net/vmw_vsock/af_vsock.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 /* Implementation notes:
  17  *
  18  * - There are two kinds of sockets: those created by user action (such as
  19  * calling socket(2)) and those created by incoming connection request packets.
  20  *
  21  * - There are two "global" tables, one for bound sockets (sockets that have
  22  * specified an address that they are responsible for) and one for connected
  23  * sockets (sockets that have established a connection with another socket).
  24  * These tables are "global" in that all sockets on the system are placed
  25  * within them. - Note, though, that the bound table contains an extra entry
  26  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  27  * that list. The bound table is used solely for lookup of sockets when packets
  28  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  29  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  30  * sockets out of the bound hash buckets will reduce the chance of collisions
  31  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  32  * socket type in the hash table lookups.
  33  *
  34  * - Sockets created by user action will either be "client" sockets that
  35  * initiate a connection or "server" sockets that listen for connections; we do
  36  * not support simultaneous connects (two "client" sockets connecting).
  37  *
  38  * - "Server" sockets are referred to as listener sockets throughout this
  39  * implementation because they are in the VSOCK_SS_LISTEN state.  When a
  40  * connection request is received (the second kind of socket mentioned above),
  41  * we create a new socket and refer to it as a pending socket.  These pending
  42  * sockets are placed on the pending connection list of the listener socket.
  43  * When future packets are received for the address the listener socket is
  44  * bound to, we check if the source of the packet is from one that has an
  45  * existing pending connection.  If it does, we process the packet for the
  46  * pending socket.  When that socket reaches the connected state, it is removed
  47  * from the listener socket's pending list and enqueued in the listener
  48  * socket's accept queue.  Callers of accept(2) will accept connected sockets
  49  * from the listener socket's accept queue.  If the socket cannot be accepted
  50  * for some reason then it is marked rejected.  Once the connection is
  51  * accepted, it is owned by the user process and the responsibility for cleanup
  52  * falls with that user process.
  53  *
  54  * - It is possible that these pending sockets will never reach the connected
  55  * state; in fact, we may never receive another packet after the connection
  56  * request.  Because of this, we must schedule a cleanup function to run in the
  57  * future, after some amount of time passes where a connection should have been
  58  * established.  This function ensures that the socket is off all lists so it
  59  * cannot be retrieved, then drops all references to the socket so it is cleaned
  60  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  61  * function will also cleanup rejected sockets, those that reach the connected
  62  * state but leave it before they have been accepted.
  63  *
  64  * - Lock ordering for pending or accept queue sockets is:
  65  *
  66  *     lock_sock(listener);
  67  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  68  *
  69  * Using explicit nested locking keeps lockdep happy since normally only one
  70  * lock of a given class may be taken at a time.
  71  *
  72  * - Sockets created by user action will be cleaned up when the user process
  73  * calls close(2), causing our release implementation to be called. Our release
  74  * implementation will perform some cleanup then drop the last reference so our
  75  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  76  * perform additional cleanup that's common for both types of sockets.
  77  *
  78  * - A socket's reference count is what ensures that the structure won't be
  79  * freed.  Each entry in a list (such as the "global" bound and connected tables
  80  * and the listener socket's pending list and connected queue) ensures a
  81  * reference.  When we defer work until process context and pass a socket as our
  82  * argument, we must ensure the reference count is increased to ensure the
  83  * socket isn't freed before the function is run; the deferred function will
  84  * then drop the reference.
  85  */
  86
  87 #include <linux/types.h>
  88 #include <linux/bitops.h>
  89 #include <linux/cred.h>
  90 #include <linux/init.h>
  91 #include <linux/io.h>
  92 #include <linux/kernel.h>
  93 #include <linux/kmod.h>
  94 #include <linux/list.h>
  95 #include <linux/miscdevice.h>
  96 #include <linux/module.h>
  97 #include <linux/mutex.h>
  98 #include <linux/net.h>
  99 #include <linux/poll.h>
 100 #include <linux/random.h>
 101 #include <linux/skbuff.h>
 102 #include <linux/smp.h>
 103 #include <linux/socket.h>
 104 #include <linux/stddef.h>
 105 #include <linux/unistd.h>
 106 #include <linux/wait.h>
 107 #include <linux/workqueue.h>
 108 #include <net/sock.h>
 109 #include <net/af_vsock.h>
 110
 111 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 112 static void vsock_sk_destruct(struct sock *sk);
 113 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 114
 115 /* Protocol family. */
 116 static struct proto vsock_proto = {
 117         .name = "AF_VSOCK",
 118         .owner = THIS_MODULE,
 119         .obj_size = sizeof(struct vsock_sock),
 120 };
 121
 122 /* The default peer timeout indicates how long we will wait for a peer response
 123  * to a control message.
 124  */
 125 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 126
 127 static const struct vsock_transport *transport;
 128 static DEFINE_MUTEX(vsock_register_mutex);
 129
 130 /**** EXPORTS ****/
 131
 132 /* Get the ID of the local context.  This is transport dependent. */
 133
 134 int vm_sockets_get_local_cid(void)
 135 {
 136         return transport->get_local_cid();
 137 }
 138 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 139
 140 /**** UTILS ****/
 141
 142 /* Each bound VSocket is stored in the bind hash table and each connected
 143  * VSocket is stored in the connected hash table.
 144  *
 145  * Unbound sockets are all put on the same list attached to the end of the hash
 146  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 147  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 148  * represents the list that addr hashes to).
 149  *
 150  * Specifically, we initialize the vsock_bind_table array to a size of
 151  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 152  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 153  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 154  * mods with VSOCK_HASH_SIZE to ensure this.
