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 SS_LISTEN state.  When a connection
  40  * request is received (the second kind of socket mentioned above), we create a
  41  * new socket and refer to it as a pending socket.  These pending sockets are
  42  * placed on the pending connection list of the listener socket.  When future
  43  * packets are received for the address the listener socket is bound to, we
  44  * check if the source of the packet is from one that has an existing pending
  45  * connection.  If it does, we process the packet for the pending socket.  When
  46  * that socket reaches the connected state, it is removed from the listener
  47  * socket's pending list and enqueued in the listener socket's accept queue.
  48  * Callers of accept(2) will accept connected sockets from the listener socket's
  49  * accept queue.  If the socket cannot be accepted for some reason then it is
  50  * marked rejected.  Once the connection is accepted, it is owned by the user
  51  * process and the responsibility for cleanup falls with that user process.
  52  *
  53  * - It is possible that these pending sockets will never reach the connected
  54  * state; in fact, we may never receive another packet after the connection
  55  * request.  Because of this, we must schedule a cleanup function to run in the
  56  * future, after some amount of time passes where a connection should have been
  57  * established.  This function ensures that the socket is off all lists so it
  58  * cannot be retrieved, then drops all references to the socket so it is cleaned
  59  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  60  * function will also cleanup rejected sockets, those that reach the connected
  61  * state but leave it before they have been accepted.
  62  *
  63  * - Sockets created by user action will be cleaned up when the user process
  64  * calls close(2), causing our release implementation to be called. Our release
  65  * implementation will perform some cleanup then drop the last reference so our
  66  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  67  * perform additional cleanup that's common for both types of sockets.
  68  *
  69  * - A socket's reference count is what ensures that the structure won't be
  70  * freed.  Each entry in a list (such as the "global" bound and connected tables
  71  * and the listener socket's pending list and connected queue) ensures a
  72  * reference.  When we defer work until process context and pass a socket as our
  73  * argument, we must ensure the reference count is increased to ensure the
  74  * socket isn't freed before the function is run; the deferred function will
  75  * then drop the reference.
  76  */
  77
  78 #include <linux/types.h>
  79 #include <linux/bitops.h>
  80 #include <linux/cred.h>
  81 #include <linux/init.h>
  82 #include <linux/io.h>
  83 #include <linux/kernel.h>
  84 #include <linux/kmod.h>
  85 #include <linux/list.h>
  86 #include <linux/miscdevice.h>
  87 #include <linux/module.h>
  88 #include <linux/mutex.h>
  89 #include <linux/net.h>
  90 #include <linux/poll.h>
  91 #include <linux/skbuff.h>
  92 #include <linux/smp.h>
  93 #include <linux/socket.h>
  94 #include <linux/stddef.h>
  95 #include <linux/unistd.h>
  96 #include <linux/wait.h>
  97 #include <linux/workqueue.h>
  98 #include <net/sock.h>
  99 #include <net/af_vsock.h>
 100
 101 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 102 static void vsock_sk_destruct(struct sock *sk);
 103 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 104
 105 /* Protocol family. */
 106 static struct proto vsock_proto = {
 107         .name = "AF_VSOCK",
 108         .owner = THIS_MODULE,
 109         .obj_size = sizeof(struct vsock_sock),
 110 };
 111
 112 /* The default peer timeout indicates how long we will wait for a peer response
 113  * to a control message.
 114  */
 115 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 116
 117 #define SS_LISTEN 255
 118
 119 static const struct vsock_transport *transport;
 120 static DEFINE_MUTEX(vsock_register_mutex);
 121
 122 /**** EXPORTS ****/
 123
 124 /* Get the ID of the local context.  This is transport dependent. */
 125
 126 int vm_sockets_get_local_cid(void)
 127 {
 128         return transport->get_local_cid();
 129 }
 130 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 131
 132 /**** UTILS ****/
 133
 134 /* Each bound VSocket is stored in the bind hash table and each connected
 135  * VSocket is stored in the connected hash table.
 136  *
 137  * Unbound sockets are all put on the same list attached to the end of the hash
 138  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 139  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 140  * represents the list that addr hashes to).
 141  *
 142  * Specifically, we initialize the vsock_bind_table array to a size of
 143  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 144  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 145  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 146  * mods with VSOCK_HASH_SIZE to ensure this.
 147  */
 148 #define VSOCK_HASH_SIZE         251
 149 #define MAX_PORT_RETRIES        24
 150
 151 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 152 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 153 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 154
 155 /* XXX This can probably be implemented in a better way. */
 156 #define VSOCK_CONN_HASH(src, dst)                               \
 157         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 158 #define vsock_connected_sockets(src, dst)               \
 159         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 160 #define vsock_connected_sockets_vsk(vsk)                                \
 161         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 162
 163 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 164 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 165 static DEFINE_SPINLOCK(vsock_table_lock);
 166
 167 /* Autobind this socket to the local address if necessary. */
 168 static int vsock_auto_bind(struct vsock_sock *vsk)
 169 {
 170         struct sock *sk = sk_vsock(vsk);
 171         struct sockaddr_vm local_addr;
 172
 173         if (vsock_addr_bound(&vsk->local_addr))
 174                 return 0;
 175         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 176         return __vsock_bind(sk, &local_addr);
 177 }
 178
 179 static void vsock_init_tables(void)
 180 {
 181         int i;
 182
 183         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 184                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 185
 186         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 187                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 188 }
 189
 190 static void __vsock_insert_bound(struct list_head *list,
 191                                  struct vsock_sock *vsk)
 192 {
 193         sock_hold(&vsk->sk);
 194         list_add(&vsk->bound_table, list);
 195 }
 196
 197 static void __vsock_insert_connected(struct list_head *list,
 198                                      struct vsock_sock *vsk)
 199 {
 200         sock_hold(&vsk->sk);
 201         list_add(&vsk->connected_table, list);
 202 }
 203
 204 static void __vsock_remove_bound(struct vsock_sock *vsk)
 205 {
 206         list_del_init(&vsk->bound_table);
 207         sock_put(&vsk->sk);
 208 }
 209
 210 static void __vsock_remove_connected(struct vsock_sock *vsk)
 211 {
 212         list_del_init(&vsk->connected_table);
 213         sock_put(&vsk->sk);
 214 }
 215
 216 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 217 {
 218         struct vsock_sock *vsk;
 219
 220         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 221                 if (addr->svm_port == vsk->local_addr.svm_port)
 222                         return sk_vsock(vsk);
 223
 224         return NULL;
 225 }
 226
 227 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 228                                                   struct sockaddr_vm *dst)
 229 {
 230         struct vsock_sock *vsk;
 231
 232         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 233                             connected_table) {
 234                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 235                     dst->svm_port == vsk->local_addr.svm_port) {
 236                         return sk_vsock(vsk);
 237                 }
 238         }
 239
 240         return NULL;
 241 }
 242
 243 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 244 {
 245         return !list_empty(&vsk->bound_table);
