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 TCP_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  * - sk->sk_state uses the TCP state constants because they are widely used by
  87  * other address families and exposed to userspace tools like ss(8):
  88  *
  89  *   TCP_CLOSE - unconnected
  90  *   TCP_SYN_SENT - connecting
  91  *   TCP_ESTABLISHED - connected
  92  *   TCP_CLOSING - disconnecting
  93  *   TCP_LISTEN - listening
  94  */
  95
  96 #include <linux/types.h>
  97 #include <linux/bitops.h>
  98 #include <linux/cred.h>
  99 #include <linux/init.h>
 100 #include <linux/io.h>
 101 #include <linux/kernel.h>
 102 #include <linux/sched/signal.h>
 103 #include <linux/kmod.h>
 104 #include <linux/list.h>
 105 #include <linux/miscdevice.h>
 106 #include <linux/module.h>
 107 #include <linux/mutex.h>
 108 #include <linux/net.h>
 109 #include <linux/poll.h>
 110 #include <linux/random.h>
 111 #include <linux/skbuff.h>
 112 #include <linux/smp.h>
 113 #include <linux/socket.h>
 114 #include <linux/stddef.h>
 115 #include <linux/unistd.h>
 116 #include <linux/wait.h>
 117 #include <linux/workqueue.h>
 118 #include <net/sock.h>
 119 #include <net/af_vsock.h>
 120
 121 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 122 static void vsock_sk_destruct(struct sock *sk);
 123 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 124
 125 /* Protocol family. */
 126 static struct proto vsock_proto = {
 127         .name = "AF_VSOCK",
 128         .owner = THIS_MODULE,
 129         .obj_size = sizeof(struct vsock_sock),
 130 };
 131
 132 /* The default peer timeout indicates how long we will wait for a peer response
 133  * to a control message.
 134  */
 135 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 136
 137 static const struct vsock_transport *transport;
 138 static DEFINE_MUTEX(vsock_register_mutex);
 139
 140 /**** EXPORTS ****/
 141
 142 /* Get the ID of the local context.  This is transport dependent. */
 143
 144 int vm_sockets_get_local_cid(void)
 145 {
 146         return transport->get_local_cid();
 147 }
 148 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 149
 150 /**** UTILS ****/
 151
 152 /* Each bound VSocket is stored in the bind hash table and each connected
 153  * VSocket is stored in the connected hash table.
 154  *
 155  * Unbound sockets are all put on the same list attached to the end of the hash
 156  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 157  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 158  * represents the list that addr hashes to).
 159  *
 160  * Specifically, we initialize the vsock_bind_table array to a size of
 161  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 162  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 163  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 164  * mods with VSOCK_HASH_SIZE to ensure this.
 165  */
 166 #define MAX_PORT_RETRIES        24
 167
 168 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 169 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 170 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 171
 172 /* XXX This can probably be implemented in a better way. */
 173 #define VSOCK_CONN_HASH(src, dst)                               \
 174         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 175 #define vsock_connected_sockets(src, dst)               \
 176         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 177 #define vsock_connected_sockets_vsk(vsk)                                \
 178         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 179
 180 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 181 EXPORT_SYMBOL_GPL(vsock_bind_table);
 182 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 183 EXPORT_SYMBOL_GPL(vsock_connected_table);
 184 DEFINE_SPINLOCK(vsock_table_lock);
 185 EXPORT_SYMBOL_GPL(vsock_table_lock);
 186
 187 /* Autobind this socket to the local address if necessary. */
 188 static int vsock_auto_bind(struct vsock_sock *vsk)
 189 {
 190         struct sock *sk = sk_vsock(vsk);
 191         struct sockaddr_vm local_addr;
 192
 193         if (vsock_addr_bound(&vsk->local_addr))
 194                 return 0;
 195         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 196         return __vsock_bind(sk, &local_addr);
 197 }
 198
 199 static int __init vsock_init_tables(void)
 200 {
 201         int i;
 202
 203         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 204                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 205
 206         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 207                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 208         return 0;
 209 }
 210
 211 static void __vsock_insert_bound(struct list_head *list,
 212                                  struct vsock_sock *vsk)
 213 {
 214         sock_hold(&vsk->sk);
 215         list_add(&vsk->bound_table, list);
 216 }
 217
 218 static void __vsock_insert_connected(struct list_head *list,
 219                                      struct vsock_sock *vsk)
 220 {
 221         sock_hold(&vsk->sk);
 222         list_add(&vsk->connected_table, list);
 223 }
 224
 225 static void __vsock_remove_bound(struct vsock_sock *vsk)
 226 {
 227         list_del_init(&vsk->bound_table);
 228         sock_put(&vsk->sk);
 229 }
 230
 231 static void __vsock_remove_connected(struct vsock_sock *vsk)
 232 {
 233         list_del_init(&vsk->connected_table);
 234         sock_put(&vsk->sk);
 235 }
 236
 237 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 238 {
 239         struct vsock_sock *vsk;
 240
 241         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 242                 if (addr->svm_port == vsk->local_addr.svm_port)
 243                         return sk_vsock(vsk);
 244
 245         return NULL;
 246 }
 247
 248 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 249                                                   struct sockaddr_vm *dst)
 250 {
 251         struct vsock_sock *vsk;
 252
 253         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 254                             connected_table) {
 255                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 256                     dst->svm_port == vsk->local_addr.svm_port) {
 257                         return sk_vsock(vsk);
 258                 }
 259         }
 260
 261         return NULL;
 262 }
 263
 264 static void vsock_insert_unbound(struct vsock_sock *vsk)
 265 {
 266         spin_lock_bh(&vsock_table_lock);
 267         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 268         spin_unlock_bh(&vsock_table_lock);
 269 }
 270
 271 void vsock_insert_connected(struct vsock_sock *vsk)
 272 {
 273         struct list_head *list = vsock_connected_sockets(
 274                 &vsk->remote_addr, &vsk->local_addr);
 275
 276         spin_lock_bh(&vsock_table_lock);
 277         __vsock_insert_connected(list, vsk);
 278         spin_unlock_bh(&vsock_table_lock);
 279 }
 280 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 281
 282 void vsock_remove_bound(struct vsock_sock *vsk)
 283 {
 284         spin_lock_bh(&vsock_table_lock);
 285         __vsock_remove_bound(vsk);
 286         spin_unlock_bh(&vsock_table_lock);
 287 }
 288 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 289
 290 void vsock_remove_connected(struct vsock_sock *vsk)
 291 {
 292         spin_lock_bh(&vsock_table_lock);
 293         __vsock_remove_connected(vsk);
 294         spin_unlock_bh(&vsock_table_lock);
 295 }
 296 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 297
 298 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 299 {
 300         struct sock *sk;
 301
 302         spin_lock_bh(&vsock_table_lock);
 303         sk = __vsock_find_bound_socket(addr);
 304         if (sk)
 305                 sock_hold(sk);
 306
 307         spin_unlock_bh(&vsock_table_lock);
 308
 309         return sk;
 310 }
 311 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 312
 313 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 314                                          struct sockaddr_vm *dst)
 315 {
 316         struct sock *sk;
 317
 318         spin_lock_bh(&vsock_table_lock);
 319         sk = __vsock_find_connected_socket(src, dst);
 320         if (sk)
 321                 sock_hold(sk);
 322
 323         spin_unlock_bh(&vsock_table_lock);
 324
 325         return sk;
 326 }
 327 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 328
 329 static bool vsock_in_bound_table(struct vsock_sock *vsk)
