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/skbuff.h>
 101 #include <linux/smp.h>
 102 #include <linux/socket.h>
 103 #include <linux/stddef.h>
 104 #include <linux/unistd.h>
 105 #include <linux/wait.h>
 106 #include <linux/workqueue.h>
 107 #include <net/sock.h>
 108 #include <net/af_vsock.h>
 109
 110 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 111 static void vsock_sk_destruct(struct sock *sk);
 112 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 113
 114 /* Protocol family. */
 115 static struct proto vsock_proto = {
 116         .name = "AF_VSOCK",
 117         .owner = THIS_MODULE,
 118         .obj_size = sizeof(struct vsock_sock),
 119 };
 120
 121 /* The default peer timeout indicates how long we will wait for a peer response
 122  * to a control message.
 123  */
 124 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 125
 126 static const struct vsock_transport *transport;
 127 static DEFINE_MUTEX(vsock_register_mutex);
 128
 129 /**** EXPORTS ****/
 130
 131 /* Get the ID of the local context.  This is transport dependent. */
 132
 133 int vm_sockets_get_local_cid(void)
 134 {
 135         return transport->get_local_cid();
 136 }
 137 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 138
 139 /**** UTILS ****/
 140
 141 /* Each bound VSocket is stored in the bind hash table and each connected
 142  * VSocket is stored in the connected hash table.
 143  *
 144  * Unbound sockets are all put on the same list attached to the end of the hash
 145  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 146  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 147  * represents the list that addr hashes to).
 148  *
 149  * Specifically, we initialize the vsock_bind_table array to a size of
 150  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 151  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 152  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 153  * mods with VSOCK_HASH_SIZE to ensure this.
 154  */
 155 #define VSOCK_HASH_SIZE         251
 156 #define MAX_PORT_RETRIES        24
 157
 158 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 159 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 160 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 161
 162 /* XXX This can probably be implemented in a better way. */
 163 #define VSOCK_CONN_HASH(src, dst)                               \
 164         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 165 #define vsock_connected_sockets(src, dst)               \
 166         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 167 #define vsock_connected_sockets_vsk(vsk)                                \
 168         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 169
 170 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 171 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 172 static DEFINE_SPINLOCK(vsock_table_lock);
 173
 174 /* Autobind this socket to the local address if necessary. */
 175 static int vsock_auto_bind(struct vsock_sock *vsk)
 176 {
 177         struct sock *sk = sk_vsock(vsk);
 178         struct sockaddr_vm local_addr;
 179
 180         if (vsock_addr_bound(&vsk->local_addr))
 181                 return 0;
 182         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 183         return __vsock_bind(sk, &local_addr);
 184 }
 185
 186 static void vsock_init_tables(void)
 187 {
 188         int i;
 189
 190         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 191                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 192
 193         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 194                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 195 }
 196
 197 static void __vsock_insert_bound(struct list_head *list,
 198                                  struct vsock_sock *vsk)
 199 {
 200         sock_hold(&vsk->sk);
 201         list_add(&vsk->bound_table, list);
 202 }
 203
 204 static void __vsock_insert_connected(struct list_head *list,
 205                                      struct vsock_sock *vsk)
 206 {
 207         sock_hold(&vsk->sk);
 208         list_add(&vsk->connected_table, list);
 209 }
 210
 211 static void __vsock_remove_bound(struct vsock_sock *vsk)
 212 {
 213         list_del_init(&vsk->bound_table);
 214         sock_put(&vsk->sk);
 215 }
 216
 217 static void __vsock_remove_connected(struct vsock_sock *vsk)
 218 {
 219         list_del_init(&vsk->connected_table);
 220         sock_put(&vsk->sk);
 221 }
 222
 223 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 224 {
 225         struct vsock_sock *vsk;
 226
 227         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 228                 if (addr->svm_port == vsk->local_addr.svm_port)
 229                         return sk_vsock(vsk);
 230
 231         return NULL;
 232 }
 233
 234 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 235                                                   struct sockaddr_vm *dst)
 236 {
 237         struct vsock_sock *vsk;
 238
 239         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 240                             connected_table) {
 241                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 242                     dst->svm_port == vsk->local_addr.svm_port) {
 243                         return sk_vsock(vsk);
 244                 }
 245         }
 246
 247         return NULL;
 248 }
 249
 250 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 251 {
 252         return !list_empty(&vsk->bound_table);
 253 }
 254
 255 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 256 {
 257         return !list_empty(&vsk->connected_table);
 258 }
 259
 260 static void vsock_insert_unbound(struct vsock_sock *vsk)
 261 {
 262         spin_lock_bh(&vsock_table_lock);
 263         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 264         spin_unlock_bh(&vsock_table_lock);
 265 }
 266
 267 void vsock_insert_connected(struct vsock_sock *vsk)
 268 {
 269         struct list_head *list = vsock_connected_sockets(
 270                 &vsk->remote_addr, &vsk->local_addr);
 271
 272         spin_lock_bh(&vsock_table_lock);
 273         __vsock_insert_connected(list, vsk);
 274         spin_unlock_bh(&vsock_table_lock);
 275 }
 276 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 277
 278 void vsock_remove_bound(struct vsock_sock *vsk)
 279 {
 280         spin_lock_bh(&vsock_table_lock);
 281         __vsock_remove_bound(vsk);
 282         spin_unlock_bh(&vsock_table_lock);
 283 }
 284 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 285
 286 void vsock_remove_connected(struct vsock_sock *vsk)
 287 {
 288         spin_lock_bh(&vsock_table_lock);
 289         __vsock_remove_connected(vsk);
 290         spin_unlock_bh(&vsock_table_lock);
 291 }
 292 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 293
 294 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 295 {
 296         struct sock *sk;
 297
 298         spin_lock_bh(&vsock_table_lock);
 299         sk = __vsock_find_bound_socket(addr);
 300         if (sk)
 301                 sock_hold(sk);
 302
 303         spin_unlock_bh(&vsock_table_lock);
 304
 305         return sk;
 306 }
 307 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 308
 309 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 310                                          struct sockaddr_vm *dst)
 311 {
 312         struct sock *sk;
 313
 314         spin_lock_bh(&vsock_table_lock);
 315         sk = __vsock_find_connected_socket(src, dst);
 316         if (sk)
 317                 sock_hold(sk);
 318
 319         spin_unlock_bh(&vsock_table_lock);
 320
 321         return sk;
 322 }
 323 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 324
 325 static bool vsock_in_bound_table(struct vsock_sock *vsk)
 326 {
 327         bool ret;
 328
 329         spin_lock_bh(&vsock_table_lock);
 330         ret = __vsock_in_bound_table(vsk);
