net/vmw_vsock/af_vsock.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VMware vSockets Driver
   4  *
   5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   6  */
   7
   8 /* Implementation notes:
   9  *
  10  * - There are two kinds of sockets: those created by user action (such as
  11  * calling socket(2)) and those created by incoming connection request packets.
  12  *
  13  * - There are two "global" tables, one for bound sockets (sockets that have
  14  * specified an address that they are responsible for) and one for connected
  15  * sockets (sockets that have established a connection with another socket).
  16  * These tables are "global" in that all sockets on the system are placed
  17  * within them. - Note, though, that the bound table contains an extra entry
  18  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  19  * that list. The bound table is used solely for lookup of sockets when packets
  20  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  21  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  22  * sockets out of the bound hash buckets will reduce the chance of collisions
  23  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  24  * socket type in the hash table lookups.
  25  *
  26  * - Sockets created by user action will either be "client" sockets that
  27  * initiate a connection or "server" sockets that listen for connections; we do
  28  * not support simultaneous connects (two "client" sockets connecting).
  29  *
  30  * - "Server" sockets are referred to as listener sockets throughout this
  31  * implementation because they are in the TCP_LISTEN state.  When a
  32  * connection request is received (the second kind of socket mentioned above),
  33  * we create a new socket and refer to it as a pending socket.  These pending
  34  * sockets are placed on the pending connection list of the listener socket.
  35  * When future packets are received for the address the listener socket is
  36  * bound to, we check if the source of the packet is from one that has an
  37  * existing pending connection.  If it does, we process the packet for the
  38  * pending socket.  When that socket reaches the connected state, it is removed
  39  * from the listener socket's pending list and enqueued in the listener
  40  * socket's accept queue.  Callers of accept(2) will accept connected sockets
  41  * from the listener socket's accept queue.  If the socket cannot be accepted
  42  * for some reason then it is marked rejected.  Once the connection is
  43  * accepted, it is owned by the user process and the responsibility for cleanup
  44  * falls with that user process.
  45  *
  46  * - It is possible that these pending sockets will never reach the connected
  47  * state; in fact, we may never receive another packet after the connection
  48  * request.  Because of this, we must schedule a cleanup function to run in the
  49  * future, after some amount of time passes where a connection should have been
  50  * established.  This function ensures that the socket is off all lists so it
  51  * cannot be retrieved, then drops all references to the socket so it is cleaned
  52  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  53  * function will also cleanup rejected sockets, those that reach the connected
  54  * state but leave it before they have been accepted.
  55  *
  56  * - Lock ordering for pending or accept queue sockets is:
  57  *
  58  *     lock_sock(listener);
  59  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  60  *
  61  * Using explicit nested locking keeps lockdep happy since normally only one
  62  * lock of a given class may be taken at a time.
  63  *
  64  * - Sockets created by user action will be cleaned up when the user process
  65  * calls close(2), causing our release implementation to be called. Our release
  66  * implementation will perform some cleanup then drop the last reference so our
  67  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  68  * perform additional cleanup that's common for both types of sockets.
  69  *
  70  * - A socket's reference count is what ensures that the structure won't be
  71  * freed.  Each entry in a list (such as the "global" bound and connected tables
  72  * and the listener socket's pending list and connected queue) ensures a
  73  * reference.  When we defer work until process context and pass a socket as our
  74  * argument, we must ensure the reference count is increased to ensure the
  75  * socket isn't freed before the function is run; the deferred function will
  76  * then drop the reference.
  77  *
  78  * - sk->sk_state uses the TCP state constants because they are widely used by
  79  * other address families and exposed to userspace tools like ss(8):
  80  *
  81  *   TCP_CLOSE - unconnected
  82  *   TCP_SYN_SENT - connecting
  83  *   TCP_ESTABLISHED - connected
  84  *   TCP_CLOSING - disconnecting
  85  *   TCP_LISTEN - listening
  86  */
  87
  88 #include <linux/types.h>
  89 #include <linux/bitops.h>
  90 #include <linux/cred.h>
  91 #include <linux/init.h>
  92 #include <linux/io.h>
  93 #include <linux/kernel.h>
  94 #include <linux/sched/signal.h>
  95 #include <linux/kmod.h>
  96 #include <linux/list.h>
  97 #include <linux/miscdevice.h>
  98 #include <linux/module.h>
  99 #include <linux/mutex.h>
 100 #include <linux/net.h>
 101 #include <linux/poll.h>
 102 #include <linux/random.h>
 103 #include <linux/skbuff.h>
 104 #include <linux/smp.h>
 105 #include <linux/socket.h>
 106 #include <linux/stddef.h>
 107 #include <linux/unistd.h>
 108 #include <linux/wait.h>
 109 #include <linux/workqueue.h>
 110 #include <net/sock.h>
 111 #include <net/af_vsock.h>
 112
 113 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 114 static void vsock_sk_destruct(struct sock *sk);
 115 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 116
 117 /* Protocol family. */
 118 static struct proto vsock_proto = {
 119         .name = "AF_VSOCK",
 120         .owner = THIS_MODULE,
 121         .obj_size = sizeof(struct vsock_sock),
 122 };
 123
 124 /* The default peer timeout indicates how long we will wait for a peer response
 125  * to a control message.
 126  */
 127 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 128
 129 static const struct vsock_transport *transport;
 130 static DEFINE_MUTEX(vsock_register_mutex);
 131
 132 /**** EXPORTS ****/
 133
 134 /* Get the ID of the local context.  This is transport dependent. */
 135
 136 int vm_sockets_get_local_cid(void)
 137 {
 138         return transport->get_local_cid();
 139 }
 140 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 141
 142 /**** UTILS ****/
 143
 144 /* Each bound VSocket is stored in the bind hash table and each connected
 145  * VSocket is stored in the connected hash table.
 146  *
 147  * Unbound sockets are all put on the same list attached to the end of the hash
 148  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 149  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 150  * represents the list that addr hashes to).
 151  *
 152  * Specifically, we initialize the vsock_bind_table array to a size of
 153  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 154  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 155  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 156  * mods with VSOCK_HASH_SIZE to ensure this.