 330 {
 331         bool ret;
 332
 333         spin_lock_bh(&vsock_table_lock);
 334         ret = __vsock_in_bound_table(vsk);
 335         spin_unlock_bh(&vsock_table_lock);
 336
 337         return ret;
 338 }
 339
 340 static bool vsock_in_connected_table(struct vsock_sock *vsk)
 341 {
 342         bool ret;
 343
 344         spin_lock_bh(&vsock_table_lock);
 345         ret = __vsock_in_connected_table(vsk);
 346         spin_unlock_bh(&vsock_table_lock);
 347
 348         return ret;
 349 }
 350
 351 void vsock_remove_sock(struct vsock_sock *vsk)
 352 {
 353         if (vsock_in_bound_table(vsk))
 354                 vsock_remove_bound(vsk);
 355
 356         if (vsock_in_connected_table(vsk))
 357                 vsock_remove_connected(vsk);
 358 }
 359 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 360
 361 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 362 {
 363         int i;
 364
 365         spin_lock_bh(&vsock_table_lock);
 366
 367         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 368                 struct vsock_sock *vsk;
 369                 list_for_each_entry(vsk, &vsock_connected_table[i],
 370                                     connected_table)
 371                         fn(sk_vsock(vsk));
 372         }
 373
 374         spin_unlock_bh(&vsock_table_lock);
 375 }
 376 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 377
 378 void vsock_add_pending(struct sock *listener, struct sock *pending)
 379 {
 380         struct vsock_sock *vlistener;
 381         struct vsock_sock *vpending;
 382
 383         vlistener = vsock_sk(listener);
 384         vpending = vsock_sk(pending);
 385
 386         sock_hold(pending);
 387         sock_hold(listener);
 388         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 389 }
 390 EXPORT_SYMBOL_GPL(vsock_add_pending);
 391
 392 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 393 {
 394         struct vsock_sock *vpending = vsock_sk(pending);
 395
 396         list_del_init(&vpending->pending_links);
 397         sock_put(listener);
 398         sock_put(pending);
 399 }
 400 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 401
 402 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 403 {
 404         struct vsock_sock *vlistener;
 405         struct vsock_sock *vconnected;
 406
 407         vlistener = vsock_sk(listener);
 408         vconnected = vsock_sk(connected);
 409
 410         sock_hold(connected);
 411         sock_hold(listener);
 412         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 413 }
 414 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 415
 416 static struct sock *vsock_dequeue_accept(struct sock *listener)
 417 {
 418         struct vsock_sock *vlistener;
 419         struct vsock_sock *vconnected;
 420
 421         vlistener = vsock_sk(listener);
 422
 423         if (list_empty(&vlistener->accept_queue))
 424                 return NULL;
 425
 426         vconnected = list_entry(vlistener->accept_queue.next,
 427                                 struct vsock_sock, accept_queue);
 428
 429         list_del_init(&vconnected->accept_queue);
 430         sock_put(listener);
 431         /* The caller will need a reference on the connected socket so we let
 432          * it call sock_put().
 433          */
 434
 435         return sk_vsock(vconnected);
 436 }
 437
 438 static bool vsock_is_accept_queue_empty(struct sock *sk)
 439 {
 440         struct vsock_sock *vsk = vsock_sk(sk);
 441         return list_empty(&vsk->accept_queue);
 442 }
 443
 444 static bool vsock_is_pending(struct sock *sk)
 445 {
 446         struct vsock_sock *vsk = vsock_sk(sk);
 447         return !list_empty(&vsk->pending_links);
 448 }
 449
 450 static int vsock_send_shutdown(struct sock *sk, int mode)
 451 {
 452         return transport->shutdown(vsock_sk(sk), mode);
 453 }
 454
 455 static void vsock_pending_work(struct work_struct *work)
 456 {
 457         struct sock *sk;
 458         struct sock *listener;
 459         struct vsock_sock *vsk;
 460         bool cleanup;
 461
 462         vsk = container_of(work, struct vsock_sock, pending_work.work);
 463         sk = sk_vsock(vsk);
 464         listener = vsk->listener;
 465         cleanup = true;
 466
 467         lock_sock(listener);
 468         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 469
 470         if (vsock_is_pending(sk)) {
 471                 vsock_remove_pending(listener, sk);
 472
 473                 listener->sk_ack_backlog--;
 474         } else if (!vsk->rejected) {
 475                 /* We are not on the pending list and accept() did not reject
 476                  * us, so we must have been accepted by our user process.  We
 477                  * just need to drop our references to the sockets and be on
 478                  * our way.
 479                  */
 480                 cleanup = false;
 481                 goto out;
 482         }
 483
 484         /* We need to remove ourself from the global connected sockets list so
 485          * incoming packets can't find this socket, and to reduce the reference
 486          * count.
 487          */
 488         if (vsock_in_connected_table(vsk))
 489                 vsock_remove_connected(vsk);
 490
 491         sk->sk_state = TCP_CLOSE;
 492
 493 out:
 494         release_sock(sk);
 495         release_sock(listener);
 496         if (cleanup)
 497                 sock_put(sk);
 498
 499         sock_put(sk);
 500         sock_put(listener);
 501 }
 502
 503 /**** SOCKET OPERATIONS ****/
 504
 505 static int __vsock_bind_stream(struct vsock_sock *vsk,
 506                                struct sockaddr_vm *addr)
 507 {
 508         static u32 port = 0;
 509         struct sockaddr_vm new_addr;
 510
 511         if (!port)
 512                 port = LAST_RESERVED_PORT + 1 +
 513                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 514
 515         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 516
 517         if (addr->svm_port == VMADDR_PORT_ANY) {
 518                 bool found = false;
 519                 unsigned int i;
 520
 521                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 522                         if (port <= LAST_RESERVED_PORT)
 523                                 port = LAST_RESERVED_PORT + 1;
 524
 525                         new_addr.svm_port = port++;
 526
 527                         if (!__vsock_find_bound_socket(&new_addr)) {
 528                                 found = true;
 529                                 break;
 530                         }
 531                 }
 532
 533                 if (!found)
 534                         return -EADDRNOTAVAIL;
 535         } else {
 536                 /* If port is in reserved range, ensure caller
 537                  * has necessary privileges.
 538                  */
 539                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 540                     !capable(CAP_NET_BIND_SERVICE)) {
 541                         return -EACCES;
 542                 }
 543
 544                 if (__vsock_find_bound_socket(&new_addr))
 545                         return -EADDRINUSE;
 546         }
 547
 548         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 549
 550         /* Remove stream sockets from the unbound list and add them to the hash
 551          * table for easy lookup by its address.  The unbound list is simply an
 552          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 553          */
 554         __vsock_remove_bound(vsk);
 555         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 556
 557         return 0;
 558 }
 559
 560 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 561                               struct sockaddr_vm *addr)
 562 {
 563         return transport->dgram_bind(vsk, addr);
 564 }
 565
 566 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 567 {
 568         struct vsock_sock *vsk = vsock_sk(sk);
 569         u32 cid;
 570         int retval;
 571
 572         /* First ensure this socket isn't already bound. */
 573         if (vsock_addr_bound(&vsk->local_addr))
 574                 return -EINVAL;
 575
 576         /* Now bind to the provided address or select appropriate values if
 577          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 578          * like AF_INET prevents binding to a non-local IP address (in most
 579          * cases), we only allow binding to the local CID.
