net/vmw_vsock/af_vsock.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 /* Implementation notes:
  17  *
  18  * - There are two kinds of sockets: those created by user action (such as
  19  * calling socket(2)) and those created by incoming connection request packets.
  20  *
  21  * - There are two "global" tables, one for bound sockets (sockets that have
  22  * specified an address that they are responsible for) and one for connected
  23  * sockets (sockets that have established a connection with another socket).
  24  * These tables are "global" in that all sockets on the system are placed
  25  * within them. - Note, though, that the bound table contains an extra entry
  26  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  27  * that list. The bound table is used solely for lookup of sockets when packets
  28  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  29  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  30  * sockets out of the bound hash buckets will reduce the chance of collisions
  31  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  32  * socket type in the hash table lookups.
  33  *
  34  * - Sockets created by user action will either be "client" sockets that
  35  * initiate a connection or "server" sockets that listen for connections; we do
  36  * not support simultaneous connects (two "client" sockets connecting).
  37  *
  38  * - "Server" sockets are referred to as listener sockets throughout this
  39  * implementation because they are in the VSOCK_SS_LISTEN state.  When a
  40  * connection request is received (the second kind of socket mentioned above),
  41  * we create a new socket and refer to it as a pending socket.  These pending
  42  * sockets are placed on the pending connection list of the listener socket.
  43  * When future packets are received for the address the listener socket is
  44  * bound to, we check if the source of the packet is from one that has an
  45  * existing pending connection.  If it does, we process the packet for the
  46  * pending socket.  When that socket reaches the connected state, it is removed
  47  * from the listener socket's pending list and enqueued in the listener
  48  * socket's accept queue.  Callers of accept(2) will accept connected sockets
  49  * from the listener socket's accept queue.  If the socket cannot be accepted
  50  * for some reason then it is marked rejected.  Once the connection is
  51  * accepted, it is owned by the user process and the responsibility for cleanup
  52  * falls with that user process.
  53  *
  54  * - It is possible that these pending sockets will never reach the connected
  55  * state; in fact, we may never receive another packet after the connection
  56  * request.  Because of this, we must schedule a cleanup function to run in the
  57  * future, after some amount of time passes where a connection should have been
  58  * established.  This function ensures that the socket is off all lists so it
  59  * cannot be retrieved, then drops all references to the socket so it is cleaned
  60  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  61  * function will also cleanup rejected sockets, those that reach the connected
  62  * state but leave it before they have been accepted.
  63  *
  64  * - Lock ordering for pending or accept queue sockets is:
  65  *
  66  *     lock_sock(listener);
  67  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  68  *
  69  * Using explicit nested locking keeps lockdep happy since normally only one
  70  * lock of a given class may be taken at a time.
  71  *
  72  * - Sockets created by user action will be cleaned up when the user process
  73  * calls close(2), causing our release implementation to be called. Our release
  74  * implementation will perform some cleanup then drop the last reference so our
  75  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  76  * perform additional cleanup that's common for both types of sockets.
  77  *
  78  * - A socket's reference count is what ensures that the structure won't be
  79  * freed.  Each entry in a list (such as the "global" bound and connected tables
  80  * and the listener socket's pending list and connected queue) ensures a
  81  * reference.  When we defer work until process context and pass a socket as our
  82  * argument, we must ensure the reference count is increased to ensure the
  83  * socket isn't freed before the function is run; the deferred function will
  84  * then drop the reference.
  85  */
  86
  87 #include <linux/types.h>
  88 #include <linux/bitops.h>
  89 #include <linux/cred.h>
  90 #include <linux/init.h>
  91 #include <linux/io.h>
  92 #include <linux/kernel.h>
  93 #include <linux/sched/signal.h>
  94 #include <linux/kmod.h>
  95 #include <linux/list.h>
  96 #include <linux/miscdevice.h>
  97 #include <linux/module.h>
  98 #include <linux/mutex.h>
  99 #include <linux/net.h>
 100 #include <linux/poll.h>
 101 #include <linux/skbuff.h>
 102 #include <linux/smp.h>
 103 #include <linux/socket.h>
 104 #include <linux/stddef.h>
 105 #include <linux/unistd.h>
 106 #include <linux/wait.h>
 107 #include <linux/workqueue.h>
 108 #include <net/sock.h>
 109 #include <net/af_vsock.h>
 110
 111 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 112 static void vsock_sk_destruct(struct sock *sk);
 113 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 114
 115 /* Protocol family. */
 116 static struct proto vsock_proto = {
 117         .name = "AF_VSOCK",
 118         .owner = THIS_MODULE,
 119         .obj_size = sizeof(struct vsock_sock),
 120 };
 121
 122 /* The default peer timeout indicates how long we will wait for a peer response
 123  * to a control message.
 124  */
 125 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 126
 127 static const struct vsock_transport *transport;
 128 static DEFINE_MUTEX(vsock_register_mutex);
 129
 130 /**** EXPORTS ****/
 131
 132 /* Get the ID of the local context.  This is transport dependent. */
 133
 134 int vm_sockets_get_local_cid(void)
 135 {
 136         return transport->get_local_cid();
 137 }
 138 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 139
 140 /**** UTILS ****/
 141
 142 /* Each bound VSocket is stored in the bind hash table and each connected
 143  * VSocket is stored in the connected hash table.
 144  *
 145  * Unbound sockets are all put on the same list attached to the end of the hash
 146  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 147  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 148  * represents the list that addr hashes to).
 149  *
 150  * Specifically, we initialize the vsock_bind_table array to a size of
 151  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 152  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 153  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 154  * mods with VSOCK_HASH_SIZE to ensure this.