 157  */
 158 #define MAX_PORT_RETRIES        24
 159
 160 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 161 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 162 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 163
 164 /* XXX This can probably be implemented in a better way. */
 165 #define VSOCK_CONN_HASH(src, dst)                               \
 166         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 167 #define vsock_connected_sockets(src, dst)               \
 168         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 169 #define vsock_connected_sockets_vsk(vsk)                                \
 170         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 171
 172 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 173 EXPORT_SYMBOL_GPL(vsock_bind_table);
 174 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 175 EXPORT_SYMBOL_GPL(vsock_connected_table);
 176 DEFINE_SPINLOCK(vsock_table_lock);
 177 EXPORT_SYMBOL_GPL(vsock_table_lock);
 178
 179 /* Autobind this socket to the local address if necessary. */
 180 static int vsock_auto_bind(struct vsock_sock *vsk)
 181 {
 182         struct sock *sk = sk_vsock(vsk);
 183         struct sockaddr_vm local_addr;
 184
 185         if (vsock_addr_bound(&vsk->local_addr))
 186                 return 0;
 187         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 188         return __vsock_bind(sk, &local_addr);
 189 }
 190
 191 static int __init vsock_init_tables(void)
 192 {
 193         int i;
 194
 195         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 196                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 197
 198         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 199                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 200         return 0;
 201 }
 202
 203 static void __vsock_insert_bound(struct list_head *list,
 204                                  struct vsock_sock *vsk)
 205 {
 206         sock_hold(&vsk->sk);
 207         list_add(&vsk->bound_table, list);
 208 }
 209
 210 static void __vsock_insert_connected(struct list_head *list,
 211                                      struct vsock_sock *vsk)
 212 {
 213         sock_hold(&vsk->sk);
 214         list_add(&vsk->connected_table, list);
 215 }
 216
 217 static void __vsock_remove_bound(struct vsock_sock *vsk)
 218 {
 219         list_del_init(&vsk->bound_table);
 220         sock_put(&vsk->sk);
 221 }
 222
 223 static void __vsock_remove_connected(struct vsock_sock *vsk)
 224 {
 225         list_del_init(&vsk->connected_table);
 226         sock_put(&vsk->sk);
 227 }
 228
 229 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 230 {
 231         struct vsock_sock *vsk;
 232
 233         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 234                 if (addr->svm_port == vsk->local_addr.svm_port)
 235                         return sk_vsock(vsk);
 236
 237         return NULL;
 238 }
 239
 240 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 241                                                   struct sockaddr_vm *dst)
 242 {
 243         struct vsock_sock *vsk;
 244
 245         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 246                             connected_table) {
 247                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 248                     dst->svm_port == vsk->local_addr.svm_port) {
 249                         return sk_vsock(vsk);
 250                 }
 251         }
 252
 253         return NULL;
 254 }
 255
 256 static void vsock_insert_unbound(struct vsock_sock *vsk)
 257 {
 258         spin_lock_bh(&vsock_table_lock);
 259         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 260         spin_unlock_bh(&vsock_table_lock);
 261 }
 262
 263 void vsock_insert_connected(struct vsock_sock *vsk)
 264 {
 265         struct list_head *list = vsock_connected_sockets(
 266                 &vsk->remote_addr, &vsk->local_addr);
 267
 268         spin_lock_bh(&vsock_table_lock);
 269         __vsock_insert_connected(list, vsk);
 270         spin_unlock_bh(&vsock_table_lock);
 271 }
 272 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 273
 274 void vsock_remove_bound(struct vsock_sock *vsk)
 275 {
 276         spin_lock_bh(&vsock_table_lock);
 277         if (__vsock_in_bound_table(vsk))
 278                 __vsock_remove_bound(vsk);
 279         spin_unlock_bh(&vsock_table_lock);
 280 }
 281 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 282
 283 void vsock_remove_connected(struct vsock_sock *vsk)
 284 {
 285         spin_lock_bh(&vsock_table_lock);
 286         if (__vsock_in_connected_table(vsk))
 287                 __vsock_remove_connected(vsk);
 288         spin_unlock_bh(&vsock_table_lock);
 289 }
 290 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 291
 292 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 293 {
 294         struct sock *sk;
 295
 296         spin_lock_bh(&vsock_table_lock);
 297         sk = __vsock_find_bound_socket(addr);
 298         if (sk)
 299                 sock_hold(sk);
 300
 301         spin_unlock_bh(&vsock_table_lock);
 302
 303         return sk;
 304 }
 305 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 306
 307 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 308                                          struct sockaddr_vm *dst)
 309 {
 310         struct sock *sk;
 311
 312         spin_lock_bh(&vsock_table_lock);
 313         sk = __vsock_find_connected_socket(src, dst);
 314         if (sk)
 315                 sock_hold(sk);
 316
 317         spin_unlock_bh(&vsock_table_lock);
 318
 319         return sk;
 320 }
 321 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 322
 323 void vsock_remove_sock(struct vsock_sock *vsk)
 324 {
 325         vsock_remove_bound(vsk);
 326         vsock_remove_connected(vsk);
 327 }
 328 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 329
 330 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 331 {
 332         int i;
 333
 334         spin_lock_bh(&vsock_table_lock);
 335
 336         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 337                 struct vsock_sock *vsk;
 338                 list_for_each_entry(vsk, &vsock_connected_table[i],
 339                                     connected_table)
 340                         fn(sk_vsock(vsk));
 341         }
 342
 343         spin_unlock_bh(&vsock_table_lock);
 344 }
 345 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 346
 347 void vsock_add_pending(struct sock *listener, struct sock *pending)
 348 {
 349         struct vsock_sock *vlistener;
 350         struct vsock_sock *vpending;
 351
 352         vlistener = vsock_sk(listener);
 353         vpending = vsock_sk(pending);
 354
 355         sock_hold(pending);
 356         sock_hold(listener);
 357         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 358 }
 359 EXPORT_SYMBOL_GPL(vsock_add_pending);
 360
 361 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 362 {
 363         struct vsock_sock *vpending = vsock_sk(pending);
 364
 365         list_del_init(&vpending->pending_links);
 366         sock_put(listener);
 367         sock_put(pending);
 368 }
 369 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 370
 371 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 372 {
 373         struct vsock_sock *vlistener;
 374         struct vsock_sock *vconnected;
 375
 376         vlistener = vsock_sk(listener);
 377         vconnected = vsock_sk(connected);
 378
 379         sock_hold(connected);
 380         sock_hold(listener);
 381         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 382 }
 383 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 384
 385 static struct sock *vsock_dequeue_accept(struct sock *listener)
 386 {
 387         struct vsock_sock *vlistener;
 388         struct vsock_sock *vconnected;
 389
 390         vlistener = vsock_sk(listener);
 391
 392         if (list_empty(&vlistener->accept_queue))
 393                 return NULL;
 394
 395         vconnected = list_entry(vlistener->accept_queue.next,
 396                                 struct vsock_sock, accept_queue);
 397
 398         list_del_init(&vconnected->accept_queue);
 399         sock_put(listener);
 400         /* The caller will need a reference on the connected socket so we let
 401          * it call sock_put().
 402          */
 403
 404         return sk_vsock(vconnected);
 405 }
 406
 407 static bool vsock_is_accept_queue_empty(struct sock *sk)
 408 {
 409         struct vsock_sock *vsk = vsock_sk(sk);
 410         return list_empty(&vsk->accept_queue);
 411 }
 412
 413 static bool vsock_is_pending(struct sock *sk)
 414 {
 415         struct vsock_sock *vsk = vsock_sk(sk);
 416         return !list_empty(&vsk->pending_links);
 417 }
 418
 419 static int vsock_send_shutdown(struct sock *sk, int mode)
 420 {
 421         return transport->shutdown(vsock_sk(sk), mode);
 422 }
 423
 424 static void vsock_pending_work(struct work_struct *work)
 425 {
 426         struct sock *sk;
 427         struct sock *listener;
 428         struct vsock_sock *vsk;
 429         bool cleanup;
 430
 431         vsk = container_of(work, struct vsock_sock, pending_work.work);
 432         sk = sk_vsock(vsk);
 433         listener = vsk->listener;
 434         cleanup = true;
 435
 436         lock_sock(listener);
 437         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 438
 439         if (vsock_is_pending(sk)) {
 440                 vsock_remove_pending(listener, sk);
 441
 442                 listener->sk_ack_backlog--;
 443         } else if (!vsk->rejected) {
 444                 /* We are not on the pending list and accept() did not reject
 445                  * us, so we must have been accepted by our user process.  We
 446                  * just need to drop our references to the sockets and be on
 447                  * our way.
 448                  */
 449                 cleanup = false;
 450                 goto out;
 451         }
 452
 453         /* We need to remove ourself from the global connected sockets list so
 454          * incoming packets can't find this socket, and to reduce the reference
 455          * count.
