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