 155  */
 156 #define VSOCK_HASH_SIZE         251
 157 #define MAX_PORT_RETRIES        24
 158
 159 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 160 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 161 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 162
 163 /* XXX This can probably be implemented in a better way. */
 164 #define VSOCK_CONN_HASH(src, dst)                               \
 165         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 166 #define vsock_connected_sockets(src, dst)               \
 167         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 168 #define vsock_connected_sockets_vsk(vsk)                                \
 169         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 170
 171 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 172 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 173 static DEFINE_SPINLOCK(vsock_table_lock);
 174
 175 /* Autobind this socket to the local address if necessary. */
 176 static int vsock_auto_bind(struct vsock_sock *vsk)
 177 {
 178         struct sock *sk = sk_vsock(vsk);
 179         struct sockaddr_vm local_addr;
 180
 181         if (vsock_addr_bound(&vsk->local_addr))
 182                 return 0;
 183         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 184         return __vsock_bind(sk, &local_addr);
 185 }
 186
 187 static void vsock_init_tables(void)
 188 {
 189         int i;
 190
 191         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 192                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 193
 194         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 195                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 196 }
 197
 198 static void __vsock_insert_bound(struct list_head *list,
 199                                  struct vsock_sock *vsk)
 200 {
 201         sock_hold(&vsk->sk);
 202         list_add(&vsk->bound_table, list);
 203 }
 204
 205 static void __vsock_insert_connected(struct list_head *list,
 206                                      struct vsock_sock *vsk)
 207 {
 208         sock_hold(&vsk->sk);
 209         list_add(&vsk->connected_table, list);
 210 }
 211
 212 static void __vsock_remove_bound(struct vsock_sock *vsk)
 213 {
 214         list_del_init(&vsk->bound_table);
 215         sock_put(&vsk->sk);
 216 }
 217
 218 static void __vsock_remove_connected(struct vsock_sock *vsk)
 219 {
 220         list_del_init(&vsk->connected_table);
 221         sock_put(&vsk->sk);
 222 }
 223
 224 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 225 {
 226         struct vsock_sock *vsk;
 227
 228         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 229                 if (addr->svm_port == vsk->local_addr.svm_port)
 230                         return sk_vsock(vsk);
 231
 232         return NULL;
 233 }
 234
 235 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 236                                                   struct sockaddr_vm *dst)
 237 {
 238         struct vsock_sock *vsk;
 239
 240         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 241                             connected_table) {
 242                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 243                     dst->svm_port == vsk->local_addr.svm_port) {
 244                         return sk_vsock(vsk);
 245                 }
 246         }
 247
 248         return NULL;
 249 }
 250
 251 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 252 {
 253         return !list_empty(&vsk->bound_table);
 254 }
 255
 256 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 257 {
 258         return !list_empty(&vsk->connected_table);
 259 }
 260
 261 static void vsock_insert_unbound(struct vsock_sock *vsk)
 262 {
 263         spin_lock_bh(&vsock_table_lock);
 264         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 265         spin_unlock_bh(&vsock_table_lock);
 266 }
 267
 268 void vsock_insert_connected(struct vsock_sock *vsk)
 269 {
 270         struct list_head *list = vsock_connected_sockets(
 271                 &vsk->remote_addr, &vsk->local_addr);
 272
 273         spin_lock_bh(&vsock_table_lock);
 274         __vsock_insert_connected(list, vsk);
 275         spin_unlock_bh(&vsock_table_lock);
 276 }
 277 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 278
 279 void vsock_remove_bound(struct vsock_sock *vsk)
 280 {
 281         spin_lock_bh(&vsock_table_lock);
 282         __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 static 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, pending_work.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
 500 /**** SOCKET OPERATIONS ****/
 501
 502 static int __vsock_bind_stream(struct vsock_sock *vsk,
 503                                struct sockaddr_vm *addr)
 504 {
 505         static u32 port = 0;
 506         struct sockaddr_vm new_addr;
 507
 508         if (!port)
 509                 port = LAST_RESERVED_PORT + 1 +
 510                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 511
 512         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 513
 514         if (addr->svm_port == VMADDR_PORT_ANY) {
 515                 bool found = false;
 516                 unsigned int i;
 517
 518                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 519                         if (port <= LAST_RESERVED_PORT)
 520                                 port = LAST_RESERVED_PORT + 1;
 521
 522                         new_addr.svm_port = port++;
 523
 524                         if (!__vsock_find_bound_socket(&new_addr)) {
 525                                 found = true;
 526                                 break;
 527                         }
 528                 }
 529
 530                 if (!found)
 531                         return -EADDRNOTAVAIL;
 532         } else {
 533                 /* If port is in reserved range, ensure caller
 534                  * has necessary privileges.
 535                  */
 536                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 537                     !capable(CAP_NET_BIND_SERVICE)) {
 538                         return -EACCES;
 539                 }
 540
 541                 if (__vsock_find_bound_socket(&new_addr))
 542                         return -EADDRINUSE;
 543         }
 544
 545         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 546
 547         /* Remove stream sockets from the unbound list and add them to the hash
 548          * table for easy lookup by its address.  The unbound list is simply an
 549          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 550          */
 551         __vsock_remove_bound(vsk);
 552         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 553
 554         return 0;
 555 }
 556
 557 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 558                               struct sockaddr_vm *addr)
 559 {
 560         return transport->dgram_bind(vsk, addr);
 561 }
 562
 563 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 564 {
 565         struct vsock_sock *vsk = vsock_sk(sk);
 566         u32 cid;
 567         int retval;
 568
 569         /* First ensure this socket isn't already bound. */
 570         if (vsock_addr_bound(&vsk->local_addr))
 571                 return -EINVAL;
 572
 573         /* Now bind to the provided address or select appropriate values if
 574          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 575          * like AF_INET prevents binding to a non-local IP address (in most
 576          * cases), we only allow binding to the local CID.
 577          */
 578         cid = transport->get_local_cid();
 579         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 580                 return -EADDRNOTAVAIL;
 581
 582         switch (sk->sk_socket->type) {
 583         case SOCK_STREAM:
 584                 spin_lock_bh(&vsock_table_lock);
 585                 retval = __vsock_bind_stream(vsk, addr);
 586                 spin_unlock_bh(&vsock_table_lock);
 587                 break;
 588
 589         case SOCK_DGRAM:
 590                 retval = __vsock_bind_dgram(vsk, addr);
 591                 break;
 592
 593         default:
 594                 retval = -EINVAL;
 595                 break;
 596         }
 597
 598         return retval;
 599 }
 600
 601 static void vsock_connect_timeout(struct work_struct *work);
 602
 603 struct sock *__vsock_create(struct net *net,
 604                             struct socket *sock,
 605                             struct sock *parent,
 606                             gfp_t priority,
 607                             unsigned short type,
 608                             int kern)
 609 {
 610         struct sock *sk;
 611         struct vsock_sock *psk;
 612         struct vsock_sock *vsk;
 613
 614         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 615         if (!sk)
 616                 return NULL;
 617
 618         sock_init_data(sock, sk);
 619
 620         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 621          * non-NULL. We make sure that our sockets always have a type by
 622          * setting it here if needed.