 456          */
 457         vsock_remove_connected(vsk);
 458
 459         sk->sk_state = TCP_CLOSE;
 460
 461 out:
 462         release_sock(sk);
 463         release_sock(listener);
 464         if (cleanup)
 465                 sock_put(sk);
 466
 467         sock_put(sk);
 468         sock_put(listener);
 469 }
 470
 471 /**** SOCKET OPERATIONS ****/
 472
 473 static int __vsock_bind_stream(struct vsock_sock *vsk,
 474                                struct sockaddr_vm *addr)
 475 {
 476         static u32 port;
 477         struct sockaddr_vm new_addr;
 478
 479         if (!port)
 480                 port = LAST_RESERVED_PORT + 1 +
 481                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 482
 483         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 484
 485         if (addr->svm_port == VMADDR_PORT_ANY) {
 486                 bool found = false;
 487                 unsigned int i;
 488
 489                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 490                         if (port <= LAST_RESERVED_PORT)
 491                                 port = LAST_RESERVED_PORT + 1;
 492
 493                         new_addr.svm_port = port++;
 494
 495                         if (!__vsock_find_bound_socket(&new_addr)) {
 496                                 found = true;
 497                                 break;
 498                         }
 499                 }
 500
 501                 if (!found)
 502                         return -EADDRNOTAVAIL;
 503         } else {
 504                 /* If port is in reserved range, ensure caller
 505                  * has necessary privileges.
 506                  */
 507                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 508                     !capable(CAP_NET_BIND_SERVICE)) {
 509                         return -EACCES;
 510                 }
 511
 512                 if (__vsock_find_bound_socket(&new_addr))
 513                         return -EADDRINUSE;
 514         }
 515
 516         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 517
 518         /* Remove stream sockets from the unbound list and add them to the hash
 519          * table for easy lookup by its address.  The unbound list is simply an
 520          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 521          */
 522         __vsock_remove_bound(vsk);
 523         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 524
 525         return 0;
 526 }
 527
 528 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 529                               struct sockaddr_vm *addr)
 530 {
 531         return transport->dgram_bind(vsk, addr);
 532 }
 533
 534 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 535 {
 536         struct vsock_sock *vsk = vsock_sk(sk);
 537         u32 cid;
 538         int retval;
 539
 540         /* First ensure this socket isn't already bound. */
 541         if (vsock_addr_bound(&vsk->local_addr))
 542                 return -EINVAL;
 543
 544         /* Now bind to the provided address or select appropriate values if
 545          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 546          * like AF_INET prevents binding to a non-local IP address (in most
 547          * cases), we only allow binding to the local CID.
 548          */
 549         cid = transport->get_local_cid();
 550         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 551                 return -EADDRNOTAVAIL;
 552
 553         switch (sk->sk_socket->type) {
 554         case SOCK_STREAM:
 555                 spin_lock_bh(&vsock_table_lock);
 556                 retval = __vsock_bind_stream(vsk, addr);
 557                 spin_unlock_bh(&vsock_table_lock);
 558                 break;
 559
 560         case SOCK_DGRAM:
 561                 retval = __vsock_bind_dgram(vsk, addr);
 562                 break;
 563
 564         default:
 565                 retval = -EINVAL;
 566                 break;
 567         }
 568
 569         return retval;
 570 }
 571
 572 static void vsock_connect_timeout(struct work_struct *work);
 573
 574 struct sock *__vsock_create(struct net *net,
 575                             struct socket *sock,
 576                             struct sock *parent,
 577                             gfp_t priority,
 578                             unsigned short type,
 579                             int kern)
 580 {
 581         struct sock *sk;
 582         struct vsock_sock *psk;
 583         struct vsock_sock *vsk;
 584
 585         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 586         if (!sk)
 587                 return NULL;
 588
 589         sock_init_data(sock, sk);
 590
 591         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 592          * non-NULL. We make sure that our sockets always have a type by
 593          * setting it here if needed.
 594          */
 595         if (!sock)
 596                 sk->sk_type = type;
 597
 598         vsk = vsock_sk(sk);
 599         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 600         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 601
 602         sk->sk_destruct = vsock_sk_destruct;
 603         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 604         sock_reset_flag(sk, SOCK_DONE);
 605
 606         INIT_LIST_HEAD(&vsk->bound_table);
 607         INIT_LIST_HEAD(&vsk->connected_table);
 608         vsk->listener = NULL;
 609         INIT_LIST_HEAD(&vsk->pending_links);
 610         INIT_LIST_HEAD(&vsk->accept_queue);
 611         vsk->rejected = false;
 612         vsk->sent_request = false;
 613         vsk->ignore_connecting_rst = false;
 614         vsk->peer_shutdown = 0;
 615         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 616         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 617
 618         psk = parent ? vsock_sk(parent) : NULL;
 619         if (parent) {
 620                 vsk->trusted = psk->trusted;
 621                 vsk->owner = get_cred(psk->owner);
 622                 vsk->connect_timeout = psk->connect_timeout;
 623         } else {
 624                 vsk->trusted = capable(CAP_NET_ADMIN);
 625                 vsk->owner = get_current_cred();
 626                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 627         }
 628
 629         if (transport->init(vsk, psk) < 0) {
 630                 sk_free(sk);
 631                 return NULL;
 632         }
 633
 634         if (sock)
 635                 vsock_insert_unbound(vsk);
 636
 637         return sk;
 638 }
 639 EXPORT_SYMBOL_GPL(__vsock_create);
 640
 641 static void __vsock_release(struct sock *sk, int level)
 642 {
 643         if (sk) {
 644                 struct sk_buff *skb;
 645                 struct sock *pending;
 646                 struct vsock_sock *vsk;
 647
 648                 vsk = vsock_sk(sk);
 649                 pending = NULL; /* Compiler warning. */
 650
 651                 /* The release call is supposed to use lock_sock_nested()
 652                  * rather than lock_sock(), if a sock lock should be acquired.
 653                  */
 654                 transport->release(vsk);
 655
 656                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 657                  * version to avoid the warning "possible recursive locking
 658                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 659                  * is the same as lock_sock(sk).
 660                  */
 661                 lock_sock_nested(sk, level);
 662                 sock_orphan(sk);
 663                 sk->sk_shutdown = SHUTDOWN_MASK;
 664
 665                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 666                         kfree_skb(skb);
 667
 668                 /* Clean up any sockets that never were accepted. */
 669                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 670                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 671                         sock_put(pending);
 672                 }
 673
 674                 release_sock(sk);
 675                 sock_put(sk);
 676         }
 677 }
 678
 679 static void vsock_sk_destruct(struct sock *sk)
 680 {
 681         struct vsock_sock *vsk = vsock_sk(sk);
 682
 683         transport->destruct(vsk);
 684
 685         /* When clearing these addresses, there's no need to set the family and
 686          * possibly register the address family with the kernel.