 331         spin_unlock_bh(&vsock_table_lock);
 332
 333         return ret;
 334 }
 335
 336 static bool vsock_in_connected_table(struct vsock_sock *vsk)
 337 {
 338         bool ret;
 339
 340         spin_lock_bh(&vsock_table_lock);
 341         ret = __vsock_in_connected_table(vsk);
 342         spin_unlock_bh(&vsock_table_lock);
 343
 344         return ret;
 345 }
 346
 347 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 348 {
 349         int i;
 350
 351         spin_lock_bh(&vsock_table_lock);
 352
 353         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 354                 struct vsock_sock *vsk;
 355                 list_for_each_entry(vsk, &vsock_connected_table[i],
 356                                     connected_table)
 357                         fn(sk_vsock(vsk));
 358         }
 359
 360         spin_unlock_bh(&vsock_table_lock);
 361 }
 362 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 363
 364 void vsock_add_pending(struct sock *listener, struct sock *pending)
 365 {
 366         struct vsock_sock *vlistener;
 367         struct vsock_sock *vpending;
 368
 369         vlistener = vsock_sk(listener);
 370         vpending = vsock_sk(pending);
 371
 372         sock_hold(pending);
 373         sock_hold(listener);
 374         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 375 }
 376 EXPORT_SYMBOL_GPL(vsock_add_pending);
 377
 378 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 379 {
 380         struct vsock_sock *vpending = vsock_sk(pending);
 381
 382         list_del_init(&vpending->pending_links);
 383         sock_put(listener);
 384         sock_put(pending);
 385 }
 386 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 387
 388 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 389 {
 390         struct vsock_sock *vlistener;
 391         struct vsock_sock *vconnected;
 392
 393         vlistener = vsock_sk(listener);
 394         vconnected = vsock_sk(connected);
 395
 396         sock_hold(connected);
 397         sock_hold(listener);
 398         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 399 }
 400 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 401
 402 static struct sock *vsock_dequeue_accept(struct sock *listener)
 403 {
 404         struct vsock_sock *vlistener;
 405         struct vsock_sock *vconnected;
 406
 407         vlistener = vsock_sk(listener);
 408
 409         if (list_empty(&vlistener->accept_queue))
 410                 return NULL;
 411
 412         vconnected = list_entry(vlistener->accept_queue.next,
 413                                 struct vsock_sock, accept_queue);
 414
 415         list_del_init(&vconnected->accept_queue);
 416         sock_put(listener);
 417         /* The caller will need a reference on the connected socket so we let
 418          * it call sock_put().
 419          */
 420
 421         return sk_vsock(vconnected);
 422 }
 423
 424 static bool vsock_is_accept_queue_empty(struct sock *sk)
 425 {
 426         struct vsock_sock *vsk = vsock_sk(sk);
 427         return list_empty(&vsk->accept_queue);
 428 }
 429
 430 static bool vsock_is_pending(struct sock *sk)
 431 {
 432         struct vsock_sock *vsk = vsock_sk(sk);
 433         return !list_empty(&vsk->pending_links);
 434 }
 435
 436 static int vsock_send_shutdown(struct sock *sk, int mode)
 437 {
 438         return transport->shutdown(vsock_sk(sk), mode);
 439 }
 440
 441 void vsock_pending_work(struct work_struct *work)
 442 {
 443         struct sock *sk;
 444         struct sock *listener;
 445         struct vsock_sock *vsk;
 446         bool cleanup;
 447
 448         vsk = container_of(work, struct vsock_sock, dwork.work);
 449         sk = sk_vsock(vsk);
 450         listener = vsk->listener;
 451         cleanup = true;
 452
 453         lock_sock(listener);
 454         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 455
 456         if (vsock_is_pending(sk)) {
 457                 vsock_remove_pending(listener, sk);
 458         } else if (!vsk->rejected) {
 459                 /* We are not on the pending list and accept() did not reject
 460                  * us, so we must have been accepted by our user process.  We
 461                  * just need to drop our references to the sockets and be on
 462                  * our way.
 463                  */
 464                 cleanup = false;
 465                 goto out;
 466         }
 467
 468         listener->sk_ack_backlog--;
 469
 470         /* We need to remove ourself from the global connected sockets list so
 471          * incoming packets can't find this socket, and to reduce the reference
 472          * count.
 473          */
 474         if (vsock_in_connected_table(vsk))
 475                 vsock_remove_connected(vsk);
 476
 477         sk->sk_state = SS_FREE;
 478
 479 out:
 480         release_sock(sk);
 481         release_sock(listener);
 482         if (cleanup)
 483                 sock_put(sk);
 484
 485         sock_put(sk);
 486         sock_put(listener);
 487 }
 488 EXPORT_SYMBOL_GPL(vsock_pending_work);
 489
 490 /**** SOCKET OPERATIONS ****/
 491
 492 static int __vsock_bind_stream(struct vsock_sock *vsk,
 493                                struct sockaddr_vm *addr)
 494 {
 495         static u32 port = LAST_RESERVED_PORT + 1;
 496         struct sockaddr_vm new_addr;
 497
 498         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 499
 500         if (addr->svm_port == VMADDR_PORT_ANY) {
 501                 bool found = false;
 502                 unsigned int i;
 503
 504                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 505                         if (port <= LAST_RESERVED_PORT)
 506                                 port = LAST_RESERVED_PORT + 1;
 507
 508                         new_addr.svm_port = port++;
 509
 510                         if (!__vsock_find_bound_socket(&new_addr)) {
 511                                 found = true;
 512                                 break;
 513                         }
 514                 }
 515
 516                 if (!found)
 517                         return -EADDRNOTAVAIL;
 518         } else {
 519                 /* If port is in reserved range, ensure caller
 520                  * has necessary privileges.
 521                  */
 522                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 523                     !capable(CAP_NET_BIND_SERVICE)) {
 524                         return -EACCES;
 525                 }
 526
 527                 if (__vsock_find_bound_socket(&new_addr))
 528                         return -EADDRINUSE;
 529         }
 530
 531         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 532
 533         /* Remove stream sockets from the unbound list and add them to the hash
 534          * table for easy lookup by its address.  The unbound list is simply an
 535          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 536          */
 537         __vsock_remove_bound(vsk);
 538         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 539
 540         return 0;
 541 }
 542
 543 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 544                               struct sockaddr_vm *addr)
 545 {
 546         return transport->dgram_bind(vsk, addr);
 547 }
 548
 549 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 550 {
 551         struct vsock_sock *vsk = vsock_sk(sk);
 552         u32 cid;
 553         int retval;
 554
 555         /* First ensure this socket isn't already bound. */
 556         if (vsock_addr_bound(&vsk->local_addr))
 557                 return -EINVAL;
 558
 559         /* Now bind to the provided address or select appropriate values if
 560          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 561          * like AF_INET prevents binding to a non-local IP address (in most
 562          * cases), we only allow binding to the local CID.
