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