 623          */
 624         if (!sock)
 625                 sk->sk_type = type;
 626
 627         vsk = vsock_sk(sk);
 628         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 629         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 630
 631         sk->sk_destruct = vsock_sk_destruct;
 632         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 633         sk->sk_state = 0;
 634         sock_reset_flag(sk, SOCK_DONE);
 635
 636         INIT_LIST_HEAD(&vsk->bound_table);
 637         INIT_LIST_HEAD(&vsk->connected_table);
 638         vsk->listener = NULL;
 639         INIT_LIST_HEAD(&vsk->pending_links);
 640         INIT_LIST_HEAD(&vsk->accept_queue);
 641         vsk->rejected = false;
 642         vsk->sent_request = false;
 643         vsk->ignore_connecting_rst = false;
 644         vsk->peer_shutdown = 0;
 645         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 646         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 647
 648         psk = parent ? vsock_sk(parent) : NULL;
 649         if (parent) {
 650                 vsk->trusted = psk->trusted;
 651                 vsk->owner = get_cred(psk->owner);
 652                 vsk->connect_timeout = psk->connect_timeout;
 653                 security_sk_clone(parent, sk);
 654         } else {
 655                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 656                 vsk->owner = get_current_cred();
 657                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 658         }
 659
 660         if (transport->init(vsk, psk) < 0) {
 661                 sk_free(sk);
 662                 return NULL;
 663         }
 664
 665         if (sock)
 666                 vsock_insert_unbound(vsk);
 667
 668         return sk;
 669 }
 670 EXPORT_SYMBOL_GPL(__vsock_create);
 671
 672 static void __vsock_release(struct sock *sk)
 673 {
 674         if (sk) {
 675                 struct sk_buff *skb;
 676                 struct sock *pending;
 677                 struct vsock_sock *vsk;
 678
 679                 vsk = vsock_sk(sk);
 680                 pending = NULL; /* Compiler warning. */
 681
 682                 transport->release(vsk);
 683
 684                 lock_sock(sk);
 685                 sock_orphan(sk);
 686                 sk->sk_shutdown = SHUTDOWN_MASK;
 687
 688                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 689                         kfree_skb(skb);
 690
 691                 /* Clean up any sockets that never were accepted. */
 692                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 693                         __vsock_release(pending);
 694                         sock_put(pending);
 695                 }
 696
 697                 release_sock(sk);
 698                 sock_put(sk);
 699         }
 700 }
 701
 702 static void vsock_sk_destruct(struct sock *sk)
 703 {
 704         struct vsock_sock *vsk = vsock_sk(sk);
 705
 706         transport->destruct(vsk);
 707
 708         /* When clearing these addresses, there's no need to set the family and
 709          * possibly register the address family with the kernel.
 710          */
 711         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 712         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 713
 714         put_cred(vsk->owner);
 715 }
 716
 717 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 718 {
 719         int err;
 720
 721         err = sock_queue_rcv_skb(sk, skb);
 722         if (err)
 723                 kfree_skb(skb);
 724
 725         return err;
 726 }
 727
 728 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 729 {
 730         return transport->stream_has_data(vsk);
 731 }
 732 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 733
 734 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 735 {
 736         return transport->stream_has_space(vsk);
 737 }
 738 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 739
 740 static int vsock_release(struct socket *sock)
 741 {
 742         __vsock_release(sock->sk);
 743         sock->sk = NULL;
 744         sock->state = SS_FREE;
 745
 746         return 0;
 747 }
 748
 749 static int
 750 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 751 {
 752         int err;
 753         struct sock *sk;
 754         struct sockaddr_vm *vm_addr;
 755
 756         sk = sock->sk;
 757
 758         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 759                 return -EINVAL;
 760
 761         lock_sock(sk);
 762         err = __vsock_bind(sk, vm_addr);
 763         release_sock(sk);
 764
 765         return err;
 766 }
 767
 768 static int vsock_getname(struct socket *sock,
 769                          struct sockaddr *addr, int *addr_len, int peer)
 770 {
 771         int err;
 772         struct sock *sk;
 773         struct vsock_sock *vsk;
 774         struct sockaddr_vm *vm_addr;
 775
 776         sk = sock->sk;
 777         vsk = vsock_sk(sk);
 778         err = 0;
 779
 780         lock_sock(sk);
 781
 782         if (peer) {
 783                 if (sock->state != SS_CONNECTED) {
 784                         err = -ENOTCONN;
 785                         goto out;
 786                 }
 787                 vm_addr = &vsk->remote_addr;
 788         } else {
 789                 vm_addr = &vsk->local_addr;
 790         }
 791
 792         if (!vm_addr) {
 793                 err = -EINVAL;
 794                 goto out;
 795         }
 796
 797         /* sys_getsockname() and sys_getpeername() pass us a
 798          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 799          * that macro is defined in socket.c instead of .h, so we hardcode its
 800          * value here.
 801          */
 802         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 803         memcpy(addr, vm_addr, sizeof(*vm_addr));
 804         *addr_len = sizeof(*vm_addr);
 805
 806 out:
 807         release_sock(sk);
 808         return err;
 809 }
 810
 811 static int vsock_shutdown(struct socket *sock, int mode)
 812 {
 813         int err;
 814         struct sock *sk;
 815
 816         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 817          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 818          * here like the other address families do.  Note also that the
 819          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 820          * which is what we want.
 821          */
 822         mode++;
 823
 824         if ((mode & ~SHUTDOWN_MASK) || !mode)
 825                 return -EINVAL;
 826
 827         /* If this is a STREAM socket and it is not connected then bail out
 828          * immediately.  If it is a DGRAM socket then we must first kick the
 829          * socket so that it wakes up from any sleeping calls, for example
 830          * recv(), and then afterwards return the error.
