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)
 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                 transport->release(vsk);
 652
 653                 lock_sock(sk);
 654                 sock_orphan(sk);
 655                 sk->sk_shutdown = SHUTDOWN_MASK;
 656
 657                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 658                         kfree_skb(skb);
 659
 660                 /* Clean up any sockets that never were accepted. */
 661                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 662                         __vsock_release(pending);
 663                         sock_put(pending);
 664                 }
 665
 666                 release_sock(sk);
 667                 sock_put(sk);
 668         }
 669 }
 670
 671 static void vsock_sk_destruct(struct sock *sk)
 672 {
 673         struct vsock_sock *vsk = vsock_sk(sk);
 674
 675         transport->destruct(vsk);
 676
 677         /* When clearing these addresses, there's no need to set the family and
 678          * possibly register the address family with the kernel.
 679          */
 680         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 681         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 682
 683         put_cred(vsk->owner);
 684 }
 685
 686 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 687 {
 688         int err;
 689
 690         err = sock_queue_rcv_skb(sk, skb);
 691         if (err)
 692                 kfree_skb(skb);
 693
 694         return err;
 695 }
 696
 697 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 698 {
 699         return transport->stream_has_data(vsk);
 700 }
 701 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 702
 703 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 704 {
 705         return transport->stream_has_space(vsk);
 706 }
 707 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 708
 709 static int vsock_release(struct socket *sock)
 710 {
 711         __vsock_release(sock->sk);
 712         sock->sk = NULL;
 713         sock->state = SS_FREE;
 714
 715         return 0;
 716 }
 717
 718 static int
 719 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 720 {
 721         int err;
 722         struct sock *sk;
 723         struct sockaddr_vm *vm_addr;
 724
 725         sk = sock->sk;
 726
 727         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 728                 return -EINVAL;
 729
 730         lock_sock(sk);
 731         err = __vsock_bind(sk, vm_addr);
 732         release_sock(sk);
 733
 734         return err;
 735 }
 736
 737 static int vsock_getname(struct socket *sock,
 738                          struct sockaddr *addr, int peer)
 739 {
 740         int err;
 741         struct sock *sk;
 742         struct vsock_sock *vsk;
 743         struct sockaddr_vm *vm_addr;
 744
 745         sk = sock->sk;
 746         vsk = vsock_sk(sk);
 747         err = 0;
 748
 749         lock_sock(sk);
 750
 751         if (peer) {
 752                 if (sock->state != SS_CONNECTED) {
 753                         err = -ENOTCONN;
 754                         goto out;
 755                 }
 756                 vm_addr = &vsk->remote_addr;
 757         } else {
 758                 vm_addr = &vsk->local_addr;
 759         }
 760
 761         if (!vm_addr) {
 762                 err = -EINVAL;
 763                 goto out;
 764         }
 765
 766         /* sys_getsockname() and sys_getpeername() pass us a
 767          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 768          * that macro is defined in socket.c instead of .h, so we hardcode its
 769          * value here.
 770          */
 771         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 772         memcpy(addr, vm_addr, sizeof(*vm_addr));
 773         err = sizeof(*vm_addr);
 774
 775 out:
 776         release_sock(sk);
 777         return err;
 778 }
 779
 780 static int vsock_shutdown(struct socket *sock, int mode)
 781 {
 782         int err;
 783         struct sock *sk;
 784
 785         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 786          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 787          * here like the other address families do.  Note also that the
 788          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 789          * which is what we want.
 790          */
 791         mode++;
 792
 793         if ((mode & ~SHUTDOWN_MASK) || !mode)
 794                 return -EINVAL;
 795
 796         /* If this is a STREAM socket and it is not connected then bail out
 797          * immediately.  If it is a DGRAM socket then we must first kick the
 798          * socket so that it wakes up from any sleeping calls, for example
 799          * recv(), and then afterwards return the error.