 687          */
 688         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 689         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 690
 691         put_cred(vsk->owner);
 692 }
 693
 694 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 695 {
 696         int err;
 697
 698         err = sock_queue_rcv_skb(sk, skb);
 699         if (err)
 700                 kfree_skb(skb);
 701
 702         return err;
 703 }
 704
 705 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 706 {
 707         return transport->stream_has_data(vsk);
 708 }
 709 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 710
 711 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 712 {
 713         return transport->stream_has_space(vsk);
 714 }
 715 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 716
 717 static int vsock_release(struct socket *sock)
 718 {
 719         __vsock_release(sock->sk, 0);
 720         sock->sk = NULL;
 721         sock->state = SS_FREE;
 722
 723         return 0;
 724 }
 725
 726 static int
 727 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 728 {
 729         int err;
 730         struct sock *sk;
 731         struct sockaddr_vm *vm_addr;
 732
 733         sk = sock->sk;
 734
 735         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 736                 return -EINVAL;
 737
 738         lock_sock(sk);
 739         err = __vsock_bind(sk, vm_addr);
 740         release_sock(sk);
 741
 742         return err;
 743 }
 744
 745 static int vsock_getname(struct socket *sock,
 746                          struct sockaddr *addr, int peer)
 747 {
 748         int err;
 749         struct sock *sk;
 750         struct vsock_sock *vsk;
 751         struct sockaddr_vm *vm_addr;
 752
 753         sk = sock->sk;
 754         vsk = vsock_sk(sk);
 755         err = 0;
 756
 757         lock_sock(sk);
 758
 759         if (peer) {
 760                 if (sock->state != SS_CONNECTED) {
 761                         err = -ENOTCONN;
 762                         goto out;
 763                 }
 764                 vm_addr = &vsk->remote_addr;
 765         } else {
 766                 vm_addr = &vsk->local_addr;
 767         }
 768
 769         if (!vm_addr) {
 770                 err = -EINVAL;
 771                 goto out;
 772         }
 773
 774         /* sys_getsockname() and sys_getpeername() pass us a
 775          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 776          * that macro is defined in socket.c instead of .h, so we hardcode its
 777          * value here.
 778          */
 779         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 780         memcpy(addr, vm_addr, sizeof(*vm_addr));
 781         err = sizeof(*vm_addr);
 782
 783 out:
 784         release_sock(sk);
 785         return err;
 786 }
 787
 788 static int vsock_shutdown(struct socket *sock, int mode)
 789 {
 790         int err;
 791         struct sock *sk;
 792
 793         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 794          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 795          * here like the other address families do.  Note also that the
 796          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 797          * which is what we want.
 798          */
 799         mode++;
 800
 801         if ((mode & ~SHUTDOWN_MASK) || !mode)
 802                 return -EINVAL;
 803
 804         /* If this is a STREAM socket and it is not connected then bail out
 805          * immediately.  If it is a DGRAM socket then we must first kick the
 806          * socket so that it wakes up from any sleeping calls, for example
 807          * recv(), and then afterwards return the error.
 808          */
 809
 810         sk = sock->sk;
 811         if (sock->state == SS_UNCONNECTED) {
 812                 err = -ENOTCONN;
 813                 if (sk->sk_type == SOCK_STREAM)
 814                         return err;
 815         } else {
 816                 sock->state = SS_DISCONNECTING;
 817                 err = 0;
 818         }
 819
 820         /* Receive and send shutdowns are treated alike. */
 821         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 822         if (mode) {
 823                 lock_sock(sk);
 824                 sk->sk_shutdown |= mode;
 825                 sk->sk_state_change(sk);
 826                 release_sock(sk);
 827
 828                 if (sk->sk_type == SOCK_STREAM) {
 829                         sock_reset_flag(sk, SOCK_DONE);
 830                         vsock_send_shutdown(sk, mode);
 831                 }
 832         }
 833
 834         return err;
 835 }
 836
 837 static __poll_t vsock_poll(struct file *file, struct socket *sock,
 838                                poll_table *wait)
 839 {
 840         struct sock *sk;
 841         __poll_t mask;
 842         struct vsock_sock *vsk;
 843
 844         sk = sock->sk;
 845         vsk = vsock_sk(sk);
 846
 847         poll_wait(file, sk_sleep(sk), wait);
 848         mask = 0;
 849
 850         if (sk->sk_err)
 851                 /* Signify that there has been an error on this socket. */
 852                 mask |= EPOLLERR;
 853
 854         /* INET sockets treat local write shutdown and peer write shutdown as a
 855          * case of EPOLLHUP set.
 856          */
 857         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 858             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 859              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 860                 mask |= EPOLLHUP;
 861         }
 862
 863         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 864             vsk->peer_shutdown & SEND_SHUTDOWN) {
 865                 mask |= EPOLLRDHUP;
 866         }
 867
 868         if (sock->type == SOCK_DGRAM) {
 869                 /* For datagram sockets we can read if there is something in
 870                  * the queue and write as long as the socket isn't shutdown for
 871                  * sending.
 872                  */
 873                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
 874                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 875                         mask |= EPOLLIN | EPOLLRDNORM;
 876                 }
 877
 878                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 879                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
 880
 881         } else if (sock->type == SOCK_STREAM) {
 882                 lock_sock(sk);
 883
 884                 /* Listening sockets that have connections in their accept
 885                  * queue can be read.
 886                  */
 887                 if (sk->sk_state == TCP_LISTEN
 888                     && !vsock_is_accept_queue_empty(sk))
 889                         mask |= EPOLLIN | EPOLLRDNORM;
 890
 891                 /* If there is something in the queue then we can read. */
 892                 if (transport->stream_is_active(vsk) &&
 893                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 894                         bool data_ready_now = false;
 895                         int ret = transport->notify_poll_in(
 896                                         vsk, 1, &data_ready_now);
 897                         if (ret < 0) {
 898                                 mask |= EPOLLERR;
 899                         } else {
 900                                 if (data_ready_now)
 901                                         mask |= EPOLLIN | EPOLLRDNORM;
 902
 903                         }
 904                 }
 905
 906                 /* Sockets whose connections have been closed, reset, or
 907                  * terminated should also be considered read, and we check the
 908                  * shutdown flag for that.
 909                  */
 910                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 911                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 912                         mask |= EPOLLIN | EPOLLRDNORM;
 913                 }
 914
 915                 /* Connected sockets that can produce data can be written. */
 916                 if (sk->sk_state == TCP_ESTABLISHED) {
 917                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 918                                 bool space_avail_now = false;
 919                                 int ret = transport->notify_poll_out(
 920                                                 vsk, 1, &space_avail_now);
 921                                 if (ret < 0) {
 922                                         mask |= EPOLLERR;
 923                                 } else {
 924                                         if (space_avail_now)
 925                                                 /* Remove EPOLLWRBAND since INET
 926                                                  * sockets are not setting it.
 927                                                  */
 928                                                 mask |= EPOLLOUT | EPOLLWRNORM;
 929
 930                                 }
 931                         }
 932                 }
 933
 934                 /* Simulate INET socket poll behaviors, which sets
 935                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
 936                  * but local send is not shutdown.
 937                  */
 938                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
 939                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 940                                 mask |= EPOLLOUT | EPOLLWRNORM;
 941
 942                 }
 943
 944                 release_sock(sk);
 945         }
 946
 947         return mask;
 948 }
 949
 950 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 951                                size_t len)
 952 {
 953         int err;
 954         struct sock *sk;
 955         struct vsock_sock *vsk;
 956         struct sockaddr_vm *remote_addr;
 957
 958         if (msg->msg_flags & MSG_OOB)
 959                 return -EOPNOTSUPP;
 960
 961         /* For now, MSG_DONTWAIT is always assumed... */
 962         err = 0;
 963         sk = sock->sk;
 964         vsk = vsock_sk(sk);
 965
 966         lock_sock(sk);
 967
 968         err = vsock_auto_bind(vsk);
 969         if (err)
 970                 goto out;
 971
 972
 973         /* If the provided message contains an address, use that.  Otherwise
 974          * fall back on the socket's remote handle (if it has been connected).
 975          */
 976         if (msg->msg_name &&
 977             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 978                             &remote_addr) == 0) {
 979                 /* Ensure this address is of the right type and is a valid
 980                  * destination.
 981                  */
 982
 983                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 984                         remote_addr->svm_cid = transport->get_local_cid();
 985
 986                 if (!vsock_addr_bound(remote_addr)) {
 987                         err = -EINVAL;
 988                         goto out;
 989                 }
 990         } else if (sock->state == SS_CONNECTED) {
 991                 remote_addr = &vsk->remote_addr;
 992
 993                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 994                         remote_addr->svm_cid = transport->get_local_cid();
 995
 996                 /* XXX Should connect() or this function ensure remote_addr is
 997                  * bound?