 831          */
 832
 833         sk = sock->sk;
 834
 835         lock_sock(sk);
 836         if (sock->state == SS_UNCONNECTED) {
 837                 err = -ENOTCONN;
 838                 if (sk->sk_type == SOCK_STREAM)
 839                         goto out;
 840         } else {
 841                 sock->state = SS_DISCONNECTING;
 842                 err = 0;
 843         }
 844
 845         /* Receive and send shutdowns are treated alike. */
 846         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 847         if (mode) {
 848                 sk->sk_shutdown |= mode;
 849                 sk->sk_state_change(sk);
 850
 851                 if (sk->sk_type == SOCK_STREAM) {
 852                         sock_reset_flag(sk, SOCK_DONE);
 853                         vsock_send_shutdown(sk, mode);
 854                 }
 855         }
 856
 857 out:
 858         release_sock(sk);
 859         return err;
 860 }
 861
 862 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 863                                poll_table *wait)
 864 {
 865         struct sock *sk;
 866         unsigned int mask;
 867         struct vsock_sock *vsk;
 868
 869         sk = sock->sk;
 870         vsk = vsock_sk(sk);
 871
 872         poll_wait(file, sk_sleep(sk), wait);
 873         mask = 0;
 874
 875         if (sk->sk_err)
 876                 /* Signify that there has been an error on this socket. */
 877                 mask |= POLLERR;
 878
 879         /* INET sockets treat local write shutdown and peer write shutdown as a
 880          * case of POLLHUP set.
 881          */
 882         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 883             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 884              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 885                 mask |= POLLHUP;
 886         }
 887
 888         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 889             vsk->peer_shutdown & SEND_SHUTDOWN) {
 890                 mask |= POLLRDHUP;
 891         }
 892
 893         if (sock->type == SOCK_DGRAM) {
 894                 /* For datagram sockets we can read if there is something in
 895                  * the queue and write as long as the socket isn't shutdown for
 896                  * sending.
 897                  */
 898                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 899                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 900                         mask |= POLLIN | POLLRDNORM;
 901                 }
 902
 903                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 904                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 905
 906         } else if (sock->type == SOCK_STREAM) {
 907                 lock_sock(sk);
 908
 909                 /* Listening sockets that have connections in their accept
 910                  * queue can be read.
 911                  */
 912                 if (sk->sk_state == VSOCK_SS_LISTEN
 913                     && !vsock_is_accept_queue_empty(sk))
 914                         mask |= POLLIN | POLLRDNORM;
 915
 916                 /* If there is something in the queue then we can read. */
 917                 if (transport->stream_is_active(vsk) &&
 918                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 919                         bool data_ready_now = false;
 920                         int ret = transport->notify_poll_in(
 921                                         vsk, 1, &data_ready_now);
 922                         if (ret < 0) {
 923                                 mask |= POLLERR;
 924                         } else {
 925                                 if (data_ready_now)
 926                                         mask |= POLLIN | POLLRDNORM;
 927
 928                         }
 929                 }
 930
 931                 /* Sockets whose connections have been closed, reset, or
 932                  * terminated should also be considered read, and we check the
 933                  * shutdown flag for that.
 934                  */
 935                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 936                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 937                         mask |= POLLIN | POLLRDNORM;
 938                 }
 939
 940                 /* Connected sockets that can produce data can be written. */
 941                 if (sk->sk_state == SS_CONNECTED) {
 942                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 943                                 bool space_avail_now = false;
 944                                 int ret = transport->notify_poll_out(
 945                                                 vsk, 1, &space_avail_now);
 946                                 if (ret < 0) {
 947                                         mask |= POLLERR;
 948                                 } else {
 949                                         if (space_avail_now)
 950                                                 /* Remove POLLWRBAND since INET
 951                                                  * sockets are not setting it.
 952                                                  */
 953                                                 mask |= POLLOUT | POLLWRNORM;
 954
 955                                 }
 956                         }
 957                 }
 958
 959                 /* Simulate INET socket poll behaviors, which sets
 960                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 961                  * but local send is not shutdown.
 962                  */
 963                 if (sk->sk_state == SS_UNCONNECTED) {
 964                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 965                                 mask |= POLLOUT | POLLWRNORM;
 966
 967                 }
 968
 969                 release_sock(sk);
 970         }
 971
 972         return mask;
 973 }
 974
 975 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 976                                size_t len)
 977 {
 978         int err;
 979         struct sock *sk;
 980         struct vsock_sock *vsk;
 981         struct sockaddr_vm *remote_addr;
 982
 983         if (msg->msg_flags & MSG_OOB)
 984                 return -EOPNOTSUPP;
 985
 986         /* For now, MSG_DONTWAIT is always assumed... */
 987         err = 0;
 988         sk = sock->sk;
 989         vsk = vsock_sk(sk);
 990
 991         lock_sock(sk);
 992
 993         err = vsock_auto_bind(vsk);
 994         if (err)
 995                 goto out;
 996
 997
 998         /* If the provided message contains an address, use that.  Otherwise
 999          * fall back on the socket's remote handle (if it has been connected).
1000          */
1001         if (msg->msg_name &&
1002             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1003                             &remote_addr) == 0) {
1004                 /* Ensure this address is of the right type and is a valid
1005                  * destination.
1006                  */
1007
1008                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1009                         remote_addr->svm_cid = transport->get_local_cid();
1010
1011                 if (!vsock_addr_bound(remote_addr)) {
1012                         err = -EINVAL;
1013                         goto out;
1014                 }
1015         } else if (sock->state == SS_CONNECTED) {
1016                 remote_addr = &vsk->remote_addr;
1017
1018                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1019                         remote_addr->svm_cid = transport->get_local_cid();
1020
1021                 /* XXX Should connect() or this function ensure remote_addr is
1022                  * bound?