 246 }
 247
 248 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 249 {
 250         return !list_empty(&vsk->connected_table);
 251 }
 252
 253 static void vsock_insert_unbound(struct vsock_sock *vsk)
 254 {
 255         spin_lock_bh(&vsock_table_lock);
 256         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 257         spin_unlock_bh(&vsock_table_lock);
 258 }
 259
 260 void vsock_insert_connected(struct vsock_sock *vsk)
 261 {
 262         struct list_head *list = vsock_connected_sockets(
 263                 &vsk->remote_addr, &vsk->local_addr);
 264
 265         spin_lock_bh(&vsock_table_lock);
 266         __vsock_insert_connected(list, vsk);
 267         spin_unlock_bh(&vsock_table_lock);
 268 }
 269 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 270
 271 void vsock_remove_bound(struct vsock_sock *vsk)
 272 {
 273         spin_lock_bh(&vsock_table_lock);
 274         __vsock_remove_bound(vsk);
 275         spin_unlock_bh(&vsock_table_lock);
 276 }
 277 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 278
 279 void vsock_remove_connected(struct vsock_sock *vsk)
 280 {
 281         spin_lock_bh(&vsock_table_lock);
 282         __vsock_remove_connected(vsk);
 283         spin_unlock_bh(&vsock_table_lock);
 284 }
 285 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 286
 287 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 288 {
 289         struct sock *sk;
 290
 291         spin_lock_bh(&vsock_table_lock);
 292         sk = __vsock_find_bound_socket(addr);
 293         if (sk)
 294                 sock_hold(sk);
 295
 296         spin_unlock_bh(&vsock_table_lock);
 297
 298         return sk;
 299 }
 300 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 301
 302 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 303                                          struct sockaddr_vm *dst)
 304 {
 305         struct sock *sk;
 306
 307         spin_lock_bh(&vsock_table_lock);
 308         sk = __vsock_find_connected_socket(src, dst);
 309         if (sk)
 310                 sock_hold(sk);
 311
 312         spin_unlock_bh(&vsock_table_lock);
 313
 314         return sk;
 315 }
 316 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 317
 318 static bool vsock_in_bound_table(struct vsock_sock *vsk)
 319 {
 320         bool ret;
 321
 322         spin_lock_bh(&vsock_table_lock);
 323         ret = __vsock_in_bound_table(vsk);
 324         spin_unlock_bh(&vsock_table_lock);
 325
 326         return ret;
 327 }
 328
 329 static bool vsock_in_connected_table(struct vsock_sock *vsk)
 330 {
 331         bool ret;
 332
 333         spin_lock_bh(&vsock_table_lock);
 334         ret = __vsock_in_connected_table(vsk);
 335         spin_unlock_bh(&vsock_table_lock);
 336
 337         return ret;
 338 }
 339
 340 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 341 {
 342         int i;
 343
 344         spin_lock_bh(&vsock_table_lock);
 345
 346         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 347                 struct vsock_sock *vsk;
 348                 list_for_each_entry(vsk, &vsock_connected_table[i],
 349                                     connected_table)
 350                         fn(sk_vsock(vsk));
 351         }
 352
 353         spin_unlock_bh(&vsock_table_lock);
 354 }
 355 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 356
 357 void vsock_add_pending(struct sock *listener, struct sock *pending)
 358 {
 359         struct vsock_sock *vlistener;
 360         struct vsock_sock *vpending;
 361
 362         vlistener = vsock_sk(listener);
 363         vpending = vsock_sk(pending);
 364
 365         sock_hold(pending);
 366         sock_hold(listener);
 367         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 368 }
 369 EXPORT_SYMBOL_GPL(vsock_add_pending);
 370
 371 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 372 {
 373         struct vsock_sock *vpending = vsock_sk(pending);
 374
 375         list_del_init(&vpending->pending_links);
 376         sock_put(listener);
 377         sock_put(pending);
 378 }
 379 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 380
 381 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 382 {
 383         struct vsock_sock *vlistener;
 384         struct vsock_sock *vconnected;
 385
 386         vlistener = vsock_sk(listener);
 387         vconnected = vsock_sk(connected);
 388
 389         sock_hold(connected);
 390         sock_hold(listener);
 391         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 392 }
 393 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 394
 395 static struct sock *vsock_dequeue_accept(struct sock *listener)
 396 {
 397         struct vsock_sock *vlistener;
 398         struct vsock_sock *vconnected;
 399
 400         vlistener = vsock_sk(listener);
 401
 402         if (list_empty(&vlistener->accept_queue))
 403                 return NULL;
 404
 405         vconnected = list_entry(vlistener->accept_queue.next,
 406                                 struct vsock_sock, accept_queue);
 407
 408         list_del_init(&vconnected->accept_queue);
 409         sock_put(listener);
 410         /* The caller will need a reference on the connected socket so we let
 411          * it call sock_put().
 412          */
 413
 414         return sk_vsock(vconnected);
 415 }
 416
 417 static bool vsock_is_accept_queue_empty(struct sock *sk)
 418 {
 419         struct vsock_sock *vsk = vsock_sk(sk);
 420         return list_empty(&vsk->accept_queue);
 421 }
 422
 423 static bool vsock_is_pending(struct sock *sk)
 424 {
 425         struct vsock_sock *vsk = vsock_sk(sk);
 426         return !list_empty(&vsk->pending_links);
 427 }
 428
 429 static int vsock_send_shutdown(struct sock *sk, int mode)
 430 {
 431         return transport->shutdown(vsock_sk(sk), mode);
 432 }
 433
 434 void vsock_pending_work(struct work_struct *work)
 435 {
 436         struct sock *sk;
 437         struct sock *listener;
 438         struct vsock_sock *vsk;
 439         bool cleanup;
 440
 441         vsk = container_of(work, struct vsock_sock, dwork.work);
 442         sk = sk_vsock(vsk);
 443         listener = vsk->listener;
 444         cleanup = true;
 445
 446         lock_sock(listener);
 447         lock_sock(sk);
 448
 449         if (vsock_is_pending(sk)) {
 450                 vsock_remove_pending(listener, sk);
 451         } else if (!vsk->rejected) {
 452                 /* We are not on the pending list and accept() did not reject
 453                  * us, so we must have been accepted by our user process.  We
 454                  * just need to drop our references to the sockets and be on
 455                  * our way.
 456                  */
 457                 cleanup = false;
 458                 goto out;
 459         }
 460
 461         listener->sk_ack_backlog--;
 462
 463         /* We need to remove ourself from the global connected sockets list so
 464          * incoming packets can't find this socket, and to reduce the reference
 465          * count.
 466          */
 467         if (vsock_in_connected_table(vsk))
 468                 vsock_remove_connected(vsk);
 469
 470         sk->sk_state = SS_FREE;
 471
 472 out:
 473         release_sock(sk);
 474         release_sock(listener);
 475         if (cleanup)
 476                 sock_put(sk);
 477
 478         sock_put(sk);
 479         sock_put(listener);
 480 }
 481 EXPORT_SYMBOL_GPL(vsock_pending_work);
 482
 483 /**** SOCKET OPERATIONS ****/
 484
 485 static int __vsock_bind_stream(struct vsock_sock *vsk,
 486                                struct sockaddr_vm *addr)
 487 {
 488         static u32 port = LAST_RESERVED_PORT + 1;
 489         struct sockaddr_vm new_addr;
 490
 491         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 492
 493         if (addr->svm_port == VMADDR_PORT_ANY) {
 494                 bool found = false;
 495                 unsigned int i;
 496
 497                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 498                         if (port <= LAST_RESERVED_PORT)
 499                                 port = LAST_RESERVED_PORT + 1;
 500
 501                         new_addr.svm_port = port++;
 502
 503                         if (!__vsock_find_bound_socket(&new_addr)) {
 504                                 found = true;
 505                                 break;
 506                         }
 507                 }
 508
 509                 if (!found)
 510                         return -EADDRNOTAVAIL;
 511         } else {
 512                 /* If port is in reserved range, ensure caller
 513                  * has necessary privileges.