 155  */
 156 #define VSOCK_HASH_SIZE         251
 157 #define MAX_PORT_RETRIES        24
 158
 159 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 160 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 161 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 162
 163 /* XXX This can probably be implemented in a better way. */
 164 #define VSOCK_CONN_HASH(src, dst)                               \
 165         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 166 #define vsock_connected_sockets(src, dst)               \
 167         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 168 #define vsock_connected_sockets_vsk(vsk)                                \
 169         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 170
 171 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 172 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 173 static DEFINE_SPINLOCK(vsock_table_lock);
 174
 175 /* Autobind this socket to the local address if necessary. */
 176 static int vsock_auto_bind(struct vsock_sock *vsk)
 177 {
 178         struct sock *sk = sk_vsock(vsk);
 179         struct sockaddr_vm local_addr;
 180
 181         if (vsock_addr_bound(&vsk->local_addr))
 182                 return 0;
 183         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 184         return __vsock_bind(sk, &local_addr);
 185 }
 186
 187 static void vsock_init_tables(void)
 188 {
 189         int i;
 190
 191         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 192                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 193
 194         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 195                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 196 }
 197
 198 static void __vsock_insert_bound(struct list_head *list,
 199                                  struct vsock_sock *vsk)
 200 {
 201         sock_hold(&vsk->sk);
 202         list_add(&vsk->bound_table, list);
 203 }
 204
 205 static void __vsock_insert_connected(struct list_head *list,
 206                                      struct vsock_sock *vsk)
 207 {
 208         sock_hold(&vsk->sk);
 209         list_add(&vsk->connected_table, list);
 210 }
 211
 212 static void __vsock_remove_bound(struct vsock_sock *vsk)
 213 {
 214         list_del_init(&vsk->bound_table);
 215         sock_put(&vsk->sk);
 216 }
 217
 218 static void __vsock_remove_connected(struct vsock_sock *vsk)
 219 {
 220         list_del_init(&vsk->connected_table);
 221         sock_put(&vsk->sk);
 222 }
 223
 224 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 225 {
 226         struct vsock_sock *vsk;
 227
 228         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 229                 if (addr->svm_port == vsk->local_addr.svm_port)
 230                         return sk_vsock(vsk);
 231
 232         return NULL;
 233 }
 234
 235 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 236                                                   struct sockaddr_vm *dst)
 237 {
 238         struct vsock_sock *vsk;
 239
 240         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 241                             connected_table) {
 242                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 243                     dst->svm_port == vsk->local_addr.svm_port) {
 244                         return sk_vsock(vsk);
 245                 }
 246         }
 247
 248         return NULL;
 249 }
 250
 251 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 252 {
 253         return !list_empty(&vsk->bound_table);
 254 }
 255
 256 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 257 {
 258         return !list_empty(&vsk->connected_table);
 259 }
 260
 261 static void vsock_insert_unbound(struct vsock_sock *vsk)
 262 {
 263         spin_lock_bh(&vsock_table_lock);
 264         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 265         spin_unlock_bh(&vsock_table_lock);
 266 }
 267
 268 void vsock_insert_connected(struct vsock_sock *vsk)
 269 {
 270         struct list_head *list = vsock_connected_sockets(
 271                 &vsk->remote_addr, &vsk->local_addr);
 272
 273         spin_lock_bh(&vsock_table_lock);
 274         __vsock_insert_connected(list, vsk);
 275         spin_unlock_bh(&vsock_table_lock);
 276 }
 277 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 278
 279 void vsock_remove_bound(struct vsock_sock *vsk)
 280 {
 281         spin_lock_bh(&vsock_table_lock);
 282         __vsock_remove_bound(vsk);
 283         spin_unlock_bh(&vsock_table_lock);
 284 }
 285 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 286
 287 void vsock_remove_connected(struct vsock_sock *vsk)
 288 {
 289         spin_lock_bh(&vsock_table_lock);
 290         __vsock_remove_connected(vsk);
 291         spin_unlock_bh(&vsock_table_lock);
 292 }
 293 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 294
 295 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 296 {
 297         struct sock *sk;
 298
 299         spin_lock_bh(&vsock_table_lock);
 300         sk = __vsock_find_bound_socket(addr);
 301         if (sk)
 302                 sock_hold(sk);
 303
 304         spin_unlock_bh(&vsock_table_lock);
 305
 306         return sk;
 307 }
 308 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 309
 310 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 311                                          struct sockaddr_vm *dst)
 312 {
 313         struct sock *sk;
 314
 315         spin_lock_bh(&vsock_table_lock);
 316         sk = __vsock_find_connected_socket(src, dst);
 317         if (sk)
 318                 sock_hold(sk);
 319
 320         spin_unlock_bh(&vsock_table_lock);
 321
 322         return sk;
 323 }
 324 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 325
 326 static bool vsock_in_bound_table(struct vsock_sock *vsk)
 327 {
 328         bool ret;
 329
 330         spin_lock_bh(&vsock_table_lock);
 331         ret = __vsock_in_bound_table(vsk);
 332         spin_unlock_bh(&vsock_table_lock);
 333
 334         return ret;
 335 }
 336
 337 static bool vsock_in_connected_table(struct vsock_sock *vsk)
 338 {
 339         bool ret;
 340
 341         spin_lock_bh(&vsock_table_lock);
 342         ret = __vsock_in_connected_table(vsk);
 343         spin_unlock_bh(&vsock_table_lock);
 344
 345         return ret;
 346 }
 347
 348 void vsock_remove_sock(struct vsock_sock *vsk)
 349 {
 350         if (vsock_in_bound_table(vsk))
 351                 vsock_remove_bound(vsk);
 352
 353         if (vsock_in_connected_table(vsk))
 354                 vsock_remove_connected(vsk);
 355 }
 356 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 357
 358 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 359 {
 360         int i;
 361
 362         spin_lock_bh(&vsock_table_lock);
 363
 364         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 365                 struct vsock_sock *vsk;
 366                 list_for_each_entry(vsk, &vsock_connected_table[i],
 367                                     connected_table)
 368                         fn(sk_vsock(vsk));
 369         }
 370
 371         spin_unlock_bh(&vsock_table_lock);
 372 }
 373 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 374
 375 void vsock_add_pending(struct sock *listener, struct sock *pending)
 376 {
 377         struct vsock_sock *vlistener;
 378         struct vsock_sock *vpending;
 379
 380         vlistener = vsock_sk(listener);
 381         vpending = vsock_sk(pending);
 382
 383         sock_hold(pending);
 384         sock_hold(listener);
 385         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 386 }
 387 EXPORT_SYMBOL_GPL(vsock_add_pending);
 388
 389 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 390 {
 391         struct vsock_sock *vpending = vsock_sk(pending);
 392
 393         list_del_init(&vpending->pending_links);
 394         sock_put(listener);
 395         sock_put(pending);
 396 }
 397 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 398
 399 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 400 {
 401         struct vsock_sock *vlistener;
 402         struct vsock_sock *vconnected;
 403
 404         vlistener = vsock_sk(listener);
 405         vconnected = vsock_sk(connected);
 406
 407         sock_hold(connected);
 408         sock_hold(listener);
 409         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 410 }
 411 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 412
 413 static struct sock *vsock_dequeue_accept(struct sock *listener)
 414 {
 415         struct vsock_sock *vlistener;
 416         struct vsock_sock *vconnected;
 417
 418         vlistener = vsock_sk(listener);
 419
 420         if (list_empty(&vlistener->accept_queue))
 421                 return NULL;
 422
 423         vconnected = list_entry(vlistener->accept_queue.next,
 424                                 struct vsock_sock, accept_queue);
 425
 426         list_del_init(&vconnected->accept_queue);
 427         sock_put(listener);
 428         /* The caller will need a reference on the connected socket so we let
 429          * it call sock_put().
 430          */
 431
 432         return sk_vsock(vconnected);
 433 }
 434
 435 static bool vsock_is_accept_queue_empty(struct sock *sk)
 436 {
 437         struct vsock_sock *vsk = vsock_sk(sk);
 438         return list_empty(&vsk->accept_queue);
 439 }
 440
 441 static bool vsock_is_pending(struct sock *sk)
 442 {
 443         struct vsock_sock *vsk = vsock_sk(sk);
 444         return !list_empty(&vsk->pending_links);
 445 }
 446
 447 static int vsock_send_shutdown(struct sock *sk, int mode)
 448 {
 449         return transport->shutdown(vsock_sk(sk), mode);
 450 }
 451
 452 void vsock_pending_work(struct work_struct *work)
 453 {
 454         struct sock *sk;
 455         struct sock *listener;
 456         struct vsock_sock *vsk;
 457         bool cleanup;
 458
 459         vsk = container_of(work, struct vsock_sock, dwork.work);
 460         sk = sk_vsock(vsk);
 461         listener = vsk->listener;
 462         cleanup = true;
 463
 464         lock_sock(listener);
 465         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 466
 467         if (vsock_is_pending(sk)) {
 468                 vsock_remove_pending(listener, sk);
 469
 470                 listener->sk_ack_backlog--;
 471         } else if (!vsk->rejected) {
 472                 /* We are not on the pending list and accept() did not reject
 473                  * us, so we must have been accepted by our user process.  We
 474                  * just need to drop our references to the sockets and be on
 475                  * our way.
 476                  */
 477                 cleanup = false;
 478                 goto out;
 479         }
 480
 481         /* We need to remove ourself from the global connected sockets list so
 482          * incoming packets can't find this socket, and to reduce the reference
 483          * count.