 998                  */
 999                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1000                         err = -EINVAL;
1001                         goto out;
1002                 }
1003         } else {
1004                 err = -EINVAL;
1005                 goto out;
1006         }
1007
1008         if (!transport->dgram_allow(remote_addr->svm_cid,
1009                                     remote_addr->svm_port)) {
1010                 err = -EINVAL;
1011                 goto out;
1012         }
1013
1014         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1015
1016 out:
1017         release_sock(sk);
1018         return err;
1019 }
1020
1021 static int vsock_dgram_connect(struct socket *sock,
1022                                struct sockaddr *addr, int addr_len, int flags)
1023 {
1024         int err;
1025         struct sock *sk;
1026         struct vsock_sock *vsk;
1027         struct sockaddr_vm *remote_addr;
1028
1029         sk = sock->sk;
1030         vsk = vsock_sk(sk);
1031
1032         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1033         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1034                 lock_sock(sk);
1035                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1036                                 VMADDR_PORT_ANY);
1037                 sock->state = SS_UNCONNECTED;
1038                 release_sock(sk);
1039                 return 0;
1040         } else if (err != 0)
1041                 return -EINVAL;
1042
1043         lock_sock(sk);
1044
1045         err = vsock_auto_bind(vsk);
1046         if (err)
1047                 goto out;
1048
1049         if (!transport->dgram_allow(remote_addr->svm_cid,
1050                                     remote_addr->svm_port)) {
1051                 err = -EINVAL;
1052                 goto out;
1053         }
1054
1055         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1056         sock->state = SS_CONNECTED;
1057
1058 out:
1059         release_sock(sk);
1060         return err;
1061 }
1062
1063 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1064                                size_t len, int flags)
1065 {
1066         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1067 }
1068
1069 static const struct proto_ops vsock_dgram_ops = {
1070         .family = PF_VSOCK,
1071         .owner = THIS_MODULE,
1072         .release = vsock_release,
1073         .bind = vsock_bind,
1074         .connect = vsock_dgram_connect,
1075         .socketpair = sock_no_socketpair,
1076         .accept = sock_no_accept,
1077         .getname = vsock_getname,
1078         .poll = vsock_poll,
1079         .ioctl = sock_no_ioctl,
1080         .listen = sock_no_listen,
1081         .shutdown = vsock_shutdown,
1082         .setsockopt = sock_no_setsockopt,
1083         .getsockopt = sock_no_getsockopt,
1084         .sendmsg = vsock_dgram_sendmsg,
1085         .recvmsg = vsock_dgram_recvmsg,
1086         .mmap = sock_no_mmap,
1087         .sendpage = sock_no_sendpage,
1088 };
1089
1090 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1091 {
1092         if (!transport->cancel_pkt)
1093                 return -EOPNOTSUPP;
1094
1095         return transport->cancel_pkt(vsk);
1096 }
1097
1098 static void vsock_connect_timeout(struct work_struct *work)
1099 {
1100         struct sock *sk;
1101         struct vsock_sock *vsk;
1102         int cancel = 0;
1103
1104         vsk = container_of(work, struct vsock_sock, connect_work.work);
1105         sk = sk_vsock(vsk);
1106
1107         lock_sock(sk);
1108         if (sk->sk_state == TCP_SYN_SENT &&
1109             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1110                 sk->sk_state = TCP_CLOSE;
1111                 sk->sk_err = ETIMEDOUT;
1112                 sk->sk_error_report(sk);
1113                 cancel = 1;
1114         }
1115         release_sock(sk);
1116         if (cancel)
1117                 vsock_transport_cancel_pkt(vsk);
1118
1119         sock_put(sk);
1120 }
1121
1122 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1123                                 int addr_len, int flags)
1124 {
1125         int err;
1126         struct sock *sk;
1127         struct vsock_sock *vsk;
1128         struct sockaddr_vm *remote_addr;
1129         long timeout;
1130         DEFINE_WAIT(wait);
1131
1132         err = 0;
1133         sk = sock->sk;
1134         vsk = vsock_sk(sk);
1135
1136         lock_sock(sk);
1137
1138         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1139         switch (sock->state) {
1140         case SS_CONNECTED:
1141                 err = -EISCONN;
1142                 goto out;
1143         case SS_DISCONNECTING:
1144                 err = -EINVAL;
1145                 goto out;
1146         case SS_CONNECTING:
1147                 /* This continues on so we can move sock into the SS_CONNECTED
1148                  * state once the connection has completed (at which point err
1149                  * will be set to zero also).  Otherwise, we will either wait
1150                  * for the connection or return -EALREADY should this be a
1151                  * non-blocking call.
1152                  */
1153                 err = -EALREADY;
1154                 break;
1155         default:
1156                 if ((sk->sk_state == TCP_LISTEN) ||
1157                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1158                         err = -EINVAL;
1159                         goto out;
1160                 }
1161
1162                 /* The hypervisor and well-known contexts do not have socket
1163                  * endpoints.
1164                  */
1165                 if (!transport->stream_allow(remote_addr->svm_cid,
1166                                              remote_addr->svm_port)) {
1167                         err = -ENETUNREACH;
1168                         goto out;
1169                 }
1170
1171                 /* Set the remote address that we are connecting to. */
1172                 memcpy(&vsk->remote_addr, remote_addr,
1173                        sizeof(vsk->remote_addr));
1174
1175                 err = vsock_auto_bind(vsk);
1176                 if (err)
1177                         goto out;
1178
1179                 sk->sk_state = TCP_SYN_SENT;
1180
1181                 err = transport->connect(vsk);
1182                 if (err < 0)
1183                         goto out;
1184
1185                 /* Mark sock as connecting and set the error code to in
1186                  * progress in case this is a non-blocking connect.
1187                  */
1188                 sock->state = SS_CONNECTING;
1189                 err = -EINPROGRESS;
1190         }
1191
1192         /* The receive path will handle all communication until we are able to
1193          * enter the connected state.  Here we wait for the connection to be
1194          * completed or a notification of an error.
1195          */
1196         timeout = vsk->connect_timeout;
1197         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1198
1199         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1200                 if (flags & O_NONBLOCK) {
1201                         /* If we're not going to block, we schedule a timeout
1202                          * function to generate a timeout on the connection
1203                          * attempt, in case the peer doesn't respond in a
1204                          * timely manner. We hold on to the socket until the
1205                          * timeout fires.