 514                  */
 515                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 516                     !capable(CAP_NET_BIND_SERVICE)) {
 517                         return -EACCES;
 518                 }
 519
 520                 if (__vsock_find_bound_socket(&new_addr))
 521                         return -EADDRINUSE;
 522         }
 523
 524         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 525
 526         /* Remove stream sockets from the unbound list and add them to the hash
 527          * table for easy lookup by its address.  The unbound list is simply an
 528          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 529          */
 530         __vsock_remove_bound(vsk);
 531         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 532
 533         return 0;
 534 }
 535
 536 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 537                               struct sockaddr_vm *addr)
 538 {
 539         return transport->dgram_bind(vsk, addr);
 540 }
 541
 542 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 543 {
 544         struct vsock_sock *vsk = vsock_sk(sk);
 545         u32 cid;
 546         int retval;
 547
 548         /* First ensure this socket isn't already bound. */
 549         if (vsock_addr_bound(&vsk->local_addr))
 550                 return -EINVAL;
 551
 552         /* Now bind to the provided address or select appropriate values if
 553          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 554          * like AF_INET prevents binding to a non-local IP address (in most
 555          * cases), we only allow binding to the local CID.
 556          */
 557         cid = transport->get_local_cid();
 558         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 559                 return -EADDRNOTAVAIL;
 560
 561         switch (sk->sk_socket->type) {
 562         case SOCK_STREAM:
 563                 spin_lock_bh(&vsock_table_lock);
 564                 retval = __vsock_bind_stream(vsk, addr);
 565                 spin_unlock_bh(&vsock_table_lock);
 566                 break;
 567
 568         case SOCK_DGRAM:
 569                 retval = __vsock_bind_dgram(vsk, addr);
 570                 break;
 571
 572         default:
 573                 retval = -EINVAL;
 574                 break;
 575         }
 576
 577         return retval;
 578 }
 579
 580 struct sock *__vsock_create(struct net *net,
 581                             struct socket *sock,
 582                             struct sock *parent,
 583                             gfp_t priority,
 584                             unsigned short type)
 585 {
 586         struct sock *sk;
 587         struct vsock_sock *psk;
 588         struct vsock_sock *vsk;
 589
 590         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto);
 591         if (!sk)
 592                 return NULL;
 593
 594         sock_init_data(sock, sk);
 595
 596         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 597          * non-NULL. We make sure that our sockets always have a type by
 598          * setting it here if needed.
 599          */
 600         if (!sock)
 601                 sk->sk_type = type;
 602
 603         vsk = vsock_sk(sk);
 604         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 605         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 606
 607         sk->sk_destruct = vsock_sk_destruct;
 608         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 609         sk->sk_state = 0;
 610         sock_reset_flag(sk, SOCK_DONE);
 611
 612         INIT_LIST_HEAD(&vsk->bound_table);
 613         INIT_LIST_HEAD(&vsk->connected_table);
 614         vsk->listener = NULL;
 615         INIT_LIST_HEAD(&vsk->pending_links);
 616         INIT_LIST_HEAD(&vsk->accept_queue);
 617         vsk->rejected = false;
 618         vsk->sent_request = false;
 619         vsk->ignore_connecting_rst = false;
 620         vsk->peer_shutdown = 0;
 621
 622         psk = parent ? vsock_sk(parent) : NULL;
 623         if (parent) {
 624                 vsk->trusted = psk->trusted;
 625                 vsk->owner = get_cred(psk->owner);
 626                 vsk->connect_timeout = psk->connect_timeout;
 627         } else {
 628                 vsk->trusted = capable(CAP_NET_ADMIN);
 629                 vsk->owner = get_current_cred();
 630                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 631         }
 632
 633         if (transport->init(vsk, psk) < 0) {
 634                 sk_free(sk);
 635                 return NULL;
 636         }
 637
 638         if (sock)
 639                 vsock_insert_unbound(vsk);
 640
 641         return sk;
 642 }
 643 EXPORT_SYMBOL_GPL(__vsock_create);
 644
 645 static void __vsock_release(struct sock *sk)
 646 {
 647         if (sk) {
 648                 struct sk_buff *skb;
 649                 struct sock *pending;
 650                 struct vsock_sock *vsk;
 651
 652                 vsk = vsock_sk(sk);
 653                 pending = NULL; /* Compiler warning. */
 654
 655                 if (vsock_in_bound_table(vsk))
 656                         vsock_remove_bound(vsk);
 657
 658                 if (vsock_in_connected_table(vsk))
 659                         vsock_remove_connected(vsk);
 660
 661                 transport->release(vsk);
 662
 663                 lock_sock(sk);
 664                 sock_orphan(sk);
 665                 sk->sk_shutdown = SHUTDOWN_MASK;
 666
 667                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 668                         kfree_skb(skb);
 669
 670                 /* Clean up any sockets that never were accepted. */
 671                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 672                         __vsock_release(pending);
 673                         sock_put(pending);
 674                 }
 675
 676                 release_sock(sk);
 677                 sock_put(sk);
 678         }
 679 }
 680
 681 static void vsock_sk_destruct(struct sock *sk)
 682 {
 683         struct vsock_sock *vsk = vsock_sk(sk);
 684
 685         transport->destruct(vsk);
 686
 687         /* When clearing these addresses, there's no need to set the family and
 688          * possibly register the address family with the kernel.
 689          */
 690         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 691         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 692
 693         put_cred(vsk->owner);
 694 }
 695
 696 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 697 {
 698         int err;
 699
 700         err = sock_queue_rcv_skb(sk, skb);
 701         if (err)
 702                 kfree_skb(skb);
 703
 704         return err;
 705 }
 706
 707 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 708 {
 709         return transport->stream_has_data(vsk);
 710 }
 711 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 712
 713 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 714 {
 715         return transport->stream_has_space(vsk);
 716 }
 717 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 718
 719 static int vsock_release(struct socket *sock)
 720 {
 721         __vsock_release(sock->sk);
 722         sock->sk = NULL;
 723         sock->state = SS_FREE;
 724
 725         return 0;
 726 }
 727
 728 static int
 729 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 730 {
 731         int err;
 732         struct sock *sk;
 733         struct sockaddr_vm *vm_addr;
 734
 735         sk = sock->sk;
 736
 737         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 738                 return -EINVAL;
 739
 740         lock_sock(sk);
 741         err = __vsock_bind(sk, vm_addr);
 742         release_sock(sk);
 743
 744         return err;
 745 }
 746
 747 static int vsock_getname(struct socket *sock,
 748                          struct sockaddr *addr, int *addr_len, int peer)
 749 {
 750         int err;
 751         struct sock *sk;
 752         struct vsock_sock *vsk;
 753         struct sockaddr_vm *vm_addr;
 754
 755         sk = sock->sk;
 756         vsk = vsock_sk(sk);
 757         err = 0;
 758
 759         lock_sock(sk);
 760
 761         if (peer) {
 762                 if (sock->state != SS_CONNECTED) {
 763                         err = -ENOTCONN;
 764                         goto out;
 765                 }
 766                 vm_addr = &vsk->remote_addr;
 767         } else {
 768                 vm_addr = &vsk->local_addr;
 769         }
 770
 771         if (!vm_addr) {
 772                 err = -EINVAL;
 773                 goto out;
 774         }
 775
 776         /* sys_getsockname() and sys_getpeername() pass us a
 777          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 778          * that macro is defined in socket.c instead of .h, so we hardcode its
 779          * value here.
