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 TCP_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  * - sk->sk_state uses the TCP state constants because they are widely used by
  87  * other address families and exposed to userspace tools like ss(8):
  88  *
  89  *   TCP_CLOSE - unconnected
  90  *   TCP_SYN_SENT - connecting
  91  *   TCP_ESTABLISHED - connected
  92  *   TCP_CLOSING - disconnecting
  93  *   TCP_LISTEN - listening
  94  */
  95
  96 #include <linux/types.h>
  97 #include <linux/bitops.h>
  98 #include <linux/cred.h>
  99 #include <linux/init.h>
 100 #include <linux/io.h>
 101 #include <linux/kernel.h>
 102 #include <linux/sched/signal.h>
 103 #include <linux/kmod.h>
 104 #include <linux/list.h>
 105 #include <linux/miscdevice.h>
 106 #include <linux/module.h>
 107 #include <linux/mutex.h>
 108 #include <linux/net.h>
 109 #include <linux/poll.h>
 110 #include <linux/random.h>
 111 #include <linux/skbuff.h>
 112 #include <linux/smp.h>
 113 #include <linux/socket.h>
 114 #include <linux/stddef.h>
 115 #include <linux/unistd.h>
 116 #include <linux/wait.h>
 117 #include <linux/workqueue.h>
 118 #include <net/sock.h>
 119 #include <net/af_vsock.h>
 120
 121 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 122 static void vsock_sk_destruct(struct sock *sk);
 123 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 124
 125 /* Protocol family. */
 126 static struct proto vsock_proto = {
 127         .name = "AF_VSOCK",
 128         .owner = THIS_MODULE,
 129         .obj_size = sizeof(struct vsock_sock),
 130 };
 131
 132 /* The default peer timeout indicates how long we will wait for a peer response
 133  * to a control message.
 134  */
 135 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 136
 137 static const struct vsock_transport *transport;
 138 static DEFINE_MUTEX(vsock_register_mutex);
 139
 140 /**** EXPORTS ****/
 141
 142 /* Get the ID of the local context.  This is transport dependent. */
 143
 144 int vm_sockets_get_local_cid(void)
 145 {
 146         return transport->get_local_cid();
 147 }
 148 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 149
 150 /**** UTILS ****/
 151
 152 /* Each bound VSocket is stored in the bind hash table and each connected
 153  * VSocket is stored in the connected hash table.
 154  *
 155  * Unbound sockets are all put on the same list attached to the end of the hash
 156  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 157  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 158  * represents the list that addr hashes to).
 159  *
 160  * Specifically, we initialize the vsock_bind_table array to a size of
 161  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 162  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 163  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 164  * mods with VSOCK_HASH_SIZE to ensure this.
 165  */
 166 #define VSOCK_HASH_SIZE         251
 167 #define MAX_PORT_RETRIES        24
 168
 169 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 170 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 171 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 172
 173 /* XXX This can probably be implemented in a better way. */
 174 #define VSOCK_CONN_HASH(src, dst)                               \
 175         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 176 #define vsock_connected_sockets(src, dst)               \
 177         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 178 #define vsock_connected_sockets_vsk(vsk)                                \
 179         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 180
 181 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 182 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 183 static DEFINE_SPINLOCK(vsock_table_lock);
 184
 185 /* Autobind this socket to the local address if necessary. */
 186 static int vsock_auto_bind(struct vsock_sock *vsk)
 187 {
 188         struct sock *sk = sk_vsock(vsk);
 189         struct sockaddr_vm local_addr;
 190
 191         if (vsock_addr_bound(&vsk->local_addr))
 192                 return 0;
 193         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 194         return __vsock_bind(sk, &local_addr);
 195 }
 196
 197 static void vsock_init_tables(void)
 198 {
 199         int i;
 200
 201         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 202                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 203
 204         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 205                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 206 }
 207
 208 static void __vsock_insert_bound(struct list_head *list,
 209                                  struct vsock_sock *vsk)
 210 {
 211         sock_hold(&vsk->sk);
 212         list_add(&vsk->bound_table, list);
 213 }
 214
 215 static void __vsock_insert_connected(struct list_head *list,
 216                                      struct vsock_sock *vsk)
 217 {
 218         sock_hold(&vsk->sk);
 219         list_add(&vsk->connected_table, list);
 220 }
 221
 222 static void __vsock_remove_bound(struct vsock_sock *vsk)
 223 {
 224         list_del_init(&vsk->bound_table);
 225         sock_put(&vsk->sk);
 226 }
 227
 228 static void __vsock_remove_connected(struct vsock_sock *vsk)
 229 {
 230         list_del_init(&vsk->connected_table);
 231         sock_put(&vsk->sk);
 232 }
 233
 234 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 235 {
 236         struct vsock_sock *vsk;
 237
 238         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 239                 if (addr->svm_port == vsk->local_addr.svm_port)
 240                         return sk_vsock(vsk);
 241
 242         return NULL;
 243 }
 244
 245 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 246                                                   struct sockaddr_vm *dst)
 247 {
 248         struct vsock_sock *vsk;
 249
 250         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 251                             connected_table) {
 252                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 253                     dst->svm_port == vsk->local_addr.svm_port) {
 254                         return sk_vsock(vsk);
 255                 }
 256         }
 257
 258         return NULL;
 259 }
 260
 261 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 262 {
 263         return !list_empty(&vsk->bound_table);
 264 }
 265
 266 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 267 {
 268         return !list_empty(&vsk->connected_table);
 269 }
 270
 271 static void vsock_insert_unbound(struct vsock_sock *vsk)
 272 {
 273         spin_lock_bh(&vsock_table_lock);
 274         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 275         spin_unlock_bh(&vsock_table_lock);
 276 }
 277
 278 void vsock_insert_connected(struct vsock_sock *vsk)
 279 {
 280         struct list_head *list = vsock_connected_sockets(
 281                 &vsk->remote_addr, &vsk->local_addr);
 282
 283         spin_lock_bh(&vsock_table_lock);
 284         __vsock_insert_connected(list, vsk);
 285         spin_unlock_bh(&vsock_table_lock);
 286 }
 287 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 288
 289 void vsock_remove_bound(struct vsock_sock *vsk)
 290 {
 291         spin_lock_bh(&vsock_table_lock);
 292         if (__vsock_in_bound_table(vsk))
 293                 __vsock_remove_bound(vsk);
 294         spin_unlock_bh(&vsock_table_lock);
 295 }
 296 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 297
 298 void vsock_remove_connected(struct vsock_sock *vsk)
 299 {
 300         spin_lock_bh(&vsock_table_lock);
 301         if (__vsock_in_connected_table(vsk))
 302                 __vsock_remove_connected(vsk);
 303         spin_unlock_bh(&vsock_table_lock);
 304 }
 305 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 306
 307 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 308 {
 309         struct sock *sk;
 310
 311         spin_lock_bh(&vsock_table_lock);
 312         sk = __vsock_find_bound_socket(addr);
 313         if (sk)
 314                 sock_hold(sk);
 315
 316         spin_unlock_bh(&vsock_table_lock);
 317
 318         return sk;
 319 }
 320 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 321