 484          */
 485         if (vsock_in_connected_table(vsk))
 486                 vsock_remove_connected(vsk);
 487
 488         sk->sk_state = SS_FREE;
 489
 490 out:
 491         release_sock(sk);
 492         release_sock(listener);
 493         if (cleanup)
 494                 sock_put(sk);
 495
 496         sock_put(sk);
 497         sock_put(listener);
 498 }
 499 EXPORT_SYMBOL_GPL(vsock_pending_work);
 500
 501 /**** SOCKET OPERATIONS ****/
 502
 503 static int __vsock_bind_stream(struct vsock_sock *vsk,
 504                                struct sockaddr_vm *addr)
 505 {
 506         static u32 port = LAST_RESERVED_PORT + 1;
 507         struct sockaddr_vm new_addr;
 508
 509         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 510
 511         if (addr->svm_port == VMADDR_PORT_ANY) {
 512                 bool found = false;
 513                 unsigned int i;
 514
 515                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 516                         if (port <= LAST_RESERVED_PORT)
 517                                 port = LAST_RESERVED_PORT + 1;
 518
 519                         new_addr.svm_port = port++;
 520
 521                         if (!__vsock_find_bound_socket(&new_addr)) {
 522                                 found = true;
 523                                 break;
 524                         }
 525                 }
 526
 527                 if (!found)
 528                         return -EADDRNOTAVAIL;
 529         } else {
 530                 /* If port is in reserved range, ensure caller
 531                  * has necessary privileges.
 532                  */
 533                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 534                     !capable(CAP_NET_BIND_SERVICE)) {
 535                         return -EACCES;
 536                 }
 537
 538                 if (__vsock_find_bound_socket(&new_addr))
 539                         return -EADDRINUSE;
 540         }
 541
 542         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 543
 544         /* Remove stream sockets from the unbound list and add them to the hash
 545          * table for easy lookup by its address.  The unbound list is simply an
 546          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 547          */
 548         __vsock_remove_bound(vsk);
 549         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 550
 551         return 0;
 552 }
 553
 554 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 555                               struct sockaddr_vm *addr)
 556 {
 557         return transport->dgram_bind(vsk, addr);
 558 }
 559
 560 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 561 {
 562         struct vsock_sock *vsk = vsock_sk(sk);
 563         u32 cid;
 564         int retval;
 565
 566         /* First ensure this socket isn't already bound. */
 567         if (vsock_addr_bound(&vsk->local_addr))
 568                 return -EINVAL;
 569
 570         /* Now bind to the provided address or select appropriate values if
 571          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 572          * like AF_INET prevents binding to a non-local IP address (in most
 573          * cases), we only allow binding to the local CID.
 574          */
 575         cid = transport->get_local_cid();
 576         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 577                 return -EADDRNOTAVAIL;
 578
 579         switch (sk->sk_socket->type) {
 580         case SOCK_STREAM:
 581                 spin_lock_bh(&vsock_table_lock);
 582                 retval = __vsock_bind_stream(vsk, addr);
 583                 spin_unlock_bh(&vsock_table_lock);
 584                 break;
 585
 586         case SOCK_DGRAM:
 587                 retval = __vsock_bind_dgram(vsk, addr);
 588                 break;
 589
 590         default:
 591                 retval = -EINVAL;
 592                 break;
 593         }
 594
 595         return retval;
 596 }
 597
 598 struct sock *__vsock_create(struct net *net,
 599                             struct socket *sock,
 600                             struct sock *parent,
 601                             gfp_t priority,
 602                             unsigned short type,
 603                             int kern)
 604 {
 605         struct sock *sk;
 606         struct vsock_sock *psk;
 607         struct vsock_sock *vsk;
 608
 609         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 610         if (!sk)
 611                 return NULL;
 612
 613         sock_init_data(sock, sk);
 614
 615         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 616          * non-NULL. We make sure that our sockets always have a type by
 617          * setting it here if needed.
 618          */
 619         if (!sock)
 620                 sk->sk_type = type;
 621
 622         vsk = vsock_sk(sk);
 623         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 624         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 625
 626         sk->sk_destruct = vsock_sk_destruct;
 627         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 628         sk->sk_state = 0;
 629         sock_reset_flag(sk, SOCK_DONE);
 630
 631         INIT_LIST_HEAD(&vsk->bound_table);
 632         INIT_LIST_HEAD(&vsk->connected_table);
 633         vsk->listener = NULL;
 634         INIT_LIST_HEAD(&vsk->pending_links);
 635         INIT_LIST_HEAD(&vsk->accept_queue);
 636         vsk->rejected = false;
 637         vsk->sent_request = false;
 638         vsk->ignore_connecting_rst = false;
 639         vsk->peer_shutdown = 0;
 640
 641         psk = parent ? vsock_sk(parent) : NULL;
 642         if (parent) {
 643                 vsk->trusted = psk->trusted;
 644                 vsk->owner = get_cred(psk->owner);
 645                 vsk->connect_timeout = psk->connect_timeout;
 646         } else {
 647                 vsk->trusted = capable(CAP_NET_ADMIN);
 648                 vsk->owner = get_current_cred();
 649                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 650         }
 651
 652         if (transport->init(vsk, psk) < 0) {
 653                 sk_free(sk);
 654                 return NULL;
 655         }
 656
 657         if (sock)
 658                 vsock_insert_unbound(vsk);
 659
 660         return sk;
 661 }
 662 EXPORT_SYMBOL_GPL(__vsock_create);
 663
 664 static void __vsock_release(struct sock *sk)
 665 {
 666         if (sk) {
 667                 struct sk_buff *skb;
 668                 struct sock *pending;
 669                 struct vsock_sock *vsk;
 670
 671                 vsk = vsock_sk(sk);
 672                 pending = NULL; /* Compiler warning. */
 673
 674                 transport->release(vsk);
 675
 676                 lock_sock(sk);
 677                 sock_orphan(sk);
 678                 sk->sk_shutdown = SHUTDOWN_MASK;
 679
 680                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 681                         kfree_skb(skb);
 682
 683                 /* Clean up any sockets that never were accepted. */
 684                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 685                         __vsock_release(pending);
 686                         sock_put(pending);
 687                 }
 688
 689                 release_sock(sk);
 690                 sock_put(sk);
 691         }
 692 }
 693
 694 static void vsock_sk_destruct(struct sock *sk)
 695 {
 696         struct vsock_sock *vsk = vsock_sk(sk);
 697
 698         transport->destruct(vsk);
 699
 700         /* When clearing these addresses, there's no need to set the family and
 701          * possibly register the address family with the kernel.