 563          */
 564         cid = transport->get_local_cid();
 565         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 566                 return -EADDRNOTAVAIL;
 567
 568         switch (sk->sk_socket->type) {
 569         case SOCK_STREAM:
 570                 spin_lock_bh(&vsock_table_lock);
 571                 retval = __vsock_bind_stream(vsk, addr);
 572                 spin_unlock_bh(&vsock_table_lock);
 573                 break;
 574
 575         case SOCK_DGRAM:
 576                 retval = __vsock_bind_dgram(vsk, addr);
 577                 break;
 578
 579         default:
 580                 retval = -EINVAL;
 581                 break;
 582         }
 583
 584         return retval;
 585 }
 586
 587 struct sock *__vsock_create(struct net *net,
 588                             struct socket *sock,
 589                             struct sock *parent,
 590                             gfp_t priority,
 591                             unsigned short type,
 592                             int kern)
 593 {
 594         struct sock *sk;
 595         struct vsock_sock *psk;
 596         struct vsock_sock *vsk;
 597
 598         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 599         if (!sk)
 600                 return NULL;
 601
 602         sock_init_data(sock, sk);
 603
 604         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 605          * non-NULL. We make sure that our sockets always have a type by
 606          * setting it here if needed.
 607          */
 608         if (!sock)
 609                 sk->sk_type = type;
 610
 611         vsk = vsock_sk(sk);
 612         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 613         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 614
 615         sk->sk_destruct = vsock_sk_destruct;
 616         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 617         sk->sk_state = 0;
 618         sock_reset_flag(sk, SOCK_DONE);
 619
 620         INIT_LIST_HEAD(&vsk->bound_table);
 621         INIT_LIST_HEAD(&vsk->connected_table);
 622         vsk->listener = NULL;
 623         INIT_LIST_HEAD(&vsk->pending_links);
 624         INIT_LIST_HEAD(&vsk->accept_queue);
 625         vsk->rejected = false;
 626         vsk->sent_request = false;
 627         vsk->ignore_connecting_rst = false;
 628         vsk->peer_shutdown = 0;
 629
 630         psk = parent ? vsock_sk(parent) : NULL;
 631         if (parent) {
 632                 vsk->trusted = psk->trusted;
 633                 vsk->owner = get_cred(psk->owner);
 634                 vsk->connect_timeout = psk->connect_timeout;
 635         } else {
 636                 vsk->trusted = capable(CAP_NET_ADMIN);
 637                 vsk->owner = get_current_cred();
 638                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 639         }
 640
 641         if (transport->init(vsk, psk) < 0) {
 642                 sk_free(sk);
 643                 return NULL;
 644         }
 645
 646         if (sock)
 647                 vsock_insert_unbound(vsk);
 648
 649         return sk;
 650 }
 651 EXPORT_SYMBOL_GPL(__vsock_create);
 652
 653 static void __vsock_release(struct sock *sk)
 654 {
 655         if (sk) {
 656                 struct sk_buff *skb;
 657                 struct sock *pending;
 658                 struct vsock_sock *vsk;
 659
 660                 vsk = vsock_sk(sk);
 661                 pending = NULL; /* Compiler warning. */
 662
 663                 if (vsock_in_bound_table(vsk))
 664                         vsock_remove_bound(vsk);
 665
 666                 if (vsock_in_connected_table(vsk))
 667                         vsock_remove_connected(vsk);
 668
 669                 transport->release(vsk);
 670
 671                 lock_sock(sk);
 672                 sock_orphan(sk);
 673                 sk->sk_shutdown = SHUTDOWN_MASK;
 674
 675                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 676                         kfree_skb(skb);
 677
 678                 /* Clean up any sockets that never were accepted. */
 679                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 680                         __vsock_release(pending);
 681                         sock_put(pending);
 682                 }
 683
 684                 release_sock(sk);
 685                 sock_put(sk);
 686         }
 687 }
 688
 689 static void vsock_sk_destruct(struct sock *sk)
 690 {
 691         struct vsock_sock *vsk = vsock_sk(sk);
 692
 693         transport->destruct(vsk);
 694
 695         /* When clearing these addresses, there's no need to set the family and
 696          * possibly register the address family with the kernel.
 697          */
 698         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 699         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 700
 701         put_cred(vsk->owner);
 702 }
 703
 704 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 705 {
 706         int err;
 707
 708         err = sock_queue_rcv_skb(sk, skb);
 709         if (err)
 710                 kfree_skb(skb);
 711
 712         return err;
 713 }
 714
 715 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 716 {
 717         return transport->stream_has_data(vsk);
 718 }
 719 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 720
 721 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 722 {
 723         return transport->stream_has_space(vsk);
 724 }
 725 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 726
 727 static int vsock_release(struct socket *sock)
 728 {
 729         __vsock_release(sock->sk);
 730         sock->sk = NULL;
 731         sock->state = SS_FREE;
 732
 733         return 0;
 734 }
 735
 736 static int
 737 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 738 {
 739         int err;
 740         struct sock *sk;
 741         struct sockaddr_vm *vm_addr;
 742
 743         sk = sock->sk;
 744
 745         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 746                 return -EINVAL;
 747
 748         lock_sock(sk);
 749         err = __vsock_bind(sk, vm_addr);
 750         release_sock(sk);
 751
 752         return err;
 753 }
 754
 755 static int vsock_getname(struct socket *sock,
 756                          struct sockaddr *addr, int *addr_len, int peer)
 757 {
 758         int err;
 759         struct sock *sk;
 760         struct vsock_sock *vsk;
 761         struct sockaddr_vm *vm_addr;
 762
 763         sk = sock->sk;
 764         vsk = vsock_sk(sk);
 765         err = 0;
 766
 767         lock_sock(sk);
 768
 769         if (peer) {
 770                 if (sock->state != SS_CONNECTED) {
 771                         err = -ENOTCONN;
 772                         goto out;
 773                 }
 774                 vm_addr = &vsk->remote_addr;
 775         } else {
 776                 vm_addr = &vsk->local_addr;
 777         }
 778
 779         if (!vm_addr) {
 780                 err = -EINVAL;
 781                 goto out;
 782         }
 783
 784         /* sys_getsockname() and sys_getpeername() pass us a
 785          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 786          * that macro is defined in socket.c instead of .h, so we hardcode its
 787          * value here.
 788          */
 789         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 790         memcpy(addr, vm_addr, sizeof(*vm_addr));
 791         *addr_len = sizeof(*vm_addr);
 792
 793 out:
 794         release_sock(sk);
 795         return err;
 796 }
 797
 798 static int vsock_shutdown(struct socket *sock, int mode)
 799 {
 800         int err;
 801         struct sock *sk;
 802
 803         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 804          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 805          * here like the other address families do.  Note also that the
 806          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 807          * which is what we want.
 808          */
 809         mode++;
 810
 811         if ((mode & ~SHUTDOWN_MASK) || !mode)
 812                 return -EINVAL;
 813
 814         /* If this is a STREAM socket and it is not connected then bail out
 815          * immediately.  If it is a DGRAM socket then we must first kick the
 816          * socket so that it wakes up from any sleeping calls, for example
 817          * recv(), and then afterwards return the error.