 800          */
 801
 802         sk = sock->sk;
 803         if (sock->state == SS_UNCONNECTED) {
 804                 err = -ENOTCONN;
 805                 if (sk->sk_type == SOCK_STREAM)
 806                         return err;
 807         } else {
 808                 sock->state = SS_DISCONNECTING;
 809                 err = 0;
 810         }
 811
 812         /* Receive and send shutdowns are treated alike. */
 813         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 814         if (mode) {
 815                 lock_sock(sk);
 816                 sk->sk_shutdown |= mode;
 817                 sk->sk_state_change(sk);
 818                 release_sock(sk);
 819
 820                 if (sk->sk_type == SOCK_STREAM) {
 821                         sock_reset_flag(sk, SOCK_DONE);
 822                         vsock_send_shutdown(sk, mode);
 823                 }
 824         }
 825
 826         return err;
 827 }
 828
 829 static __poll_t vsock_poll(struct file *file, struct socket *sock,
 830                                poll_table *wait)
 831 {
 832         struct sock *sk;
 833         __poll_t mask;
 834         struct vsock_sock *vsk;
 835
 836         sk = sock->sk;
 837         vsk = vsock_sk(sk);
 838
 839         poll_wait(file, sk_sleep(sk), wait);
 840         mask = 0;
 841
 842         if (sk->sk_err)
 843                 /* Signify that there has been an error on this socket. */
 844                 mask |= EPOLLERR;
 845
 846         /* INET sockets treat local write shutdown and peer write shutdown as a
 847          * case of EPOLLHUP set.
 848          */
 849         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 850             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 851              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 852                 mask |= EPOLLHUP;
 853         }
 854
 855         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 856             vsk->peer_shutdown & SEND_SHUTDOWN) {
 857                 mask |= EPOLLRDHUP;
 858         }
 859
 860         if (sock->type == SOCK_DGRAM) {
 861                 /* For datagram sockets we can read if there is something in
 862                  * the queue and write as long as the socket isn't shutdown for
 863                  * sending.
 864                  */
 865                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
 866                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 867                         mask |= EPOLLIN | EPOLLRDNORM;
 868                 }
 869
 870                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 871                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
 872
 873         } else if (sock->type == SOCK_STREAM) {
 874                 lock_sock(sk);
 875
 876                 /* Listening sockets that have connections in their accept
 877                  * queue can be read.
 878                  */
 879                 if (sk->sk_state == TCP_LISTEN
 880                     && !vsock_is_accept_queue_empty(sk))
 881                         mask |= EPOLLIN | EPOLLRDNORM;
 882
 883                 /* If there is something in the queue then we can read. */
 884                 if (transport->stream_is_active(vsk) &&
 885                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
 886                         bool data_ready_now = false;
 887                         int ret = transport->notify_poll_in(
 888                                         vsk, 1, &data_ready_now);
 889                         if (ret < 0) {
 890                                 mask |= EPOLLERR;
 891                         } else {
 892                                 if (data_ready_now)
 893                                         mask |= EPOLLIN | EPOLLRDNORM;
 894
 895                         }
 896                 }
 897
 898                 /* Sockets whose connections have been closed, reset, or
 899                  * terminated should also be considered read, and we check the
 900                  * shutdown flag for that.
 901                  */
 902                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
 903                     vsk->peer_shutdown & SEND_SHUTDOWN) {
 904                         mask |= EPOLLIN | EPOLLRDNORM;
 905                 }
 906
 907                 /* Connected sockets that can produce data can be written. */
 908                 if (sk->sk_state == TCP_ESTABLISHED) {
 909                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 910                                 bool space_avail_now = false;
 911                                 int ret = transport->notify_poll_out(
 912                                                 vsk, 1, &space_avail_now);
 913                                 if (ret < 0) {
 914                                         mask |= EPOLLERR;
 915                                 } else {
 916                                         if (space_avail_now)
 917                                                 /* Remove EPOLLWRBAND since INET
 918                                                  * sockets are not setting it.
 919                                                  */
 920                                                 mask |= EPOLLOUT | EPOLLWRNORM;
 921
 922                                 }
 923                         }
 924                 }
 925
 926                 /* Simulate INET socket poll behaviors, which sets
 927                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
 928                  * but local send is not shutdown.
 929                  */
 930                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
 931                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 932                                 mask |= EPOLLOUT | EPOLLWRNORM;
 933
 934                 }
 935
 936                 release_sock(sk);
 937         }
 938
 939         return mask;
 940 }
 941
 942 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 943                                size_t len)
 944 {
 945         int err;
 946         struct sock *sk;
 947         struct vsock_sock *vsk;
 948         struct sockaddr_vm *remote_addr;
 949
 950         if (msg->msg_flags & MSG_OOB)
 951                 return -EOPNOTSUPP;
 952
 953         /* For now, MSG_DONTWAIT is always assumed... */
 954         err = 0;
 955         sk = sock->sk;
 956         vsk = vsock_sk(sk);
 957
 958         lock_sock(sk);
 959
 960         err = vsock_auto_bind(vsk);
 961         if (err)
 962                 goto out;
 963
 964
 965         /* If the provided message contains an address, use that.  Otherwise
 966          * fall back on the socket's remote handle (if it has been connected).