 702          */
 703         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 704         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 705
 706         put_cred(vsk->owner);
 707 }
 708
 709 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 710 {
 711         int err;
 712
 713         err = sock_queue_rcv_skb(sk, skb);
 714         if (err)
 715                 kfree_skb(skb);
 716
 717         return err;
 718 }
 719
 720 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 721 {
 722         return transport->stream_has_data(vsk);
 723 }
 724 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 725
 726 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 727 {
 728         return transport->stream_has_space(vsk);
 729 }
 730 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 731
 732 static int vsock_release(struct socket *sock)
 733 {
 734         __vsock_release(sock->sk);
 735         sock->sk = NULL;
 736         sock->state = SS_FREE;
 737
 738         return 0;
 739 }
 740
 741 static int
 742 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 743 {
 744         int err;
 745         struct sock *sk;
 746         struct sockaddr_vm *vm_addr;
 747
 748         sk = sock->sk;
 749
 750         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 751                 return -EINVAL;
 752
 753         lock_sock(sk);
 754         err = __vsock_bind(sk, vm_addr);
 755         release_sock(sk);
 756
 757         return err;
 758 }
 759
 760 static int vsock_getname(struct socket *sock,
 761                          struct sockaddr *addr, int *addr_len, int peer)
 762 {
 763         int err;
 764         struct sock *sk;
 765         struct vsock_sock *vsk;
 766         struct sockaddr_vm *vm_addr;
 767
 768         sk = sock->sk;
 769         vsk = vsock_sk(sk);
 770         err = 0;
 771
 772         lock_sock(sk);
 773
 774         if (peer) {
 775                 if (sock->state != SS_CONNECTED) {
 776                         err = -ENOTCONN;
 777                         goto out;
 778                 }
 779                 vm_addr = &vsk->remote_addr;
 780         } else {
 781                 vm_addr = &vsk->local_addr;
 782         }
 783
 784         if (!vm_addr) {
 785                 err = -EINVAL;
 786                 goto out;
 787         }
 788
 789         /* sys_getsockname() and sys_getpeername() pass us a
 790          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 791          * that macro is defined in socket.c instead of .h, so we hardcode its
 792          * value here.
 793          */
 794         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 795         memcpy(addr, vm_addr, sizeof(*vm_addr));
 796         *addr_len = sizeof(*vm_addr);
 797
 798 out:
 799         release_sock(sk);
 800         return err;
 801 }
 802
 803 static int vsock_shutdown(struct socket *sock, int mode)
 804 {
 805         int err;
 806         struct sock *sk;
 807
 808         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 809          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 810          * here like the other address families do.  Note also that the
 811          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 812          * which is what we want.
 813          */
 814         mode++;
 815
 816         if ((mode & ~SHUTDOWN_MASK) || !mode)
 817                 return -EINVAL;
 818
 819         /* If this is a STREAM socket and it is not connected then bail out
 820          * immediately.  If it is a DGRAM socket then we must first kick the
 821          * socket so that it wakes up from any sleeping calls, for example
 822          * recv(), and then afterwards return the error.
 823          */
 824
 825         sk = sock->sk;
 826         if (sock->state == SS_UNCONNECTED) {
 827                 err = -ENOTCONN;
 828                 if (sk->sk_type == SOCK_STREAM)
 829                         return err;
 830         } else {
 831                 sock->state = SS_DISCONNECTING;
 832                 err = 0;
 833         }
 834
 835         /* Receive and send shutdowns are treated alike. */
 836         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 837         if (mode) {
 838                 lock_sock(sk);
 839                 sk->sk_shutdown |= mode;
 840                 sk->sk_state_change(sk);
 841                 release_sock(sk);
 842
 843                 if (sk->sk_type == SOCK_STREAM) {
 844                         sock_reset_flag(sk, SOCK_DONE);
 845                         vsock_send_shutdown(sk, mode);
 846                 }
 847         }
 848
 849         return err;
 850 }
 851
 852 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 853                                poll_table *wait)
 854 {
 855         struct sock *sk;
 856         unsigned int mask;
 857         struct vsock_sock *vsk;
 858
 859         sk = sock->sk;
 860         vsk = vsock_sk(sk);
 861
 862         poll_wait(file, sk_sleep(sk), wait);
 863         mask = 0;
 864
 865         if (sk->sk_err)
 866                 /* Signify that there has been an error on this socket. */
 867                 mask |= POLLERR;
 868
 869         /* INET sockets treat local write shutdown and peer write shutdown as a
 870          * case of POLLHUP set.
 871          */
 872         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 873             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 874              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 875                 mask |= POLLHUP;
 876         }
 877
 878         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 879             vsk->peer_shutdown & SEND_SHUTDOWN) {
 880                 mask |= POLLRDHUP;
 881         }
 882
 883         if (sock->type == SOCK_DGRAM) {
 884                 /* For datagram sockets we can read if there is something in
 885                  * the queue and write as long as the socket isn't shutdown for
 886                  * sending.
 887                  */
 888                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 889                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 890                         mask |= POLLIN | POLLRDNORM;
 891                 }
 892
 893                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 894                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 895
 896         } else if (sock->type == SOCK_STREAM) {
 897                 lock_sock(sk);
 898
 899                 /* Listening sockets that have connections in their accept
 900                  * queue can be read.
 901                  */
 902                 if (sk->sk_state == VSOCK_SS_LISTEN
 903                     && !vsock_is_accept_queue_empty(sk))
 904                         mask |= POLLIN | POLLRDNORM;
 905
 906                 /* If there is something in the queue then we can read. */
 907                 if (transport->stream_is_active(vsk) &&
 908                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 909                         bool data_ready_now = false;
 910                         int ret = transport->notify_poll_in(
 911                                         vsk, 1, &data_ready_now);
 912                         if (ret < 0) {
 913                                 mask |= POLLERR;
 914                         } else {
 915                                 if (data_ready_now)
 916                                         mask |= POLLIN | POLLRDNORM;
 917
 918                         }
 919                 }
 920
 921                 /* Sockets whose connections have been closed, reset, or
 922                  * terminated should also be considered read, and we check the
 923                  * shutdown flag for that.
 924                  */
 925                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 926                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 927                         mask |= POLLIN | POLLRDNORM;
 928                 }
 929
 930                 /* Connected sockets that can produce data can be written. */
 931                 if (sk->sk_state == SS_CONNECTED) {
 932                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 933                                 bool space_avail_now = false;
 934                                 int ret = transport->notify_poll_out(
 935                                                 vsk, 1, &space_avail_now);
 936                                 if (ret < 0) {
 937                                         mask |= POLLERR;
 938                                 } else {
 939                                         if (space_avail_now)
 940                                                 /* Remove POLLWRBAND since INET
 941                                                  * sockets are not setting it.
 942                                                  */
 943                                                 mask |= POLLOUT | POLLWRNORM;
 944
 945                                 }
 946                         }
 947                 }
 948
 949                 /* Simulate INET socket poll behaviors, which sets
 950                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 951                  * but local send is not shutdown.
 952                  */
 953                 if (sk->sk_state == SS_UNCONNECTED) {
 954                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 955                                 mask |= POLLOUT | POLLWRNORM;
 956
 957                 }
 958
 959                 release_sock(sk);
 960         }
 961
 962         return mask;
 963 }
 964
 965 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 966                                size_t len)
 967 {
 968         int err;
 969         struct sock *sk;
 970         struct vsock_sock *vsk;
 971         struct sockaddr_vm *remote_addr;
 972
 973         if (msg->msg_flags & MSG_OOB)
 974                 return -EOPNOTSUPP;
 975
 976         /* For now, MSG_DONTWAIT is always assumed... */
 977         err = 0;
 978         sk = sock->sk;
 979         vsk = vsock_sk(sk);
 980
 981         lock_sock(sk);
 982
 983         err = vsock_auto_bind(vsk);
 984         if (err)
 985                 goto out;
 986
 987
 988         /* If the provided message contains an address, use that.  Otherwise
 989          * fall back on the socket's remote handle (if it has been connected).
 990          */
 991         if (msg->msg_name &&
 992             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 993                             &remote_addr) == 0) {
 994                 /* Ensure this address is of the right type and is a valid
 995                  * destination.
 996                  */
 997
 998                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 999                         remote_addr->svm_cid = transport->get_local_cid();
1000
1001                 if (!vsock_addr_bound(remote_addr)) {
1002                         err = -EINVAL;
1003                         goto out;
1004                 }
1005         } else if (sock->state == SS_CONNECTED) {
1006                 remote_addr = &vsk->remote_addr;
1007
1008                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1009                         remote_addr->svm_cid = transport->get_local_cid();
1010
1011                 /* XXX Should connect() or this function ensure remote_addr is
1012                  * bound?