 818          */
 819
 820         sk = sock->sk;
 821         if (sock->state == SS_UNCONNECTED) {
 822                 err = -ENOTCONN;
 823                 if (sk->sk_type == SOCK_STREAM)
 824                         return err;
 825         } else {
 826                 sock->state = SS_DISCONNECTING;
 827                 err = 0;
 828         }
 829
 830         /* Receive and send shutdowns are treated alike. */
 831         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 832         if (mode) {
 833                 lock_sock(sk);
 834                 sk->sk_shutdown |= mode;
 835                 sk->sk_state_change(sk);
 836                 release_sock(sk);
 837
 838                 if (sk->sk_type == SOCK_STREAM) {
 839                         sock_reset_flag(sk, SOCK_DONE);
 840                         vsock_send_shutdown(sk, mode);
 841                 }
 842         }
 843
 844         return err;
 845 }
 846
 847 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 848                                poll_table *wait)
 849 {
 850         struct sock *sk;
 851         unsigned int mask;
 852         struct vsock_sock *vsk;
 853
 854         sk = sock->sk;
 855         vsk = vsock_sk(sk);
 856
 857         poll_wait(file, sk_sleep(sk), wait);
 858         mask = 0;
 859
 860         if (sk->sk_err)
 861                 /* Signify that there has been an error on this socket. */
 862                 mask |= POLLERR;
 863
 864         /* INET sockets treat local write shutdown and peer write shutdown as a
 865          * case of POLLHUP set.
 866          */
 867         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 868             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 869              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 870                 mask |= POLLHUP;
 871         }
 872
 873         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 874             vsk->peer_shutdown & SEND_SHUTDOWN) {
 875                 mask |= POLLRDHUP;
 876         }
 877
 878         if (sock->type == SOCK_DGRAM) {
 879                 /* For datagram sockets we can read if there is something in
 880                  * the queue and write as long as the socket isn't shutdown for
 881                  * sending.
 882                  */
 883                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 884                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 885                         mask |= POLLIN | POLLRDNORM;
 886                 }
 887
 888                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 889                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 890
 891         } else if (sock->type == SOCK_STREAM) {
 892                 lock_sock(sk);
 893
 894                 /* Listening sockets that have connections in their accept
 895                  * queue can be read.
 896                  */
 897                 if (sk->sk_state == VSOCK_SS_LISTEN
 898                     && !vsock_is_accept_queue_empty(sk))
 899                         mask |= POLLIN | POLLRDNORM;
 900
 901                 /* If there is something in the queue then we can read. */
 902                 if (transport->stream_is_active(vsk) &&
 903                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 904                         bool data_ready_now = false;
 905                         int ret = transport->notify_poll_in(
 906                                         vsk, 1, &data_ready_now);
 907                         if (ret < 0) {
 908                                 mask |= POLLERR;
 909                         } else {
 910                                 if (data_ready_now)
 911                                         mask |= POLLIN | POLLRDNORM;
 912
 913                         }
 914                 }
 915
 916                 /* Sockets whose connections have been closed, reset, or
 917                  * terminated should also be considered read, and we check the
 918                  * shutdown flag for that.
 919                  */
 920                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 921                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 922                         mask |= POLLIN | POLLRDNORM;
 923                 }
 924
 925                 /* Connected sockets that can produce data can be written. */
 926                 if (sk->sk_state == SS_CONNECTED) {
 927                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 928                                 bool space_avail_now = false;
 929                                 int ret = transport->notify_poll_out(
 930                                                 vsk, 1, &space_avail_now);
 931                                 if (ret < 0) {
 932                                         mask |= POLLERR;
 933                                 } else {
 934                                         if (space_avail_now)
 935                                                 /* Remove POLLWRBAND since INET
 936                                                  * sockets are not setting it.
 937                                                  */
 938                                                 mask |= POLLOUT | POLLWRNORM;
 939
 940                                 }
 941                         }
 942                 }
 943
 944                 /* Simulate INET socket poll behaviors, which sets
 945                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 946                  * but local send is not shutdown.
 947                  */
 948                 if (sk->sk_state == SS_UNCONNECTED) {
 949                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 950                                 mask |= POLLOUT | POLLWRNORM;
 951
 952                 }
 953
 954                 release_sock(sk);
 955         }
 956
 957         return mask;
 958 }
 959
 960 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 961                                size_t len)
 962 {
 963         int err;
 964         struct sock *sk;
 965         struct vsock_sock *vsk;
 966         struct sockaddr_vm *remote_addr;
 967
 968         if (msg->msg_flags & MSG_OOB)
 969                 return -EOPNOTSUPP;
 970
 971         /* For now, MSG_DONTWAIT is always assumed... */
 972         err = 0;
 973         sk = sock->sk;
 974         vsk = vsock_sk(sk);
 975
 976         lock_sock(sk);
 977
 978         err = vsock_auto_bind(vsk);
 979         if (err)
 980                 goto out;
 981
 982
 983         /* If the provided message contains an address, use that.  Otherwise
 984          * fall back on the socket's remote handle (if it has been connected).
 985          */
 986         if (msg->msg_name &&
 987             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 988                             &remote_addr) == 0) {
 989                 /* Ensure this address is of the right type and is a valid
 990                  * destination.
 991                  */
 992
 993                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 994                         remote_addr->svm_cid = transport->get_local_cid();
 995
 996                 if (!vsock_addr_bound(remote_addr)) {
 997                         err = -EINVAL;
 998                         goto out;
 999                 }
1000         } else if (sock->state == SS_CONNECTED) {
1001                 remote_addr = &vsk->remote_addr;
1002
1003                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1004                         remote_addr->svm_cid = transport->get_local_cid();
1005
1006                 /* XXX Should connect() or this function ensure remote_addr is
1007                  * bound?