 780          */
 781         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 782         memcpy(addr, vm_addr, sizeof(*vm_addr));
 783         *addr_len = sizeof(*vm_addr);
 784
 785 out:
 786         release_sock(sk);
 787         return err;
 788 }
 789
 790 static int vsock_shutdown(struct socket *sock, int mode)
 791 {
 792         int err;
 793         struct sock *sk;
 794
 795         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 796          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 797          * here like the other address families do.  Note also that the
 798          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 799          * which is what we want.
 800          */
 801         mode++;
 802
 803         if ((mode & ~SHUTDOWN_MASK) || !mode)
 804                 return -EINVAL;
 805
 806         /* If this is a STREAM socket and it is not connected then bail out
 807          * immediately.  If it is a DGRAM socket then we must first kick the
 808          * socket so that it wakes up from any sleeping calls, for example
 809          * recv(), and then afterwards return the error.
 810          */
 811
 812         sk = sock->sk;
 813         if (sock->state == SS_UNCONNECTED) {
 814                 err = -ENOTCONN;
 815                 if (sk->sk_type == SOCK_STREAM)
 816                         return err;
 817         } else {
 818                 sock->state = SS_DISCONNECTING;
 819                 err = 0;
 820         }
 821
 822         /* Receive and send shutdowns are treated alike. */
 823         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 824         if (mode) {
 825                 lock_sock(sk);
 826                 sk->sk_shutdown |= mode;
 827                 sk->sk_state_change(sk);
 828                 release_sock(sk);
 829
 830                 if (sk->sk_type == SOCK_STREAM) {
 831                         sock_reset_flag(sk, SOCK_DONE);
 832                         vsock_send_shutdown(sk, mode);
 833                 }
 834         }
 835
 836         return err;
 837 }
 838
 839 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 840                                poll_table *wait)
 841 {
 842         struct sock *sk;
 843         unsigned int mask;
 844         struct vsock_sock *vsk;
 845
 846         sk = sock->sk;
 847         vsk = vsock_sk(sk);
 848
 849         poll_wait(file, sk_sleep(sk), wait);
 850         mask = 0;
 851
 852         if (sk->sk_err)
 853                 /* Signify that there has been an error on this socket. */
 854                 mask |= POLLERR;
 855
 856         /* INET sockets treat local write shutdown and peer write shutdown as a
 857          * case of POLLHUP set.
 858          */
 859         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 860             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 861              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 862                 mask |= POLLHUP;
 863         }
 864
 865         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 866             vsk->peer_shutdown & SEND_SHUTDOWN) {
 867                 mask |= POLLRDHUP;
 868         }
 869
 870         if (sock->type == SOCK_DGRAM) {
 871                 /* For datagram sockets we can read if there is something in
 872                  * the queue and write as long as the socket isn't shutdown for
 873                  * sending.
 874                  */
 875                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 876                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 877                         mask |= POLLIN | POLLRDNORM;
 878                 }
 879
 880                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 881                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 882
 883         } else if (sock->type == SOCK_STREAM) {
 884                 lock_sock(sk);
 885
 886                 /* Listening sockets that have connections in their accept
 887                  * queue can be read.
 888                  */
 889                 if (sk->sk_state == SS_LISTEN
 890                     && !vsock_is_accept_queue_empty(sk))
 891                         mask |= POLLIN | POLLRDNORM;
 892
 893                 /* If there is something in the queue then we can read. */
 894                 if (transport->stream_is_active(vsk) &&
 895                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 896                         bool data_ready_now = false;
 897                         int ret = transport->notify_poll_in(
 898                                         vsk, 1, &data_ready_now);
 899                         if (ret < 0) {
 900                                 mask |= POLLERR;
 901                         } else {
 902                                 if (data_ready_now)
 903                                         mask |= POLLIN | POLLRDNORM;
 904
 905                         }
 906                 }
 907
 908                 /* Sockets whose connections have been closed, reset, or
 909                  * terminated should also be considered read, and we check the
 910                  * shutdown flag for that.
 911                  */
 912                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 913                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 914                         mask |= POLLIN | POLLRDNORM;
 915                 }
 916
 917                 /* Connected sockets that can produce data can be written. */
 918                 if (sk->sk_state == SS_CONNECTED) {
 919                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 920                                 bool space_avail_now = false;
 921                                 int ret = transport->notify_poll_out(
 922                                                 vsk, 1, &space_avail_now);
 923                                 if (ret < 0) {
 924                                         mask |= POLLERR;
 925                                 } else {
 926                                         if (space_avail_now)
 927                                                 /* Remove POLLWRBAND since INET
 928                                                  * sockets are not setting it.
 929                                                  */
 930                                                 mask |= POLLOUT | POLLWRNORM;
 931
 932                                 }
 933                         }
 934                 }
 935
 936                 /* Simulate INET socket poll behaviors, which sets
 937                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 938                  * but local send is not shutdown.
 939                  */
 940                 if (sk->sk_state == SS_UNCONNECTED) {
 941                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 942                                 mask |= POLLOUT | POLLWRNORM;
 943
 944                 }
 945
 946                 release_sock(sk);
 947         }
 948
 949         return mask;
 950 }
 951
 952 static int vsock_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
 953                                struct msghdr *msg, size_t len)
 954 {
 955         int err;
 956         struct sock *sk;
 957         struct vsock_sock *vsk;
 958         struct sockaddr_vm *remote_addr;
 959
 960         if (msg->msg_flags & MSG_OOB)
 961                 return -EOPNOTSUPP;
 962
 963         /* For now, MSG_DONTWAIT is always assumed... */
 964         err = 0;
 965         sk = sock->sk;
 966         vsk = vsock_sk(sk);
 967
 968         lock_sock(sk);
 969
 970         err = vsock_auto_bind(vsk);
 971         if (err)
 972                 goto out;
 973
 974
 975         /* If the provided message contains an address, use that.  Otherwise
 976          * fall back on the socket's remote handle (if it has been connected).
 977          */
 978         if (msg->msg_name &&
 979             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 980                             &remote_addr) == 0) {
 981                 /* Ensure this address is of the right type and is a valid
 982                  * destination.
 983                  */
 984
 985                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 986                         remote_addr->svm_cid = transport->get_local_cid();
 987
 988                 if (!vsock_addr_bound(remote_addr)) {
 989                         err = -EINVAL;
 990                         goto out;
 991                 }
 992         } else if (sock->state == SS_CONNECTED) {
 993                 remote_addr = &vsk->remote_addr;
 994
 995                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 996                         remote_addr->svm_cid = transport->get_local_cid();
 997
 998                 /* XXX Should connect() or this function ensure remote_addr is
 999                  * bound?