 967          */
 968         if (msg->msg_name &&
 969             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
 970                             &remote_addr) == 0) {
 971                 /* Ensure this address is of the right type and is a valid
 972                  * destination.
 973                  */
 974
 975                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 976                         remote_addr->svm_cid = transport->get_local_cid();
 977
 978                 if (!vsock_addr_bound(remote_addr)) {
 979                         err = -EINVAL;
 980                         goto out;
 981                 }
 982         } else if (sock->state == SS_CONNECTED) {
 983                 remote_addr = &vsk->remote_addr;
 984
 985                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
 986                         remote_addr->svm_cid = transport->get_local_cid();
 987
 988                 /* XXX Should connect() or this function ensure remote_addr is
 989                  * bound?
 990                  */
 991                 if (!vsock_addr_bound(&vsk->remote_addr)) {
 992                         err = -EINVAL;
 993                         goto out;
 994                 }
 995         } else {
 996                 err = -EINVAL;
 997                 goto out;
 998         }
 999
1000         if (!transport->dgram_allow(remote_addr->svm_cid,
1001                                     remote_addr->svm_port)) {
1002                 err = -EINVAL;
1003                 goto out;
1004         }
1005
1006         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1007
1008 out:
1009         release_sock(sk);
1010         return err;
1011 }
1012
1013 static int vsock_dgram_connect(struct socket *sock,
1014                                struct sockaddr *addr, int addr_len, int flags)
1015 {
1016         int err;
1017         struct sock *sk;
1018         struct vsock_sock *vsk;
1019         struct sockaddr_vm *remote_addr;
1020
1021         sk = sock->sk;
1022         vsk = vsock_sk(sk);
1023
1024         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1025         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1026                 lock_sock(sk);
1027                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1028                                 VMADDR_PORT_ANY);
1029                 sock->state = SS_UNCONNECTED;
1030                 release_sock(sk);
1031                 return 0;
1032         } else if (err != 0)
1033                 return -EINVAL;
1034
1035         lock_sock(sk);
1036
1037         err = vsock_auto_bind(vsk);
1038         if (err)
1039                 goto out;
1040
1041         if (!transport->dgram_allow(remote_addr->svm_cid,
1042                                     remote_addr->svm_port)) {
1043                 err = -EINVAL;
1044                 goto out;
1045         }
1046
1047         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1048         sock->state = SS_CONNECTED;
1049
1050 out:
1051         release_sock(sk);
1052         return err;
1053 }
1054
1055 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1056                                size_t len, int flags)
1057 {
1058         return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags);
1059 }
1060
1061 static const struct proto_ops vsock_dgram_ops = {
1062         .family = PF_VSOCK,
1063         .owner = THIS_MODULE,
1064         .release = vsock_release,
1065         .bind = vsock_bind,
1066         .connect = vsock_dgram_connect,
1067         .socketpair = sock_no_socketpair,
1068         .accept = sock_no_accept,
1069         .getname = vsock_getname,
1070         .poll = vsock_poll,
1071         .ioctl = sock_no_ioctl,
1072         .listen = sock_no_listen,
1073         .shutdown = vsock_shutdown,
1074         .setsockopt = sock_no_setsockopt,
1075         .getsockopt = sock_no_getsockopt,
1076         .sendmsg = vsock_dgram_sendmsg,
1077         .recvmsg = vsock_dgram_recvmsg,
1078         .mmap = sock_no_mmap,
1079         .sendpage = sock_no_sendpage,
1080 };
1081
1082 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1083 {
1084         if (!transport->cancel_pkt)
1085                 return -EOPNOTSUPP;
1086
1087         return transport->cancel_pkt(vsk);
1088 }
1089