 322 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 323                                          struct sockaddr_vm *dst)
 324 {
 325         struct sock *sk;
 326
 327         spin_lock_bh(&vsock_table_lock);
 328         sk = __vsock_find_connected_socket(src, dst);
 329         if (sk)
 330                 sock_hold(sk);
 331
 332         spin_unlock_bh(&vsock_table_lock);
 333
 334         return sk;
 335 }
 336 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 337
 338 void vsock_remove_sock(struct vsock_sock *vsk)
 339 {
 340         vsock_remove_bound(vsk);
 341         vsock_remove_connected(vsk);
 342 }
 343 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 344
 345 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 346 {
 347         int i;
 348
 349         spin_lock_bh(&vsock_table_lock);
 350
 351         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 352                 struct vsock_sock *vsk;
 353                 list_for_each_entry(vsk, &vsock_connected_table[i],
 354                                     connected_table)
 355                         fn(sk_vsock(vsk));
 356         }
 357
 358         spin_unlock_bh(&vsock_table_lock);
 359 }
 360 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 361
 362 void vsock_add_pending(struct sock *listener, struct sock *pending)
 363 {
 364         struct vsock_sock *vlistener;
 365         struct vsock_sock *vpending;
 366
 367         vlistener = vsock_sk(listener);
 368         vpending = vsock_sk(pending);
 369
 370         sock_hold(pending);
 371         sock_hold(listener);
 372         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 373 }
 374 EXPORT_SYMBOL_GPL(vsock_add_pending);
 375
 376 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 377 {
 378         struct vsock_sock *vpending = vsock_sk(pending);
 379
 380         list_del_init(&vpending->pending_links);
 381         sock_put(listener);
 382         sock_put(pending);
 383 }
 384 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 385
 386 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 387 {
 388         struct vsock_sock *vlistener;
 389         struct vsock_sock *vconnected;
 390
 391         vlistener = vsock_sk(listener);
 392         vconnected = vsock_sk(connected);
 393
 394         sock_hold(connected);
 395         sock_hold(listener);
 396         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 397 }
 398 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 399
 400 static struct sock *vsock_dequeue_accept(struct sock *listener)
 401 {
 402         struct vsock_sock *vlistener;
 403         struct vsock_sock *vconnected;
 404
 405         vlistener = vsock_sk(listener);
 406
 407         if (list_empty(&vlistener->accept_queue))
 408                 return NULL;
 409
 410         vconnected = list_entry(vlistener->accept_queue.next,
 411                                 struct vsock_sock, accept_queue);
 412
 413         list_del_init(&vconnected->accept_queue);
 414         sock_put(listener);
 415         /* The caller will need a reference on the connected socket so we let
 416          * it call sock_put().
 417          */
 418
 419         return sk_vsock(vconnected);
 420 }
 421
 422 static bool vsock_is_accept_queue_empty(struct sock *sk)
 423 {
 424         struct vsock_sock *vsk = vsock_sk(sk);
 425         return list_empty(&vsk->accept_queue);
 426 }
 427
 428 static bool vsock_is_pending(struct sock *sk)
 429 {
 430         struct vsock_sock *vsk = vsock_sk(sk);
 431         return !list_empty(&vsk->pending_links);
 432 }
 433
 434 static int vsock_send_shutdown(struct sock *sk, int mode)
 435 {
 436         return transport->shutdown(vsock_sk(sk), mode);
 437 }
 438
 439 static void vsock_pending_work(struct work_struct *work)
 440 {
 441         struct sock *sk;
 442         struct sock *listener;
 443         struct vsock_sock *vsk;
 444         bool cleanup;
 445
 446         vsk = container_of(work, struct vsock_sock, pending_work.work);
 447         sk = sk_vsock(vsk);
 448         listener = vsk->listener;
 449         cleanup = true;
 450
 451         lock_sock(listener);
 452         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 453
 454         if (vsock_is_pending(sk)) {
 455                 vsock_remove_pending(listener, sk);
 456
 457                 listener->sk_ack_backlog--;
 458         } else if (!vsk->rejected) {
 459                 /* We are not on the pending list and accept() did not reject
 460                  * us, so we must have been accepted by our user process.  We
 461                  * just need to drop our references to the sockets and be on
 462                  * our way.
 463                  */
 464                 cleanup = false;
 465                 goto out;
 466         }
 467
 468         /* We need to remove ourself from the global connected sockets list so
 469          * incoming packets can't find this socket, and to reduce the reference
 470          * count.
 471          */
 472         vsock_remove_connected(vsk);
 473
 474         sk->sk_state = TCP_CLOSE;
 475
 476 out:
 477         release_sock(sk);
 478         release_sock(listener);
 479         if (cleanup)
 480                 sock_put(sk);
 481
 482         sock_put(sk);
 483         sock_put(listener);
 484 }
 485
 486 /**** SOCKET OPERATIONS ****/
 487
 488 static int __vsock_bind_stream(struct vsock_sock *vsk,
 489                                struct sockaddr_vm *addr)
 490 {
 491         static u32 port = 0;
 492         struct sockaddr_vm new_addr;
 493
 494         if (!port)
 495                 port = LAST_RESERVED_PORT + 1 +
 496                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 497
 498         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 499
 500         if (addr->svm_port == VMADDR_PORT_ANY) {
 501                 bool found = false;
 502                 unsigned int i;
 503
 504                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 505                         if (port <= LAST_RESERVED_PORT)
 506                                 port = LAST_RESERVED_PORT + 1;
 507
 508                         new_addr.svm_port = port++;
 509
 510                         if (!__vsock_find_bound_socket(&new_addr)) {
 511                                 found = true;
 512                                 break;
 513                         }
 514                 }
 515
 516                 if (!found)
 517                         return -EADDRNOTAVAIL;
 518         } else {
 519                 /* If port is in reserved range, ensure caller
 520                  * has necessary privileges.
 521                  */
 522                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 523                     !capable(CAP_NET_BIND_SERVICE)) {
 524                         return -EACCES;
 525                 }
 526
 527                 if (__vsock_find_bound_socket(&new_addr))
 528                         return -EADDRINUSE;
 529         }
 530
 531         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 532
 533         /* Remove stream sockets from the unbound list and add them to the hash
 534          * table for easy lookup by its address.  The unbound list is simply an
 535          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 536          */
 537         __vsock_remove_bound(vsk);
 538         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 539
 540         return 0;
 541 }
 542
 543 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 544                               struct sockaddr_vm *addr)
 545 {
 546         return transport->dgram_bind(vsk, addr);
 547 }
 548
 549 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 550 {
 551         struct vsock_sock *vsk = vsock_sk(sk);
 552         u32 cid;
 553         int retval;
 554
 555         /* First ensure this socket isn't already bound. */
 556         if (vsock_addr_bound(&vsk->local_addr))
 557                 return -EINVAL;
 558
 559         /* Now bind to the provided address or select appropriate values if
 560          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 561          * like AF_INET prevents binding to a non-local IP address (in most
 562          * cases), we only allow binding to the local CID.