 580          */
 581         cid = transport->get_local_cid();
 582         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 583                 return -EADDRNOTAVAIL;
 584
 585         switch (sk->sk_socket->type) {
 586         case SOCK_STREAM:
 587                 spin_lock_bh(&vsock_table_lock);
 588                 retval = __vsock_bind_stream(vsk, addr);
 589                 spin_unlock_bh(&vsock_table_lock);
 590                 break;
 591
 592         case SOCK_DGRAM:
 593                 retval = __vsock_bind_dgram(vsk, addr);
 594                 break;
 595
 596         default:
 597                 retval = -EINVAL;
 598                 break;
 599         }
 600
 601         return retval;
 602 }
 603
 604 static void vsock_connect_timeout(struct work_struct *work);
 605
 606 struct sock *__vsock_create(struct net *net,
 607                             struct socket *sock,
 608                             struct sock *parent,
 609                             gfp_t priority,
 610                             unsigned short type,
 611                             int kern)
 612 {
 613         struct sock *sk;
 614         struct vsock_sock *psk;
 615         struct vsock_sock *vsk;
 616
 617         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 618         if (!sk)
 619                 return NULL;
 620
 621         sock_init_data(sock, sk);
 622
 623         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 624          * non-NULL. We make sure that our sockets always have a type by
 625          * setting it here if needed.
 626          */
 627         if (!sock)
 628                 sk->sk_type = type;
 629
 630         vsk = vsock_sk(sk);
 631         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 632         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 633
 634         sk->sk_destruct = vsock_sk_destruct;
 635         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 636         sock_reset_flag(sk, SOCK_DONE);
 637
 638         INIT_LIST_HEAD(&vsk->bound_table);
 639         INIT_LIST_HEAD(&vsk->connected_table);
 640         vsk->listener = NULL;
 641         INIT_LIST_HEAD(&vsk->pending_links);
 642         INIT_LIST_HEAD(&vsk->accept_queue);
 643         vsk->rejected = false;
 644         vsk->sent_request = false;
 645         vsk->ignore_connecting_rst = false;
 646         vsk->peer_shutdown = 0;
 647         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 648         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 649
 650         psk = parent ? vsock_sk(parent) : NULL;
 651         if (parent) {
 652                 vsk->trusted = psk->trusted;
 653                 vsk->owner = get_cred(psk->owner);
 654                 vsk->connect_timeout = psk->connect_timeout;
 655         } else {
 656                 vsk->trusted = capable(CAP_NET_ADMIN);
 657                 vsk->owner = get_current_cred();
 658                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 659         }
 660
 661         if (transport->init(vsk, psk) < 0) {
 662                 sk_free(sk);
 663                 return NULL;
 664         }
 665
 666         if (sock)
 667                 vsock_insert_unbound(vsk);
 668
 669         return sk;
 670 }
 671 EXPORT_SYMBOL_GPL(__vsock_create);
 672
 673 static void __vsock_release(struct sock *sk)
 674 {
 675         if (sk) {
 676                 struct sk_buff *skb;
 677                 struct sock *pending;
 678                 struct vsock_sock *vsk;
 679
 680                 vsk = vsock_sk(sk);
 681                 pending = NULL; /* Compiler warning. */
 682
 683                 transport->release(vsk);
 684
 685                 lock_sock(sk);
 686                 sock_orphan(sk);
 687                 sk->sk_shutdown = SHUTDOWN_MASK;
 688
 689                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 690                         kfree_skb(skb);
 691
 692                 /* Clean up any sockets that never were accepted. */
 693                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 694                         __vsock_release(pending);
 695                         sock_put(pending);
 696                 }
 697
 698                 release_sock(sk);
 699                 sock_put(sk);
 700         }
 701 }
 702
 703 static void vsock_sk_destruct(struct sock *sk)
 704 {
 705         struct vsock_sock *vsk = vsock_sk(sk);
 706
 707         transport->destruct(vsk);
 708
 709         /* When clearing these addresses, there's no need to set the family and
 710          * possibly register the address family with the kernel.
 711          */
 712         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 713         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 714
 715         put_cred(vsk->owner);
 716 }
 717
 718 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 719 {
 720         int err;
 721
 722         err = sock_queue_rcv_skb(sk, skb);
 723         if (err)
 724                 kfree_skb(skb);
 725
 726         return err;
 727 }
 728
 729 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 730 {
 731         return transport->stream_has_data(vsk);
 732 }
 733 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 734
 735 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 736 {
 737         return transport->stream_has_space(vsk);
 738 }
 739 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 740
 741 static int vsock_release(struct socket *sock)
 742 {
 743         __vsock_release(sock->sk);
 744         sock->sk = NULL;
 745         sock->state = SS_FREE;
 746
 747         return 0;
 748 }
 749
 750 static int
 751 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 752 {
 753         int err;
 754         struct sock *sk;
 755         struct sockaddr_vm *vm_addr;
 756
 757         sk = sock->sk;
 758
 759         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 760                 return -EINVAL;
 761
 762         lock_sock(sk);
 763         err = __vsock_bind(sk, vm_addr);
 764         release_sock(sk);
 765
 766         return err;
 767 }
 768
 769 static int vsock_getname(struct socket *sock,
 770                          struct sockaddr *addr, int peer)
 771 {
 772         int err;
 773         struct sock *sk;
 774         struct vsock_sock *vsk;
 775         struct sockaddr_vm *vm_addr;
 776
 777         sk = sock->sk;
 778         vsk = vsock_sk(sk);
 779         err = 0;
 780
 781         lock_sock(sk);
 782
 783         if (peer) {
 784                 if (sock->state != SS_CONNECTED) {
 785                         err = -ENOTCONN;
 786                         goto out;
 787                 }
 788                 vm_addr = &vsk->remote_addr;
 789         } else {
 790                 vm_addr = &vsk->local_addr;
 791         }
 792
 793         if (!vm_addr) {
 794                 err = -EINVAL;
 795                 goto out;
 796         }
 797
 798         /* sys_getsockname() and sys_getpeername() pass us a
 799          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 800          * that macro is defined in socket.c instead of .h, so we hardcode its
 801          * value here.
 802          */
 803         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 804         memcpy(addr, vm_addr, sizeof(*vm_addr));
 805         err = sizeof(*vm_addr);
 806
 807 out:
 808         release_sock(sk);
 809         return err;
 810 }
 811
 812 static int vsock_shutdown(struct socket *sock, int mode)
 813 {
 814         int err;
 815         struct sock *sk;
 816
 817         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 818          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 819          * here like the other address families do.  Note also that the
 820          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 821          * which is what we want.
 822          */
 823         mode++;
 824
 825         if ((mode & ~SHUTDOWN_MASK) || !mode)
 826                 return -EINVAL;
 827
 828         /* If this is a STREAM socket and it is not connected then bail out
 829          * immediately.  If it is a DGRAM socket then we must first kick the
 830          * socket so that it wakes up from any sleeping calls, for example
 831          * recv(), and then afterwards return the error.