1000                  */
1001                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1002                         err = -EINVAL;
1003                         goto out;
1004                 }
1005         } else {
1006                 err = -EINVAL;
1007                 goto out;
1008         }
1009
1010         if (!transport->dgram_allow(remote_addr->svm_cid,
1011                                     remote_addr->svm_port)) {
1012                 err = -EINVAL;
1013                 goto out;
1014         }
1015
1016         err = transport->dgram_enqueue(vsk, remote_addr, msg->msg_iov, len);
1017
1018 out:
1019         release_sock(sk);
1020         return err;
1021 }
1022
1023 static int vsock_dgram_connect(struct socket *sock,
1024                                struct sockaddr *addr, int addr_len, int flags)
1025 {
1026         int err;
1027         struct sock *sk;
1028         struct vsock_sock *vsk;
1029         struct sockaddr_vm *remote_addr;
1030
1031         sk = sock->sk;
1032         vsk = vsock_sk(sk);
1033
1034         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1035         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1036                 lock_sock(sk);
1037                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1038                                 VMADDR_PORT_ANY);
1039                 sock->state = SS_UNCONNECTED;
1040                 release_sock(sk);
1041                 return 0;
1042         } else if (err != 0)
1043                 return -EINVAL;
1044
1045         lock_sock(sk);
1046
1047         err = vsock_auto_bind(vsk);
1048         if (err)
1049                 goto out;
1050
1051         if (!transport->dgram_allow(remote_addr->svm_cid,
1052                                     remote_addr->svm_port)) {
1053                 err = -EINVAL;
1054                 goto out;
1055         }
1056
1057         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1058         sock->state = SS_CONNECTED;
1059
1060 out:
1061         release_sock(sk);
1062         return err;
1063 }
1064
1065 static int vsock_dgram_recvmsg(struct kiocb *kiocb, struct socket *sock,
1066                                struct msghdr *msg, size_t len, int flags)
1067 {
1068         return transport->dgram_dequeue(kiocb, vsock_sk(sock->sk), msg, len,
1069                                         flags);
1070 }
1071
1072 static const struct proto_ops vsock_dgram_ops = {
1073         .family = PF_VSOCK,
1074         .owner = THIS_MODULE,
1075         .release = vsock_release,
1076         .bind = vsock_bind,
1077         .connect = vsock_dgram_connect,
1078         .socketpair = sock_no_socketpair,
1079         .accept = sock_no_accept,
1080         .getname = vsock_getname,
1081         .poll = vsock_poll,
1082         .ioctl = sock_no_ioctl,
1083         .listen = sock_no_listen,
1084         .shutdown = vsock_shutdown,
1085         .setsockopt = sock_no_setsockopt,
1086         .getsockopt = sock_no_getsockopt,
1087         .sendmsg = vsock_dgram_sendmsg,
1088         .recvmsg = vsock_dgram_recvmsg,
1089         .mmap = sock_no_mmap,
1090         .sendpage = sock_no_sendpage,
1091 };
1092
1093 static void vsock_connect_timeout(struct work_struct *work)
1094 {
1095         struct sock *sk;
1096         struct vsock_sock *vsk;
1097
1098         vsk = container_of(work, struct vsock_sock, dwork.work);
1099         sk = sk_vsock(vsk);
1100
1101         lock_sock(sk);
1102         if (sk->sk_state == SS_CONNECTING &&
1103             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1104                 sk->sk_state = SS_UNCONNECTED;
1105                 sk->sk_err = ETIMEDOUT;
1106                 sk->sk_error_report(sk);
1107         }
1108         release_sock(sk);
1109
1110         sock_put(sk);
1111 }
1112
1113 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1114                                 int addr_len, int flags)
1115 {
1116         int err;
1117         struct sock *sk;
1118         struct vsock_sock *vsk;
1119         struct sockaddr_vm *remote_addr;
1120         long timeout;
1121         DEFINE_WAIT(wait);
1122
1123         err = 0;
1124         sk = sock->sk;
1125         vsk = vsock_sk(sk);
1126
1127         lock_sock(sk);
1128
1129         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1130         switch (sock->state) {
1131         case SS_CONNECTED:
1132                 err = -EISCONN;
1133                 goto out;
1134         case SS_DISCONNECTING:
1135                 err = -EINVAL;
1136                 goto out;
1137         case SS_CONNECTING:
1138                 /* This continues on so we can move sock into the SS_CONNECTED
1139                  * state once the connection has completed (at which point err
1140                  * will be set to zero also).  Otherwise, we will either wait
1141                  * for the connection or return -EALREADY should this be a
1142                  * non-blocking call.
1143                  */
1144                 err = -EALREADY;
1145                 break;
1146         default:
1147                 if ((sk->sk_state == SS_LISTEN) ||
1148                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1149                         err = -EINVAL;
1150                         goto out;
1151                 }
1152
1153                 /* The hypervisor and well-known contexts do not have socket
1154                  * endpoints.
1155                  */
1156                 if (!transport->stream_allow(remote_addr->svm_cid,
1157                                              remote_addr->svm_port)) {
1158                         err = -ENETUNREACH;
1159                         goto out;
1160                 }
1161
1162                 /* Set the remote address that we are connecting to. */
1163                 memcpy(&vsk->remote_addr, remote_addr,
1164                        sizeof(vsk->remote_addr));
1165
1166                 err = vsock_auto_bind(vsk);
1167                 if (err)
1168                         goto out;
1169
1170                 sk->sk_state = SS_CONNECTING;
1171
1172                 err = transport->connect(vsk);
1173                 if (err < 0)
1174                         goto out;
1175
1176                 /* Mark sock as connecting and set the error code to in
1177                  * progress in case this is a non-blocking connect.
1178                  */
1179                 sock->state = SS_CONNECTING;
1180                 err = -EINPROGRESS;
1181         }
1182
1183         /* The receive path will handle all communication until we are able to
1184          * enter the connected state.  Here we wait for the connection to be
1185          * completed or a notification of an error.
1186          */
1187         timeout = vsk->connect_timeout;
1188         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1189
1190         while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) {
1191                 if (flags & O_NONBLOCK) {
1192                         /* If we're not going to block, we schedule a timeout
1193                          * function to generate a timeout on the connection
1194                          * attempt, in case the peer doesn't respond in a
1195                          * timely manner. We hold on to the socket until the
1196                          * timeout fires.
1197                          */
1198                         sock_hold(sk);
1199                         INIT_DELAYED_WORK(&vsk->dwork,
1200                                           vsock_connect_timeout);
1201                         schedule_delayed_work(&vsk->dwork, timeout);
1202
1203                         /* Skip ahead to preserve error code set above. */
1204                         goto out_wait;
1205                 }
1206
1207                 release_sock(sk);
1208                 timeout = schedule_timeout(timeout);
1209                 lock_sock(sk);
1210
1211                 if (signal_pending(current)) {
1212                         err = sock_intr_errno(timeout);
1213                         goto out_wait_error;
1214                 } else if (timeout == 0) {
1215                         err = -ETIMEDOUT;
1216                         goto out_wait_error;
1217                 }
1218
1219                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1220         }
1221
1222         if (sk->sk_err) {
1223                 err = -sk->sk_err;
1224                 goto out_wait_error;
1225         } else
1226                 err = 0;
1227
1228 out_wait:
1229         finish_wait(sk_sleep(sk), &wait);
1230 out:
1231         release_sock(sk);
1232         return err;
1233
1234 out_wait_error:
1235         sk->sk_state = SS_UNCONNECTED;
1236         sock->state = SS_UNCONNECTED;
1237         goto out_wait;
1238 }
1239
1240 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1241 {
1242         struct sock *listener;
1243         int err;
1244         struct sock *connected;
1245         struct vsock_sock *vconnected;
1246         long timeout;
1247         DEFINE_WAIT(wait);
1248
1249         err = 0;
1250         listener = sock->sk;
1251
1252         lock_sock(listener);
1253
1254         if (sock->type != SOCK_STREAM) {
1255                 err = -EOPNOTSUPP;
1256                 goto out;
1257         }
1258
1259         if (listener->sk_state != SS_LISTEN) {
1260                 err = -EINVAL;
1261                 goto out;
1262         }
1263
1264         /* Wait for children sockets to appear; these are the new sockets
1265          * created upon connection establishment.