 563          */
 564         cid = transport->get_local_cid();
 565         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 566                 return -EADDRNOTAVAIL;
 567
 568         switch (sk->sk_socket->type) {
 569         case SOCK_STREAM:
 570                 spin_lock_bh(&vsock_table_lock);
 571                 retval = __vsock_bind_stream(vsk, addr);
 572                 spin_unlock_bh(&vsock_table_lock);
 573                 break;
 574
 575         case SOCK_DGRAM:
 576                 retval = __vsock_bind_dgram(vsk, addr);
 577                 break;
 578
 579         default:
 580                 retval = -EINVAL;
 581                 break;
 582         }
 583
 584         return retval;
 585 }
 586
 587 static void vsock_connect_timeout(struct work_struct *work);
 588
 589 struct sock *__vsock_create(struct net *net,
 590                             struct socket *sock,
 591                             struct sock *parent,
 592                             gfp_t priority,
 593                             unsigned short type,
 594                             int kern)
 595 {
 596         struct sock *sk;
 597         struct vsock_sock *psk;
 598         struct vsock_sock *vsk;
 599
 600         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 601         if (!sk)
 602                 return NULL;
 603
 604         sock_init_data(sock, sk);
 605
 606         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 607          * non-NULL. We make sure that our sockets always have a type by
 608          * setting it here if needed.
 609          */
 610         if (!sock)
 611                 sk->sk_type = type;
 612
 613         vsk = vsock_sk(sk);
 614         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 615         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 616
 617         sk->sk_destruct = vsock_sk_destruct;
 618         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 619         sock_reset_flag(sk, SOCK_DONE);
 620
 621         INIT_LIST_HEAD(&vsk->bound_table);
 622         INIT_LIST_HEAD(&vsk->connected_table);
 623         vsk->listener = NULL;
 624         INIT_LIST_HEAD(&vsk->pending_links);
 625         INIT_LIST_HEAD(&vsk->accept_queue);
 626         vsk->rejected = false;
 627         vsk->sent_request = false;
 628         vsk->ignore_connecting_rst = false;
 629         vsk->peer_shutdown = 0;
 630         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 631         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 632
 633         psk = parent ? vsock_sk(parent) : NULL;
 634         if (parent) {
 635                 vsk->trusted = psk->trusted;
 636                 vsk->owner = get_cred(psk->owner);
 637                 vsk->connect_timeout = psk->connect_timeout;
 638                 security_sk_clone(parent, sk);
 639         } else {
 640                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 641                 vsk->owner = get_current_cred();
 642                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 643         }
 644
 645         if (transport->init(vsk, psk) < 0) {
 646                 sk_free(sk);
 647                 return NULL;
 648         }
 649
 650         if (sock)
 651                 vsock_insert_unbound(vsk);
 652
 653         return sk;
 654 }
 655 EXPORT_SYMBOL_GPL(__vsock_create);
 656
 657 static void __vsock_release(struct sock *sk, int level)
 658 {
 659         if (sk) {
 660                 struct sk_buff *skb;
 661                 struct sock *pending;
 662                 struct vsock_sock *vsk;
 663
 664                 vsk = vsock_sk(sk);
 665                 pending = NULL; /* Compiler warning. */
 666
 667                 /* The release call is supposed to use lock_sock_nested()
 668                  * rather than lock_sock(), if a sock lock should be acquired.
 669                  */
 670                 transport->release(vsk);
 671
 672                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 673                  * version to avoid the warning "possible recursive locking
 674                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 675                  * is the same as lock_sock(sk).
 676                  */
 677                 lock_sock_nested(sk, level);
 678                 sock_orphan(sk);
 679                 sk->sk_shutdown = SHUTDOWN_MASK;
 680
 681                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 682                         kfree_skb(skb);
 683
 684                 /* Clean up any sockets that never were accepted. */
 685                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 686                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 687                         sock_put(pending);
 688                 }
 689
 690                 release_sock(sk);
 691                 sock_put(sk);
 692         }
 693 }
 694
 695 static void vsock_sk_destruct(struct sock *sk)
 696 {
 697         struct vsock_sock *vsk = vsock_sk(sk);
 698
 699         transport->destruct(vsk);
 700
 701         /* When clearing these addresses, there's no need to set the family and
 702          * possibly register the address family with the kernel.
 703          */
 704         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 705         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 706
 707         put_cred(vsk->owner);
 708 }
 709
 710 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 711 {
 712         int err;
 713
 714         err = sock_queue_rcv_skb(sk, skb);
 715         if (err)
 716                 kfree_skb(skb);
 717
 718         return err;
 719 }
 720
 721 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 722 {
 723         return transport->stream_has_data(vsk);
 724 }
 725 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 726
 727 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 728 {
 729         return transport->stream_has_space(vsk);
 730 }
 731 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 732
 733 static int vsock_release(struct socket *sock)
 734 {
 735         __vsock_release(sock->sk, 0);
 736         sock->sk = NULL;
 737         sock->state = SS_FREE;
 738
 739         return 0;
 740 }
 741
 742 static int
 743 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 744 {
 745         int err;
 746         struct sock *sk;
 747         struct sockaddr_vm *vm_addr;
 748
 749         sk = sock->sk;
 750
 751         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 752                 return -EINVAL;
 753
 754         lock_sock(sk);
 755         err = __vsock_bind(sk, vm_addr);
 756         release_sock(sk);
 757
 758         return err;
 759 }
 760
 761 static int vsock_getname(struct socket *sock,
 762                          struct sockaddr *addr, int *addr_len, int peer)
 763 {
 764         int err;
 765         struct sock *sk;
 766         struct vsock_sock *vsk;
 767         struct sockaddr_vm *vm_addr;
 768
 769         sk = sock->sk;
 770         vsk = vsock_sk(sk);
 771         err = 0;
 772
 773         lock_sock(sk);
 774
 775         if (peer) {
 776                 if (sock->state != SS_CONNECTED) {
 777                         err = -ENOTCONN;
 778                         goto out;
 779                 }
 780                 vm_addr = &vsk->remote_addr;
 781         } else {
 782                 vm_addr = &vsk->local_addr;
 783         }
 784
 785         if (!vm_addr) {
 786                 err = -EINVAL;
 787                 goto out;
 788         }
 789
 790         /* sys_getsockname() and sys_getpeername() pass us a
 791          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 792          * that macro is defined in socket.c instead of .h, so we hardcode its
 793          * value here.
 794          */
 795         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 796         memcpy(addr, vm_addr, sizeof(*vm_addr));
 797         *addr_len = sizeof(*vm_addr);
 798
 799 out:
 800         release_sock(sk);
 801         return err;
 802 }
 803
 804 static int vsock_shutdown(struct socket *sock, int mode)
 805 {
 806         int err;
 807         struct sock *sk;
 808
 809         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 810          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 811          * here like the other address families do.  Note also that the
 812          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 813          * which is what we want.
 814          */
 815         mode++;
 816
 817         if ((mode & ~SHUTDOWN_MASK) || !mode)
 818                 return -EINVAL;
 819
 820         /* If this is a STREAM socket and it is not connected then bail out
 821          * immediately.  If it is a DGRAM socket then we must first kick the
 822          * socket so that it wakes up from any sleeping calls, for example
 823          * recv(), and then afterwards return the error.