 832          */
 833
 834         sk = sock->sk;
 835         if (sock->state == SS_UNCONNECTED) {
 836                 err = -ENOTCONN;
 837                 if (sk->sk_type == SOCK_STREAM)
 838                         return err;
 839         } else {
 840                 sock->state = SS_DISCONNECTING;
 841                 err = 0;
 842         }
 843
 844         /* Receive and send shutdowns are treated alike. */
 845         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 846         if (mode) {
 847                 lock_sock(sk);
 848                 sk->sk_shutdown |= mode;
 849                 sk->sk_state_change(sk);
 850                 release_sock(sk);
 851
 852                 if (sk->sk_type == SOCK_STREAM) {
 853                         sock_reset_flag(sk, SOCK_DONE);
 854                         vsock_send_shutdown(sk, mode);
 855                 }
 856         }
 857
 858         return err;
 859 }
 860
 861 static __poll_t vsock_poll(struct file *file, struct socket *sock,
 862                                poll_table *wait)
 863 {
 864         struct sock *sk;
 865         __poll_t mask;
 866         struct vsock_sock *vsk;
 867
 868         sk = sock->sk;
 869         vsk = vsock_sk(sk);
 870
 871         poll_wait(file, sk_sleep(sk), wait);
 872         mask = 0;
 873
 874         if (sk->sk_err)
 875                 /* Signify that there has been an error on this socket. */
 876                 mask |= EPOLLERR;
 877
 878         /* INET sockets treat local write shutdown and peer write shutdown as a
 879          * case of EPOLLHUP set.
 880          */
 881         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 882             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 883              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 884                 mask |= EPOLLHUP;
 885         }
 886
 887         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 888             vsk->peer_shutdown & SEND_SHUTDOWN) {
 889                 mask |= EPOLLRDHUP;
 890         }
 891
 892         if (sock->type == SOCK_DGRAM) {
 893                 /* For datagram sockets we can read if there is something in
 894                  * the queue and write as long as the socket isn't shutdown for
 895                  * sending.
 896                  */
 897                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 898                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 899                         mask |= EPOLLIN | EPOLLRDNORM;
 900                 }
 901
 902                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 903                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
 904
 905         } else if (sock->type == SOCK_STREAM) {
 906                 lock_sock(sk);
 907
 908                 /* Listening sockets that have connections in their accept
 909                  * queue can be read.
 910                  */
 911                 if (sk->sk_state == TCP_LISTEN
 912                     && !vsock_is_accept_queue_empty(sk))
 913                         mask |= EPOLLIN | EPOLLRDNORM;
 914
 915                 /* If there is something in the queue then we can read. */
 916                 if (transport->stream_is_active(vsk) &&
 917                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 918                         bool data_ready_now = false;
 919                         int ret = transport->notify_poll_in(
 920                                         vsk, 1, &data_ready_now);
 921                         if (ret < 0) {
 922                                 mask |= EPOLLERR;
 923                         } else {
 924                                 if (data_ready_now)
 925                                         mask |= EPOLLIN | EPOLLRDNORM;
 926
 927                         }
 928                 }
 929
 930                 /* Sockets whose connections have been closed, reset, or
 931                  * terminated should also be considered read, and we check the
 932                  * shutdown flag for that.
 933                  */
 934                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 935                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 936                         mask |= EPOLLIN | EPOLLRDNORM;
 937                 }
 938
 939                 /* Connected sockets that can produce data can be written. */
 940                 if (sk->sk_state == TCP_ESTABLISHED) {
 941                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 942                                 bool space_avail_now = false;
 943                                 int ret = transport->notify_poll_out(
 944                                                 vsk, 1, &space_avail_now);
 945                                 if (ret < 0) {
 946                                         mask |= EPOLLERR;
 947                                 } else {
 948                                         if (space_avail_now)
 949                                                 /* Remove EPOLLWRBAND since INET
 950                                                  * sockets are not setting it.
 951                                                  */
 952                                                 mask |= EPOLLOUT | EPOLLWRNORM;
 953
 954                                 }
 955                         }
 956                 }
 957
 958                 /* Simulate INET socket poll behaviors, which sets
 959                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
 960                  * but local send is not shutdown.
 961                  */
 962                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
 963                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 964                                 mask |= EPOLLOUT | EPOLLWRNORM;
 965
 966                 }
 967
 968                 release_sock(sk);
 969         }
 970
 971         return mask;
 972 }
 973
 974 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 975                                size_t len)
 976 {
 977         int err;
 978         struct sock *sk;
 979         struct vsock_sock *vsk;
 980         struct sockaddr_vm *remote_addr;
 981
 982         if (msg->msg_flags & MSG_OOB)
 983                 return -EOPNOTSUPP;
 984
 985         /* For now, MSG_DONTWAIT is always assumed... */
 986         err = 0;
 987         sk = sock->sk;
 988         vsk = vsock_sk(sk);
 989
 990         lock_sock(sk);
 991
 992         err = vsock_auto_bind(vsk);
 993         if (err)
 994                 goto out;
 995
 996
 997         /* If the provided message contains an address, use that.  Otherwise
 998          * fall back on the socket's remote handle (if it has been connected).
 999          */
1000         if (msg->msg_name &&
1001             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1002                             &remote_addr) == 0) {
1003                 /* Ensure this address is of the right type and is a valid
1004                  * destination.
1005                  */
1006
1007                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1008                         remote_addr->svm_cid = transport->get_local_cid();
1009
1010                 if (!vsock_addr_bound(remote_addr)) {
1011                         err = -EINVAL;
1012                         goto out;
1013                 }
1014         } else if (sock->state == SS_CONNECTED) {
1015                 remote_addr = &vsk->remote_addr;
1016
1017                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1018                         remote_addr->svm_cid = transport->get_local_cid();
1019
1020                 /* XXX Should connect() or this function ensure remote_addr is
1021                  * bound?