1013                  */
1014                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1015                         err = -EINVAL;
1016                         goto out;
1017                 }
1018         } else {
1019                 err = -EINVAL;
1020                 goto out;
1021         }
1022
1023         if (!transport->dgram_allow(remote_addr->svm_cid,
1024                                     remote_addr->svm_port)) {
1025                 err = -EINVAL;
1026                 goto out;
1027         }
1028
1029         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1030
1031 out:
1032         release_sock(sk);
1033         return err;
1034 }
1035
1036 static int vsock_dgram_connect(struct socket *sock,
1037                                struct sockaddr *addr, int addr_len, int flags)
1038 {
1039         int err;
1040         struct sock *sk;
1041         struct vsock_sock *vsk;
1042         struct sockaddr_vm *remote_addr;
1043
1044         sk = sock->sk;
1045         vsk = vsock_sk(sk);
1046
1047         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1048         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1049                 lock_sock(sk);
1050                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1051                                 VMADDR_PORT_ANY);
1052                 sock->state = SS_UNCONNECTED;
1053                 release_sock(sk);
1054                 return 0;
1055         } else if (err != 0)
1056                 return -EINVAL;
1057
1058         lock_sock(sk);
1059
1060         err = vsock_auto_bind(vsk);
1061         if (err)
1062                 goto out;
1063
1064         if (!transport->dgram_allow(remote_addr->svm_cid,
1065                                     remote_addr->svm_port)) {
1066                 err = -EINVAL;
1067                 goto out;
1068         }
1069
1070         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1071         sock->state = SS_CONNECTED;
1072
1073 out:
1074         release_sock(sk);
1075         return err;
1076 }
1077
1078 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1079                                size_t len, int flags)
1080 {
1081         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1082 }
1083
1084 static const struct proto_ops vsock_dgram_ops = {
1085         .family = PF_VSOCK,
1086         .owner = THIS_MODULE,
1087         .release = vsock_release,
1088         .bind = vsock_bind,
1089         .connect = vsock_dgram_connect,
1090         .socketpair = sock_no_socketpair,
1091         .accept = sock_no_accept,
1092         .getname = vsock_getname,
1093         .poll = vsock_poll,
1094         .ioctl = sock_no_ioctl,
1095         .listen = sock_no_listen,
1096         .shutdown = vsock_shutdown,
1097         .setsockopt = sock_no_setsockopt,
1098         .getsockopt = sock_no_getsockopt,
1099         .sendmsg = vsock_dgram_sendmsg,
1100         .recvmsg = vsock_dgram_recvmsg,
1101         .mmap = sock_no_mmap,
1102         .sendpage = sock_no_sendpage,
1103 };
1104
1105 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1106 {
1107         if (!transport->cancel_pkt)
1108                 return -EOPNOTSUPP;
1109
1110         return transport->cancel_pkt(vsk);
1111 }
1112
1113 static void vsock_connect_timeout(struct work_struct *work)
1114 {
1115         struct sock *sk;
1116         struct vsock_sock *vsk;
1117         int cancel = 0;
1118
1119         vsk = container_of(work, struct vsock_sock, dwork.work);
1120         sk = sk_vsock(vsk);
1121
1122         lock_sock(sk);
1123         if (sk->sk_state == SS_CONNECTING &&
1124             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1125                 sk->sk_state = SS_UNCONNECTED;
1126                 sk->sk_err = ETIMEDOUT;
1127                 sk->sk_error_report(sk);
1128                 cancel = 1;
1129         }
1130         release_sock(sk);
1131         if (cancel)
1132                 vsock_transport_cancel_pkt(vsk);
1133
1134         sock_put(sk);
1135 }
1136
1137 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1138                                 int addr_len, int flags)
1139 {
1140         int err;
1141         struct sock *sk;
1142         struct vsock_sock *vsk;
1143         struct sockaddr_vm *remote_addr;
1144         long timeout;
1145         DEFINE_WAIT(wait);
1146
1147         err = 0;
1148         sk = sock->sk;
1149         vsk = vsock_sk(sk);
1150
1151         lock_sock(sk);
1152
1153         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1154         switch (sock->state) {
1155         case SS_CONNECTED:
1156                 err = -EISCONN;
1157                 goto out;
1158         case SS_DISCONNECTING:
1159                 err = -EINVAL;
1160                 goto out;
1161         case SS_CONNECTING:
1162                 /* This continues on so we can move sock into the SS_CONNECTED
1163                  * state once the connection has completed (at which point err
1164                  * will be set to zero also).  Otherwise, we will either wait
1165                  * for the connection or return -EALREADY should this be a
1166                  * non-blocking call.
1167                  */
1168                 err = -EALREADY;
1169                 break;
1170         default:
1171                 if ((sk->sk_state == VSOCK_SS_LISTEN) ||
1172                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1173                         err = -EINVAL;
1174                         goto out;
1175                 }
1176
1177                 /* The hypervisor and well-known contexts do not have socket
1178                  * endpoints.
1179                  */
1180                 if (!transport->stream_allow(remote_addr->svm_cid,
1181                                              remote_addr->svm_port)) {
1182                         err = -ENETUNREACH;
1183                         goto out;
1184                 }
1185
1186                 /* Set the remote address that we are connecting to. */
1187                 memcpy(&vsk->remote_addr, remote_addr,
1188                        sizeof(vsk->remote_addr));
1189
1190                 err = vsock_auto_bind(vsk);
1191                 if (err)
1192                         goto out;
1193
1194                 sk->sk_state = SS_CONNECTING;
1195
1196                 err = transport->connect(vsk);
1197                 if (err < 0)
1198                         goto out;
1199
1200                 /* Mark sock as connecting and set the error code to in
1201                  * progress in case this is a non-blocking connect.
1202                  */
1203                 sock->state = SS_CONNECTING;
1204                 err = -EINPROGRESS;
1205         }
1206
1207         /* The receive path will handle all communication until we are able to
1208          * enter the connected state.  Here we wait for the connection to be
1209          * completed or a notification of an error.
1210          */
1211         timeout = vsk->connect_timeout;
1212         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1213
1214         while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1215                 if (flags & O_NONBLOCK) {
1216                         /* If we're not going to block, we schedule a timeout
1217                          * function to generate a timeout on the connection
1218                          * attempt, in case the peer doesn't respond in a
1219                          * timely manner. We hold on to the socket until the
1220                          * timeout fires.