 824          */
 825
 826         sk = sock->sk;
 827
 828         lock_sock(sk);
 829         if (sock->state == SS_UNCONNECTED) {
 830                 err = -ENOTCONN;
 831                 if (sk->sk_type == SOCK_STREAM)
 832                         goto out;
 833         } else {
 834                 sock->state = SS_DISCONNECTING;
 835                 err = 0;
 836         }
 837
 838         /* Receive and send shutdowns are treated alike. */
 839         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 840         if (mode) {
 841                 sk->sk_shutdown |= mode;
 842                 sk->sk_state_change(sk);
 843
 844                 if (sk->sk_type == SOCK_STREAM) {
 845                         sock_reset_flag(sk, SOCK_DONE);
 846                         vsock_send_shutdown(sk, mode);
 847                 }
 848         }
 849
 850 out:
 851         release_sock(sk);
 852         return err;
 853 }
 854
 855 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 856                                poll_table *wait)
 857 {
 858         struct sock *sk;
 859         unsigned int mask;
 860         struct vsock_sock *vsk;
 861
 862         sk = sock->sk;
 863         vsk = vsock_sk(sk);
 864
 865         poll_wait(file, sk_sleep(sk), wait);
 866         mask = 0;
 867
 868         if (sk->sk_err)
 869                 /* Signify that there has been an error on this socket. */
 870                 mask |= POLLERR;
 871
 872         /* INET sockets treat local write shutdown and peer write shutdown as a
 873          * case of POLLHUP set.
 874          */
 875         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 876             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 877              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 878                 mask |= POLLHUP;
 879         }
 880
 881         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 882             vsk->peer_shutdown & SEND_SHUTDOWN) {
 883                 mask |= POLLRDHUP;
 884         }
 885
 886         if (sock->type == SOCK_DGRAM) {
 887                 /* For datagram sockets we can read if there is something in
 888                  * the queue and write as long as the socket isn't shutdown for
 889                  * sending.
 890                  */
 891                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
 892                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 893                         mask |= POLLIN | POLLRDNORM;
 894                 }
 895
 896                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 897                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 898
 899         } else if (sock->type == SOCK_STREAM) {
 900                 lock_sock(sk);
 901
 902                 /* Listening sockets that have connections in their accept
 903                  * queue can be read.
 904                  */
 905                 if (sk->sk_state == TCP_LISTEN
 906                     && !vsock_is_accept_queue_empty(sk))
 907                         mask |= POLLIN | POLLRDNORM;
 908
 909                 /* If there is something in the queue then we can read. */
 910                 if (transport->stream_is_active(vsk) &&
 911                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 912                         bool data_ready_now = false;
 913                         int ret = transport->notify_poll_in(
 914                                         vsk, 1, &data_ready_now);
 915                         if (ret < 0) {
 916                                 mask |= POLLERR;
 917                         } else {
 918                                 if (data_ready_now)
 919                                         mask |= POLLIN | POLLRDNORM;
 920
 921                         }
 922                 }
 923
 924                 /* Sockets whose connections have been closed, reset, or
 925                  * terminated should also be considered read, and we check the
 926                  * shutdown flag for that.
 927                  */
 928                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 929                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 930                         mask |= POLLIN | POLLRDNORM;
 931                 }
 932
 933                 /* Connected sockets that can produce data can be written. */
 934                 if (sk->sk_state == TCP_ESTABLISHED) {
 935                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 936                                 bool space_avail_now = false;
 937                                 int ret = transport->notify_poll_out(
 938                                                 vsk, 1, &space_avail_now);
 939                                 if (ret < 0) {
 940                                         mask |= POLLERR;
 941                                 } else {
 942                                         if (space_avail_now)
 943                                                 /* Remove POLLWRBAND since INET
 944                                                  * sockets are not setting it.
 945                                                  */
 946                                                 mask |= POLLOUT | POLLWRNORM;
 947
 948                                 }
 949                         }
 950                 }
 951
 952                 /* Simulate INET socket poll behaviors, which sets
 953                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 954                  * but local send is not shutdown.
 955                  */
 956                 if (sk->sk_state == TCP_CLOSE) {
 957                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 958                                 mask |= POLLOUT | POLLWRNORM;
 959
 960                 }
 961
 962                 release_sock(sk);
 963         }
 964
 965         return mask;
 966 }
 967
 968 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 969                                size_t len)
 970 {
 971         int err;
 972         struct sock *sk;
 973         struct vsock_sock *vsk;
 974         struct sockaddr_vm *remote_addr;
 975
 976         if (msg->msg_flags & MSG_OOB)
 977                 return -EOPNOTSUPP;
 978
 979         /* For now, MSG_DONTWAIT is always assumed... */
 980         err = 0;
 981         sk = sock->sk;
 982         vsk = vsock_sk(sk);
 983
 984         lock_sock(sk);
 985
 986         err = vsock_auto_bind(vsk);
 987         if (err)
 988                 goto out;
 989
 990
 991         /* If the provided message contains an address, use that.  Otherwise
 992          * fall back on the socket's remote handle (if it has been connected).
 993          */
 994         if (msg->msg_name &&
 995             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 996                             &remote_addr) == 0) {
 997                 /* Ensure this address is of the right type and is a valid
 998                  * destination.
 999                  */
1000
1001                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1002                         remote_addr->svm_cid = transport->get_local_cid();
1003
1004                 if (!vsock_addr_bound(remote_addr)) {
1005                         err = -EINVAL;
1006                         goto out;
1007                 }
1008         } else if (sock->state == SS_CONNECTED) {
1009                 remote_addr = &vsk->remote_addr;
1010
1011                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1012                         remote_addr->svm_cid = transport->get_local_cid();
1013
1014                 /* XXX Should connect() or this function ensure remote_addr is
1015                  * bound?