1008                  */
1009                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1010                         err = -EINVAL;
1011                         goto out;
1012                 }
1013         } else {
1014                 err = -EINVAL;
1015                 goto out;
1016         }
1017
1018         if (!transport->dgram_allow(remote_addr->svm_cid,
1019                                     remote_addr->svm_port)) {
1020                 err = -EINVAL;
1021                 goto out;
1022         }
1023
1024         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1025
1026 out:
1027         release_sock(sk);
1028         return err;
1029 }
1030
1031 static int vsock_dgram_connect(struct socket *sock,
1032                                struct sockaddr *addr, int addr_len, int flags)
1033 {
1034         int err;
1035         struct sock *sk;
1036         struct vsock_sock *vsk;
1037         struct sockaddr_vm *remote_addr;
1038
1039         sk = sock->sk;
1040         vsk = vsock_sk(sk);
1041
1042         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1043         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1044                 lock_sock(sk);
1045                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1046                                 VMADDR_PORT_ANY);
1047                 sock->state = SS_UNCONNECTED;
1048                 release_sock(sk);
1049                 return 0;
1050         } else if (err != 0)
1051                 return -EINVAL;
1052
1053         lock_sock(sk);
1054
1055         err = vsock_auto_bind(vsk);
1056         if (err)
1057                 goto out;
1058
1059         if (!transport->dgram_allow(remote_addr->svm_cid,
1060                                     remote_addr->svm_port)) {
1061                 err = -EINVAL;
1062                 goto out;
1063         }
1064
1065         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1066         sock->state = SS_CONNECTED;
1067
1068 out:
1069         release_sock(sk);
1070         return err;
1071 }
1072
1073 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1074                                size_t len, int flags)
1075 {
1076         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1077 }
1078
1079 static const struct proto_ops vsock_dgram_ops = {
1080         .family = PF_VSOCK,
1081         .owner = THIS_MODULE,
1082         .release = vsock_release,
1083         .bind = vsock_bind,
1084         .connect = vsock_dgram_connect,
1085         .socketpair = sock_no_socketpair,
1086         .accept = sock_no_accept,
1087         .getname = vsock_getname,
1088         .poll = vsock_poll,
1089         .ioctl = sock_no_ioctl,
1090         .listen = sock_no_listen,
1091         .shutdown = vsock_shutdown,
1092         .setsockopt = sock_no_setsockopt,
1093         .getsockopt = sock_no_getsockopt,
1094         .sendmsg = vsock_dgram_sendmsg,
1095         .recvmsg = vsock_dgram_recvmsg,
1096         .mmap = sock_no_mmap,
1097         .sendpage = sock_no_sendpage,
1098 };
1099
1100 static void vsock_connect_timeout(struct work_struct *work)
1101 {
1102         struct sock *sk;
1103         struct vsock_sock *vsk;
1104
1105         vsk = container_of(work, struct vsock_sock, dwork.work);
1106         sk = sk_vsock(vsk);
1107
1108         lock_sock(sk);
1109         if (sk->sk_state == SS_CONNECTING &&
1110             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1111                 sk->sk_state = SS_UNCONNECTED;
1112                 sk->sk_err = ETIMEDOUT;
1113                 sk->sk_error_report(sk);
1114         }
1115         release_sock(sk);
1116
1117         sock_put(sk);
1118 }
1119
1120 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1121                                 int addr_len, int flags)
1122 {
1123         int err;
1124         struct sock *sk;
1125         struct vsock_sock *vsk;
1126         struct sockaddr_vm *remote_addr;
1127         long timeout;
1128         DEFINE_WAIT(wait);
1129
1130         err = 0;
1131         sk = sock->sk;
1132         vsk = vsock_sk(sk);
1133
1134         lock_sock(sk);
1135
1136         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1137         switch (sock->state) {
1138         case SS_CONNECTED:
1139                 err = -EISCONN;
1140                 goto out;
1141         case SS_DISCONNECTING:
1142                 err = -EINVAL;
1143                 goto out;
1144         case SS_CONNECTING:
1145                 /* This continues on so we can move sock into the SS_CONNECTED
1146                  * state once the connection has completed (at which point err
1147                  * will be set to zero also).  Otherwise, we will either wait
1148                  * for the connection or return -EALREADY should this be a
1149                  * non-blocking call.
1150                  */
1151                 err = -EALREADY;
1152                 break;
1153         default:
1154                 if ((sk->sk_state == VSOCK_SS_LISTEN) ||
1155                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1156                         err = -EINVAL;
1157                         goto out;
1158                 }
1159
1160                 /* The hypervisor and well-known contexts do not have socket
1161                  * endpoints.
1162                  */
1163                 if (!transport->stream_allow(remote_addr->svm_cid,
1164                                              remote_addr->svm_port)) {
1165                         err = -ENETUNREACH;
1166                         goto out;
1167                 }
1168
1169                 /* Set the remote address that we are connecting to. */
1170                 memcpy(&vsk->remote_addr, remote_addr,
1171                        sizeof(vsk->remote_addr));
1172
1173                 err = vsock_auto_bind(vsk);
1174                 if (err)
1175                         goto out;
1176
1177                 sk->sk_state = SS_CONNECTING;
1178
1179                 err = transport->connect(vsk);
1180                 if (err < 0)
1181                         goto out;
1182
1183                 /* Mark sock as connecting and set the error code to in
1184                  * progress in case this is a non-blocking connect.
1185                  */
1186                 sock->state = SS_CONNECTING;
1187                 err = -EINPROGRESS;
1188         }
1189
1190         /* The receive path will handle all communication until we are able to
1191          * enter the connected state.  Here we wait for the connection to be
1192          * completed or a notification of an error.
1193          */
1194         timeout = vsk->connect_timeout;
1195         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1196
1197         while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1198                 if (flags & O_NONBLOCK) {
1199                         /* If we're not going to block, we schedule a timeout
1200                          * function to generate a timeout on the connection
1201                          * attempt, in case the peer doesn't respond in a
1202                          * timely manner. We hold on to the socket until the
1203                          * timeout fires.
1204                          */
1205                         sock_hold(sk);
1206                         INIT_DELAYED_WORK(&vsk->dwork,
1207                                           vsock_connect_timeout);
1208                         schedule_delayed_work(&vsk->dwork, timeout);
1209
1210                         /* Skip ahead to preserve error code set above. */
1211                         goto out_wait;
1212                 }
1213
1214                 release_sock(sk);
1215                 timeout = schedule_timeout(timeout);
1216                 lock_sock(sk);
1217
1218                 if (signal_pending(current)) {
1219                         err = sock_intr_errno(timeout);
1220                         sk->sk_state = SS_UNCONNECTED;
1221                         sock->state = SS_UNCONNECTED;
1222                         goto out_wait;
1223                 } else if (timeout == 0) {
1224                         err = -ETIMEDOUT;
1225                         sk->sk_state = SS_UNCONNECTED;
1226                         sock->state = SS_UNCONNECTED;
1227                         goto out_wait;
1228                 }
1229
1230                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1231         }
1232
1233         if (sk->sk_err) {
1234                 err = -sk->sk_err;
1235                 sk->sk_state = SS_UNCONNECTED;
1236                 sock->state = SS_UNCONNECTED;
1237         } else {
1238                 err = 0;
1239         }
1240
1241 out_wait:
1242         finish_wait(sk_sleep(sk), &wait);
1243 out:
1244         release_sock(sk);
1245         return err;
1246 }
1247
1248 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1249 {
1250         struct sock *listener;
1251         int err;
1252         struct sock *connected;
1253         struct vsock_sock *vconnected;
1254         long timeout;
1255         DEFINE_WAIT(wait);
1256
1257         err = 0;
1258         listener = sock->sk;
1259
1260         lock_sock(listener);
1261
1262         if (sock->type != SOCK_STREAM) {
1263                 err = -EOPNOTSUPP;
1264                 goto out;
1265         }
1266
1267         if (listener->sk_state != VSOCK_SS_LISTEN) {
1268                 err = -EINVAL;
1269                 goto out;
1270         }
1271
1272         /* Wait for children sockets to appear; these are the new sockets
1273          * created upon connection establishment.