1221                          */
1222                         sock_hold(sk);
1223                         INIT_DELAYED_WORK(&vsk->dwork,
1224                                           vsock_connect_timeout);
1225                         schedule_delayed_work(&vsk->dwork, timeout);
1226
1227                         /* Skip ahead to preserve error code set above. */
1228                         goto out_wait;
1229                 }
1230
1231                 release_sock(sk);
1232                 timeout = schedule_timeout(timeout);
1233                 lock_sock(sk);
1234
1235                 if (signal_pending(current)) {
1236                         err = sock_intr_errno(timeout);
1237                         sk->sk_state = SS_UNCONNECTED;
1238                         sock->state = SS_UNCONNECTED;
1239                         vsock_transport_cancel_pkt(vsk);
1240                         goto out_wait;
1241                 } else if (timeout == 0) {
1242                         err = -ETIMEDOUT;
1243                         sk->sk_state = SS_UNCONNECTED;
1244                         sock->state = SS_UNCONNECTED;
1245                         vsock_transport_cancel_pkt(vsk);
1246                         goto out_wait;
1247                 }
1248
1249                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1250         }
1251
1252         if (sk->sk_err) {
1253                 err = -sk->sk_err;
1254                 sk->sk_state = SS_UNCONNECTED;
1255                 sock->state = SS_UNCONNECTED;
1256         } else {
1257                 err = 0;
1258         }
1259
1260 out_wait:
1261         finish_wait(sk_sleep(sk), &wait);
1262 out:
1263         release_sock(sk);
1264         return err;
1265 }
1266
1267 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1268                         bool kern)
1269 {
1270         struct sock *listener;
1271         int err;
1272         struct sock *connected;
1273         struct vsock_sock *vconnected;
1274         long timeout;
1275         DEFINE_WAIT(wait);
1276
1277         err = 0;
1278         listener = sock->sk;
1279
1280         lock_sock(listener);
1281
1282         if (sock->type != SOCK_STREAM) {
1283                 err = -EOPNOTSUPP;
1284                 goto out;
1285         }
1286
1287         if (listener->sk_state != VSOCK_SS_LISTEN) {
1288                 err = -EINVAL;
1289                 goto out;
1290         }
1291
1292         /* Wait for children sockets to appear; these are the new sockets
1293          * created upon connection establishment.
1294          */
1295         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1296         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1297
1298         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1299                listener->sk_err == 0) {
1300                 release_sock(listener);
1301                 timeout = schedule_timeout(timeout);
1302                 finish_wait(sk_sleep(listener), &wait);
1303                 lock_sock(listener);
1304
1305                 if (signal_pending(current)) {
1306                         err = sock_intr_errno(timeout);
1307                         goto out;
1308                 } else if (timeout == 0) {
1309                         err = -EAGAIN;
1310                         goto out;
1311                 }
1312
1313                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1314         }
1315         finish_wait(sk_sleep(listener), &wait);
1316
1317         if (listener->sk_err)
1318                 err = -listener->sk_err;
1319
1320         if (connected) {
1321                 listener->sk_ack_backlog--;
1322
1323                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1324                 vconnected = vsock_sk(connected);
1325
1326                 /* If the listener socket has received an error, then we should
1327                  * reject this socket and return.  Note that we simply mark the
1328                  * socket rejected, drop our reference, and let the cleanup
1329                  * function handle the cleanup; the fact that we found it in
1330                  * the listener's accept queue guarantees that the cleanup
1331                  * function hasn't run yet.
1332                  */
1333                 if (err) {
1334                         vconnected->rejected = true;
1335                 } else {
1336                         newsock->state = SS_CONNECTED;
1337                         sock_graft(connected, newsock);
1338                 }
1339
1340                 release_sock(connected);
1341                 sock_put(connected);
1342         }
1343
1344 out:
1345         release_sock(listener);
1346         return err;
1347 }
1348
1349 static int vsock_listen(struct socket *sock, int backlog)
1350 {
1351         int err;
1352         struct sock *sk;
1353         struct vsock_sock *vsk;
1354
1355         sk = sock->sk;
1356
1357         lock_sock(sk);
1358
1359         if (sock->type != SOCK_STREAM) {
1360                 err = -EOPNOTSUPP;
1361                 goto out;
1362         }
1363
1364         if (sock->state != SS_UNCONNECTED) {
1365                 err = -EINVAL;
1366                 goto out;
1367         }
1368
1369         vsk = vsock_sk(sk);
1370
1371         if (!vsock_addr_bound(&vsk->local_addr)) {
1372                 err = -EINVAL;
1373                 goto out;
1374         }
1375
1376         sk->sk_max_ack_backlog = backlog;
1377         sk->sk_state = VSOCK_SS_LISTEN;
1378
1379         err = 0;
1380
1381 out:
1382         release_sock(sk);
1383         return err;
1384 }
1385
1386 static int vsock_stream_setsockopt(struct socket *sock,
1387                                    int level,
1388                                    int optname,
1389                                    char __user *optval,
1390                                    unsigned int optlen)
1391 {
1392         int err;
1393         struct sock *sk;
1394         struct vsock_sock *vsk;
1395         u64 val;
1396
1397         if (level != AF_VSOCK)
1398                 return -ENOPROTOOPT;
1399
1400 #define COPY_IN(_v)                                       \
1401         do {                                              \
1402                 if (optlen < sizeof(_v)) {                \
1403                         err = -EINVAL;                    \
1404                         goto exit;                        \
1405                 }                                         \
1406                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1407                         err = -EFAULT;                                  \
1408                         goto exit;                                      \
1409                 }                                                       \
1410         } while (0)
1411
1412         err = 0;
1413         sk = sock->sk;
1414         vsk = vsock_sk(sk);
1415
1416         lock_sock(sk);
1417
1418         switch (optname) {
1419         case SO_VM_SOCKETS_BUFFER_SIZE:
1420                 COPY_IN(val);
1421                 transport->set_buffer_size(vsk, val);
1422                 break;
1423
1424         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1425                 COPY_IN(val);
1426                 transport->set_max_buffer_size(vsk, val);
1427                 break;
1428
1429         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1430                 COPY_IN(val);
1431                 transport->set_min_buffer_size(vsk, val);
1432                 break;
1433
1434         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1435                 struct timeval tv;
1436                 COPY_IN(tv);
1437                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1438                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1439                         vsk->connect_timeout = tv.tv_sec * HZ +
1440                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1441                         if (vsk->connect_timeout == 0)
1442                                 vsk->connect_timeout =
1443                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1444
1445                 } else {
1446                         err = -ERANGE;
1447                 }
1448                 break;
1449         }
1450
1451         default:
1452                 err = -ENOPROTOOPT;
1453                 break;
1454         }
1455
1456 #undef COPY_IN
1457
1458 exit:
1459         release_sock(sk);
1460         return err;
1461 }
1462
1463 static int vsock_stream_getsockopt(struct socket *sock,
1464                                    int level, int optname,
1465                                    char __user *optval,
1466                                    int __user *optlen)
1467 {
1468         int err;
1469         int len;
1470         struct sock *sk;
1471         struct vsock_sock *vsk;
1472         u64 val;
1473
1474         if (level != AF_VSOCK)
1475                 return -ENOPROTOOPT;
1476
1477         err = get_user(len, optlen);
1478         if (err != 0)
1479                 return err;
1480
1481 #define COPY_OUT(_v)                            \
1482         do {                                    \
1483                 if (len < sizeof(_v))           \
1484                         return -EINVAL;         \
1485                                                 \
1486                 len = sizeof(_v);               \
1487                 if (copy_to_user(optval, &_v, len) != 0)        \
1488                         return -EFAULT;                         \
1489                                                                 \
1490         } while (0)
1491
1492         err = 0;
1493         sk = sock->sk;
1494         vsk = vsock_sk(sk);
1495
1496         switch (optname) {
1497         case SO_VM_SOCKETS_BUFFER_SIZE:
1498                 val = transport->get_buffer_size(vsk);
1499                 COPY_OUT(val);
1500                 break;
1501
1502         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1503                 val = transport->get_max_buffer_size(vsk);
1504                 COPY_OUT(val);
1505                 break;
1506
1507         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1508                 val = transport->get_min_buffer_size(vsk);
1509                 COPY_OUT(val);
1510                 break;
1511
1512         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1513                 struct timeval tv;
1514                 tv.tv_sec = vsk->connect_timeout / HZ;
1515                 tv.tv_usec =
1516                     (vsk->connect_timeout -
1517                      tv.tv_sec * HZ) * (1000000 / HZ);
1518                 COPY_OUT(tv);