1090 static void vsock_connect_timeout(struct work_struct *work)
1091 {
1092         struct sock *sk;
1093         struct vsock_sock *vsk;
1094         int cancel = 0;
1095
1096         vsk = container_of(work, struct vsock_sock, connect_work.work);
1097         sk = sk_vsock(vsk);
1098
1099         lock_sock(sk);
1100         if (sk->sk_state == TCP_SYN_SENT &&
1101             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1102                 sk->sk_state = TCP_CLOSE;
1103                 sk->sk_err = ETIMEDOUT;
1104                 sk->sk_error_report(sk);
1105                 cancel = 1;
1106         }
1107         release_sock(sk);
1108         if (cancel)
1109                 vsock_transport_cancel_pkt(vsk);
1110
1111         sock_put(sk);
1112 }
1113
1114 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr,
1115                                 int addr_len, int flags)
1116 {
1117         int err;
1118         struct sock *sk;
1119         struct vsock_sock *vsk;
1120         struct sockaddr_vm *remote_addr;
1121         long timeout;
1122         DEFINE_WAIT(wait);
1123
1124         err = 0;
1125         sk = sock->sk;
1126         vsk = vsock_sk(sk);
1127
1128         lock_sock(sk);
1129
1130         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1131         switch (sock->state) {
1132         case SS_CONNECTED:
1133                 err = -EISCONN;
1134                 goto out;
1135         case SS_DISCONNECTING:
1136                 err = -EINVAL;
1137                 goto out;
1138         case SS_CONNECTING:
1139                 /* This continues on so we can move sock into the SS_CONNECTED
1140                  * state once the connection has completed (at which point err
1141                  * will be set to zero also).  Otherwise, we will either wait
1142                  * for the connection or return -EALREADY should this be a
1143                  * non-blocking call.
1144                  */
1145                 err = -EALREADY;
1146                 break;
1147         default:
1148                 if ((sk->sk_state == TCP_LISTEN) ||
1149                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1150                         err = -EINVAL;
1151                         goto out;
1152                 }
1153
1154                 /* The hypervisor and well-known contexts do not have socket
1155                  * endpoints.
1156                  */
1157                 if (!transport->stream_allow(remote_addr->svm_cid,
1158                                              remote_addr->svm_port)) {
1159                         err = -ENETUNREACH;
1160                         goto out;
1161                 }
1162
1163                 /* Set the remote address that we are connecting to. */
1164                 memcpy(&vsk->remote_addr, remote_addr,
1165                        sizeof(vsk->remote_addr));
1166
1167                 err = vsock_auto_bind(vsk);
1168                 if (err)
1169                         goto out;
1170
1171                 sk->sk_state = TCP_SYN_SENT;
1172
1173                 err = transport->connect(vsk);
1174                 if (err < 0)
1175                         goto out;
1176
1177                 /* Mark sock as connecting and set the error code to in
1178                  * progress in case this is a non-blocking connect.
1179                  */
1180                 sock->state = SS_CONNECTING;
1181                 err = -EINPROGRESS;
1182         }
1183
1184         /* The receive path will handle all communication until we are able to
1185          * enter the connected state.  Here we wait for the connection to be
1186          * completed or a notification of an error.
1187          */
1188         timeout = vsk->connect_timeout;
1189         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1190
1191         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1192                 if (flags & O_NONBLOCK) {
1193                         /* If we're not going to block, we schedule a timeout
1194                          * function to generate a timeout on the connection
1195                          * attempt, in case the peer doesn't respond in a
1196                          * timely manner. We hold on to the socket until the
1197                          * timeout fires.