1266          */
1267         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1268         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1269
1270         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1271                listener->sk_err == 0) {
1272                 release_sock(listener);
1273                 timeout = schedule_timeout(timeout);
1274                 lock_sock(listener);
1275
1276                 if (signal_pending(current)) {
1277                         err = sock_intr_errno(timeout);
1278                         goto out_wait;
1279                 } else if (timeout == 0) {
1280                         err = -EAGAIN;
1281                         goto out_wait;
1282                 }
1283
1284                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1285         }
1286
1287         if (listener->sk_err)
1288                 err = -listener->sk_err;
1289
1290         if (connected) {
1291                 listener->sk_ack_backlog--;
1292
1293                 lock_sock(connected);
1294                 vconnected = vsock_sk(connected);
1295
1296                 /* If the listener socket has received an error, then we should
1297                  * reject this socket and return.  Note that we simply mark the
1298                  * socket rejected, drop our reference, and let the cleanup
1299                  * function handle the cleanup; the fact that we found it in
1300                  * the listener's accept queue guarantees that the cleanup
1301                  * function hasn't run yet.
1302                  */
1303                 if (err) {
1304                         vconnected->rejected = true;
1305                         release_sock(connected);
1306                         sock_put(connected);
1307                         goto out_wait;
1308                 }
1309
1310                 newsock->state = SS_CONNECTED;
1311                 sock_graft(connected, newsock);
1312                 release_sock(connected);
1313                 sock_put(connected);
1314         }
1315
1316 out_wait:
1317         finish_wait(sk_sleep(listener), &wait);
1318 out:
1319         release_sock(listener);
1320         return err;
1321 }
1322
1323 static int vsock_listen(struct socket *sock, int backlog)
1324 {
1325         int err;
1326         struct sock *sk;
1327         struct vsock_sock *vsk;
1328
1329         sk = sock->sk;
1330
1331         lock_sock(sk);
1332
1333         if (sock->type != SOCK_STREAM) {
1334                 err = -EOPNOTSUPP;
1335                 goto out;
1336         }
1337
1338         if (sock->state != SS_UNCONNECTED) {
1339                 err = -EINVAL;
1340                 goto out;
1341         }
1342
1343         vsk = vsock_sk(sk);
1344
1345         if (!vsock_addr_bound(&vsk->local_addr)) {
1346                 err = -EINVAL;
1347                 goto out;
1348         }
1349
1350         sk->sk_max_ack_backlog = backlog;
1351         sk->sk_state = SS_LISTEN;
1352
1353         err = 0;
1354
1355 out:
1356         release_sock(sk);
1357         return err;
1358 }
1359
1360 static int vsock_stream_setsockopt(struct socket *sock,
1361                                    int level,
1362                                    int optname,
1363                                    char __user *optval,
1364                                    unsigned int optlen)
1365 {
1366         int err;
1367         struct sock *sk;
1368         struct vsock_sock *vsk;
1369         u64 val;
1370
1371         if (level != AF_VSOCK)
1372                 return -ENOPROTOOPT;
1373
1374 #define COPY_IN(_v)                                       \
1375         do {                                              \
1376                 if (optlen < sizeof(_v)) {                \
1377                         err = -EINVAL;                    \
1378                         goto exit;                        \
1379                 }                                         \
1380                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1381                         err = -EFAULT;                                  \
1382                         goto exit;                                      \
1383                 }                                                       \
1384         } while (0)
1385
1386         err = 0;
1387         sk = sock->sk;
1388         vsk = vsock_sk(sk);
1389
1390         lock_sock(sk);
1391
1392         switch (optname) {
1393         case SO_VM_SOCKETS_BUFFER_SIZE:
1394                 COPY_IN(val);
1395                 transport->set_buffer_size(vsk, val);
1396                 break;
1397
1398         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1399                 COPY_IN(val);
1400                 transport->set_max_buffer_size(vsk, val);
1401                 break;
1402
1403         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1404                 COPY_IN(val);
1405                 transport->set_min_buffer_size(vsk, val);
1406                 break;
1407
1408         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1409                 struct timeval tv;
1410                 COPY_IN(tv);
1411                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1412                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1413                         vsk->connect_timeout = tv.tv_sec * HZ +
1414                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1415                         if (vsk->connect_timeout == 0)
1416                                 vsk->connect_timeout =
1417                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1418
1419                 } else {
1420                         err = -ERANGE;
1421                 }
1422                 break;
1423         }
1424
1425         default:
1426                 err = -ENOPROTOOPT;
1427                 break;
1428         }
1429
1430 #undef COPY_IN
1431
1432 exit:
1433         release_sock(sk);
1434         return err;
1435 }
1436
1437 static int vsock_stream_getsockopt(struct socket *sock,
1438                                    int level, int optname,
1439                                    char __user *optval,
1440                                    int __user *optlen)
1441 {
1442         int err;
1443         int len;
1444         struct sock *sk;
1445         struct vsock_sock *vsk;
1446         u64 val;
1447
1448         if (level != AF_VSOCK)
1449                 return -ENOPROTOOPT;
1450
1451         err = get_user(len, optlen);
1452         if (err != 0)
1453                 return err;
1454
1455 #define COPY_OUT(_v)                            \
1456         do {                                    \
1457                 if (len < sizeof(_v))           \
1458                         return -EINVAL;         \
1459                                                 \
1460                 len = sizeof(_v);               \
1461                 if (copy_to_user(optval, &_v, len) != 0)        \
1462                         return -EFAULT;                         \
1463                                                                 \
1464         } while (0)
1465
1466         err = 0;
1467         sk = sock->sk;
1468         vsk = vsock_sk(sk);
1469
1470         switch (optname) {
1471         case SO_VM_SOCKETS_BUFFER_SIZE:
1472                 val = transport->get_buffer_size(vsk);
1473                 COPY_OUT(val);
1474                 break;
1475
1476         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1477                 val = transport->get_max_buffer_size(vsk);
1478                 COPY_OUT(val);
1479                 break;
1480
1481         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1482                 val = transport->get_min_buffer_size(vsk);
1483                 COPY_OUT(val);
1484                 break;
1485
1486         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1487                 struct timeval tv;
1488                 tv.tv_sec = vsk->connect_timeout / HZ;
1489                 tv.tv_usec =
1490                     (vsk->connect_timeout -
1491                      tv.tv_sec * HZ) * (1000000 / HZ);
1492                 COPY_OUT(tv);
1493                 break;
1494         }
1495         default:
1496                 return -ENOPROTOOPT;
1497         }
1498
1499         err = put_user(len, optlen);
1500         if (err != 0)
1501                 return -EFAULT;
1502
1503 #undef COPY_OUT
1504
1505         return 0;
1506 }
1507
1508 static int vsock_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1509                                 struct msghdr *msg, size_t len)
1510 {
1511         struct sock *sk;
1512         struct vsock_sock *vsk;
1513         ssize_t total_written;
1514         long timeout;
1515         int err;
1516         struct vsock_transport_send_notify_data send_data;
1517
1518         DEFINE_WAIT(wait);
1519
1520         sk = sock->sk;
1521         vsk = vsock_sk(sk);
1522         total_written = 0;
1523         err = 0;
1524
1525         if (msg->msg_flags & MSG_OOB)
1526                 return -EOPNOTSUPP;
1527
1528         lock_sock(sk);
1529
1530         /* Callers should not provide a destination with stream sockets. */
1531         if (msg->msg_namelen) {
1532                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1533                 goto out;
1534         }
1535
1536         /* Send data only if both sides are not shutdown in the direction. */
1537         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1538             vsk->peer_shutdown & RCV_SHUTDOWN) {
1539                 err = -EPIPE;
1540                 goto out;
1541         }
1542
1543         if (sk->sk_state != SS_CONNECTED ||
1544             !vsock_addr_bound(&vsk->local_addr)) {
1545                 err = -ENOTCONN;
1546                 goto out;
1547         }
1548
1549         if (!vsock_addr_bound(&vsk->remote_addr)) {
1550                 err = -EDESTADDRREQ;
1551                 goto out;
1552         }
1553