1016                  */
1017                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1018                         err = -EINVAL;
1019                         goto out;
1020                 }
1021         } else {
1022                 err = -EINVAL;
1023                 goto out;
1024         }
1025
1026         if (!transport->dgram_allow(remote_addr->svm_cid,
1027                                     remote_addr->svm_port)) {
1028                 err = -EINVAL;
1029                 goto out;
1030         }
1031
1032         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1033
1034 out:
1035         release_sock(sk);
1036         return err;
1037 }
1038
1039 static int vsock_dgram_connect(struct socket *sock,
1040                                struct sockaddr *addr, int addr_len, int flags)
1041 {
1042         int err;
1043         struct sock *sk;
1044         struct vsock_sock *vsk;
1045         struct sockaddr_vm *remote_addr;
1046
1047         sk = sock->sk;
1048         vsk = vsock_sk(sk);
1049
1050         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1051         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1052                 lock_sock(sk);
1053                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1054                                 VMADDR_PORT_ANY);
1055                 sock->state = SS_UNCONNECTED;
1056                 release_sock(sk);
1057                 return 0;
1058         } else if (err != 0)
1059                 return -EINVAL;
1060
1061         lock_sock(sk);
1062
1063         err = vsock_auto_bind(vsk);
1064         if (err)
1065                 goto out;
1066
1067         if (!transport->dgram_allow(remote_addr->svm_cid,
1068                                     remote_addr->svm_port)) {
1069                 err = -EINVAL;
1070                 goto out;
1071         }
1072
1073         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1074         sock->state = SS_CONNECTED;
1075
1076 out:
1077         release_sock(sk);
1078         return err;
1079 }
1080
1081 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1082                                size_t len, int flags)
1083 {
1084         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1085 }
1086
1087 static const struct proto_ops vsock_dgram_ops = {
1088         .family = PF_VSOCK,
1089         .owner = THIS_MODULE,
1090         .release = vsock_release,
1091         .bind = vsock_bind,
1092         .connect = vsock_dgram_connect,
1093         .socketpair = sock_no_socketpair,
1094         .accept = sock_no_accept,
1095         .getname = vsock_getname,
1096         .poll = vsock_poll,
1097         .ioctl = sock_no_ioctl,
1098         .listen = sock_no_listen,
1099         .shutdown = vsock_shutdown,
1100         .setsockopt = sock_no_setsockopt,
1101         .getsockopt = sock_no_getsockopt,
1102         .sendmsg = vsock_dgram_sendmsg,
1103         .recvmsg = vsock_dgram_recvmsg,
1104         .mmap = sock_no_mmap,
1105         .sendpage = sock_no_sendpage,
1106 };
1107
1108 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1109 {
1110         if (!transport->cancel_pkt)
1111                 return -EOPNOTSUPP;
1112
1113         return transport->cancel_pkt(vsk);
1114 }
1115
1116 static void vsock_connect_timeout(struct work_struct *work)
1117 {
1118         struct sock *sk;
1119         struct vsock_sock *vsk;
1120
1121         vsk = container_of(work, struct vsock_sock, connect_work.work);
1122         sk = sk_vsock(vsk);
1123
1124         lock_sock(sk);
1125         if (sk->sk_state == TCP_SYN_SENT &&
1126             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1127                 sk->sk_state = TCP_CLOSE;
1128                 sk->sk_err = ETIMEDOUT;
1129                 sk->sk_error_report(sk);
1130                 vsock_transport_cancel_pkt(vsk);
1131         }
1132         release_sock(sk);
1133
1134         sock_put(sk);
1135 }
1136
1137 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1138                                 int addr_len, int flags)
1139 {
1140         int err;
1141         struct sock *sk;
1142         struct vsock_sock *vsk;
1143         struct sockaddr_vm *remote_addr;
1144         long timeout;
1145         DEFINE_WAIT(wait);
1146
1147         err = 0;
1148         sk = sock->sk;
1149         vsk = vsock_sk(sk);
1150
1151         lock_sock(sk);
1152
1153         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1154         switch (sock->state) {
1155         case SS_CONNECTED:
1156                 err = -EISCONN;
1157                 goto out;
1158         case SS_DISCONNECTING:
1159                 err = -EINVAL;
1160                 goto out;
1161         case SS_CONNECTING:
1162                 /* This continues on so we can move sock into the SS_CONNECTED
1163                  * state once the connection has completed (at which point err
1164                  * will be set to zero also).  Otherwise, we will either wait
1165                  * for the connection or return -EALREADY should this be a
1166                  * non-blocking call.
1167                  */
1168                 err = -EALREADY;
1169                 if (flags & O_NONBLOCK)
1170                         goto out;
1171                 break;
1172         default:
1173                 if ((sk->sk_state == TCP_LISTEN) ||
1174                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1175                         err = -EINVAL;
1176                         goto out;
1177                 }
1178
1179                 /* The hypervisor and well-known contexts do not have socket
1180                  * endpoints.
1181                  */
1182                 if (!transport->stream_allow(remote_addr->svm_cid,
1183                                              remote_addr->svm_port)) {
1184                         err = -ENETUNREACH;
1185                         goto out;
1186                 }
1187
1188                 /* Set the remote address that we are connecting to. */
1189                 memcpy(&vsk->remote_addr, remote_addr,
1190                        sizeof(vsk->remote_addr));
1191
1192                 err = vsock_auto_bind(vsk);
1193                 if (err)
1194                         goto out;
1195
1196                 sk->sk_state = TCP_SYN_SENT;
1197
1198                 err = transport->connect(vsk);
1199                 if (err < 0)
1200                         goto out;
1201
1202                 /* Mark sock as connecting and set the error code to in
1203                  * progress in case this is a non-blocking connect.
1204                  */
1205                 sock->state = SS_CONNECTING;
1206                 err = -EINPROGRESS;
1207         }
1208
1209         /* The receive path will handle all communication until we are able to
1210          * enter the connected state.  Here we wait for the connection to be
1211          * completed or a notification of an error.
1212          */
1213         timeout = vsk->connect_timeout;
1214         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1215
1216         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1217                 if (flags & O_NONBLOCK) {
1218                         /* If we're not going to block, we schedule a timeout
1219                          * function to generate a timeout on the connection
1220                          * attempt, in case the peer doesn't respond in a
1221                          * timely manner. We hold on to the socket until the
1222                          * timeout fires.
1223                          */
1224                         sock_hold(sk);
1225                         schedule_delayed_work(&vsk->connect_work, timeout);
1226
1227                         /* Skip ahead to preserve error code set above. */
1228                         goto out_wait;
1229                 }
1230
1231                 release_sock(sk);
1232                 timeout = schedule_timeout(timeout);
1233                 lock_sock(sk);
1234
1235                 if (signal_pending(current)) {
1236                         err = sock_intr_errno(timeout);
1237                         sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1238                         sock->state = SS_UNCONNECTED;
1239                         vsock_transport_cancel_pkt(vsk);
1240                         vsock_remove_connected(vsk);
1241                         goto out_wait;
1242                 } else if (timeout == 0) {
1243                         err = -ETIMEDOUT;
1244                         sk->sk_state = TCP_CLOSE;
1245                         sock->state = SS_UNCONNECTED;
1246                         vsock_transport_cancel_pkt(vsk);
1247                         goto out_wait;
1248                 }
1249
1250                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1251         }
1252
1253         if (sk->sk_err) {
1254                 err = -sk->sk_err;
1255                 sk->sk_state = TCP_CLOSE;
1256                 sock->state = SS_UNCONNECTED;
1257         } else {
1258                 err = 0;
1259         }
1260
1261 out_wait:
1262         finish_wait(sk_sleep(sk), &wait);
1263 out:
1264         release_sock(sk);
1265         return err;
1266 }
1267
1268 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1269                         bool kern)
1270 {
1271         struct sock *listener;
1272         int err;
1273         struct sock *connected;
1274         struct vsock_sock *vconnected;
1275         long timeout;
1276         DEFINE_WAIT(wait);
1277
1278         err = 0;
1279         listener = sock->sk;
1280
1281         lock_sock(listener);
1282
1283         if (sock->type != SOCK_STREAM) {
1284                 err = -EOPNOTSUPP;
1285                 goto out;
1286         }
1287
1288         if (listener->sk_state != TCP_LISTEN) {
1289                 err = -EINVAL;
1290                 goto out;
1291         }
1292
1293         /* Wait for children sockets to appear; these are the new sockets
1294          * created upon connection establishment.
