1535         err = 0;
1536         sk = sock->sk;
1537         vsk = vsock_sk(sk);
1538
1539         switch (optname) {
1540         case SO_VM_SOCKETS_BUFFER_SIZE:
1541                 val = transport->get_buffer_size(vsk);
1542                 COPY_OUT(val);
1543                 break;
1544
1545         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1546                 val = transport->get_max_buffer_size(vsk);
1547                 COPY_OUT(val);
1548                 break;
1549
1550         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1551                 val = transport->get_min_buffer_size(vsk);
1552                 COPY_OUT(val);
1553                 break;
1554
1555         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1556                 struct timeval tv;
1557                 tv.tv_sec = vsk->connect_timeout / HZ;
1558                 tv.tv_usec =
1559                     (vsk->connect_timeout -
1560                      tv.tv_sec * HZ) * (1000000 / HZ);
1561                 COPY_OUT(tv);
1562                 break;
1563         }
1564         default:
1565                 return -ENOPROTOOPT;
1566         }
1567
1568         err = put_user(len, optlen);
1569         if (err != 0)
1570                 return -EFAULT;
1571
1572 #undef COPY_OUT
1573
1574         return 0;
1575 }
1576
1577 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1578                                 size_t len)
1579 {
1580         struct sock *sk;
1581         struct vsock_sock *vsk;
1582         ssize_t total_written;
1583         long timeout;
1584         int err;
1585         struct vsock_transport_send_notify_data send_data;
1586
1587         DEFINE_WAIT(wait);
1588
1589         sk = sock->sk;
1590         vsk = vsock_sk(sk);
1591         total_written = 0;
1592         err = 0;
1593
1594         if (msg->msg_flags & MSG_OOB)
1595                 return -EOPNOTSUPP;
1596
1597         lock_sock(sk);
1598
1599         /* Callers should not provide a destination with stream sockets. */
1600         if (msg->msg_namelen) {
1601                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1602                 goto out;
1603         }
1604
1605         /* Send data only if both sides are not shutdown in the direction. */
1606         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1607             vsk->peer_shutdown & RCV_SHUTDOWN) {
1608                 err = -EPIPE;
1609                 goto out;
1610         }
1611
1612         if (sk->sk_state != SS_CONNECTED ||
1613             !vsock_addr_bound(&vsk->local_addr)) {
1614                 err = -ENOTCONN;
1615                 goto out;
1616         }
1617
1618         if (!vsock_addr_bound(&vsk->remote_addr)) {
1619                 err = -EDESTADDRREQ;
1620                 goto out;
1621         }
1622
1623         /* Wait for room in the produce queue to enqueue our user's data. */
1624         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1625
1626         err = transport->notify_send_init(vsk, &send_data);
1627         if (err < 0)
1628                 goto out;
1629
1630         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1631
1632         while (total_written < len) {
1633                 ssize_t written;
1634
1635                 while (vsock_stream_has_space(vsk) == 0 &&
1636                        sk->sk_err == 0 &&
1637                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1638                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1639
1640                         /* Don't wait for non-blocking sockets. */
1641                         if (timeout == 0) {
1642                                 err = -EAGAIN;
1643                                 goto out_wait;
1644                         }
1645
1646                         err = transport->notify_send_pre_block(vsk, &send_data);
1647                         if (err < 0)
1648                                 goto out_wait;
1649
1650                         release_sock(sk);
1651                         timeout = schedule_timeout(timeout);
1652                         lock_sock(sk);
1653                         if (signal_pending(current)) {
1654                                 err = sock_intr_errno(timeout);
1655                                 goto out_wait;
1656                         } else if (timeout == 0) {
1657                                 err = -EAGAIN;
1658                                 goto out_wait;
1659                         }
1660
1661                         prepare_to_wait(sk_sleep(sk), &wait,
1662                                         TASK_INTERRUPTIBLE);
1663                 }
1664
1665                 /* These checks occur both as part of and after the loop
1666                  * conditional since we need to check before and after
1667                  * sleeping.