1022                  */
1023                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1024                         err = -EINVAL;
1025                         goto out;
1026                 }
1027         } else {
1028                 err = -EINVAL;
1029                 goto out;
1030         }
1031
1032         if (!transport->dgram_allow(remote_addr->svm_cid,
1033                                     remote_addr->svm_port)) {
1034                 err = -EINVAL;
1035                 goto out;
1036         }
1037
1038         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1039
1040 out:
1041         release_sock(sk);
1042         return err;
1043 }
1044
1045 static int vsock_dgram_connect(struct socket *sock,
1046                                struct sockaddr *addr, int addr_len, int flags)
1047 {
1048         int err;
1049         struct sock *sk;
1050         struct vsock_sock *vsk;
1051         struct sockaddr_vm *remote_addr;
1052
1053         sk = sock->sk;
1054         vsk = vsock_sk(sk);
1055
1056         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1057         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1058                 lock_sock(sk);
1059                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1060                                 VMADDR_PORT_ANY);
1061                 sock->state = SS_UNCONNECTED;
1062                 release_sock(sk);
1063                 return 0;
1064         } else if (err != 0)
1065                 return -EINVAL;
1066
1067         lock_sock(sk);
1068
1069         err = vsock_auto_bind(vsk);
1070         if (err)
1071                 goto out;
1072
1073         if (!transport->dgram_allow(remote_addr->svm_cid,
1074                                     remote_addr->svm_port)) {
1075                 err = -EINVAL;
1076                 goto out;
1077         }
1078
1079         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1080         sock->state = SS_CONNECTED;
1081
1082 out:
1083         release_sock(sk);
1084         return err;
1085 }
1086
1087 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1088                                size_t len, int flags)
1089 {
1090         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1091 }
1092
1093 static const struct proto_ops vsock_dgram_ops = {
1094         .family = PF_VSOCK,
1095         .owner = THIS_MODULE,
1096         .release = vsock_release,
1097         .bind = vsock_bind,
1098         .connect = vsock_dgram_connect,
1099         .socketpair = sock_no_socketpair,
1100         .accept = sock_no_accept,
1101         .getname = vsock_getname,
1102         .poll = vsock_poll,
1103         .ioctl = sock_no_ioctl,
1104         .listen = sock_no_listen,
1105         .shutdown = vsock_shutdown,
1106         .setsockopt = sock_no_setsockopt,
1107         .getsockopt = sock_no_getsockopt,
1108         .sendmsg = vsock_dgram_sendmsg,
1109         .recvmsg = vsock_dgram_recvmsg,
1110         .mmap = sock_no_mmap,
1111         .sendpage = sock_no_sendpage,
1112 };
1113
1114 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1115 {
1116         if (!transport->cancel_pkt)
1117                 return -EOPNOTSUPP;
1118
1119         return transport->cancel_pkt(vsk);
1120 }
1121
1122 static void vsock_connect_timeout(struct work_struct *work)
1123 {
1124         struct sock *sk;
1125         struct vsock_sock *vsk;
1126         int cancel = 0;
1127
1128         vsk = container_of(work, struct vsock_sock, connect_work.work);
1129         sk = sk_vsock(vsk);
1130
1131         lock_sock(sk);
1132         if (sk->sk_state == TCP_SYN_SENT &&
1133             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1134                 sk->sk_state = TCP_CLOSE;
1135                 sk->sk_err = ETIMEDOUT;
1136                 sk->sk_error_report(sk);
1137                 cancel = 1;
1138         }
1139         release_sock(sk);
1140         if (cancel)
1141                 vsock_transport_cancel_pkt(vsk);
1142
1143         sock_put(sk);
1144 }
1145
1146 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1147                                 int addr_len, int flags)
1148 {
1149         int err;
1150         struct sock *sk;
1151         struct vsock_sock *vsk;
1152         struct sockaddr_vm *remote_addr;
1153         long timeout;
1154         DEFINE_WAIT(wait);
1155
1156         err = 0;
1157         sk = sock->sk;
1158         vsk = vsock_sk(sk);
1159
1160         lock_sock(sk);
1161
1162         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1163         switch (sock->state) {
1164         case SS_CONNECTED:
1165                 err = -EISCONN;
1166                 goto out;
1167         case SS_DISCONNECTING:
1168                 err = -EINVAL;
1169                 goto out;
1170         case SS_CONNECTING:
1171                 /* This continues on so we can move sock into the SS_CONNECTED
1172                  * state once the connection has completed (at which point err
1173                  * will be set to zero also).  Otherwise, we will either wait
1174                  * for the connection or return -EALREADY should this be a
1175                  * non-blocking call.
1176                  */
1177                 err = -EALREADY;
1178                 break;
1179         default:
1180                 if ((sk->sk_state == TCP_LISTEN) ||
1181                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1182                         err = -EINVAL;
1183                         goto out;
1184                 }
1185
1186                 /* The hypervisor and well-known contexts do not have socket
1187                  * endpoints.
1188                  */
1189                 if (!transport->stream_allow(remote_addr->svm_cid,
1190                                              remote_addr->svm_port)) {
1191                         err = -ENETUNREACH;
1192                         goto out;
1193                 }
1194
1195                 /* Set the remote address that we are connecting to. */
1196                 memcpy(&vsk->remote_addr, remote_addr,
1197                        sizeof(vsk->remote_addr));
1198
1199                 err = vsock_auto_bind(vsk);
1200                 if (err)
1201                         goto out;
1202
1203                 sk->sk_state = TCP_SYN_SENT;
1204
1205                 err = transport->connect(vsk);
1206                 if (err < 0)
1207                         goto out;
1208
1209                 /* Mark sock as connecting and set the error code to in
1210                  * progress in case this is a non-blocking connect.
1211                  */
1212                 sock->state = SS_CONNECTING;
1213                 err = -EINPROGRESS;
1214         }
1215
1216         /* The receive path will handle all communication until we are able to
1217          * enter the connected state.  Here we wait for the connection to be
1218          * completed or a notification of an error.
1219          */
1220         timeout = vsk->connect_timeout;
1221         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1222
1223         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1224                 if (flags & O_NONBLOCK) {
1225                         /* If we're not going to block, we schedule a timeout
1226                          * function to generate a timeout on the connection
1227                          * attempt, in case the peer doesn't respond in a
1228                          * timely manner. We hold on to the socket until the
1229                          * timeout fires.
1230                          */
1231                         sock_hold(sk);
1232                         schedule_delayed_work(&vsk->connect_work, timeout);
1233
1234                         /* Skip ahead to preserve error code set above. */
1235                         goto out_wait;
1236                 }
1237
1238                 release_sock(sk);
1239                 timeout = schedule_timeout(timeout);
1240                 lock_sock(sk);
1241
1242                 if (signal_pending(current)) {
1243                         err = sock_intr_errno(timeout);
1244                         sk->sk_state = TCP_CLOSE;
1245                         sock->state = SS_UNCONNECTED;
1246                         vsock_transport_cancel_pkt(vsk);
1247                         goto out_wait;
1248                 } else if (timeout == 0) {
1249                         err = -ETIMEDOUT;
1250                         sk->sk_state = TCP_CLOSE;
1251                         sock->state = SS_UNCONNECTED;
1252                         vsock_transport_cancel_pkt(vsk);
1253                         goto out_wait;
1254                 }
1255
1256                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1257         }
1258
1259         if (sk->sk_err) {
1260                 err = -sk->sk_err;
1261                 sk->sk_state = TCP_CLOSE;
1262                 sock->state = SS_UNCONNECTED;
1263         } else {
1264                 err = 0;
1265         }
1266
1267 out_wait:
1268         finish_wait(sk_sleep(sk), &wait);
1269 out:
1270         release_sock(sk);
1271         return err;
1272 }
1273
1274 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1275                         bool kern)
1276 {
1277         struct sock *listener;
1278         int err;
1279         struct sock *connected;
1280         struct vsock_sock *vconnected;
1281         long timeout;
1282         DEFINE_WAIT(wait);
1283
1284         err = 0;
1285         listener = sock->sk;
1286
1287         lock_sock(listener);
1288
1289         if (sock->type != SOCK_STREAM) {
1290                 err = -EOPNOTSUPP;
1291                 goto out;
1292         }
1293
1294         if (listener->sk_state != TCP_LISTEN) {
1295                 err = -EINVAL;
1296                 goto out;
1297         }
1298
1299         /* Wait for children sockets to appear; these are the new sockets
1300          * created upon connection establishment.