1023                  */
1024                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1025                         err = -EINVAL;
1026                         goto out;
1027                 }
1028         } else {
1029                 err = -EINVAL;
1030                 goto out;
1031         }
1032
1033         if (!transport->dgram_allow(remote_addr->svm_cid,
1034                                     remote_addr->svm_port)) {
1035                 err = -EINVAL;
1036                 goto out;
1037         }
1038
1039         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1040
1041 out:
1042         release_sock(sk);
1043         return err;
1044 }
1045
1046 static int vsock_dgram_connect(struct socket *sock,
1047                                struct sockaddr *addr, int addr_len, int flags)
1048 {
1049         int err;
1050         struct sock *sk;
1051         struct vsock_sock *vsk;
1052         struct sockaddr_vm *remote_addr;
1053
1054         sk = sock->sk;
1055         vsk = vsock_sk(sk);
1056
1057         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1058         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1059                 lock_sock(sk);
1060                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1061                                 VMADDR_PORT_ANY);
1062                 sock->state = SS_UNCONNECTED;
1063                 release_sock(sk);
1064                 return 0;
1065         } else if (err != 0)
1066                 return -EINVAL;
1067
1068         lock_sock(sk);
1069
1070         err = vsock_auto_bind(vsk);
1071         if (err)
1072                 goto out;
1073
1074         if (!transport->dgram_allow(remote_addr->svm_cid,
1075                                     remote_addr->svm_port)) {
1076                 err = -EINVAL;
1077                 goto out;
1078         }
1079
1080         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1081         sock->state = SS_CONNECTED;
1082
1083 out:
1084         release_sock(sk);
1085         return err;
1086 }
1087
1088 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1089                                size_t len, int flags)
1090 {
1091         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1092 }
1093
1094 static const struct proto_ops vsock_dgram_ops = {
1095         .family = PF_VSOCK,
1096         .owner = THIS_MODULE,
1097         .release = vsock_release,
1098         .bind = vsock_bind,
1099         .connect = vsock_dgram_connect,
1100         .socketpair = sock_no_socketpair,
1101         .accept = sock_no_accept,
1102         .getname = vsock_getname,
1103         .poll = vsock_poll,
1104         .ioctl = sock_no_ioctl,
1105         .listen = sock_no_listen,
1106         .shutdown = vsock_shutdown,
1107         .setsockopt = sock_no_setsockopt,
1108         .getsockopt = sock_no_getsockopt,
1109         .sendmsg = vsock_dgram_sendmsg,
1110         .recvmsg = vsock_dgram_recvmsg,
1111         .mmap = sock_no_mmap,
1112         .sendpage = sock_no_sendpage,
1113 };
1114
1115 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1116 {
1117         if (!transport->cancel_pkt)
1118                 return -EOPNOTSUPP;
1119
1120         return transport->cancel_pkt(vsk);
1121 }
1122
1123 static void vsock_connect_timeout(struct work_struct *work)
1124 {
1125         struct sock *sk;
1126         struct vsock_sock *vsk;
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 == SS_CONNECTING &&
1133             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1134                 sk->sk_state = SS_UNCONNECTED;
1135                 sk->sk_err = ETIMEDOUT;
1136                 sk->sk_error_report(sk);
1137                 vsock_transport_cancel_pkt(vsk);
1138         }
1139         release_sock(sk);
1140
1141         sock_put(sk);
1142 }
1143
1144 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1145                                 int addr_len, int flags)
1146 {
1147         int err;
1148         struct sock *sk;
1149         struct vsock_sock *vsk;
1150         struct sockaddr_vm *remote_addr;
1151         long timeout;
1152         DEFINE_WAIT(wait);
1153
1154         err = 0;
1155         sk = sock->sk;
1156         vsk = vsock_sk(sk);
1157
1158         lock_sock(sk);
1159
1160         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1161         switch (sock->state) {
1162         case SS_CONNECTED:
1163                 err = -EISCONN;
1164                 goto out;
1165         case SS_DISCONNECTING:
1166                 err = -EINVAL;
1167                 goto out;
1168         case SS_CONNECTING:
1169                 /* This continues on so we can move sock into the SS_CONNECTED
1170                  * state once the connection has completed (at which point err
1171                  * will be set to zero also).  Otherwise, we will either wait
1172                  * for the connection or return -EALREADY should this be a
1173                  * non-blocking call.
1174                  */
1175                 err = -EALREADY;
1176                 if (flags & O_NONBLOCK)
1177                         goto out;
1178                 break;
1179         default:
1180                 if ((sk->sk_state == VSOCK_SS_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 = SS_CONNECTING;
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 != SS_CONNECTED && 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 = SS_UNCONNECTED;
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 = SS_UNCONNECTED;
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 = SS_UNCONNECTED;
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 {
1276         struct sock *listener;
1277         int err;
1278         struct sock *connected;
1279         struct vsock_sock *vconnected;
1280         long timeout;
1281         DEFINE_WAIT(wait);
1282
1283         err = 0;
1284         listener = sock->sk;
1285
1286         lock_sock(listener);
1287
1288         if (sock->type != SOCK_STREAM) {
1289                 err = -EOPNOTSUPP;
1290                 goto out;
1291         }
1292
1293         if (listener->sk_state != VSOCK_SS_LISTEN) {
1294                 err = -EINVAL;
1295                 goto out;
1296         }
1297
1298         /* Wait for children sockets to appear; these are the new sockets
1299          * created upon connection establishment.
1300          */
1301         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1302         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1303
1304         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1305                listener->sk_err == 0) {
1306                 release_sock(listener);
1307                 timeout = schedule_timeout(timeout);
1308                 finish_wait(sk_sleep(listener), &wait);
1309                 lock_sock(listener);
1310
1311                 if (signal_pending(current)) {
1312                         err = sock_intr_errno(timeout);
1313                         goto out;
1314                 } else if (timeout == 0) {
1315                         err = -EAGAIN;
1316                         goto out;
1317                 }
1318
1319                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1320         }
1321         finish_wait(sk_sleep(listener), &wait);
1322
1323         if (listener->sk_err)
1324                 err = -listener->sk_err;
1325
1326         if (connected) {
1327                 listener->sk_ack_backlog--;
1328
1329                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1330                 vconnected = vsock_sk(connected);
1331
1332                 /* If the listener socket has received an error, then we should
1333                  * reject this socket and return.  Note that we simply mark the
1334                  * socket rejected, drop our reference, and let the cleanup
1335                  * function handle the cleanup; the fact that we found it in
1336                  * the listener's accept queue guarantees that the cleanup
1337                  * function hasn't run yet.