1668                  */
1669                 if (sk->sk_err) {
1670                         err = -sk->sk_err;
1671                         goto out_wait;
1672                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1673                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1674                         err = -EPIPE;
1675                         goto out_wait;
1676                 }
1677
1678                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1679                 if (err < 0)
1680                         goto out_wait;
1681
1682                 /* Note that enqueue will only write as many bytes as are free
1683                  * in the produce queue, so we don't need to ensure len is
1684                  * smaller than the queue size.  It is the caller's
1685                  * responsibility to check how many bytes we were able to send.
1686                  */
1687
1688                 written = transport->stream_enqueue(
1689                                 vsk, msg,
1690                                 len - total_written);
1691                 if (written < 0) {
1692                         err = -ENOMEM;
1693                         goto out_wait;
1694                 }
1695
1696                 total_written += written;
1697
1698                 err = transport->notify_send_post_enqueue(
1699                                 vsk, written, &send_data);
1700                 if (err < 0)
1701                         goto out_wait;
1702
1703         }
1704
1705 out_wait:
1706         if (total_written > 0)
1707                 err = total_written;
1708         finish_wait(sk_sleep(sk), &wait);
1709 out:
1710         release_sock(sk);
1711         return err;
1712 }
1713
1714
1715 static int
1716 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1717                      int flags)
1718 {
1719         struct sock *sk;
1720         struct vsock_sock *vsk;
1721         int err;
1722         size_t target;
1723         ssize_t copied;
1724         long timeout;
1725         struct vsock_transport_recv_notify_data recv_data;
1726
1727         DEFINE_WAIT(wait);
1728
1729         sk = sock->sk;
1730         vsk = vsock_sk(sk);
1731         err = 0;
1732
1733         lock_sock(sk);
1734
1735         if (sk->sk_state != SS_CONNECTED) {
1736                 /* Recvmsg is supposed to return 0 if a peer performs an
1737                  * orderly shutdown. Differentiate between that case and when a
1738                  * peer has not connected or a local shutdown occured with the
1739                  * SOCK_DONE flag.
1740                  */
1741                 if (sock_flag(sk, SOCK_DONE))
1742                         err = 0;
1743                 else
1744                         err = -ENOTCONN;
1745
1746                 goto out;
1747         }
1748
1749         if (flags & MSG_OOB) {
1750                 err = -EOPNOTSUPP;
1751                 goto out;
1752         }
1753
1754         /* We don't check peer_shutdown flag here since peer may actually shut
1755          * down, but there can be data in the queue that a local socket can
1756          * receive.
1757          */
1758         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1759                 err = 0;
1760                 goto out;
1761         }
1762
1763         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1764          * is not an error.  We may as well bail out now.
1765          */
1766         if (!len) {
1767                 err = 0;
1768                 goto out;
1769         }
1770
1771         /* We must not copy less than target bytes into the user's buffer
1772          * before returning successfully, so we wait for the consume queue to
1773          * have that much data to consume before dequeueing.  Note that this
1774          * makes it impossible to handle cases where target is greater than the
1775          * queue size.
1776          */
1777         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1778         if (target >= transport->stream_rcvhiwat(vsk)) {
1779                 err = -ENOMEM;
1780                 goto out;
1781         }
1782         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1783         copied = 0;
1784
1785         err = transport->notify_recv_init(vsk, target, &recv_data);
1786         if (err < 0)
1787                 goto out;
1788
1789         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1790
1791         while (1) {
1792                 s64 ready = vsock_stream_has_data(vsk);
1793
1794                 if (ready < 0) {
1795                         /* Invalid queue pair content. XXX This should be
1796                          * changed to a connection reset in a later change.