1554         /* Wait for room in the produce queue to enqueue our user's data. */
1555         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1556
1557         err = transport->notify_send_init(vsk, &send_data);
1558         if (err < 0)
1559                 goto out;
1560
1561         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1562
1563         while (total_written < len) {
1564                 ssize_t written;
1565
1566                 while (vsock_stream_has_space(vsk) == 0 &&
1567                        sk->sk_err == 0 &&
1568                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1569                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1570
1571                         /* Don't wait for non-blocking sockets. */
1572                         if (timeout == 0) {
1573                                 err = -EAGAIN;
1574                                 goto out_wait;
1575                         }
1576
1577                         err = transport->notify_send_pre_block(vsk, &send_data);
1578                         if (err < 0)
1579                                 goto out_wait;
1580
1581                         release_sock(sk);
1582                         timeout = schedule_timeout(timeout);
1583                         lock_sock(sk);
1584                         if (signal_pending(current)) {
1585                                 err = sock_intr_errno(timeout);
1586                                 goto out_wait;
1587                         } else if (timeout == 0) {
1588                                 err = -EAGAIN;
1589                                 goto out_wait;
1590                         }
1591
1592                         prepare_to_wait(sk_sleep(sk), &wait,
1593                                         TASK_INTERRUPTIBLE);
1594                 }
1595
1596                 /* These checks occur both as part of and after the loop
1597                  * conditional since we need to check before and after
1598                  * sleeping.
1599                  */
1600                 if (sk->sk_err) {
1601                         err = -sk->sk_err;
1602                         goto out_wait;
1603                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1604                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1605                         err = -EPIPE;
1606                         goto out_wait;
1607                 }
1608
1609                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1610                 if (err < 0)
1611                         goto out_wait;
1612
1613                 /* Note that enqueue will only write as many bytes as are free
1614                  * in the produce queue, so we don't need to ensure len is
1615                  * smaller than the queue size.  It is the caller's
1616                  * responsibility to check how many bytes we were able to send.
1617                  */
1618
1619                 written = transport->stream_enqueue(
1620                                 vsk, msg->msg_iov,
1621                                 len - total_written);
1622                 if (written < 0) {
1623                         err = -ENOMEM;
1624                         goto out_wait;
1625                 }
1626
1627                 total_written += written;
1628
1629                 err = transport->notify_send_post_enqueue(
1630                                 vsk, written, &send_data);
1631                 if (err < 0)
1632                         goto out_wait;
1633
1634         }
1635
1636 out_wait:
1637         if (total_written > 0)
1638                 err = total_written;
1639         finish_wait(sk_sleep(sk), &wait);
1640 out:
1641         release_sock(sk);
1642         return err;
1643 }
1644
1645
1646 static int
1647 vsock_stream_recvmsg(struct kiocb *kiocb,
1648                      struct socket *sock,
1649                      struct msghdr *msg, size_t len, int flags)
1650 {
1651         struct sock *sk;
1652         struct vsock_sock *vsk;
1653         int err;
1654         size_t target;
1655         ssize_t copied;
1656         long timeout;
1657         struct vsock_transport_recv_notify_data recv_data;
1658
1659         DEFINE_WAIT(wait);
1660
1661         sk = sock->sk;
1662         vsk = vsock_sk(sk);
1663         err = 0;
1664
1665         msg->msg_namelen = 0;
1666
1667         lock_sock(sk);
1668
1669         if (sk->sk_state != SS_CONNECTED) {
1670                 /* Recvmsg is supposed to return 0 if a peer performs an
1671                  * orderly shutdown. Differentiate between that case and when a
1672                  * peer has not connected or a local shutdown occured with the
1673                  * SOCK_DONE flag.
1674                  */
1675                 if (sock_flag(sk, SOCK_DONE))
1676                         err = 0;
1677                 else
1678                         err = -ENOTCONN;
1679
1680                 goto out;
1681         }
1682
1683         if (flags & MSG_OOB) {
1684                 err = -EOPNOTSUPP;
1685                 goto out;
1686         }
1687
1688         /* We don't check peer_shutdown flag here since peer may actually shut
1689          * down, but there can be data in the queue that a local socket can
1690          * receive.
1691          */
1692         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1693                 err = 0;
1694                 goto out;
1695         }
1696
1697         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1698          * is not an error.  We may as well bail out now.
1699          */
1700         if (!len) {
1701                 err = 0;
1702                 goto out;
1703         }
1704
1705         /* We must not copy less than target bytes into the user's buffer
1706          * before returning successfully, so we wait for the consume queue to
1707          * have that much data to consume before dequeueing.  Note that this
1708          * makes it impossible to handle cases where target is greater than the
1709          * queue size.
1710          */
1711         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1712         if (target >= transport->stream_rcvhiwat(vsk)) {
1713                 err = -ENOMEM;
1714                 goto out;
1715         }
1716         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1717         copied = 0;
1718
1719         err = transport->notify_recv_init(vsk, target, &recv_data);
1720         if (err < 0)
1721                 goto out;
1722
1723         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1724
1725         while (1) {
1726                 s64 ready = vsock_stream_has_data(vsk);
1727
1728                 if (ready < 0) {
1729                         /* Invalid queue pair content. XXX This should be
1730                          * changed to a connection reset in a later change.
1731                          */
1732
1733                         err = -ENOMEM;
1734                         goto out_wait;
1735                 } else if (ready > 0) {
1736                         ssize_t read;
1737
1738                         err = transport->notify_recv_pre_dequeue(
1739                                         vsk, target, &recv_data);
1740                         if (err < 0)
1741                                 break;
1742
1743                         read = transport->stream_dequeue(
1744                                         vsk, msg->msg_iov,
1745                                         len - copied, flags);
1746                         if (read < 0) {
1747                                 err = -ENOMEM;
1748                                 break;
1749                         }
1750
1751                         copied += read;
1752
1753                         err = transport->notify_recv_post_dequeue(
1754                                         vsk, target, read,
1755                                         !(flags & MSG_PEEK), &recv_data);
1756                         if (err < 0)
1757                                 goto out_wait;
1758
1759                         if (read >= target || flags & MSG_PEEK)
1760                                 break;
1761
1762                         target -= read;
1763                 } else {
1764                         if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1765                             || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1766                                 break;
1767                         }
1768                         /* Don't wait for non-blocking sockets. */
1769                         if (timeout == 0) {
1770                                 err = -EAGAIN;
1771                                 break;
1772                         }
1773
1774                         err = transport->notify_recv_pre_block(
1775                                         vsk, target, &recv_data);
1776                         if (err < 0)
1777                                 break;
1778
1779                         release_sock(sk);
1780                         timeout = schedule_timeout(timeout);
1781                         lock_sock(sk);
1782
1783                         if (signal_pending(current)) {
1784                                 err = sock_intr_errno(timeout);
1785                                 break;
1786                         } else if (timeout == 0) {
1787                                 err = -EAGAIN;
1788                                 break;
1789                         }
1790
1791                         prepare_to_wait(sk_sleep(sk), &wait,
1792                                         TASK_INTERRUPTIBLE);
1793                 }
1794         }
1795
1796         if (sk->sk_err)
1797                 err = -sk->sk_err;
1798         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1799                 err = 0;
1800
1801         if (copied > 0) {
1802                 /* We only do these additional bookkeeping/notification steps
1803                  * if we actually copied something out of the queue pair
1804                  * instead of just peeking ahead.