1295          */
1296         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1297         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1298
1299         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1300                listener->sk_err == 0) {
1301                 release_sock(listener);
1302                 timeout = schedule_timeout(timeout);
1303                 finish_wait(sk_sleep(listener), &wait);
1304                 lock_sock(listener);
1305
1306                 if (signal_pending(current)) {
1307                         err = sock_intr_errno(timeout);
1308                         goto out;
1309                 } else if (timeout == 0) {
1310                         err = -EAGAIN;
1311                         goto out;
1312                 }
1313
1314                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1315         }
1316         finish_wait(sk_sleep(listener), &wait);
1317
1318         if (listener->sk_err)
1319                 err = -listener->sk_err;
1320
1321         if (connected) {
1322                 listener->sk_ack_backlog--;
1323
1324                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1325                 vconnected = vsock_sk(connected);
1326
1327                 /* If the listener socket has received an error, then we should
1328                  * reject this socket and return.  Note that we simply mark the
1329                  * socket rejected, drop our reference, and let the cleanup
1330                  * function handle the cleanup; the fact that we found it in
1331                  * the listener's accept queue guarantees that the cleanup
1332                  * function hasn't run yet.
1333                  */
1334                 if (err) {
1335                         vconnected->rejected = true;
1336                 } else {
1337                         newsock->state = SS_CONNECTED;
1338                         sock_graft(connected, newsock);
1339                 }
1340
1341                 release_sock(connected);
1342                 sock_put(connected);
1343         }
1344
1345 out:
1346         release_sock(listener);
1347         return err;
1348 }
1349
1350 static int vsock_listen(struct socket *sock, int backlog)
1351 {
1352         int err;
1353         struct sock *sk;
1354         struct vsock_sock *vsk;
1355
1356         sk = sock->sk;
1357
1358         lock_sock(sk);
1359
1360         if (sock->type != SOCK_STREAM) {
1361                 err = -EOPNOTSUPP;
1362                 goto out;
1363         }
1364
1365         if (sock->state != SS_UNCONNECTED) {
1366                 err = -EINVAL;
1367                 goto out;
1368         }
1369
1370         vsk = vsock_sk(sk);
1371
1372         if (!vsock_addr_bound(&vsk->local_addr)) {
1373                 err = -EINVAL;
1374                 goto out;
1375         }
1376
1377         sk->sk_max_ack_backlog = backlog;
1378         sk->sk_state = TCP_LISTEN;
1379
1380         err = 0;
1381
1382 out:
1383         release_sock(sk);
1384         return err;
1385 }
1386
1387 static int vsock_stream_setsockopt(struct socket *sock,
1388                                    int level,
1389                                    int optname,
1390                                    char __user *optval,
1391                                    unsigned int optlen)
1392 {
1393         int err;
1394         struct sock *sk;
1395         struct vsock_sock *vsk;
1396         u64 val;
1397
1398         if (level != AF_VSOCK)
1399                 return -ENOPROTOOPT;
1400
1401 #define COPY_IN(_v)                                       \
1402         do {                                              \
1403                 if (optlen < sizeof(_v)) {                \
1404                         err = -EINVAL;                    \
1405                         goto exit;                        \
1406                 }                                         \
1407                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1408                         err = -EFAULT;                                  \
1409                         goto exit;                                      \
1410                 }                                                       \
1411         } while (0)
1412
1413         err = 0;
1414         sk = sock->sk;
1415         vsk = vsock_sk(sk);
1416
1417         lock_sock(sk);
1418
1419         switch (optname) {
1420         case SO_VM_SOCKETS_BUFFER_SIZE:
1421                 COPY_IN(val);
1422                 transport->set_buffer_size(vsk, val);
1423                 break;
1424
1425         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1426                 COPY_IN(val);
1427                 transport->set_max_buffer_size(vsk, val);
1428                 break;
1429
1430         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1431                 COPY_IN(val);
1432                 transport->set_min_buffer_size(vsk, val);
1433                 break;
1434
1435         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1436                 struct timeval tv;
1437                 COPY_IN(tv);
1438                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1439                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1440                         vsk->connect_timeout = tv.tv_sec * HZ +
1441                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1442                         if (vsk->connect_timeout == 0)
1443                                 vsk->connect_timeout =
1444                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1445
1446                 } else {
1447                         err = -ERANGE;
1448                 }
1449                 break;
1450         }
1451
1452         default:
1453                 err = -ENOPROTOOPT;
1454                 break;
1455         }
1456
1457 #undef COPY_IN
1458
1459 exit:
1460         release_sock(sk);
1461         return err;
1462 }
1463
1464 static int vsock_stream_getsockopt(struct socket *sock,
1465                                    int level, int optname,
1466                                    char __user *optval,
1467                                    int __user *optlen)
1468 {
1469         int err;
1470         int len;
1471         struct sock *sk;
1472         struct vsock_sock *vsk;
1473         u64 val;
1474
1475         if (level != AF_VSOCK)
1476                 return -ENOPROTOOPT;
1477
1478         err = get_user(len, optlen);
1479         if (err != 0)
1480                 return err;
1481
1482 #define COPY_OUT(_v)                            \
1483         do {                                    \
1484                 if (len < sizeof(_v))           \
1485                         return -EINVAL;         \
1486                                                 \
1487                 len = sizeof(_v);               \
1488                 if (copy_to_user(optval, &_v, len) != 0)        \
1489                         return -EFAULT;                         \
1490                                                                 \
1491         } while (0)
1492
1493         err = 0;
1494         sk = sock->sk;
1495         vsk = vsock_sk(sk);
1496
1497         switch (optname) {
1498         case SO_VM_SOCKETS_BUFFER_SIZE:
1499                 val = transport->get_buffer_size(vsk);
1500                 COPY_OUT(val);
1501                 break;
1502
1503         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1504                 val = transport->get_max_buffer_size(vsk);
1505                 COPY_OUT(val);
1506                 break;
1507
1508         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1509                 val = transport->get_min_buffer_size(vsk);
1510                 COPY_OUT(val);
1511                 break;
1512
1513         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1514                 struct timeval tv;
1515                 tv.tv_sec = vsk->connect_timeout / HZ;
1516                 tv.tv_usec =
1517                     (vsk->connect_timeout -
1518                      tv.tv_sec * HZ) * (1000000 / HZ);
1519                 COPY_OUT(tv);
1520                 break;
1521         }
1522         default:
1523                 return -ENOPROTOOPT;
1524         }
1525
1526         err = put_user(len, optlen);
1527         if (err != 0)
1528                 return -EFAULT;
1529
1530 #undef COPY_OUT
1531
1532         return 0;
1533 }
1534
1535 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1536                                 size_t len)
1537 {
1538         struct sock *sk;
1539         struct vsock_sock *vsk;
1540         ssize_t total_written;
1541         long timeout;
1542         int err;
1543         struct vsock_transport_send_notify_data send_data;
1544         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1545
1546         sk = sock->sk;
1547         vsk = vsock_sk(sk);
1548         total_written = 0;
1549         err = 0;
1550
1551         if (msg->msg_flags & MSG_OOB)
1552                 return -EOPNOTSUPP;
1553
1554         lock_sock(sk);
1555
1556         /* Callers should not provide a destination with stream sockets. */
1557         if (msg->msg_namelen) {
1558                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1559                 goto out;
1560         }
1561
1562         /* Send data only if both sides are not shutdown in the direction. */