1274          */
1275         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1276         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1277
1278         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1279                listener->sk_err == 0) {
1280                 release_sock(listener);
1281                 timeout = schedule_timeout(timeout);
1282                 finish_wait(sk_sleep(listener), &wait);
1283                 lock_sock(listener);
1284
1285                 if (signal_pending(current)) {
1286                         err = sock_intr_errno(timeout);
1287                         goto out;
1288                 } else if (timeout == 0) {
1289                         err = -EAGAIN;
1290                         goto out;
1291                 }
1292
1293                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1294         }
1295         finish_wait(sk_sleep(listener), &wait);
1296
1297         if (listener->sk_err)
1298                 err = -listener->sk_err;
1299
1300         if (connected) {
1301                 listener->sk_ack_backlog--;
1302
1303                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1304                 vconnected = vsock_sk(connected);
1305
1306                 /* If the listener socket has received an error, then we should
1307                  * reject this socket and return.  Note that we simply mark the
1308                  * socket rejected, drop our reference, and let the cleanup
1309                  * function handle the cleanup; the fact that we found it in
1310                  * the listener's accept queue guarantees that the cleanup
1311                  * function hasn't run yet.
1312                  */
1313                 if (err) {
1314                         vconnected->rejected = true;
1315                 } else {
1316                         newsock->state = SS_CONNECTED;
1317                         sock_graft(connected, newsock);
1318                 }
1319
1320                 release_sock(connected);
1321                 sock_put(connected);
1322         }
1323
1324 out:
1325         release_sock(listener);
1326         return err;
1327 }
1328
1329 static int vsock_listen(struct socket *sock, int backlog)
1330 {
1331         int err;
1332         struct sock *sk;
1333         struct vsock_sock *vsk;
1334
1335         sk = sock->sk;
1336
1337         lock_sock(sk);
1338
1339         if (sock->type != SOCK_STREAM) {
1340                 err = -EOPNOTSUPP;
1341                 goto out;
1342         }
1343
1344         if (sock->state != SS_UNCONNECTED) {
1345                 err = -EINVAL;
1346                 goto out;
1347         }
1348
1349         vsk = vsock_sk(sk);
1350
1351         if (!vsock_addr_bound(&vsk->local_addr)) {
1352                 err = -EINVAL;
1353                 goto out;
1354         }
1355
1356         sk->sk_max_ack_backlog = backlog;
1357         sk->sk_state = VSOCK_SS_LISTEN;
1358
1359         err = 0;
1360
1361 out:
1362         release_sock(sk);
1363         return err;
1364 }
1365
1366 static int vsock_stream_setsockopt(struct socket *sock,
1367                                    int level,
1368                                    int optname,
1369                                    char __user *optval,
1370                                    unsigned int optlen)
1371 {
1372         int err;
1373         struct sock *sk;
1374         struct vsock_sock *vsk;
1375         u64 val;
1376
1377         if (level != AF_VSOCK)
1378                 return -ENOPROTOOPT;
1379
1380 #define COPY_IN(_v)                                       \
1381         do {                                              \
1382                 if (optlen < sizeof(_v)) {                \
1383                         err = -EINVAL;                    \
1384                         goto exit;                        \
1385                 }                                         \
1386                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1387                         err = -EFAULT;                                  \
1388                         goto exit;                                      \
1389                 }                                                       \
1390         } while (0)
1391
1392         err = 0;
1393         sk = sock->sk;
1394         vsk = vsock_sk(sk);
1395
1396         lock_sock(sk);
1397
1398         switch (optname) {
1399         case SO_VM_SOCKETS_BUFFER_SIZE:
1400                 COPY_IN(val);
1401                 transport->set_buffer_size(vsk, val);
1402                 break;
1403
1404         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1405                 COPY_IN(val);
1406                 transport->set_max_buffer_size(vsk, val);
1407                 break;
1408
1409         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1410                 COPY_IN(val);
1411                 transport->set_min_buffer_size(vsk, val);
1412                 break;
1413
1414         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1415                 struct timeval tv;
1416                 COPY_IN(tv);
1417                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1418                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1419                         vsk->connect_timeout = tv.tv_sec * HZ +
1420                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1421                         if (vsk->connect_timeout == 0)
1422                                 vsk->connect_timeout =
1423                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1424
1425                 } else {
1426                         err = -ERANGE;
1427                 }
1428                 break;
1429         }
1430
1431         default:
1432                 err = -ENOPROTOOPT;
1433                 break;
1434         }
1435
1436 #undef COPY_IN
1437
1438 exit:
1439         release_sock(sk);
1440         return err;
1441 }
1442
1443 static int vsock_stream_getsockopt(struct socket *sock,
1444                                    int level, int optname,
1445                                    char __user *optval,
1446                                    int __user *optlen)
1447 {
1448         int err;
1449         int len;
1450         struct sock *sk;
1451         struct vsock_sock *vsk;
1452         u64 val;
1453
1454         if (level != AF_VSOCK)
1455                 return -ENOPROTOOPT;
1456
1457         err = get_user(len, optlen);
1458         if (err != 0)
1459                 return err;
1460
1461 #define COPY_OUT(_v)                            \
1462         do {                                    \
1463                 if (len < sizeof(_v))           \
1464                         return -EINVAL;         \
1465                                                 \
1466                 len = sizeof(_v);               \
1467                 if (copy_to_user(optval, &_v, len) != 0)        \
1468                         return -EFAULT;                         \
1469                                                                 \
1470         } while (0)
1471
1472         err = 0;
1473         sk = sock->sk;
1474         vsk = vsock_sk(sk);
1475
1476         switch (optname) {
1477         case SO_VM_SOCKETS_BUFFER_SIZE:
1478                 val = transport->get_buffer_size(vsk);
1479                 COPY_OUT(val);
1480                 break;
1481
1482         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1483                 val = transport->get_max_buffer_size(vsk);
1484                 COPY_OUT(val);
1485                 break;
1486
1487         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1488                 val = transport->get_min_buffer_size(vsk);
1489                 COPY_OUT(val);
1490                 break;
1491
1492         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1493                 struct timeval tv;
1494                 tv.tv_sec = vsk->connect_timeout / HZ;
1495                 tv.tv_usec =
1496                     (vsk->connect_timeout -
1497                      tv.tv_sec * HZ) * (1000000 / HZ);
1498                 COPY_OUT(tv);
1499                 break;
1500         }
1501         default:
1502                 return -ENOPROTOOPT;
1503         }
1504
1505         err = put_user(len, optlen);
1506         if (err != 0)
1507                 return -EFAULT;
1508
1509 #undef COPY_OUT
1510
1511         return 0;
1512 }
1513
1514 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1515                                 size_t len)
1516 {
1517         struct sock *sk;
1518         struct vsock_sock *vsk;
1519         ssize_t total_written;
1520         long timeout;
1521         int err;
1522         struct vsock_transport_send_notify_data send_data;
1523
1524         DEFINE_WAIT(wait);
1525
1526         sk = sock->sk;
1527         vsk = vsock_sk(sk);
1528         total_written = 0;
1529         err = 0;
1530
1531         if (msg->msg_flags & MSG_OOB)
1532                 return -EOPNOTSUPP;
1533
1534         lock_sock(sk);
1535
1536         /* Callers should not provide a destination with stream sockets. */
1537         if (msg->msg_namelen) {
1538                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1539                 goto out;
1540         }
1541
1542         /* Send data only if both sides are not shutdown in the direction. */
1543         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1544             vsk->peer_shutdown & RCV_SHUTDOWN) {