1797                          */
1798
1799                         err = -ENOMEM;
1800                         goto out_wait;
1801                 } else if (ready > 0) {
1802                         ssize_t read;
1803
1804                         err = transport->notify_recv_pre_dequeue(
1805                                         vsk, target, &recv_data);
1806                         if (err < 0)
1807                                 break;
1808
1809                         read = transport->stream_dequeue(
1810                                         vsk, msg,
1811                                         len - copied, flags);
1812                         if (read < 0) {
1813                                 err = -ENOMEM;
1814                                 break;
1815                         }
1816
1817                         copied += read;
1818
1819                         err = transport->notify_recv_post_dequeue(
1820                                         vsk, target, read,
1821                                         !(flags & MSG_PEEK), &recv_data);
1822                         if (err < 0)
1823                                 goto out_wait;
1824
1825                         if (read >= target || flags & MSG_PEEK)
1826                                 break;
1827
1828                         target -= read;
1829                 } else {
1830                         if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1831                             || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1832                                 break;
1833                         }
1834                         /* Don't wait for non-blocking sockets. */
1835                         if (timeout == 0) {
1836                                 err = -EAGAIN;
1837                                 break;
1838                         }
1839
1840                         err = transport->notify_recv_pre_block(
1841                                         vsk, target, &recv_data);
1842                         if (err < 0)
1843                                 break;
1844
1845                         release_sock(sk);
1846                         timeout = schedule_timeout(timeout);
1847                         lock_sock(sk);
1848
1849                         if (signal_pending(current)) {
1850                                 err = sock_intr_errno(timeout);
1851                                 break;
1852                         } else if (timeout == 0) {
1853                                 err = -EAGAIN;
1854                                 break;
1855                         }
1856
1857                         prepare_to_wait(sk_sleep(sk), &wait,
1858                                         TASK_INTERRUPTIBLE);
1859                 }
1860         }
1861
1862         if (sk->sk_err)
1863                 err = -sk->sk_err;
1864         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1865                 err = 0;
1866
1867         if (copied > 0) {
1868                 /* We only do these additional bookkeeping/notification steps
1869                  * if we actually copied something out of the queue pair
1870                  * instead of just peeking ahead.
1871                  */
1872
1873                 if (!(flags & MSG_PEEK)) {
1874                         /* If the other side has shutdown for sending and there
1875                          * is nothing more to read, then modify the socket
1876                          * state.
1877                          */
1878                         if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1879                                 if (vsock_stream_has_data(vsk) <= 0) {
1880                                         sk->sk_state = SS_UNCONNECTED;
1881                                         sock_set_flag(sk, SOCK_DONE);
1882                                         sk->sk_state_change(sk);
1883                                 }
1884                         }
1885                 }
1886                 err = copied;
1887         }
1888
1889 out_wait:
1890         finish_wait(sk_sleep(sk), &wait);
1891 out:
1892         release_sock(sk);
1893         return err;
1894 }
1895
1896 static const struct proto_ops vsock_stream_ops = {
1897         .family = PF_VSOCK,
1898         .owner = THIS_MODULE,
1899         .release = vsock_release,
1900         .bind = vsock_bind,
1901         .connect = vsock_stream_connect,
1902         .socketpair = sock_no_socketpair,
1903         .accept = vsock_accept,
1904         .getname = vsock_getname,
1905         .poll = vsock_poll,
1906         .ioctl = sock_no_ioctl,
1907         .listen = vsock_listen,
1908         .shutdown = vsock_shutdown,
1909         .setsockopt = vsock_stream_setsockopt,
1910         .getsockopt = vsock_stream_getsockopt,
1911         .sendmsg = vsock_stream_sendmsg,
1912         .recvmsg = vsock_stream_recvmsg,
1913         .mmap = sock_no_mmap,