1805                  */
1806
1807                 if (!(flags & MSG_PEEK)) {
1808                         /* If the other side has shutdown for sending and there
1809                          * is nothing more to read, then modify the socket
1810                          * state.
1811                          */
1812                         if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1813                                 if (vsock_stream_has_data(vsk) <= 0) {
1814                                         sk->sk_state = SS_UNCONNECTED;
1815                                         sock_set_flag(sk, SOCK_DONE);
1816                                         sk->sk_state_change(sk);
1817                                 }
1818                         }
1819                 }
1820                 err = copied;
1821         }
1822
1823 out_wait:
1824         finish_wait(sk_sleep(sk), &wait);
1825 out:
1826         release_sock(sk);
1827         return err;
1828 }
1829
1830 static const struct proto_ops vsock_stream_ops = {
1831         .family = PF_VSOCK,
1832         .owner = THIS_MODULE,
1833         .release = vsock_release,
1834         .bind = vsock_bind,
1835         .connect = vsock_stream_connect,
1836         .socketpair = sock_no_socketpair,
1837         .accept = vsock_accept,
1838         .getname = vsock_getname,
1839         .poll = vsock_poll,
1840         .ioctl = sock_no_ioctl,
1841         .listen = vsock_listen,
1842         .shutdown = vsock_shutdown,
1843         .setsockopt = vsock_stream_setsockopt,
1844         .getsockopt = vsock_stream_getsockopt,
1845         .sendmsg = vsock_stream_sendmsg,
1846         .recvmsg = vsock_stream_recvmsg,
1847         .mmap = sock_no_mmap,
1848         .sendpage = sock_no_sendpage,
1849 };
1850
1851 static int vsock_create(struct net *net, struct socket *sock,
1852                         int protocol, int kern)
1853 {
1854         if (!sock)
1855                 return -EINVAL;
1856
1857         if (protocol && protocol != PF_VSOCK)
1858                 return -EPROTONOSUPPORT;
1859
1860         switch (sock->type) {
1861         case SOCK_DGRAM:
1862                 sock->ops = &vsock_dgram_ops;
1863                 break;
1864         case SOCK_STREAM:
1865                 sock->ops = &vsock_stream_ops;
1866                 break;
1867         default:
1868                 return -ESOCKTNOSUPPORT;
1869         }
1870
1871         sock->state = SS_UNCONNECTED;
1872
1873         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0) ? 0 : -ENOMEM;
1874 }
1875
1876 static const struct net_proto_family vsock_family_ops = {
1877         .family = AF_VSOCK,
1878         .create = vsock_create,
1879         .owner = THIS_MODULE,
1880 };
1881
1882 static long vsock_dev_do_ioctl(struct file *filp,
1883                                unsigned int cmd, void __user *ptr)
1884 {
1885         u32 __user *p = ptr;
1886         int retval = 0;
1887
1888         switch (cmd) {
1889         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1890                 if (put_user(transport->get_local_cid(), p) != 0)
1891                         retval = -EFAULT;
1892                 break;
1893
1894         default:
1895                 pr_err("Unknown ioctl %d\n", cmd);
1896                 retval = -EINVAL;
1897         }
1898
1899         return retval;
1900 }
1901
1902 static long vsock_dev_ioctl(struct file *filp,
1903                             unsigned int cmd, unsigned long arg)
1904 {
1905         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1906 }
1907
1908 #ifdef CONFIG_COMPAT
1909 static long vsock_dev_compat_ioctl(struct file *filp,
1910                                    unsigned int cmd, unsigned long arg)
1911 {
1912         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1913 }
1914 #endif
1915
1916 static const struct file_operations vsock_device_ops = {
1917         .owner          = THIS_MODULE,
1918         .unlocked_ioctl = vsock_dev_ioctl,
1919 #ifdef CONFIG_COMPAT
1920         .compat_ioctl   = vsock_dev_compat_ioctl,
1921 #endif
1922         .open           = nonseekable_open,
1923 };
1924
1925 static struct miscdevice vsock_device = {
1926         .name           = "vsock",
1927         .fops           = &vsock_device_ops,
1928 };
1929
1930 static int __vsock_core_init(void)
1931 {
1932         int err;
1933
1934         vsock_init_tables();
1935
1936         vsock_device.minor = MISC_DYNAMIC_MINOR;
1937         err = misc_register(&vsock_device);
1938         if (err) {
1939                 pr_err("Failed to register misc device\n");
1940                 return -ENOENT;
1941         }
1942
1943         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1944         if (err) {
1945                 pr_err("Cannot register vsock protocol\n");
1946                 goto err_misc_deregister;
1947         }
1948
1949         err = sock_register(&vsock_family_ops);
1950         if (err) {
1951                 pr_err("could not register af_vsock (%d) address family: %d\n",
1952                        AF_VSOCK, err);
1953                 goto err_unregister_proto;
1954         }
1955
1956         return 0;
1957
1958 err_unregister_proto:
1959         proto_unregister(&vsock_proto);
1960 err_misc_deregister:
1961         misc_deregister(&vsock_device);
1962         return err;
1963 }
1964
1965 int vsock_core_init(const struct vsock_transport *t)
1966 {
1967         int retval = mutex_lock_interruptible(&vsock_register_mutex);
1968         if (retval)
1969                 return retval;
1970
1971         if (transport) {
1972                 retval = -EBUSY;
1973                 goto out;
1974         }
1975
1976         transport = t;
1977         retval = __vsock_core_init();
1978         if (retval)
1979                 transport = NULL;
1980
1981 out:
1982         mutex_unlock(&vsock_register_mutex);
1983         return retval;
1984 }
1985 EXPORT_SYMBOL_GPL(vsock_core_init);
1986
1987 void vsock_core_exit(void)
1988 {
1989         mutex_lock(&vsock_register_mutex);
1990
1991         misc_deregister(&vsock_device);
1992         sock_unregister(AF_VSOCK);
1993         proto_unregister(&vsock_proto);
1994
1995         /* We do not want the assignment below re-ordered. */
1996         mb();
1997         transport = NULL;
1998
1999         mutex_unlock(&vsock_register_mutex);
2000 }
2001 EXPORT_SYMBOL_GPL(vsock_core_exit);
2002
2003 MODULE_AUTHOR("VMware, Inc.");
2004 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2005 MODULE_VERSION("1.0.0.0-k");
2006 MODULE_LICENSE("GPL v2");