1545                 err = -EPIPE;
1546                 goto out;
1547         }
1548
1549         if (sk->sk_state != SS_CONNECTED ||
1550             !vsock_addr_bound(&vsk->local_addr)) {
1551                 err = -ENOTCONN;
1552                 goto out;
1553         }
1554
1555         if (!vsock_addr_bound(&vsk->remote_addr)) {
1556                 err = -EDESTADDRREQ;
1557                 goto out;
1558         }
1559
1560         /* Wait for room in the produce queue to enqueue our user's data. */
1561         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1562
1563         err = transport->notify_send_init(vsk, &send_data);
1564         if (err < 0)
1565                 goto out;
1566
1567
1568         while (total_written < len) {
1569                 ssize_t written;
1570
1571                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1572                 while (vsock_stream_has_space(vsk) == 0 &&
1573                        sk->sk_err == 0 &&
1574                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1575                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1576
1577                         /* Don't wait for non-blocking sockets. */
1578                         if (timeout == 0) {
1579                                 err = -EAGAIN;
1580                                 finish_wait(sk_sleep(sk), &wait);
1581                                 goto out_err;
1582                         }
1583
1584                         err = transport->notify_send_pre_block(vsk, &send_data);
1585                         if (err < 0) {
1586                                 finish_wait(sk_sleep(sk), &wait);
1587                                 goto out_err;
1588                         }
1589
1590                         release_sock(sk);
1591                         timeout = schedule_timeout(timeout);
1592                         lock_sock(sk);
1593                         if (signal_pending(current)) {
1594                                 err = sock_intr_errno(timeout);
1595                                 finish_wait(sk_sleep(sk), &wait);
1596                                 goto out_err;
1597                         } else if (timeout == 0) {
1598                                 err = -EAGAIN;
1599                                 finish_wait(sk_sleep(sk), &wait);
1600                                 goto out_err;
1601                         }
1602
1603                         prepare_to_wait(sk_sleep(sk), &wait,
1604                                         TASK_INTERRUPTIBLE);
1605                 }
1606                 finish_wait(sk_sleep(sk), &wait);
1607
1608                 /* These checks occur both as part of and after the loop
1609                  * conditional since we need to check before and after
1610                  * sleeping.
1611                  */
1612                 if (sk->sk_err) {
1613                         err = -sk->sk_err;
1614                         goto out_err;
1615                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1616                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1617                         err = -EPIPE;
1618                         goto out_err;
1619                 }
1620
1621                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1622                 if (err < 0)
1623                         goto out_err;
1624
1625                 /* Note that enqueue will only write as many bytes as are free
1626                  * in the produce queue, so we don't need to ensure len is
1627                  * smaller than the queue size.  It is the caller's
1628                  * responsibility to check how many bytes we were able to send.
1629                  */
1630
1631                 written = transport->stream_enqueue(
1632                                 vsk, msg,
1633                                 len - total_written);
1634                 if (written < 0) {
1635                         err = -ENOMEM;
1636                         goto out_err;
1637                 }
1638
1639                 total_written += written;
1640
1641                 err = transport->notify_send_post_enqueue(
1642                                 vsk, written, &send_data);
1643                 if (err < 0)
1644                         goto out_err;
1645
1646         }
1647
1648 out_err:
1649         if (total_written > 0)
1650                 err = total_written;
1651 out:
1652         release_sock(sk);
1653         return err;
1654 }
1655
1656
1657 static int
1658 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1659                      int flags)
1660 {
1661         struct sock *sk;
1662         struct vsock_sock *vsk;
1663         int err;
1664         size_t target;
1665         ssize_t copied;
1666         long timeout;
1667         struct vsock_transport_recv_notify_data recv_data;
1668
1669         DEFINE_WAIT(wait);
1670
1671         sk = sock->sk;
1672         vsk = vsock_sk(sk);
1673         err = 0;
1674
1675         lock_sock(sk);
1676
1677         if (sk->sk_state != SS_CONNECTED) {
1678                 /* Recvmsg is supposed to return 0 if a peer performs an
1679                  * orderly shutdown. Differentiate between that case and when a
1680                  * peer has not connected or a local shutdown occured with the
1681                  * SOCK_DONE flag.
1682                  */
1683                 if (sock_flag(sk, SOCK_DONE))
1684                         err = 0;
1685                 else
1686                         err = -ENOTCONN;
1687
1688                 goto out;
1689         }
1690
1691         if (flags & MSG_OOB) {
1692                 err = -EOPNOTSUPP;
1693                 goto out;
1694         }
1695
1696         /* We don't check peer_shutdown flag here since peer may actually shut
1697          * down, but there can be data in the queue that a local socket can
1698          * receive.
1699          */
1700         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1701                 err = 0;
1702                 goto out;
1703         }
1704
1705         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1706          * is not an error.  We may as well bail out now.
1707          */
1708         if (!len) {
1709                 err = 0;
1710                 goto out;
1711         }
1712
1713         /* We must not copy less than target bytes into the user's buffer
1714          * before returning successfully, so we wait for the consume queue to
1715          * have that much data to consume before dequeueing.  Note that this
1716          * makes it impossible to handle cases where target is greater than the
1717          * queue size.
1718          */
1719         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1720         if (target >= transport->stream_rcvhiwat(vsk)) {
1721                 err = -ENOMEM;
1722                 goto out;
1723         }
1724         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1725         copied = 0;
1726
1727         err = transport->notify_recv_init(vsk, target, &recv_data);
1728         if (err < 0)
1729                 goto out;
1730
1731
1732         while (1) {
1733                 s64 ready;
1734
1735                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1736                 ready = vsock_stream_has_data(vsk);
1737
1738                 if (ready == 0) {
1739                         if (sk->sk_err != 0 ||
1740                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1741                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1742                                 finish_wait(sk_sleep(sk), &wait);
1743                                 break;
1744                         }
1745                         /* Don't wait for non-blocking sockets. */
1746                         if (timeout == 0) {
1747                                 err = -EAGAIN;
1748                                 finish_wait(sk_sleep(sk), &wait);
1749                                 break;
1750                         }
1751
1752                         err = transport->notify_recv_pre_block(
1753                                         vsk, target, &recv_data);
1754                         if (err < 0) {
1755                                 finish_wait(sk_sleep(sk), &wait);
1756                                 break;
1757                         }
1758                         release_sock(sk);
1759                         timeout = schedule_timeout(timeout);
1760                         lock_sock(sk);
1761
1762                         if (signal_pending(current)) {
1763                                 err = sock_intr_errno(timeout);
1764                                 finish_wait(sk_sleep(sk), &wait);
1765                                 break;
1766                         } else if (timeout == 0) {
1767                                 err = -EAGAIN;
1768                                 finish_wait(sk_sleep(sk), &wait);
1769                                 break;
1770                         }
1771                 } else {
1772                         ssize_t read;
1773
1774                         finish_wait(sk_sleep(sk), &wait);
1775
1776                         if (ready < 0) {
1777                                 /* Invalid queue pair content. XXX This should
1778                                 * be changed to a connection reset in a later
1779                                 * change.