1301          */
1302         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1303         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1304
1305         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1306                listener->sk_err == 0) {
1307                 release_sock(listener);
1308                 timeout = schedule_timeout(timeout);
1309                 finish_wait(sk_sleep(listener), &wait);
1310                 lock_sock(listener);
1311
1312                 if (signal_pending(current)) {
1313                         err = sock_intr_errno(timeout);
1314                         goto out;
1315                 } else if (timeout == 0) {
1316                         err = -EAGAIN;
1317                         goto out;
1318                 }
1319
1320                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1321         }
1322         finish_wait(sk_sleep(listener), &wait);
1323
1324         if (listener->sk_err)
1325                 err = -listener->sk_err;
1326
1327         if (connected) {
1328                 listener->sk_ack_backlog--;
1329
1330                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1331                 vconnected = vsock_sk(connected);
1332
1333                 /* If the listener socket has received an error, then we should
1334                  * reject this socket and return.  Note that we simply mark the
1335                  * socket rejected, drop our reference, and let the cleanup
1336                  * function handle the cleanup; the fact that we found it in
1337                  * the listener's accept queue guarantees that the cleanup
1338                  * function hasn't run yet.
1339                  */
1340                 if (err) {
1341                         vconnected->rejected = true;
1342                 } else {
1343                         newsock->state = SS_CONNECTED;
1344                         sock_graft(connected, newsock);
1345                 }
1346
1347                 release_sock(connected);
1348                 sock_put(connected);
1349         }
1350
1351 out:
1352         release_sock(listener);
1353         return err;
1354 }
1355
1356 static int vsock_listen(struct socket *sock, int backlog)
1357 {
1358         int err;
1359         struct sock *sk;
1360         struct vsock_sock *vsk;
1361
1362         sk = sock->sk;
1363
1364         lock_sock(sk);
1365
1366         if (sock->type != SOCK_STREAM) {
1367                 err = -EOPNOTSUPP;
1368                 goto out;
1369         }
1370
1371         if (sock->state != SS_UNCONNECTED) {
1372                 err = -EINVAL;
1373                 goto out;
1374         }
1375
1376         vsk = vsock_sk(sk);
1377
1378         if (!vsock_addr_bound(&vsk->local_addr)) {
1379                 err = -EINVAL;
1380                 goto out;
1381         }
1382
1383         sk->sk_max_ack_backlog = backlog;
1384         sk->sk_state = TCP_LISTEN;
1385
1386         err = 0;
1387
1388 out:
1389         release_sock(sk);
1390         return err;
1391 }
1392
1393 static int vsock_stream_setsockopt(struct socket *sock,
1394                                    int level,
1395                                    int optname,
1396                                    char __user *optval,
1397                                    unsigned int optlen)
1398 {
1399         int err;
1400         struct sock *sk;
1401         struct vsock_sock *vsk;
1402         u64 val;
1403
1404         if (level != AF_VSOCK)
1405                 return -ENOPROTOOPT;
1406
1407 #define COPY_IN(_v)                                       \
1408         do {                                              \
1409                 if (optlen < sizeof(_v)) {                \
1410                         err = -EINVAL;                    \
1411                         goto exit;                        \
1412                 }                                         \
1413                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1414                         err = -EFAULT;                                  \
1415                         goto exit;                                      \
1416                 }                                                       \
1417         } while (0)
1418
1419         err = 0;
1420         sk = sock->sk;
1421         vsk = vsock_sk(sk);
1422
1423         lock_sock(sk);
1424
1425         switch (optname) {
1426         case SO_VM_SOCKETS_BUFFER_SIZE:
1427                 COPY_IN(val);
1428                 transport->set_buffer_size(vsk, val);
1429                 break;
1430
1431         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1432                 COPY_IN(val);
1433                 transport->set_max_buffer_size(vsk, val);
1434                 break;
1435
1436         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1437                 COPY_IN(val);
1438                 transport->set_min_buffer_size(vsk, val);
1439                 break;
1440
1441         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1442                 struct timeval tv;
1443                 COPY_IN(tv);
1444                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1445                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1446                         vsk->connect_timeout = tv.tv_sec * HZ +
1447                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1448                         if (vsk->connect_timeout == 0)
1449                                 vsk->connect_timeout =
1450                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1451
1452                 } else {
1453                         err = -ERANGE;
1454                 }
1455                 break;
1456         }
1457
1458         default:
1459                 err = -ENOPROTOOPT;
1460                 break;
1461         }
1462
1463 #undef COPY_IN
1464
1465 exit:
1466         release_sock(sk);
1467         return err;
1468 }
1469
1470 static int vsock_stream_getsockopt(struct socket *sock,
1471                                    int level, int optname,
1472                                    char __user *optval,
1473                                    int __user *optlen)
1474 {
1475         int err;
1476         int len;
1477         struct sock *sk;
1478         struct vsock_sock *vsk;
1479         u64 val;
1480
1481         if (level != AF_VSOCK)
1482                 return -ENOPROTOOPT;
1483
1484         err = get_user(len, optlen);
1485         if (err != 0)
1486                 return err;
1487
1488 #define COPY_OUT(_v)                            \
1489         do {                                    \
1490                 if (len < sizeof(_v))           \
1491                         return -EINVAL;         \
1492                                                 \
1493                 len = sizeof(_v);               \
1494                 if (copy_to_user(optval, &_v, len) != 0)        \
1495                         return -EFAULT;                         \
1496                                                                 \
1497         } while (0)
1498
1499         err = 0;
1500         sk = sock->sk;
1501         vsk = vsock_sk(sk);
1502
1503         switch (optname) {
1504         case SO_VM_SOCKETS_BUFFER_SIZE:
1505                 val = transport->get_buffer_size(vsk);
1506                 COPY_OUT(val);
1507                 break;
1508
1509         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1510                 val = transport->get_max_buffer_size(vsk);
1511                 COPY_OUT(val);
1512                 break;
1513
1514         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1515                 val = transport->get_min_buffer_size(vsk);
1516                 COPY_OUT(val);
1517                 break;
1518
1519         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1520                 struct timeval tv;
1521                 tv.tv_sec = vsk->connect_timeout / HZ;
1522                 tv.tv_usec =
1523                     (vsk->connect_timeout -
1524                      tv.tv_sec * HZ) * (1000000 / HZ);
1525                 COPY_OUT(tv);
1526                 break;
1527         }
1528         default:
1529                 return -ENOPROTOOPT;
1530         }
1531
1532         err = put_user(len, optlen);
1533         if (err != 0)
1534                 return -EFAULT;
1535
1536 #undef COPY_OUT
1537
1538         return 0;
1539 }
1540
1541 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1542                                 size_t len)
1543 {
1544         struct sock *sk;
1545         struct vsock_sock *vsk;
1546         ssize_t total_written;
1547         long timeout;
1548         int err;
1549         struct vsock_transport_send_notify_data send_data;
1550         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1551
1552         sk = sock->sk;
1553         vsk = vsock_sk(sk);
1554         total_written = 0;
1555         err = 0;
1556
1557         if (msg->msg_flags & MSG_OOB)
1558                 return -EOPNOTSUPP;
1559
1560         lock_sock(sk);
1561
1562         /* Callers should not provide a destination with stream sockets. */
1563         if (msg->msg_namelen) {
1564                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1565                 goto out;
1566         }
1567
1568         /* Send data only if both sides are not shutdown in the direction. */