1338                  */
1339                 if (err) {
1340                         vconnected->rejected = true;
1341                 } else {
1342                         newsock->state = SS_CONNECTED;
1343                         sock_graft(connected, newsock);
1344                 }
1345
1346                 release_sock(connected);
1347                 sock_put(connected);
1348         }
1349
1350 out:
1351         release_sock(listener);
1352         return err;
1353 }
1354
1355 static int vsock_listen(struct socket *sock, int backlog)
1356 {
1357         int err;
1358         struct sock *sk;
1359         struct vsock_sock *vsk;
1360
1361         sk = sock->sk;
1362
1363         lock_sock(sk);
1364
1365         if (sock->type != SOCK_STREAM) {
1366                 err = -EOPNOTSUPP;
1367                 goto out;
1368         }
1369
1370         if (sock->state != SS_UNCONNECTED) {
1371                 err = -EINVAL;
1372                 goto out;
1373         }
1374
1375         vsk = vsock_sk(sk);
1376
1377         if (!vsock_addr_bound(&vsk->local_addr)) {
1378                 err = -EINVAL;
1379                 goto out;
1380         }
1381
1382         sk->sk_max_ack_backlog = backlog;
1383         sk->sk_state = VSOCK_SS_LISTEN;
1384
1385         err = 0;
1386
1387 out:
1388         release_sock(sk);
1389         return err;
1390 }
1391
1392 static int vsock_stream_setsockopt(struct socket *sock,
1393                                    int level,
1394                                    int optname,
1395                                    char __user *optval,
1396                                    unsigned int optlen)
1397 {
1398         int err;
1399         struct sock *sk;
1400         struct vsock_sock *vsk;
1401         u64 val;
1402
1403         if (level != AF_VSOCK)
1404                 return -ENOPROTOOPT;
1405
1406 #define COPY_IN(_v)                                       \
1407         do {                                              \
1408                 if (optlen < sizeof(_v)) {                \
1409                         err = -EINVAL;                    \
1410                         goto exit;                        \
1411                 }                                         \
1412                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1413                         err = -EFAULT;                                  \
1414                         goto exit;                                      \
1415                 }                                                       \
1416         } while (0)
1417
1418         err = 0;
1419         sk = sock->sk;
1420         vsk = vsock_sk(sk);
1421
1422         lock_sock(sk);
1423
1424         switch (optname) {
1425         case SO_VM_SOCKETS_BUFFER_SIZE:
1426                 COPY_IN(val);
1427                 transport->set_buffer_size(vsk, val);
1428                 break;
1429
1430         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1431                 COPY_IN(val);
1432                 transport->set_max_buffer_size(vsk, val);
1433                 break;
1434
1435         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1436                 COPY_IN(val);
1437                 transport->set_min_buffer_size(vsk, val);
1438                 break;
1439
1440         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1441                 struct timeval tv;
1442                 COPY_IN(tv);
1443                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1444                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1445                         vsk->connect_timeout = tv.tv_sec * HZ +
1446                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1447                         if (vsk->connect_timeout == 0)
1448                                 vsk->connect_timeout =
1449                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1450
1451                 } else {
1452                         err = -ERANGE;
1453                 }
1454                 break;
1455         }
1456
1457         default:
1458                 err = -ENOPROTOOPT;
1459                 break;
1460         }
1461
1462 #undef COPY_IN
1463
1464 exit:
1465         release_sock(sk);
1466         return err;
1467 }
1468
1469 static int vsock_stream_getsockopt(struct socket *sock,
1470                                    int level, int optname,
1471                                    char __user *optval,
1472                                    int __user *optlen)
1473 {
1474         int err;
1475         int len;
1476         struct sock *sk;
1477         struct vsock_sock *vsk;
1478         u64 val;
1479
1480         if (level != AF_VSOCK)
1481                 return -ENOPROTOOPT;
1482
1483         err = get_user(len, optlen);
1484         if (err != 0)
1485                 return err;
1486
1487 #define COPY_OUT(_v)                            \
1488         do {                                    \
1489                 if (len < sizeof(_v))           \
1490                         return -EINVAL;         \
1491                                                 \
1492                 len = sizeof(_v);               \
1493                 if (copy_to_user(optval, &_v, len) != 0)        \
1494                         return -EFAULT;                         \
1495                                                                 \
1496         } while (0)
1497
1498         err = 0;
1499         sk = sock->sk;
1500         vsk = vsock_sk(sk);
1501
1502         switch (optname) {
1503         case SO_VM_SOCKETS_BUFFER_SIZE:
1504                 val = transport->get_buffer_size(vsk);
1505                 COPY_OUT(val);
1506                 break;
1507
1508         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1509                 val = transport->get_max_buffer_size(vsk);
1510                 COPY_OUT(val);
1511                 break;
1512
1513         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1514                 val = transport->get_min_buffer_size(vsk);
1515                 COPY_OUT(val);
1516                 break;
1517
1518         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1519                 struct timeval tv;
1520                 tv.tv_sec = vsk->connect_timeout / HZ;
1521                 tv.tv_usec =
1522                     (vsk->connect_timeout -
1523                      tv.tv_sec * HZ) * (1000000 / HZ);
1524                 COPY_OUT(tv);
1525                 break;
1526         }
1527         default:
1528                 return -ENOPROTOOPT;
1529         }
1530
1531         err = put_user(len, optlen);
1532         if (err != 0)
1533                 return -EFAULT;
1534
1535 #undef COPY_OUT
1536
1537         return 0;
1538 }
1539
1540 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1541                                 size_t len)
1542 {
1543         struct sock *sk;
1544         struct vsock_sock *vsk;
1545         ssize_t total_written;
1546         long timeout;
1547         int err;
1548         struct vsock_transport_send_notify_data send_data;
1549         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1550
1551         sk = sock->sk;
1552         vsk = vsock_sk(sk);
1553         total_written = 0;
1554         err = 0;
1555
1556         if (msg->msg_flags & MSG_OOB)
1557                 return -EOPNOTSUPP;
1558
1559         lock_sock(sk);
1560
1561         /* Callers should not provide a destination with stream sockets. */
1562         if (msg->msg_namelen) {
1563                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1564                 goto out;
1565         }
1566
1567         /* Send data only if both sides are not shutdown in the direction. */
1568         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1569             vsk->peer_shutdown & RCV_SHUTDOWN) {
1570                 err = -EPIPE;
1571                 goto out;
1572         }
1573
1574         if (sk->sk_state != SS_CONNECTED ||
1575             !vsock_addr_bound(&vsk->local_addr)) {
1576                 err = -ENOTCONN;
1577                 goto out;
1578         }
1579
1580         if (!vsock_addr_bound(&vsk->remote_addr)) {
1581                 err = -EDESTADDRREQ;
1582                 goto out;
1583         }
1584
1585         /* Wait for room in the produce queue to enqueue our user's data. */
1586         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1587
1588         err = transport->notify_send_init(vsk, &send_data);
1589         if (err < 0)
1590                 goto out;
1591
1592         while (total_written < len) {
1593                 ssize_t written;
1594
1595                 add_wait_queue(sk_sleep(sk), &wait);
1596                 while (vsock_stream_has_space(vsk) == 0 &&
1597                        sk->sk_err == 0 &&
1598                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1599                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1600
1601                         /* Don't wait for non-blocking sockets. */
1602                         if (timeout == 0) {
1603                                 err = -EAGAIN;
1604                                 remove_wait_queue(sk_sleep(sk), &wait);
1605                                 goto out_err;
1606                         }
1607
1608                         err = transport->notify_send_pre_block(vsk, &send_data);
1609                         if (err < 0) {
1610                                 remove_wait_queue(sk_sleep(sk), &wait);
1611                                 goto out_err;
1612                         }
1613
1614                         release_sock(sk);
1615                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1616                         lock_sock(sk);
1617                         if (signal_pending(current)) {
1618                                 err = sock_intr_errno(timeout);
1619                                 remove_wait_queue(sk_sleep(sk), &wait);
1620                                 goto out_err;
1621                         } else if (timeout == 0) {
1622                                 err = -EAGAIN;
1623                                 remove_wait_queue(sk_sleep(sk), &wait);
1624                                 goto out_err;
1625                         }
1626                 }
1627                 remove_wait_queue(sk_sleep(sk), &wait);
1628
1629                 /* These checks occur both as part of and after the loop
1630                  * conditional since we need to check before and after
1631                  * sleeping.