1563         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1564             vsk->peer_shutdown & RCV_SHUTDOWN) {
1565                 err = -EPIPE;
1566                 goto out;
1567         }
1568
1569         if (sk->sk_state != TCP_ESTABLISHED ||
1570             !vsock_addr_bound(&vsk->local_addr)) {
1571                 err = -ENOTCONN;
1572                 goto out;
1573         }
1574
1575         if (!vsock_addr_bound(&vsk->remote_addr)) {
1576                 err = -EDESTADDRREQ;
1577                 goto out;
1578         }
1579
1580         /* Wait for room in the produce queue to enqueue our user's data. */
1581         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1582
1583         err = transport->notify_send_init(vsk, &send_data);
1584         if (err < 0)
1585                 goto out;
1586
1587         while (total_written < len) {
1588                 ssize_t written;
1589
1590                 add_wait_queue(sk_sleep(sk), &wait);
1591                 while (vsock_stream_has_space(vsk) == 0 &&
1592                        sk->sk_err == 0 &&
1593                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1594                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1595
1596                         /* Don't wait for non-blocking sockets. */
1597                         if (timeout == 0) {
1598                                 err = -EAGAIN;
1599                                 remove_wait_queue(sk_sleep(sk), &wait);
1600                                 goto out_err;
1601                         }
1602
1603                         err = transport->notify_send_pre_block(vsk, &send_data);
1604                         if (err < 0) {
1605                                 remove_wait_queue(sk_sleep(sk), &wait);
1606                                 goto out_err;
1607                         }
1608
1609                         release_sock(sk);
1610                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1611                         lock_sock(sk);
1612                         if (signal_pending(current)) {
1613                                 err = sock_intr_errno(timeout);
1614                                 remove_wait_queue(sk_sleep(sk), &wait);
1615                                 goto out_err;
1616                         } else if (timeout == 0) {
1617                                 err = -EAGAIN;
1618                                 remove_wait_queue(sk_sleep(sk), &wait);
1619                                 goto out_err;
1620                         }
1621                 }
1622                 remove_wait_queue(sk_sleep(sk), &wait);
1623
1624                 /* These checks occur both as part of and after the loop
1625                  * conditional since we need to check before and after
1626                  * sleeping.
1627                  */
1628                 if (sk->sk_err) {
1629                         err = -sk->sk_err;
1630                         goto out_err;
1631                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1632                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1633                         err = -EPIPE;
1634                         goto out_err;
1635                 }
1636
1637                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1638                 if (err < 0)
1639                         goto out_err;
1640
1641                 /* Note that enqueue will only write as many bytes as are free
1642                  * in the produce queue, so we don't need to ensure len is
1643                  * smaller than the queue size.  It is the caller's
1644                  * responsibility to check how many bytes we were able to send.
1645                  */
1646
1647                 written = transport->stream_enqueue(
1648                                 vsk, msg,
1649                                 len - total_written);
1650                 if (written < 0) {
1651                         err = -ENOMEM;
1652                         goto out_err;
1653                 }
1654
1655                 total_written += written;
1656
1657                 err = transport->notify_send_post_enqueue(
1658                                 vsk, written, &send_data);
1659                 if (err < 0)
1660                         goto out_err;
1661
1662         }
1663
1664 out_err:
1665         if (total_written > 0)
1666                 err = total_written;
1667 out:
1668         release_sock(sk);
1669         return err;
1670 }
1671
1672
1673 static int
1674 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1675                      int flags)
1676 {
1677         struct sock *sk;
1678         struct vsock_sock *vsk;
1679         int err;
1680         size_t target;
1681         ssize_t copied;
1682         long timeout;
1683         struct vsock_transport_recv_notify_data recv_data;
1684
1685         DEFINE_WAIT(wait);
1686
1687         sk = sock->sk;
1688         vsk = vsock_sk(sk);
1689         err = 0;
1690
1691         lock_sock(sk);
1692
1693         if (sk->sk_state != TCP_ESTABLISHED) {
1694                 /* Recvmsg is supposed to return 0 if a peer performs an
1695                  * orderly shutdown. Differentiate between that case and when a
1696                  * peer has not connected or a local shutdown occured with the
1697                  * SOCK_DONE flag.
1698                  */
1699                 if (sock_flag(sk, SOCK_DONE))
1700                         err = 0;
1701                 else
1702                         err = -ENOTCONN;
1703
1704                 goto out;
1705         }
1706
1707         if (flags & MSG_OOB) {
1708                 err = -EOPNOTSUPP;
1709                 goto out;
1710         }
1711
1712         /* We don't check peer_shutdown flag here since peer may actually shut
1713          * down, but there can be data in the queue that a local socket can
1714          * receive.
1715          */
1716         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1717                 err = 0;
1718                 goto out;
1719         }
1720
1721         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1722          * is not an error.  We may as well bail out now.
1723          */
1724         if (!len) {
1725                 err = 0;
1726                 goto out;
1727         }
1728
1729         /* We must not copy less than target bytes into the user's buffer
1730          * before returning successfully, so we wait for the consume queue to
1731          * have that much data to consume before dequeueing.  Note that this
1732          * makes it impossible to handle cases where target is greater than the
1733          * queue size.
1734          */
1735         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1736         if (target >= transport->stream_rcvhiwat(vsk)) {
1737                 err = -ENOMEM;
1738                 goto out;
1739         }
1740         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1741         copied = 0;
1742
1743         err = transport->notify_recv_init(vsk, target, &recv_data);
1744         if (err < 0)
1745                 goto out;
1746
1747
1748         while (1) {
1749                 s64 ready;
1750
1751                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1752                 ready = vsock_stream_has_data(vsk);
1753
1754                 if (ready == 0) {
1755                         if (sk->sk_err != 0 ||
1756                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1757                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1758                                 finish_wait(sk_sleep(sk), &wait);
1759                                 break;
1760                         }
1761                         /* Don't wait for non-blocking sockets. */
1762                         if (timeout == 0) {
1763                                 err = -EAGAIN;
1764                                 finish_wait(sk_sleep(sk), &wait);
1765                                 break;
1766                         }
1767
1768                         err = transport->notify_recv_pre_block(
1769                                         vsk, target, &recv_data);
1770                         if (err < 0) {
1771                                 finish_wait(sk_sleep(sk), &wait);
1772                                 break;
1773                         }
1774                         release_sock(sk);
1775                         timeout = schedule_timeout(timeout);
1776                         lock_sock(sk);
1777
1778                         if (signal_pending(current)) {
1779                                 err = sock_intr_errno(timeout);
1780                                 finish_wait(sk_sleep(sk), &wait);
1781                                 break;
1782                         } else if (timeout == 0) {
1783                                 err = -EAGAIN;
1784                                 finish_wait(sk_sleep(sk), &wait);
1785                                 break;
1786                         }
1787                 } else {
1788                         ssize_t read;
1789
1790                         finish_wait(sk_sleep(sk), &wait);
1791
1792                         if (ready < 0) {
1793                                 /* Invalid queue pair content. XXX This should
1794                                 * be changed to a connection reset in a later
1795                                 * change.