1206                          */
1207                         sock_hold(sk);
1208                         schedule_delayed_work(&vsk->connect_work, timeout);
1209
1210                         /* Skip ahead to preserve error code set above. */
1211                         goto out_wait;
1212                 }
1213
1214                 release_sock(sk);
1215                 timeout = schedule_timeout(timeout);
1216                 lock_sock(sk);
1217
1218                 if (signal_pending(current)) {
1219                         err = sock_intr_errno(timeout);
1220                         sk->sk_state = TCP_CLOSE;
1221                         sock->state = SS_UNCONNECTED;
1222                         vsock_transport_cancel_pkt(vsk);
1223                         goto out_wait;
1224                 } else if (timeout == 0) {
1225                         err = -ETIMEDOUT;
1226                         sk->sk_state = TCP_CLOSE;
1227                         sock->state = SS_UNCONNECTED;
1228                         vsock_transport_cancel_pkt(vsk);
1229                         goto out_wait;
1230                 }
1231
1232                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1233         }
1234
1235         if (sk->sk_err) {
1236                 err = -sk->sk_err;
1237                 sk->sk_state = TCP_CLOSE;
1238                 sock->state = SS_UNCONNECTED;
1239         } else {
1240                 err = 0;
1241         }
1242
1243 out_wait:
1244         finish_wait(sk_sleep(sk), &wait);
1245 out:
1246         release_sock(sk);
1247         return err;
1248 }
1249
1250 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1251                         bool kern)
1252 {
1253         struct sock *listener;
1254         int err;
1255         struct sock *connected;
1256         struct vsock_sock *vconnected;
1257         long timeout;
1258         DEFINE_WAIT(wait);
1259
1260         err = 0;
1261         listener = sock->sk;
1262
1263         lock_sock(listener);
1264
1265         if (sock->type != SOCK_STREAM) {
1266                 err = -EOPNOTSUPP;
1267                 goto out;
1268         }
1269
1270         if (listener->sk_state != TCP_LISTEN) {
1271                 err = -EINVAL;
1272                 goto out;
1273         }
1274
1275         /* Wait for children sockets to appear; these are the new sockets
1276          * created upon connection establishment.
1277          */
1278         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1279         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1280
1281         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1282                listener->sk_err == 0) {
1283                 release_sock(listener);
1284                 timeout = schedule_timeout(timeout);
1285                 finish_wait(sk_sleep(listener), &wait);
1286                 lock_sock(listener);
1287
1288                 if (signal_pending(current)) {
1289                         err = sock_intr_errno(timeout);
1290                         goto out;
1291                 } else if (timeout == 0) {
1292                         err = -EAGAIN;
1293                         goto out;
1294                 }
1295
1296                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1297         }
1298         finish_wait(sk_sleep(listener), &wait);
1299
1300         if (listener->sk_err)
1301                 err = -listener->sk_err;
1302
1303         if (connected) {
1304                 listener->sk_ack_backlog--;
1305
1306                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1307                 vconnected = vsock_sk(connected);
1308
1309                 /* If the listener socket has received an error, then we should
1310                  * reject this socket and return.  Note that we simply mark the
1311                  * socket rejected, drop our reference, and let the cleanup
1312                  * function handle the cleanup; the fact that we found it in
1313                  * the listener's accept queue guarantees that the cleanup
1314                  * function hasn't run yet.
1315                  */
1316                 if (err) {
1317                         vconnected->rejected = true;
1318                 } else {
1319                         newsock->state = SS_CONNECTED;
1320                         sock_graft(connected, newsock);
1321                 }
1322
1323                 release_sock(connected);
1324                 sock_put(connected);
1325         }
1326
1327 out:
1328         release_sock(listener);
1329         return err;
1330 }
1331
1332 static int vsock_listen(struct socket *sock, int backlog)
1333 {
1334         int err;
1335         struct sock *sk;
1336         struct vsock_sock *vsk;
1337
1338         sk = sock->sk;
1339
1340         lock_sock(sk);
1341
1342         if (sock->type != SOCK_STREAM) {
1343                 err = -EOPNOTSUPP;
1344                 goto out;
1345         }
1346
1347         if (sock->state != SS_UNCONNECTED) {
1348                 err = -EINVAL;
1349                 goto out;
1350         }
1351
1352         vsk = vsock_sk(sk);
1353
1354         if (!vsock_addr_bound(&vsk->local_addr)) {
1355                 err = -EINVAL;
1356                 goto out;
1357         }
1358
1359         sk->sk_max_ack_backlog = backlog;
1360         sk->sk_state = TCP_LISTEN;
1361
1362         err = 0;
1363
1364 out:
1365         release_sock(sk);
1366         return err;
1367 }
1368
1369 static int vsock_stream_setsockopt(struct socket *sock,
1370                                    int level,
1371                                    int optname,
1372                                    char __user *optval,
1373                                    unsigned int optlen)
1374 {
1375         int err;
1376         struct sock *sk;
1377         struct vsock_sock *vsk;
1378         u64 val;
1379
1380         if (level != AF_VSOCK)
1381                 return -ENOPROTOOPT;
1382
1383 #define COPY_IN(_v)                                       \
1384         do {                                              \
1385                 if (optlen < sizeof(_v)) {                \
1386                         err = -EINVAL;                    \
1387                         goto exit;                        \
1388                 }                                         \
1389                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1390                         err = -EFAULT;                                  \
1391                         goto exit;                                      \
1392                 }                                                       \
1393         } while (0)
1394
1395         err = 0;
1396         sk = sock->sk;
1397         vsk = vsock_sk(sk);
1398
1399         lock_sock(sk);
1400
1401         switch (optname) {
1402         case SO_VM_SOCKETS_BUFFER_SIZE:
1403                 COPY_IN(val);
1404                 transport->set_buffer_size(vsk, val);
1405                 break;
1406
1407         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1408                 COPY_IN(val);
1409                 transport->set_max_buffer_size(vsk, val);
1410                 break;
1411
1412         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1413                 COPY_IN(val);
1414                 transport->set_min_buffer_size(vsk, val);
1415                 break;
1416
1417         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1418                 struct __kernel_old_timeval tv;
1419                 COPY_IN(tv);
1420                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1421                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1422                         vsk->connect_timeout = tv.tv_sec * HZ +
1423                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1424                         if (vsk->connect_timeout == 0)
1425                                 vsk->connect_timeout =
1426                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1427
1428                 } else {
1429                         err = -ERANGE;
1430                 }
1431                 break;
1432         }
1433
1434         default:
1435                 err = -ENOPROTOOPT;
1436                 break;
1437         }
1438
1439 #undef COPY_IN
1440
1441 exit:
1442         release_sock(sk);
1443         return err;
1444 }
1445
1446 static int vsock_stream_getsockopt(struct socket *sock,
1447                                    int level, int optname,
1448                                    char __user *optval,
1449                                    int __user *optlen)
1450 {
1451         int err;
1452         int len;
1453         struct sock *sk;
1454         struct vsock_sock *vsk;
1455         u64 val;
1456
1457         if (level != AF_VSOCK)
1458                 return -ENOPROTOOPT;
1459
1460         err = get_user(len, optlen);
1461         if (err != 0)
1462                 return err;
1463
1464 #define COPY_OUT(_v)                            \
1465         do {                                    \
1466                 if (len < sizeof(_v))           \
1467                         return -EINVAL;         \
1468                                                 \
1469                 len = sizeof(_v);               \
1470                 if (copy_to_user(optval, &_v, len) != 0)        \
1471                         return -EFAULT;                         \
1472                                                                 \
1473         } while (0)
1474
1475         err = 0;
1476         sk = sock->sk;
1477         vsk = vsock_sk(sk);
1478
1479         switch (optname) {
1480         case SO_VM_SOCKETS_BUFFER_SIZE:
1481                 val = transport->get_buffer_size(vsk);
1482                 COPY_OUT(val);
1483                 break;
1484
1485         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1486                 val = transport->get_max_buffer_size(vsk);
1487                 COPY_OUT(val);
1488                 break;
1489
1490         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1491                 val = transport->get_min_buffer_size(vsk);
1492                 COPY_OUT(val);
1493                 break;
1494
1495         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1496                 struct __kernel_old_timeval tv;
1497                 tv.tv_sec = vsk->connect_timeout / HZ;
1498                 tv.tv_usec =
1499                     (vsk->connect_timeout -
1500                      tv.tv_sec * HZ) * (1000000 / HZ);
1501                 COPY_OUT(tv);
1502                 break;
1503         }
1504         default:
1505                 return -ENOPROTOOPT;
1506         }
1507
1508         err = put_user(len, optlen);
1509         if (err != 0)
1510                 return -EFAULT;
1511
1512 #undef COPY_OUT
1513
1514         return 0;
1515 }
1516
1517 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1518                                 size_t len)
1519 {
1520         struct sock *sk;
1521         struct vsock_sock *vsk;
1522         ssize_t total_written;
1523         long timeout;
1524         int err;
1525         struct vsock_transport_send_notify_data send_data;
1526         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1527
1528         sk = sock->sk;
1529         vsk = vsock_sk(sk);
1530         total_written = 0;
1531         err = 0;
1532
1533         if (msg->msg_flags & MSG_OOB)
1534                 return -EOPNOTSUPP;
1535
1536         lock_sock(sk);
1537
1538         /* Callers should not provide a destination with stream sockets. */
1539         if (msg->msg_namelen) {
1540                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1541                 goto out;
1542         }
1543
1544         /* Send data only if both sides are not shutdown in the direction. */
1545         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1546             vsk->peer_shutdown & RCV_SHUTDOWN) {
1547                 err = -EPIPE;
1548                 goto out;
1549         }
1550
1551         if (sk->sk_state != TCP_ESTABLISHED ||
1552             !vsock_addr_bound(&vsk->local_addr)) {
1553                 err = -ENOTCONN;
1554                 goto out;
1555         }
1556
1557         if (!vsock_addr_bound(&vsk->remote_addr)) {
1558                 err = -EDESTADDRREQ;
1559                 goto out;
1560         }
1561
1562         /* Wait for room in the produce queue to enqueue our user's data. */
1563         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1564
1565         err = transport->notify_send_init(vsk, &send_data);
1566         if (err < 0)
1567                 goto out;
1568
1569         while (total_written < len) {
1570                 ssize_t written;
1571
1572                 add_wait_queue(sk_sleep(sk), &wait);
1573                 while (vsock_stream_has_space(vsk) == 0 &&
1574                        sk->sk_err == 0 &&
1575                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1576                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1577
1578                         /* Don't wait for non-blocking sockets. */
1579                         if (timeout == 0) {
1580                                 err = -EAGAIN;
1581                                 remove_wait_queue(sk_sleep(sk), &wait);
1582                                 goto out_err;
1583                         }
1584
1585                         err = transport->notify_send_pre_block(vsk, &send_data);
1586                         if (err < 0) {
1587                                 remove_wait_queue(sk_sleep(sk), &wait);
1588                                 goto out_err;
1589                         }
1590
1591                         release_sock(sk);
1592                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1593                         lock_sock(sk);
1594                         if (signal_pending(current)) {
1595                                 err = sock_intr_errno(timeout);
1596                                 remove_wait_queue(sk_sleep(sk), &wait);
1597                                 goto out_err;
1598                         } else if (timeout == 0) {
1599                                 err = -EAGAIN;
1600                                 remove_wait_queue(sk_sleep(sk), &wait);
1601                                 goto out_err;
1602                         }
1603                 }
1604                 remove_wait_queue(sk_sleep(sk), &wait);
1605
1606                 /* These checks occur both as part of and after the loop
1607                  * conditional since we need to check before and after
1608                  * sleeping.
1609                  */
1610                 if (sk->sk_err) {
1611                         err = -sk->sk_err;
1612                         goto out_err;
1613                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1614                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1615                         err = -EPIPE;
1616                         goto out_err;
1617                 }
1618
1619                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1620                 if (err < 0)
1621                         goto out_err;
1622
1623                 /* Note that enqueue will only write as many bytes as are free
1624                  * in the produce queue, so we don't need to ensure len is
1625                  * smaller than the queue size.  It is the caller's
1626                  * responsibility to check how many bytes we were able to send.
1627                  */
1628
1629                 written = transport->stream_enqueue(
1630                                 vsk, msg,
1631                                 len - total_written);
1632                 if (written < 0) {
1633                         err = -ENOMEM;
1634                         goto out_err;
1635                 }
1636
1637                 total_written += written;
1638
1639                 err = transport->notify_send_post_enqueue(
1640                                 vsk, written, &send_data);
1641                 if (err < 0)
1642                         goto out_err;
1643
1644         }
1645
1646 out_err:
1647         if (total_written > 0)
1648                 err = total_written;
1649 out:
1650         release_sock(sk);
1651         return err;
1652 }
1653
1654
1655 static int
1656 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1657                      int flags)
1658 {
1659         struct sock *sk;
1660         struct vsock_sock *vsk;
1661         int err;
1662         size_t target;
1663         ssize_t copied;
1664         long timeout;
1665         struct vsock_transport_recv_notify_data recv_data;
1666
1667         DEFINE_WAIT(wait);
1668
1669         sk = sock->sk;
1670         vsk = vsock_sk(sk);
1671         err = 0;
1672
1673         lock_sock(sk);
1674
1675         if (sk->sk_state != TCP_ESTABLISHED) {
1676                 /* Recvmsg is supposed to return 0 if a peer performs an
1677                  * orderly shutdown. Differentiate between that case and when a
1678                  * peer has not connected or a local shutdown occured with the
1679                  * SOCK_DONE flag.
1680                  */
1681                 if (sock_flag(sk, SOCK_DONE))
1682                         err = 0;
1683                 else
1684                         err = -ENOTCONN;
1685
1686                 goto out;
1687         }
1688
1689         if (flags & MSG_OOB) {
1690                 err = -EOPNOTSUPP;
1691                 goto out;
1692         }
1693
1694         /* We don't check peer_shutdown flag here since peer may actually shut
1695          * down, but there can be data in the queue that a local socket can
1696          * receive.
1697          */
1698         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1699                 err = 0;
1700                 goto out;
1701         }
1702
1703         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1704          * is not an error.  We may as well bail out now.
1705          */
1706         if (!len) {
1707                 err = 0;
1708                 goto out;
1709         }
1710
1711         /* We must not copy less than target bytes into the user's buffer
1712          * before returning successfully, so we wait for the consume queue to
1713          * have that much data to consume before dequeueing.  Note that this
1714          * makes it impossible to handle cases where target is greater than the
1715          * queue size.
1716          */
1717         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1718         if (target >= transport->stream_rcvhiwat(vsk)) {
1719                 err = -ENOMEM;
1720                 goto out;
1721         }
1722         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1723         copied = 0;
1724
1725         err = transport->notify_recv_init(vsk, target, &recv_data);
1726         if (err < 0)
1727                 goto out;
1728
1729
1730         while (1) {
1731                 s64 ready;
1732
1733                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1734                 ready = vsock_stream_has_data(vsk);
1735
1736                 if (ready == 0) {
1737                         if (sk->sk_err != 0 ||
1738                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1739                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1740                                 finish_wait(sk_sleep(sk), &wait);
1741                                 break;
1742                         }
1743                         /* Don't wait for non-blocking sockets. */
1744                         if (timeout == 0) {
1745                                 err = -EAGAIN;
1746                                 finish_wait(sk_sleep(sk), &wait);
1747                                 break;
1748                         }
1749
1750                         err = transport->notify_recv_pre_block(
1751                                         vsk, target, &recv_data);
1752                         if (err < 0) {
1753                                 finish_wait(sk_sleep(sk), &wait);
1754                                 break;
1755                         }
1756                         release_sock(sk);
1757                         timeout = schedule_timeout(timeout);
1758                         lock_sock(sk);
1759
1760                         if (signal_pending(current)) {
1761                                 err = sock_intr_errno(timeout);
1762                                 finish_wait(sk_sleep(sk), &wait);
1763                                 break;
1764                         } else if (timeout == 0) {
1765                                 err = -EAGAIN;
1766                                 finish_wait(sk_sleep(sk), &wait);
1767                                 break;
1768                         }
1769                 } else {
1770                         ssize_t read;
1771
1772                         finish_wait(sk_sleep(sk), &wait);
1773
1774                         if (ready < 0) {
1775                                 /* Invalid queue pair content. XXX This should
1776                                 * be changed to a connection reset in a later
1777                                 * change.