1632                  */
1633                 if (sk->sk_err) {
1634                         err = -sk->sk_err;
1635                         goto out_err;
1636                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1637                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1638                         err = -EPIPE;
1639                         goto out_err;
1640                 }
1641
1642                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1643                 if (err < 0)
1644                         goto out_err;
1645
1646                 /* Note that enqueue will only write as many bytes as are free
1647                  * in the produce queue, so we don't need to ensure len is
1648                  * smaller than the queue size.  It is the caller's
1649                  * responsibility to check how many bytes we were able to send.
1650                  */
1651
1652                 written = transport->stream_enqueue(
1653                                 vsk, msg,
1654                                 len - total_written);
1655                 if (written < 0) {
1656                         err = -ENOMEM;
1657                         goto out_err;
1658                 }
1659
1660                 total_written += written;
1661
1662                 err = transport->notify_send_post_enqueue(
1663                                 vsk, written, &send_data);
1664                 if (err < 0)
1665                         goto out_err;
1666
1667         }
1668
1669 out_err:
1670         if (total_written > 0)
1671                 err = total_written;
1672 out:
1673         release_sock(sk);
1674         return err;
1675 }
1676
1677
1678 static int
1679 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1680                      int flags)
1681 {
1682         struct sock *sk;
1683         struct vsock_sock *vsk;
1684         int err;
1685         size_t target;
1686         ssize_t copied;
1687         long timeout;
1688         struct vsock_transport_recv_notify_data recv_data;
1689
1690         DEFINE_WAIT(wait);
1691
1692         sk = sock->sk;
1693         vsk = vsock_sk(sk);
1694         err = 0;
1695
1696         lock_sock(sk);
1697
1698         if (sk->sk_state != SS_CONNECTED) {
1699                 /* Recvmsg is supposed to return 0 if a peer performs an
1700                  * orderly shutdown. Differentiate between that case and when a
1701                  * peer has not connected or a local shutdown occured with the
1702                  * SOCK_DONE flag.
1703                  */
1704                 if (sock_flag(sk, SOCK_DONE))
1705                         err = 0;
1706                 else
1707                         err = -ENOTCONN;
1708
1709                 goto out;
1710         }
1711
1712         if (flags & MSG_OOB) {
1713                 err = -EOPNOTSUPP;
1714                 goto out;
1715         }
1716
1717         /* We don't check peer_shutdown flag here since peer may actually shut
1718          * down, but there can be data in the queue that a local socket can
1719          * receive.
1720          */
1721         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1722                 err = 0;
1723                 goto out;
1724         }
1725
1726         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1727          * is not an error.  We may as well bail out now.
1728          */
1729         if (!len) {
1730                 err = 0;
1731                 goto out;
1732         }
1733
1734         /* We must not copy less than target bytes into the user's buffer
1735          * before returning successfully, so we wait for the consume queue to
1736          * have that much data to consume before dequeueing.  Note that this
1737          * makes it impossible to handle cases where target is greater than the
1738          * queue size.
1739          */
1740         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1741         if (target >= transport->stream_rcvhiwat(vsk)) {
1742                 err = -ENOMEM;
1743                 goto out;
1744         }
1745         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1746         copied = 0;
1747
1748         err = transport->notify_recv_init(vsk, target, &recv_data);
1749         if (err < 0)
1750                 goto out;
1751
1752
1753         while (1) {
1754                 s64 ready;
1755
1756                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1757                 ready = vsock_stream_has_data(vsk);
1758
1759                 if (ready == 0) {
1760                         if (sk->sk_err != 0 ||
1761                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1762                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1763                                 finish_wait(sk_sleep(sk), &wait);
1764                                 break;
1765                         }
1766                         /* Don't wait for non-blocking sockets. */
1767                         if (timeout == 0) {
1768                                 err = -EAGAIN;
1769                                 finish_wait(sk_sleep(sk), &wait);
1770                                 break;
1771                         }
1772
1773                         err = transport->notify_recv_pre_block(
1774                                         vsk, target, &recv_data);
1775                         if (err < 0) {
1776                                 finish_wait(sk_sleep(sk), &wait);
1777                                 break;
1778                         }
1779                         release_sock(sk);
1780                         timeout = schedule_timeout(timeout);
1781                         lock_sock(sk);
1782
1783                         if (signal_pending(current)) {
1784                                 err = sock_intr_errno(timeout);
1785                                 finish_wait(sk_sleep(sk), &wait);
1786                                 break;
1787                         } else if (timeout == 0) {
1788                                 err = -EAGAIN;
1789                                 finish_wait(sk_sleep(sk), &wait);
1790                                 break;
1791                         }
1792                 } else {
1793                         ssize_t read;
1794
1795                         finish_wait(sk_sleep(sk), &wait);
1796
1797                         if (ready < 0) {
1798                                 /* Invalid queue pair content. XXX This should
1799                                 * be changed to a connection reset in a later
1800                                 * change.