1914         .sendpage = sock_no_sendpage,
1915 };
1916
1917 static int vsock_create(struct net *net, struct socket *sock,
1918                         int protocol, int kern)
1919 {
1920         if (!sock)
1921                 return -EINVAL;
1922
1923         if (protocol && protocol != PF_VSOCK)
1924                 return -EPROTONOSUPPORT;
1925
1926         switch (sock->type) {
1927         case SOCK_DGRAM:
1928                 sock->ops = &vsock_dgram_ops;
1929                 break;
1930         case SOCK_STREAM:
1931                 sock->ops = &vsock_stream_ops;
1932                 break;
1933         default:
1934                 return -ESOCKTNOSUPPORT;
1935         }
1936
1937         sock->state = SS_UNCONNECTED;
1938
1939         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1940 }
1941
1942 static const struct net_proto_family vsock_family_ops = {
1943         .family = AF_VSOCK,
1944         .create = vsock_create,
1945         .owner = THIS_MODULE,
1946 };
1947
1948 static long vsock_dev_do_ioctl(struct file *filp,
1949                                unsigned int cmd, void __user *ptr)
1950 {
1951         u32 __user *p = ptr;
1952         int retval = 0;
1953
1954         switch (cmd) {
1955         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1956                 if (put_user(transport->get_local_cid(), p) != 0)
1957                         retval = -EFAULT;
1958                 break;
1959
1960         default:
1961                 pr_err("Unknown ioctl %d\n", cmd);
1962                 retval = -EINVAL;
1963         }
1964
1965         return retval;
1966 }
1967
1968 static long vsock_dev_ioctl(struct file *filp,
1969                             unsigned int cmd, unsigned long arg)
1970 {
1971         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1972 }
1973
1974 #ifdef CONFIG_COMPAT
1975 static long vsock_dev_compat_ioctl(struct file *filp,
1976                                    unsigned int cmd, unsigned long arg)
1977 {
1978         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1979 }
1980 #endif
1981
1982 static const struct file_operations vsock_device_ops = {
1983         .owner          = THIS_MODULE,
1984         .unlocked_ioctl = vsock_dev_ioctl,
1985 #ifdef CONFIG_COMPAT
1986         .compat_ioctl   = vsock_dev_compat_ioctl,
1987 #endif
1988         .open           = nonseekable_open,
1989 };
1990
1991 static struct miscdevice vsock_device = {
1992         .name           = "vsock",
1993         .fops           = &vsock_device_ops,
1994 };
1995
1996 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1997 {
1998         int err = mutex_lock_interruptible(&vsock_register_mutex);
1999
2000         if (err)
2001                 return err;
2002
2003         if (transport) {
2004                 err = -EBUSY;
2005                 goto err_busy;
2006         }
2007
2008         /* Transport must be the owner of the protocol so that it can't
2009          * unload while there are open sockets.
2010          */
2011         vsock_proto.owner = owner;
2012         transport = t;
2013
2014         vsock_init_tables();
2015
2016         vsock_device.minor = MISC_DYNAMIC_MINOR;
2017         err = misc_register(&vsock_device);
2018         if (err) {
2019                 pr_err("Failed to register misc device\n");
2020                 goto err_reset_transport;
2021         }
2022
2023         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2024         if (err) {
2025                 pr_err("Cannot register vsock protocol\n");
2026                 goto err_deregister_misc;
2027         }
2028
2029         err = sock_register(&vsock_family_ops);
2030         if (err) {
2031                 pr_err("could not register af_vsock (%d) address family: %d\n",
2032                        AF_VSOCK, err);
2033                 goto err_unregister_proto;
2034         }
2035
2036         mutex_unlock(&vsock_register_mutex);
2037         return 0;
2038
2039 err_unregister_proto:
2040         proto_unregister(&vsock_proto);
2041 err_deregister_misc:
2042         misc_deregister(&vsock_device);
2043 err_reset_transport:
2044         transport = NULL;
2045 err_busy:
2046         mutex_unlock(&vsock_register_mutex);
2047         return err;
2048 }
2049 EXPORT_SYMBOL_GPL(__vsock_core_init);
2050
2051 void vsock_core_exit(void)
2052 {
2053         mutex_lock(&vsock_register_mutex);
2054
2055         misc_deregister(&vsock_device);
2056         sock_unregister(AF_VSOCK);
2057         proto_unregister(&vsock_proto);
2058
2059         /* We do not want the assignment below re-ordered. */
2060         mb();
2061         transport = NULL;
2062
2063         mutex_unlock(&vsock_register_mutex);
2064 }
2065 EXPORT_SYMBOL_GPL(vsock_core_exit);
2066
2067 MODULE_AUTHOR("VMware, Inc.");
2068 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2069 MODULE_VERSION("1.0.1.0-k");
2070 MODULE_LICENSE("GPL v2");