1569         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1570             vsk->peer_shutdown & RCV_SHUTDOWN) {
1571                 err = -EPIPE;
1572                 goto out;
1573         }
1574
1575         if (sk->sk_state != TCP_ESTABLISHED ||
1576             !vsock_addr_bound(&vsk->local_addr)) {
1577                 err = -ENOTCONN;
1578                 goto out;
1579         }
1580
1581         if (!vsock_addr_bound(&vsk->remote_addr)) {
1582                 err = -EDESTADDRREQ;
1583                 goto out;
1584         }
1585
1586         /* Wait for room in the produce queue to enqueue our user's data. */
1587         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1588
1589         err = transport->notify_send_init(vsk, &send_data);
1590         if (err < 0)
1591                 goto out;
1592
1593         while (total_written < len) {
1594                 ssize_t written;
1595
1596                 add_wait_queue(sk_sleep(sk), &wait);
1597                 while (vsock_stream_has_space(vsk) == 0 &&
1598                        sk->sk_err == 0 &&
1599                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1600                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1601
1602                         /* Don't wait for non-blocking sockets. */
1603                         if (timeout == 0) {
1604                                 err = -EAGAIN;
1605                                 remove_wait_queue(sk_sleep(sk), &wait);
1606                                 goto out_err;
1607                         }
1608
1609                         err = transport->notify_send_pre_block(vsk, &send_data);
1610                         if (err < 0) {
1611                                 remove_wait_queue(sk_sleep(sk), &wait);
1612                                 goto out_err;
1613                         }
1614
1615                         release_sock(sk);
1616                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1617                         lock_sock(sk);
1618                         if (signal_pending(current)) {
1619                                 err = sock_intr_errno(timeout);
1620                                 remove_wait_queue(sk_sleep(sk), &wait);
1621                                 goto out_err;
1622                         } else if (timeout == 0) {
1623                                 err = -EAGAIN;
1624                                 remove_wait_queue(sk_sleep(sk), &wait);
1625                                 goto out_err;
1626                         }
1627                 }
1628                 remove_wait_queue(sk_sleep(sk), &wait);
1629
1630                 /* These checks occur both as part of and after the loop
1631                  * conditional since we need to check before and after
1632                  * sleeping.
1633                  */
1634                 if (sk->sk_err) {
1635                         err = -sk->sk_err;
1636                         goto out_err;
1637                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1638                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1639                         err = -EPIPE;
1640                         goto out_err;
1641                 }
1642
1643                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1644                 if (err < 0)
1645                         goto out_err;
1646
1647                 /* Note that enqueue will only write as many bytes as are free
1648                  * in the produce queue, so we don't need to ensure len is
1649                  * smaller than the queue size.  It is the caller's
1650                  * responsibility to check how many bytes we were able to send.
1651                  */
1652
1653                 written = transport->stream_enqueue(
1654                                 vsk, msg,
1655                                 len - total_written);
1656                 if (written < 0) {
1657                         err = -ENOMEM;
1658                         goto out_err;
1659                 }
1660
1661                 total_written += written;
1662
1663                 err = transport->notify_send_post_enqueue(
1664                                 vsk, written, &send_data);
1665                 if (err < 0)
1666                         goto out_err;
1667
1668         }
1669
1670 out_err:
1671         if (total_written > 0)
1672                 err = total_written;
1673 out:
1674         release_sock(sk);
1675         return err;
1676 }
1677
1678
1679 static int
1680 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1681                      int flags)
1682 {
1683         struct sock *sk;
1684         struct vsock_sock *vsk;
1685         int err;
1686         size_t target;
1687         ssize_t copied;
1688         long timeout;
1689         struct vsock_transport_recv_notify_data recv_data;
1690
1691         DEFINE_WAIT(wait);
1692
1693         sk = sock->sk;
1694         vsk = vsock_sk(sk);
1695         err = 0;
1696
1697         lock_sock(sk);
1698
1699         if (sk->sk_state != TCP_ESTABLISHED) {
1700                 /* Recvmsg is supposed to return 0 if a peer performs an
1701                  * orderly shutdown. Differentiate between that case and when a
1702                  * peer has not connected or a local shutdown occured with the
1703                  * SOCK_DONE flag.
1704                  */
1705                 if (sock_flag(sk, SOCK_DONE))
1706                         err = 0;
1707                 else
1708                         err = -ENOTCONN;
1709
1710                 goto out;
1711         }
1712
1713         if (flags & MSG_OOB) {
1714                 err = -EOPNOTSUPP;
1715                 goto out;
1716         }
1717
1718         /* We don't check peer_shutdown flag here since peer may actually shut
1719          * down, but there can be data in the queue that a local socket can
1720          * receive.
1721          */
1722         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1723                 err = 0;
1724                 goto out;
1725         }
1726
1727         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1728          * is not an error.  We may as well bail out now.
1729          */
1730         if (!len) {
1731                 err = 0;
1732                 goto out;
1733         }
1734
1735         /* We must not copy less than target bytes into the user's buffer
1736          * before returning successfully, so we wait for the consume queue to
1737          * have that much data to consume before dequeueing.  Note that this
1738          * makes it impossible to handle cases where target is greater than the
1739          * queue size.
1740          */
1741         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1742         if (target >= transport->stream_rcvhiwat(vsk)) {
1743                 err = -ENOMEM;
1744                 goto out;
1745         }
1746         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1747         copied = 0;
1748
1749         err = transport->notify_recv_init(vsk, target, &recv_data);
1750         if (err < 0)
1751                 goto out;
1752
1753
1754         while (1) {
1755                 s64 ready;
1756
1757                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1758                 ready = vsock_stream_has_data(vsk);
1759
1760                 if (ready == 0) {
1761                         if (sk->sk_err != 0 ||
1762                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1763                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1764                                 finish_wait(sk_sleep(sk), &wait);
1765                                 break;
1766                         }
1767                         /* Don't wait for non-blocking sockets. */
1768                         if (timeout == 0) {
1769                                 err = -EAGAIN;
1770                                 finish_wait(sk_sleep(sk), &wait);
1771                                 break;
1772                         }
1773
1774                         err = transport->notify_recv_pre_block(
1775                                         vsk, target, &recv_data);
1776                         if (err < 0) {
1777                                 finish_wait(sk_sleep(sk), &wait);
1778                                 break;
1779                         }
1780                         release_sock(sk);
1781                         timeout = schedule_timeout(timeout);
1782                         lock_sock(sk);
1783
1784                         if (signal_pending(current)) {
1785                                 err = sock_intr_errno(timeout);
1786                                 finish_wait(sk_sleep(sk), &wait);
1787                                 break;
1788                         } else if (timeout == 0) {
1789                                 err = -EAGAIN;
1790                                 finish_wait(sk_sleep(sk), &wait);
1791                                 break;
1792                         }
1793                 } else {
1794                         ssize_t read;
1795
1796                         finish_wait(sk_sleep(sk), &wait);
1797
1798                         if (ready < 0) {
1799                                 /* Invalid queue pair content. XXX This should
1800                                 * be changed to a connection reset in a later
1801                                 * change.