1780                                 */
1781
1782                                 err = -ENOMEM;
1783                                 goto out;
1784                         }
1785
1786                         err = transport->notify_recv_pre_dequeue(
1787                                         vsk, target, &recv_data);
1788                         if (err < 0)
1789                                 break;
1790
1791                         read = transport->stream_dequeue(
1792                                         vsk, msg,
1793                                         len - copied, flags);
1794                         if (read < 0) {
1795                                 err = -ENOMEM;
1796                                 break;
1797                         }
1798
1799                         copied += read;
1800
1801                         err = transport->notify_recv_post_dequeue(
1802                                         vsk, target, read,
1803                                         !(flags & MSG_PEEK), &recv_data);
1804                         if (err < 0)
1805                                 goto out;
1806
1807                         if (read >= target || flags & MSG_PEEK)
1808                                 break;
1809
1810                         target -= read;
1811                 }
1812         }
1813
1814         if (sk->sk_err)
1815                 err = -sk->sk_err;
1816         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1817                 err = 0;
1818
1819         if (copied > 0)
1820                 err = copied;
1821
1822 out:
1823         release_sock(sk);
1824         return err;
1825 }
1826
1827 static const struct proto_ops vsock_stream_ops = {
1828         .family = PF_VSOCK,
1829         .owner = THIS_MODULE,
1830         .release = vsock_release,
1831         .bind = vsock_bind,
1832         .connect = vsock_stream_connect,
1833         .socketpair = sock_no_socketpair,
1834         .accept = vsock_accept,
1835         .getname = vsock_getname,
1836         .poll = vsock_poll,
1837         .ioctl = sock_no_ioctl,
1838         .listen = vsock_listen,
1839         .shutdown = vsock_shutdown,
1840         .setsockopt = vsock_stream_setsockopt,
1841         .getsockopt = vsock_stream_getsockopt,
1842         .sendmsg = vsock_stream_sendmsg,
1843         .recvmsg = vsock_stream_recvmsg,
1844         .mmap = sock_no_mmap,
1845         .sendpage = sock_no_sendpage,
1846 };
1847
1848 static int vsock_create(struct net *net, struct socket *sock,
1849                         int protocol, int kern)
1850 {
1851         if (!sock)
1852                 return -EINVAL;
1853
1854         if (protocol && protocol != PF_VSOCK)
1855                 return -EPROTONOSUPPORT;
1856
1857         switch (sock->type) {
1858         case SOCK_DGRAM:
1859                 sock->ops = &vsock_dgram_ops;
1860                 break;
1861         case SOCK_STREAM:
1862                 sock->ops = &vsock_stream_ops;
1863                 break;
1864         default:
1865                 return -ESOCKTNOSUPPORT;
1866         }
1867
1868         sock->state = SS_UNCONNECTED;
1869
1870         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1871 }
1872
1873 static const struct net_proto_family vsock_family_ops = {
1874         .family = AF_VSOCK,
1875         .create = vsock_create,
1876         .owner = THIS_MODULE,
1877 };
1878
1879 static long vsock_dev_do_ioctl(struct file *filp,
1880                                unsigned int cmd, void __user *ptr)
1881 {
1882         u32 __user *p = ptr;
1883         int retval = 0;
1884
1885         switch (cmd) {
1886         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1887                 if (put_user(transport->get_local_cid(), p) != 0)
1888                         retval = -EFAULT;
1889                 break;
1890
1891         default:
1892                 pr_err("Unknown ioctl %d\n", cmd);
1893                 retval = -EINVAL;
1894         }
1895
1896         return retval;
1897 }
1898
1899 static long vsock_dev_ioctl(struct file *filp,
1900                             unsigned int cmd, unsigned long arg)
1901 {
1902         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1903 }
1904
1905 #ifdef CONFIG_COMPAT
1906 static long vsock_dev_compat_ioctl(struct file *filp,
1907                                    unsigned int cmd, unsigned long arg)
1908 {
1909         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1910 }
1911 #endif
1912
1913 static const struct file_operations vsock_device_ops = {
1914         .owner          = THIS_MODULE,
1915         .unlocked_ioctl = vsock_dev_ioctl,
1916 #ifdef CONFIG_COMPAT
1917         .compat_ioctl   = vsock_dev_compat_ioctl,
1918 #endif
1919         .open           = nonseekable_open,
1920 };
1921
1922 static struct miscdevice vsock_device = {
1923         .name           = "vsock",
1924         .fops           = &vsock_device_ops,
1925 };
1926
1927 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1928 {
1929         int err = mutex_lock_interruptible(&vsock_register_mutex);
1930
1931         if (err)
1932                 return err;
1933
1934         if (transport) {
1935                 err = -EBUSY;
1936                 goto err_busy;
1937         }
1938
1939         /* Transport must be the owner of the protocol so that it can't
1940          * unload while there are open sockets.
1941          */
1942         vsock_proto.owner = owner;
1943         transport = t;
1944
1945         vsock_init_tables();
1946
1947         vsock_device.minor = MISC_DYNAMIC_MINOR;
1948         err = misc_register(&vsock_device);
1949         if (err) {
1950                 pr_err("Failed to register misc device\n");
1951                 goto err_reset_transport;
1952         }
1953
1954         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1955         if (err) {
1956                 pr_err("Cannot register vsock protocol\n");
1957                 goto err_deregister_misc;
1958         }
1959
1960         err = sock_register(&vsock_family_ops);
1961         if (err) {
1962                 pr_err("could not register af_vsock (%d) address family: %d\n",
1963                        AF_VSOCK, err);
1964                 goto err_unregister_proto;
1965         }
1966
1967         mutex_unlock(&vsock_register_mutex);
1968         return 0;
1969
1970 err_unregister_proto:
1971         proto_unregister(&vsock_proto);
1972 err_deregister_misc:
1973         misc_deregister(&vsock_device);
1974 err_reset_transport:
1975         transport = NULL;
1976 err_busy:
1977         mutex_unlock(&vsock_register_mutex);
1978         return err;
1979 }
1980 EXPORT_SYMBOL_GPL(__vsock_core_init);
1981
1982 void vsock_core_exit(void)
1983 {
1984         mutex_lock(&vsock_register_mutex);
1985
1986         misc_deregister(&vsock_device);
1987         sock_unregister(AF_VSOCK);
1988         proto_unregister(&vsock_proto);
1989
1990         /* We do not want the assignment below re-ordered. */
1991         mb();
1992         transport = NULL;
1993
1994         mutex_unlock(&vsock_register_mutex);
1995 }
1996 EXPORT_SYMBOL_GPL(vsock_core_exit);
1997
1998 MODULE_AUTHOR("VMware, Inc.");
1999 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2000 MODULE_VERSION("1.0.1.0-k");
2001 MODULE_LICENSE("GPL v2");