1796                                 */
1797
1798                                 err = -ENOMEM;
1799                                 goto out;
1800                         }
1801
1802                         err = transport->notify_recv_pre_dequeue(
1803                                         vsk, target, &recv_data);
1804                         if (err < 0)
1805                                 break;
1806
1807                         read = transport->stream_dequeue(
1808                                         vsk, msg,
1809                                         len - copied, flags);
1810                         if (read < 0) {
1811                                 err = -ENOMEM;
1812                                 break;
1813                         }
1814
1815                         copied += read;
1816
1817                         err = transport->notify_recv_post_dequeue(
1818                                         vsk, target, read,
1819                                         !(flags & MSG_PEEK), &recv_data);
1820                         if (err < 0)
1821                                 goto out;
1822
1823                         if (read >= target || flags & MSG_PEEK)
1824                                 break;
1825
1826                         target -= read;
1827                 }
1828         }
1829
1830         if (sk->sk_err)
1831                 err = -sk->sk_err;
1832         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1833                 err = 0;
1834
1835         if (copied > 0)
1836                 err = copied;
1837
1838 out:
1839         release_sock(sk);
1840         return err;
1841 }
1842
1843 static const struct proto_ops vsock_stream_ops = {
1844         .family = PF_VSOCK,
1845         .owner = THIS_MODULE,
1846         .release = vsock_release,
1847         .bind = vsock_bind,
1848         .connect = vsock_stream_connect,
1849         .socketpair = sock_no_socketpair,
1850         .accept = vsock_accept,
1851         .getname = vsock_getname,
1852         .poll = vsock_poll,
1853         .ioctl = sock_no_ioctl,
1854         .listen = vsock_listen,
1855         .shutdown = vsock_shutdown,
1856         .setsockopt = vsock_stream_setsockopt,
1857         .getsockopt = vsock_stream_getsockopt,
1858         .sendmsg = vsock_stream_sendmsg,
1859         .recvmsg = vsock_stream_recvmsg,
1860         .mmap = sock_no_mmap,
1861         .sendpage = sock_no_sendpage,
1862 };
1863
1864 static int vsock_create(struct net *net, struct socket *sock,
1865                         int protocol, int kern)
1866 {
1867         if (!sock)
1868                 return -EINVAL;
1869
1870         if (protocol && protocol != PF_VSOCK)
1871                 return -EPROTONOSUPPORT;
1872
1873         switch (sock->type) {
1874         case SOCK_DGRAM:
1875                 sock->ops = &vsock_dgram_ops;
1876                 break;
1877         case SOCK_STREAM:
1878                 sock->ops = &vsock_stream_ops;
1879                 break;
1880         default:
1881                 return -ESOCKTNOSUPPORT;
1882         }
1883
1884         sock->state = SS_UNCONNECTED;
1885
1886         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1887 }
1888
1889 static const struct net_proto_family vsock_family_ops = {
1890         .family = AF_VSOCK,
1891         .create = vsock_create,
1892         .owner = THIS_MODULE,
1893 };
1894
1895 static long vsock_dev_do_ioctl(struct file *filp,
1896                                unsigned int cmd, void __user *ptr)
1897 {
1898         u32 __user *p = ptr;
1899         int retval = 0;
1900
1901         switch (cmd) {
1902         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1903                 if (put_user(transport->get_local_cid(), p) != 0)
1904                         retval = -EFAULT;
1905                 break;
1906
1907         default:
1908                 pr_err("Unknown ioctl %d\n", cmd);
1909                 retval = -EINVAL;
1910         }
1911
1912         return retval;
1913 }
1914
1915 static long vsock_dev_ioctl(struct file *filp,
1916                             unsigned int cmd, unsigned long arg)
1917 {
1918         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1919 }
1920
1921 #ifdef CONFIG_COMPAT
1922 static long vsock_dev_compat_ioctl(struct file *filp,
1923                                    unsigned int cmd, unsigned long arg)
1924 {
1925         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1926 }
1927 #endif
1928
1929 static const struct file_operations vsock_device_ops = {
1930         .owner          = THIS_MODULE,
1931         .unlocked_ioctl = vsock_dev_ioctl,
1932 #ifdef CONFIG_COMPAT
1933         .compat_ioctl   = vsock_dev_compat_ioctl,
1934 #endif
1935         .open           = nonseekable_open,
1936 };
1937
1938 static struct miscdevice vsock_device = {
1939         .name           = "vsock",
1940         .fops           = &vsock_device_ops,
1941 };
1942
1943 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1944 {
1945         int err = mutex_lock_interruptible(&vsock_register_mutex);
1946
1947         if (err)
1948                 return err;
1949
1950         if (transport) {
1951                 err = -EBUSY;
1952                 goto err_busy;
1953         }
1954
1955         /* Transport must be the owner of the protocol so that it can't
1956          * unload while there are open sockets.
1957          */
1958         vsock_proto.owner = owner;
1959         transport = t;
1960
1961         vsock_init_tables();
1962
1963         vsock_device.minor = MISC_DYNAMIC_MINOR;
1964         err = misc_register(&vsock_device);
1965         if (err) {
1966                 pr_err("Failed to register misc device\n");
1967                 goto err_reset_transport;
1968         }
1969
1970         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1971         if (err) {
1972                 pr_err("Cannot register vsock protocol\n");
1973                 goto err_deregister_misc;
1974         }
1975
1976         err = sock_register(&vsock_family_ops);
1977         if (err) {
1978                 pr_err("could not register af_vsock (%d) address family: %d\n",
1979                        AF_VSOCK, err);
1980                 goto err_unregister_proto;
1981         }
1982
1983         mutex_unlock(&vsock_register_mutex);
1984         return 0;
1985
1986 err_unregister_proto:
1987         proto_unregister(&vsock_proto);
1988 err_deregister_misc:
1989         misc_deregister(&vsock_device);
1990 err_reset_transport:
1991         transport = NULL;
1992 err_busy:
1993         mutex_unlock(&vsock_register_mutex);
1994         return err;
1995 }
1996 EXPORT_SYMBOL_GPL(__vsock_core_init);
1997
1998 void vsock_core_exit(void)
1999 {
2000         mutex_lock(&vsock_register_mutex);
2001
2002         misc_deregister(&vsock_device);
2003         sock_unregister(AF_VSOCK);
2004         proto_unregister(&vsock_proto);
2005
2006         /* We do not want the assignment below re-ordered. */
2007         mb();
2008         transport = NULL;
2009
2010         mutex_unlock(&vsock_register_mutex);
2011 }
2012 EXPORT_SYMBOL_GPL(vsock_core_exit);
2013
2014 const struct vsock_transport *vsock_core_get_transport(void)
2015 {
2016         /* vsock_register_mutex not taken since only the transport uses this
2017          * function and only while registered.
2018          */
2019         return transport;
2020 }
2021 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2022
2023 MODULE_AUTHOR("VMware, Inc.");
2024 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2025 MODULE_VERSION("1.0.2.0-k");
2026 MODULE_LICENSE("GPL v2");