1778                                 */
1779
1780                                 err = -ENOMEM;
1781                                 goto out;
1782                         }
1783
1784                         err = transport->notify_recv_pre_dequeue(
1785                                         vsk, target, &recv_data);
1786                         if (err < 0)
1787                                 break;
1788
1789                         read = transport->stream_dequeue(
1790                                         vsk, msg,
1791                                         len - copied, flags);
1792                         if (read < 0) {
1793                                 err = -ENOMEM;
1794                                 break;
1795                         }
1796
1797                         copied += read;
1798
1799                         err = transport->notify_recv_post_dequeue(
1800                                         vsk, target, read,
1801                                         !(flags & MSG_PEEK), &recv_data);
1802                         if (err < 0)
1803                                 goto out;
1804
1805                         if (read >= target || flags & MSG_PEEK)
1806                                 break;
1807
1808                         target -= read;
1809                 }
1810         }
1811
1812         if (sk->sk_err)
1813                 err = -sk->sk_err;
1814         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1815                 err = 0;
1816
1817         if (copied > 0)
1818                 err = copied;
1819
1820 out:
1821         release_sock(sk);
1822         return err;
1823 }
1824
1825 static const struct proto_ops vsock_stream_ops = {
1826         .family = PF_VSOCK,
1827         .owner = THIS_MODULE,
1828         .release = vsock_release,
1829         .bind = vsock_bind,
1830         .connect = vsock_stream_connect,
1831         .socketpair = sock_no_socketpair,
1832         .accept = vsock_accept,
1833         .getname = vsock_getname,
1834         .poll = vsock_poll,
1835         .ioctl = sock_no_ioctl,
1836         .listen = vsock_listen,
1837         .shutdown = vsock_shutdown,
1838         .setsockopt = vsock_stream_setsockopt,
1839         .getsockopt = vsock_stream_getsockopt,
1840         .sendmsg = vsock_stream_sendmsg,
1841         .recvmsg = vsock_stream_recvmsg,
1842         .mmap = sock_no_mmap,
1843         .sendpage = sock_no_sendpage,
1844 };
1845
1846 static int vsock_create(struct net *net, struct socket *sock,
1847                         int protocol, int kern)
1848 {
1849         if (!sock)
1850                 return -EINVAL;
1851
1852         if (protocol && protocol != PF_VSOCK)
1853                 return -EPROTONOSUPPORT;
1854
1855         switch (sock->type) {
1856         case SOCK_DGRAM:
1857                 sock->ops = &vsock_dgram_ops;
1858                 break;
1859         case SOCK_STREAM:
1860                 sock->ops = &vsock_stream_ops;
1861                 break;
1862         default:
1863                 return -ESOCKTNOSUPPORT;
1864         }
1865
1866         sock->state = SS_UNCONNECTED;
1867
1868         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1869 }
1870
1871 static const struct net_proto_family vsock_family_ops = {
1872         .family = AF_VSOCK,
1873         .create = vsock_create,
1874         .owner = THIS_MODULE,
1875 };
1876
1877 static long vsock_dev_do_ioctl(struct file *filp,
1878                                unsigned int cmd, void __user *ptr)
1879 {
1880         u32 __user *p = ptr;
1881         int retval = 0;
1882
1883         switch (cmd) {
1884         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1885                 if (put_user(transport->get_local_cid(), p) != 0)
1886                         retval = -EFAULT;
1887                 break;
1888
1889         default:
1890                 pr_err("Unknown ioctl %d\n", cmd);
1891                 retval = -EINVAL;
1892         }
1893
1894         return retval;
1895 }
1896
1897 static long vsock_dev_ioctl(struct file *filp,
1898                             unsigned int cmd, unsigned long arg)
1899 {
1900         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1901 }
1902
1903 #ifdef CONFIG_COMPAT
1904 static long vsock_dev_compat_ioctl(struct file *filp,
1905                                    unsigned int cmd, unsigned long arg)
1906 {
1907         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1908 }
1909 #endif
1910
1911 static const struct file_operations vsock_device_ops = {
1912         .owner          = THIS_MODULE,
1913         .unlocked_ioctl = vsock_dev_ioctl,
1914 #ifdef CONFIG_COMPAT
1915         .compat_ioctl   = vsock_dev_compat_ioctl,
1916 #endif
1917         .open           = nonseekable_open,
1918 };
1919
1920 static struct miscdevice vsock_device = {
1921         .name           = "vsock",
1922         .fops           = &vsock_device_ops,
1923 };
1924
1925 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1926 {
1927         int err = mutex_lock_interruptible(&vsock_register_mutex);
1928
1929         if (err)
1930                 return err;
1931
1932         if (transport) {
1933                 err = -EBUSY;
1934                 goto err_busy;
1935         }
1936
1937         /* Transport must be the owner of the protocol so that it can't
1938          * unload while there are open sockets.
1939          */
1940         vsock_proto.owner = owner;
1941         transport = t;
1942
1943         vsock_device.minor = MISC_DYNAMIC_MINOR;
1944         err = misc_register(&vsock_device);
1945         if (err) {
1946                 pr_err("Failed to register misc device\n");
1947                 goto err_reset_transport;
1948         }
1949
1950         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1951         if (err) {
1952                 pr_err("Cannot register vsock protocol\n");
1953                 goto err_deregister_misc;
1954         }
1955
1956         err = sock_register(&vsock_family_ops);
1957         if (err) {
1958                 pr_err("could not register af_vsock (%d) address family: %d\n",
1959                        AF_VSOCK, err);
1960                 goto err_unregister_proto;
1961         }
1962
1963         mutex_unlock(&vsock_register_mutex);
1964         return 0;
1965
1966 err_unregister_proto:
1967         proto_unregister(&vsock_proto);
1968 err_deregister_misc:
1969         misc_deregister(&vsock_device);
1970 err_reset_transport:
1971         transport = NULL;
1972 err_busy:
1973         mutex_unlock(&vsock_register_mutex);
1974         return err;
1975 }
1976 EXPORT_SYMBOL_GPL(__vsock_core_init);
1977
1978 void vsock_core_exit(void)
1979 {
1980         mutex_lock(&vsock_register_mutex);
1981
1982         misc_deregister(&vsock_device);
1983         sock_unregister(AF_VSOCK);
1984         proto_unregister(&vsock_proto);
1985
1986         /* We do not want the assignment below re-ordered. */
1987         mb();
1988         transport = NULL;
1989
1990         mutex_unlock(&vsock_register_mutex);
1991 }
1992 EXPORT_SYMBOL_GPL(vsock_core_exit);
1993
1994 const struct vsock_transport *vsock_core_get_transport(void)
1995 {
1996         /* vsock_register_mutex not taken since only the transport uses this
1997          * function and only while registered.
1998          */
1999         return transport;
2000 }
2001 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2002
2003 static void __exit vsock_exit(void)
2004 {
2005         /* Do nothing.  This function makes this module removable. */
2006 }
2007
2008 module_init(vsock_init_tables);
2009 module_exit(vsock_exit);
2010
2011 MODULE_AUTHOR("VMware, Inc.");
2012 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2013 MODULE_VERSION("1.0.2.0-k");
2014 MODULE_LICENSE("GPL v2");