1802                                 */
1803
1804                                 err = -ENOMEM;
1805                                 goto out;
1806                         }
1807
1808                         err = transport->notify_recv_pre_dequeue(
1809                                         vsk, target, &recv_data);
1810                         if (err < 0)
1811                                 break;
1812
1813                         read = transport->stream_dequeue(
1814                                         vsk, msg,
1815                                         len - copied, flags);
1816                         if (read < 0) {
1817                                 err = -ENOMEM;
1818                                 break;
1819                         }
1820
1821                         copied += read;
1822
1823                         err = transport->notify_recv_post_dequeue(
1824                                         vsk, target, read,
1825                                         !(flags & MSG_PEEK), &recv_data);
1826                         if (err < 0)
1827                                 goto out;
1828
1829                         if (read >= target || flags & MSG_PEEK)
1830                                 break;
1831
1832                         target -= read;
1833                 }
1834         }
1835
1836         if (sk->sk_err)
1837                 err = -sk->sk_err;
1838         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1839                 err = 0;
1840
1841         if (copied > 0)
1842                 err = copied;
1843
1844 out:
1845         release_sock(sk);
1846         return err;
1847 }
1848
1849 static const struct proto_ops vsock_stream_ops = {
1850         .family = PF_VSOCK,
1851         .owner = THIS_MODULE,
1852         .release = vsock_release,
1853         .bind = vsock_bind,
1854         .connect = vsock_stream_connect,
1855         .socketpair = sock_no_socketpair,
1856         .accept = vsock_accept,
1857         .getname = vsock_getname,
1858         .poll = vsock_poll,
1859         .ioctl = sock_no_ioctl,
1860         .listen = vsock_listen,
1861         .shutdown = vsock_shutdown,
1862         .setsockopt = vsock_stream_setsockopt,
1863         .getsockopt = vsock_stream_getsockopt,
1864         .sendmsg = vsock_stream_sendmsg,
1865         .recvmsg = vsock_stream_recvmsg,
1866         .mmap = sock_no_mmap,
1867         .sendpage = sock_no_sendpage,
1868 };
1869
1870 static int vsock_create(struct net *net, struct socket *sock,
1871                         int protocol, int kern)
1872 {
1873         if (!sock)
1874                 return -EINVAL;
1875
1876         if (protocol && protocol != PF_VSOCK)
1877                 return -EPROTONOSUPPORT;
1878
1879         switch (sock->type) {
1880         case SOCK_DGRAM:
1881                 sock->ops = &vsock_dgram_ops;
1882                 break;
1883         case SOCK_STREAM:
1884                 sock->ops = &vsock_stream_ops;
1885                 break;
1886         default:
1887                 return -ESOCKTNOSUPPORT;
1888         }
1889
1890         sock->state = SS_UNCONNECTED;
1891
1892         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1893 }
1894
1895 static const struct net_proto_family vsock_family_ops = {
1896         .family = AF_VSOCK,
1897         .create = vsock_create,
1898         .owner = THIS_MODULE,
1899 };
1900
1901 static long vsock_dev_do_ioctl(struct file *filp,
1902                                unsigned int cmd, void __user *ptr)
1903 {
1904         u32 __user *p = ptr;
1905         int retval = 0;
1906
1907         switch (cmd) {
1908         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1909                 if (put_user(transport->get_local_cid(), p) != 0)
1910                         retval = -EFAULT;
1911                 break;
1912
1913         default:
1914                 pr_err("Unknown ioctl %d\n", cmd);
1915                 retval = -EINVAL;
1916         }
1917
1918         return retval;
1919 }
1920
1921 static long vsock_dev_ioctl(struct file *filp,
1922                             unsigned int cmd, unsigned long arg)
1923 {
1924         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1925 }
1926
1927 #ifdef CONFIG_COMPAT
1928 static long vsock_dev_compat_ioctl(struct file *filp,
1929                                    unsigned int cmd, unsigned long arg)
1930 {
1931         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1932 }
1933 #endif
1934
1935 static const struct file_operations vsock_device_ops = {
1936         .owner          = THIS_MODULE,
1937         .unlocked_ioctl = vsock_dev_ioctl,
1938 #ifdef CONFIG_COMPAT
1939         .compat_ioctl   = vsock_dev_compat_ioctl,
1940 #endif
1941         .open           = nonseekable_open,
1942 };
1943
1944 static struct miscdevice vsock_device = {
1945         .name           = "vsock",
1946         .fops           = &vsock_device_ops,
1947 };
1948
1949 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1950 {
1951         int err = mutex_lock_interruptible(&vsock_register_mutex);
1952
1953         if (err)
1954                 return err;
1955
1956         if (transport) {
1957                 err = -EBUSY;
1958                 goto err_busy;
1959         }
1960
1961         /* Transport must be the owner of the protocol so that it can't
1962          * unload while there are open sockets.
1963          */
1964         vsock_proto.owner = owner;
1965         transport = t;
1966
1967         vsock_device.minor = MISC_DYNAMIC_MINOR;
1968         err = misc_register(&vsock_device);
1969         if (err) {
1970                 pr_err("Failed to register misc device\n");
1971                 goto err_reset_transport;
1972         }
1973
1974         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1975         if (err) {
1976                 pr_err("Cannot register vsock protocol\n");
1977                 goto err_deregister_misc;
1978         }
1979
1980         err = sock_register(&vsock_family_ops);
1981         if (err) {
1982                 pr_err("could not register af_vsock (%d) address family: %d\n",
1983                        AF_VSOCK, err);
1984                 goto err_unregister_proto;
1985         }
1986
1987         mutex_unlock(&vsock_register_mutex);
1988         return 0;
1989
1990 err_unregister_proto:
1991         proto_unregister(&vsock_proto);
1992 err_deregister_misc:
1993         misc_deregister(&vsock_device);
1994 err_reset_transport:
1995         transport = NULL;
1996 err_busy:
1997         mutex_unlock(&vsock_register_mutex);
1998         return err;
1999 }
2000 EXPORT_SYMBOL_GPL(__vsock_core_init);
2001
2002 void vsock_core_exit(void)
2003 {
2004         mutex_lock(&vsock_register_mutex);
2005
2006         misc_deregister(&vsock_device);
2007         sock_unregister(AF_VSOCK);
2008         proto_unregister(&vsock_proto);
2009
2010         /* We do not want the assignment below re-ordered. */
2011         mb();
2012         transport = NULL;
2013
2014         mutex_unlock(&vsock_register_mutex);
2015 }
2016 EXPORT_SYMBOL_GPL(vsock_core_exit);
2017
2018 const struct vsock_transport *vsock_core_get_transport(void)
2019 {
2020         /* vsock_register_mutex not taken since only the transport uses this
2021          * function and only while registered.
2022          */
2023         return transport;
2024 }
2025 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2026
2027 static void __exit vsock_exit(void)
2028 {
2029         /* Do nothing.  This function makes this module removable. */
2030 }
2031
2032 module_init(vsock_init_tables);
2033 module_exit(vsock_exit);
2034
2035 MODULE_AUTHOR("VMware, Inc.");
2036 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2037 MODULE_VERSION("1.0.2.0-k");
2038 MODULE_LICENSE("GPL v2");