1519                 break;
1520         }
1521         default:
1522                 return -ENOPROTOOPT;
1523         }
1524
1525         err = put_user(len, optlen);
1526         if (err != 0)
1527                 return -EFAULT;
1528
1529 #undef COPY_OUT
1530
1531         return 0;
1532 }
1533
1534 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1535                                 size_t len)
1536 {
1537         struct sock *sk;
1538         struct vsock_sock *vsk;
1539         ssize_t total_written;
1540         long timeout;
1541         int err;
1542         struct vsock_transport_send_notify_data send_data;
1543         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1544
1545         sk = sock->sk;
1546         vsk = vsock_sk(sk);
1547         total_written = 0;
1548         err = 0;
1549
1550         if (msg->msg_flags & MSG_OOB)
1551                 return -EOPNOTSUPP;
1552
1553         lock_sock(sk);
1554
1555         /* Callers should not provide a destination with stream sockets. */
1556         if (msg->msg_namelen) {
1557                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1558                 goto out;
1559         }
1560
1561         /* Send data only if both sides are not shutdown in the direction. */
1562         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1563             vsk->peer_shutdown & RCV_SHUTDOWN) {
1564                 err = -EPIPE;
1565                 goto out;
1566         }
1567
1568         if (sk->sk_state != SS_CONNECTED ||
1569             !vsock_addr_bound(&vsk->local_addr)) {
1570                 err = -ENOTCONN;
1571                 goto out;
1572         }
1573
1574         if (!vsock_addr_bound(&vsk->remote_addr)) {
1575                 err = -EDESTADDRREQ;
1576                 goto out;
1577         }
1578
1579         /* Wait for room in the produce queue to enqueue our user's data. */
1580         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1581
1582         err = transport->notify_send_init(vsk, &send_data);
1583         if (err < 0)
1584                 goto out;
1585
1586         while (total_written < len) {
1587                 ssize_t written;
1588
1589                 add_wait_queue(sk_sleep(sk), &wait);
1590                 while (vsock_stream_has_space(vsk) == 0 &&
1591                        sk->sk_err == 0 &&
1592                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1593                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1594
1595                         /* Don't wait for non-blocking sockets. */
1596                         if (timeout == 0) {
1597                                 err = -EAGAIN;
1598                                 remove_wait_queue(sk_sleep(sk), &wait);
1599                                 goto out_err;
1600                         }
1601
1602                         err = transport->notify_send_pre_block(vsk, &send_data);
1603                         if (err < 0) {
1604                                 remove_wait_queue(sk_sleep(sk), &wait);
1605                                 goto out_err;
1606                         }
1607
1608                         release_sock(sk);
1609                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1610                         lock_sock(sk);
1611                         if (signal_pending(current)) {
1612                                 err = sock_intr_errno(timeout);
1613                                 remove_wait_queue(sk_sleep(sk), &wait);
1614                                 goto out_err;
1615                         } else if (timeout == 0) {
1616                                 err = -EAGAIN;
1617                                 remove_wait_queue(sk_sleep(sk), &wait);
1618                                 goto out_err;
1619                         }
1620                 }
1621                 remove_wait_queue(sk_sleep(sk), &wait);
1622
1623                 /* These checks occur both as part of and after the loop
1624                  * conditional since we need to check before and after
1625                  * sleeping.
1626                  */
1627                 if (sk->sk_err) {
1628                         err = -sk->sk_err;
1629                         goto out_err;
1630                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1631                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1632                         err = -EPIPE;
1633                         goto out_err;
1634                 }
1635
1636                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1637                 if (err < 0)
1638                         goto out_err;
1639
1640                 /* Note that enqueue will only write as many bytes as are free
1641                  * in the produce queue, so we don't need to ensure len is
1642                  * smaller than the queue size.  It is the caller's
1643                  * responsibility to check how many bytes we were able to send.
1644                  */
1645
1646                 written = transport->stream_enqueue(
1647                                 vsk, msg,
1648                                 len - total_written);
1649                 if (written < 0) {
1650                         err = -ENOMEM;
1651                         goto out_err;
1652                 }
1653
1654                 total_written += written;
1655
1656                 err = transport->notify_send_post_enqueue(
1657                                 vsk, written, &send_data);
1658                 if (err < 0)
1659                         goto out_err;
1660
1661         }
1662
1663 out_err:
1664         if (total_written > 0)
1665                 err = total_written;
1666 out:
1667         release_sock(sk);
1668         return err;
1669 }
1670
1671
1672 static int
1673 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1674                      int flags)
1675 {
1676         struct sock *sk;
1677         struct vsock_sock *vsk;
1678         int err;
1679         size_t target;
1680         ssize_t copied;
1681         long timeout;
1682         struct vsock_transport_recv_notify_data recv_data;
1683
1684         DEFINE_WAIT(wait);
1685
1686         sk = sock->sk;
1687         vsk = vsock_sk(sk);
1688         err = 0;
1689
1690         lock_sock(sk);
1691
1692         if (sk->sk_state != SS_CONNECTED) {
1693                 /* Recvmsg is supposed to return 0 if a peer performs an
1694                  * orderly shutdown. Differentiate between that case and when a
1695                  * peer has not connected or a local shutdown occured with the
1696                  * SOCK_DONE flag.
1697                  */
1698                 if (sock_flag(sk, SOCK_DONE))
1699                         err = 0;
1700                 else
1701                         err = -ENOTCONN;
1702
1703                 goto out;
1704         }
1705
1706         if (flags & MSG_OOB) {
1707                 err = -EOPNOTSUPP;
1708                 goto out;
1709         }
1710
1711         /* We don't check peer_shutdown flag here since peer may actually shut
1712          * down, but there can be data in the queue that a local socket can
1713          * receive.
1714          */
1715         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1716                 err = 0;
1717                 goto out;
1718         }
1719
1720         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1721          * is not an error.  We may as well bail out now.
1722          */
1723         if (!len) {
1724                 err = 0;
1725                 goto out;
1726         }
1727
1728         /* We must not copy less than target bytes into the user's buffer
1729          * before returning successfully, so we wait for the consume queue to
1730          * have that much data to consume before dequeueing.  Note that this
1731          * makes it impossible to handle cases where target is greater than the
1732          * queue size.
1733          */
1734         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1735         if (target >= transport->stream_rcvhiwat(vsk)) {
1736                 err = -ENOMEM;
1737                 goto out;
1738         }
1739         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1740         copied = 0;
1741
1742         err = transport->notify_recv_init(vsk, target, &recv_data);
1743         if (err < 0)
1744                 goto out;
1745
1746
1747         while (1) {
1748                 s64 ready;
1749
1750                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1751                 ready = vsock_stream_has_data(vsk);
1752
1753                 if (ready == 0) {
1754                         if (sk->sk_err != 0 ||
1755                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1756                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1757                                 finish_wait(sk_sleep(sk), &wait);
1758                                 break;
1759                         }
1760                         /* Don't wait for non-blocking sockets. */
1761                         if (timeout == 0) {
1762                                 err = -EAGAIN;
1763                                 finish_wait(sk_sleep(sk), &wait);
1764                                 break;
1765                         }
1766
1767                         err = transport->notify_recv_pre_block(
1768                                         vsk, target, &recv_data);
1769                         if (err < 0) {
1770                                 finish_wait(sk_sleep(sk), &wait);
1771                                 break;
1772                         }
1773                         release_sock(sk);
1774                         timeout = schedule_timeout(timeout);
1775                         lock_sock(sk);
1776
1777                         if (signal_pending(current)) {
1778                                 err = sock_intr_errno(timeout);
1779                                 finish_wait(sk_sleep(sk), &wait);
1780                                 break;
1781                         } else if (timeout == 0) {
1782                                 err = -EAGAIN;
1783                                 finish_wait(sk_sleep(sk), &wait);
1784                                 break;
1785                         }
1786                 } else {
1787                         ssize_t read;
1788
1789                         finish_wait(sk_sleep(sk), &wait);
1790
1791                         if (ready < 0) {
1792                                 /* Invalid queue pair content. XXX This should
1793                                 * be changed to a connection reset in a later
1794                                 * change.