1801                                 */
1802
1803                                 err = -ENOMEM;
1804                                 goto out;
1805                         }
1806
1807                         err = transport->notify_recv_pre_dequeue(
1808                                         vsk, target, &recv_data);
1809                         if (err < 0)
1810                                 break;
1811
1812                         read = transport->stream_dequeue(
1813                                         vsk, msg,
1814                                         len - copied, flags);
1815                         if (read < 0) {
1816                                 err = -ENOMEM;
1817                                 break;
1818                         }
1819
1820                         copied += read;
1821
1822                         err = transport->notify_recv_post_dequeue(
1823                                         vsk, target, read,
1824                                         !(flags & MSG_PEEK), &recv_data);
1825                         if (err < 0)
1826                                 goto out;
1827
1828                         if (read >= target || flags & MSG_PEEK)
1829                                 break;
1830
1831                         target -= read;
1832                 }
1833         }
1834
1835         if (sk->sk_err)
1836                 err = -sk->sk_err;
1837         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1838                 err = 0;
1839
1840         if (copied > 0)
1841                 err = copied;
1842
1843 out:
1844         release_sock(sk);
1845         return err;
1846 }
1847
1848 static const struct proto_ops vsock_stream_ops = {
1849         .family = PF_VSOCK,
1850         .owner = THIS_MODULE,
1851         .release = vsock_release,
1852         .bind = vsock_bind,
1853         .connect = vsock_stream_connect,
1854         .socketpair = sock_no_socketpair,
1855         .accept = vsock_accept,
1856         .getname = vsock_getname,
1857         .poll = vsock_poll,
1858         .ioctl = sock_no_ioctl,
1859         .listen = vsock_listen,
1860         .shutdown = vsock_shutdown,
1861         .setsockopt = vsock_stream_setsockopt,
1862         .getsockopt = vsock_stream_getsockopt,
1863         .sendmsg = vsock_stream_sendmsg,
1864         .recvmsg = vsock_stream_recvmsg,
1865         .mmap = sock_no_mmap,
1866         .sendpage = sock_no_sendpage,
1867 };
1868
1869 static int vsock_create(struct net *net, struct socket *sock,
1870                         int protocol, int kern)
1871 {
1872         if (!sock)
1873                 return -EINVAL;
1874
1875         if (protocol && protocol != PF_VSOCK)
1876                 return -EPROTONOSUPPORT;
1877
1878         switch (sock->type) {
1879         case SOCK_DGRAM:
1880                 sock->ops = &vsock_dgram_ops;
1881                 break;
1882         case SOCK_STREAM:
1883                 sock->ops = &vsock_stream_ops;
1884                 break;
1885         default:
1886                 return -ESOCKTNOSUPPORT;
1887         }
1888
1889         sock->state = SS_UNCONNECTED;
1890
1891         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1892 }
1893
1894 static const struct net_proto_family vsock_family_ops = {
1895         .family = AF_VSOCK,
1896         .create = vsock_create,
1897         .owner = THIS_MODULE,
1898 };
1899
1900 static long vsock_dev_do_ioctl(struct file *filp,
1901                                unsigned int cmd, void __user *ptr)
1902 {
1903         u32 __user *p = ptr;
1904         int retval = 0;
1905
1906         switch (cmd) {
1907         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1908                 if (put_user(transport->get_local_cid(), p) != 0)
1909                         retval = -EFAULT;
1910                 break;
1911
1912         default:
1913                 pr_err("Unknown ioctl %d\n", cmd);
1914                 retval = -EINVAL;
1915         }
1916
1917         return retval;
1918 }
1919
1920 static long vsock_dev_ioctl(struct file *filp,
1921                             unsigned int cmd, unsigned long arg)
1922 {
1923         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1924 }
1925
1926 #ifdef CONFIG_COMPAT
1927 static long vsock_dev_compat_ioctl(struct file *filp,
1928                                    unsigned int cmd, unsigned long arg)
1929 {
1930         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1931 }
1932 #endif
1933
1934 static const struct file_operations vsock_device_ops = {
1935         .owner          = THIS_MODULE,
1936         .unlocked_ioctl = vsock_dev_ioctl,
1937 #ifdef CONFIG_COMPAT
1938         .compat_ioctl   = vsock_dev_compat_ioctl,
1939 #endif
1940         .open           = nonseekable_open,
1941 };
1942
1943 static struct miscdevice vsock_device = {
1944         .name           = "vsock",
1945         .fops           = &vsock_device_ops,
1946 };
1947
1948 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1949 {
1950         int err = mutex_lock_interruptible(&vsock_register_mutex);
1951
1952         if (err)
1953                 return err;
1954
1955         if (transport) {
1956                 err = -EBUSY;
1957                 goto err_busy;
1958         }
1959
1960         /* Transport must be the owner of the protocol so that it can't
1961          * unload while there are open sockets.
1962          */
1963         vsock_proto.owner = owner;
1964         transport = t;
1965
1966         vsock_init_tables();
1967
1968         vsock_device.minor = MISC_DYNAMIC_MINOR;
1969         err = misc_register(&vsock_device);
1970         if (err) {
1971                 pr_err("Failed to register misc device\n");
1972                 goto err_reset_transport;
1973         }
1974
1975         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1976         if (err) {
1977                 pr_err("Cannot register vsock protocol\n");
1978                 goto err_deregister_misc;
1979         }
1980
1981         err = sock_register(&vsock_family_ops);
1982         if (err) {
1983                 pr_err("could not register af_vsock (%d) address family: %d\n",
1984                        AF_VSOCK, err);
1985                 goto err_unregister_proto;
1986         }
1987
1988         mutex_unlock(&vsock_register_mutex);
1989         return 0;
1990
1991 err_unregister_proto:
1992         proto_unregister(&vsock_proto);
1993 err_deregister_misc:
1994         misc_deregister(&vsock_device);
1995 err_reset_transport:
1996         transport = NULL;
1997 err_busy:
1998         mutex_unlock(&vsock_register_mutex);
1999         return err;
2000 }
2001 EXPORT_SYMBOL_GPL(__vsock_core_init);
2002
2003 void vsock_core_exit(void)
2004 {
2005         mutex_lock(&vsock_register_mutex);
2006
2007         misc_deregister(&vsock_device);
2008         sock_unregister(AF_VSOCK);
2009         proto_unregister(&vsock_proto);
2010
2011         /* We do not want the assignment below re-ordered. */
2012         mb();
2013         transport = NULL;
2014
2015         mutex_unlock(&vsock_register_mutex);
2016 }
2017 EXPORT_SYMBOL_GPL(vsock_core_exit);
2018
2019 const struct vsock_transport *vsock_core_get_transport(void)
2020 {
2021         /* vsock_register_mutex not taken since only the transport uses this
2022          * function and only while registered.
2023          */
2024         return transport;
2025 }
2026 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2027
2028 MODULE_AUTHOR("VMware, Inc.");
2029 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2030 MODULE_VERSION("1.0.2.0-k");
2031 MODULE_LICENSE("GPL v2");