1795                                 */
1796
1797                                 err = -ENOMEM;
1798                                 goto out;
1799                         }
1800
1801                         err = transport->notify_recv_pre_dequeue(
1802                                         vsk, target, &recv_data);
1803                         if (err < 0)
1804                                 break;
1805
1806                         read = transport->stream_dequeue(
1807                                         vsk, msg,
1808                                         len - copied, flags);
1809                         if (read < 0) {
1810                                 err = -ENOMEM;
1811                                 break;
1812                         }
1813
1814                         copied += read;
1815
1816                         err = transport->notify_recv_post_dequeue(
1817                                         vsk, target, read,
1818                                         !(flags & MSG_PEEK), &recv_data);
1819                         if (err < 0)
1820                                 goto out;
1821
1822                         if (read >= target || flags & MSG_PEEK)
1823                                 break;
1824
1825                         target -= read;
1826                 }
1827         }
1828
1829         if (sk->sk_err)
1830                 err = -sk->sk_err;
1831         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1832                 err = 0;
1833
1834         if (copied > 0)
1835                 err = copied;
1836
1837 out:
1838         release_sock(sk);
1839         return err;
1840 }
1841
1842 static const struct proto_ops vsock_stream_ops = {
1843         .family = PF_VSOCK,
1844         .owner = THIS_MODULE,
1845         .release = vsock_release,
1846         .bind = vsock_bind,
1847         .connect = vsock_stream_connect,
1848         .socketpair = sock_no_socketpair,
1849         .accept = vsock_accept,
1850         .getname = vsock_getname,
1851         .poll = vsock_poll,
1852         .ioctl = sock_no_ioctl,
1853         .listen = vsock_listen,
1854         .shutdown = vsock_shutdown,
1855         .setsockopt = vsock_stream_setsockopt,
1856         .getsockopt = vsock_stream_getsockopt,
1857         .sendmsg = vsock_stream_sendmsg,
1858         .recvmsg = vsock_stream_recvmsg,
1859         .mmap = sock_no_mmap,
1860         .sendpage = sock_no_sendpage,
1861 };
1862
1863 static int vsock_create(struct net *net, struct socket *sock,
1864                         int protocol, int kern)
1865 {
1866         if (!sock)
1867                 return -EINVAL;
1868
1869         if (protocol && protocol != PF_VSOCK)
1870                 return -EPROTONOSUPPORT;
1871
1872         switch (sock->type) {
1873         case SOCK_DGRAM:
1874                 sock->ops = &vsock_dgram_ops;
1875                 break;
1876         case SOCK_STREAM:
1877                 sock->ops = &vsock_stream_ops;
1878                 break;
1879         default:
1880                 return -ESOCKTNOSUPPORT;
1881         }
1882
1883         sock->state = SS_UNCONNECTED;
1884
1885         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1886 }
1887
1888 static const struct net_proto_family vsock_family_ops = {
1889         .family = AF_VSOCK,
1890         .create = vsock_create,
1891         .owner = THIS_MODULE,
1892 };
1893
1894 static long vsock_dev_do_ioctl(struct file *filp,
1895                                unsigned int cmd, void __user *ptr)
1896 {
1897         u32 __user *p = ptr;
1898         int retval = 0;
1899
1900         switch (cmd) {
1901         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1902                 if (put_user(transport->get_local_cid(), p) != 0)
1903                         retval = -EFAULT;
1904                 break;
1905
1906         default:
1907                 pr_err("Unknown ioctl %d\n", cmd);
1908                 retval = -EINVAL;
1909         }
1910
1911         return retval;
1912 }
1913
1914 static long vsock_dev_ioctl(struct file *filp,
1915                             unsigned int cmd, unsigned long arg)
1916 {
1917         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1918 }
1919
1920 #ifdef CONFIG_COMPAT
1921 static long vsock_dev_compat_ioctl(struct file *filp,
1922                                    unsigned int cmd, unsigned long arg)
1923 {
1924         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1925 }
1926 #endif
1927
1928 static const struct file_operations vsock_device_ops = {
1929         .owner          = THIS_MODULE,
1930         .unlocked_ioctl = vsock_dev_ioctl,
1931 #ifdef CONFIG_COMPAT
1932         .compat_ioctl   = vsock_dev_compat_ioctl,
1933 #endif
1934         .open           = nonseekable_open,
1935 };
1936
1937 static struct miscdevice vsock_device = {
1938         .name           = "vsock",
1939         .fops           = &vsock_device_ops,
1940 };
1941
1942 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1943 {
1944         int err = mutex_lock_interruptible(&vsock_register_mutex);
1945
1946         if (err)
1947                 return err;
1948
1949         if (transport) {
1950                 err = -EBUSY;
1951                 goto err_busy;
1952         }
1953
1954         /* Transport must be the owner of the protocol so that it can't
1955          * unload while there are open sockets.
1956          */
1957         vsock_proto.owner = owner;
1958         transport = t;
1959
1960         vsock_init_tables();
1961
1962         vsock_device.minor = MISC_DYNAMIC_MINOR;
1963         err = misc_register(&vsock_device);
1964         if (err) {
1965                 pr_err("Failed to register misc device\n");
1966                 goto err_reset_transport;
1967         }
1968
1969         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1970         if (err) {
1971                 pr_err("Cannot register vsock protocol\n");
1972                 goto err_deregister_misc;
1973         }
1974
1975         err = sock_register(&vsock_family_ops);
1976         if (err) {
1977                 pr_err("could not register af_vsock (%d) address family: %d\n",
1978                        AF_VSOCK, err);
1979                 goto err_unregister_proto;
1980         }
1981
1982         mutex_unlock(&vsock_register_mutex);
1983         return 0;
1984
1985 err_unregister_proto:
1986         proto_unregister(&vsock_proto);
1987 err_deregister_misc:
1988         misc_deregister(&vsock_device);
1989 err_reset_transport:
1990         transport = NULL;
1991 err_busy:
1992         mutex_unlock(&vsock_register_mutex);
1993         return err;
1994 }
1995 EXPORT_SYMBOL_GPL(__vsock_core_init);
1996
1997 void vsock_core_exit(void)
1998 {
1999         mutex_lock(&vsock_register_mutex);
2000
2001         misc_deregister(&vsock_device);
2002         sock_unregister(AF_VSOCK);
2003         proto_unregister(&vsock_proto);
2004
2005         /* We do not want the assignment below re-ordered. */
2006         mb();
2007         transport = NULL;
2008
2009         mutex_unlock(&vsock_register_mutex);
2010 }
2011 EXPORT_SYMBOL_GPL(vsock_core_exit);
2012
2013 const struct vsock_transport *vsock_core_get_transport(void)
2014 {
2015         /* vsock_register_mutex not taken since only the transport uses this
2016          * function and only while registered.
2017          */
2018         return transport;
2019 }
2020 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2021
2022 MODULE_AUTHOR("VMware, Inc.");
2023 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2024 MODULE_VERSION("1.0.2.0-k");
2025 MODULE_LICENSE("GPL v2");