1198                          */
1199                         sock_hold(sk);
1200                         schedule_delayed_work(&vsk->connect_work, timeout);
1201
1202                         /* Skip ahead to preserve error code set above. */
1203                         goto out_wait;
1204                 }
1205
1206                 release_sock(sk);
1207                 timeout = schedule_timeout(timeout);
1208                 lock_sock(sk);
1209
1210                 if (signal_pending(current)) {
1211                         err = sock_intr_errno(timeout);
1212                         sk->sk_state = TCP_CLOSE;
1213                         sock->state = SS_UNCONNECTED;
1214                         vsock_transport_cancel_pkt(vsk);
1215                         goto out_wait;
1216                 } else if (timeout == 0) {
1217                         err = -ETIMEDOUT;
1218                         sk->sk_state = TCP_CLOSE;
1219                         sock->state = SS_UNCONNECTED;
1220                         vsock_transport_cancel_pkt(vsk);
1221                         goto out_wait;
1222                 }
1223
1224                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1225         }
1226
1227         if (sk->sk_err) {
1228                 err = -sk->sk_err;
1229                 sk->sk_state = TCP_CLOSE;
1230                 sock->state = SS_UNCONNECTED;
1231         } else {
1232                 err = 0;
1233         }
1234
1235 out_wait:
1236         finish_wait(sk_sleep(sk), &wait);
1237 out:
1238         release_sock(sk);
1239         return err;
1240 }
1241
1242 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1243                         bool kern)
1244 {
1245         struct sock *listener;
1246         int err;
1247         struct sock *connected;
1248         struct vsock_sock *vconnected;
1249         long timeout;
1250         DEFINE_WAIT(wait);
1251
1252         err = 0;
1253         listener = sock->sk;
1254
1255         lock_sock(listener);
1256
1257         if (sock->type != SOCK_STREAM) {
1258                 err = -EOPNOTSUPP;
1259                 goto out;
1260         }
1261
1262         if (listener->sk_state != TCP_LISTEN) {
1263                 err = -EINVAL;
1264                 goto out;
1265         }
1266
1267         /* Wait for children sockets to appear; these are the new sockets
1268          * created upon connection establishment.
1269          */
1270         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1271         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1272
1273         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1274                listener->sk_err == 0) {
1275                 release_sock(listener);
1276                 timeout = schedule_timeout(timeout);
1277                 finish_wait(sk_sleep(listener), &wait);
1278                 lock_sock(listener);
1279
1280                 if (signal_pending(current)) {
1281                         err = sock_intr_errno(timeout);
1282                         goto out;
1283                 } else if (timeout == 0) {
1284                         err = -EAGAIN;
1285                         goto out;
1286                 }
1287
1288                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1289         }
1290         finish_wait(sk_sleep(listener), &wait);
1291
1292         if (listener->sk_err)
1293                 err = -listener->sk_err;
1294
1295         if (connected) {
1296                 listener->sk_ack_backlog--;
1297
1298                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1299                 vconnected = vsock_sk(connected);
1300
1301                 /* If the listener socket has received an error, then we should
1302                  * reject this socket and return.  Note that we simply mark the
1303                  * socket rejected, drop our reference, and let the cleanup
1304                  * function handle the cleanup; the fact that we found it in
1305                  * the listener's accept queue guarantees that the cleanup
1306                  * function hasn't run yet.
1307                  */
1308                 if (err) {
1309                         vconnected->rejected = true;
1310                 } else {
1311                         newsock->state = SS_CONNECTED;
1312                         sock_graft(connected, newsock);
1313                 }
1314
1315                 release_sock(connected);
1316                 sock_put(connected);
1317         }
1318
1319 out:
1320         release_sock(listener);
1321         return err;
1322 }
1323
1324 static int vsock_listen(struct socket *sock, int backlog)
1325 {
1326         int err;
1327         struct sock *sk;
1328         struct vsock_sock *vsk;
1329
1330         sk = sock->sk;
1331
1332         lock_sock(sk);
1333
1334         if (sock->type != SOCK_STREAM) {
1335                 err = -EOPNOTSUPP;
1336                 goto out;
1337         }
1338
1339         if (sock->state != SS_UNCONNECTED) {
1340                 err = -EINVAL;
1341                 goto out;
1342         }
1343
1344         vsk = vsock_sk(sk);
1345
1346         if (!vsock_addr_bound(&vsk->local_addr)) {
1347                 err = -EINVAL;
1348                 goto out;
1349         }
1350
1351         sk->sk_max_ack_backlog = backlog;
1352         sk->sk_state = TCP_LISTEN;
1353
1354         err = 0;
1355
1356 out:
1357         release_sock(sk);
1358         return err;
1359 }
1360
1361 static int vsock_stream_setsockopt(struct socket *sock,
1362                                    int level,
1363                                    int optname,
1364                                    char __user *optval,
1365                                    unsigned int optlen)
1366 {
1367         int err;
1368         struct sock *sk;
1369         struct vsock_sock *vsk;
1370         u64 val;
1371
1372         if (level != AF_VSOCK)
1373                 return -ENOPROTOOPT;
1374
1375 #define COPY_IN(_v)                                       \
1376         do {                                              \
1377                 if (optlen < sizeof(_v)) {                \
1378                         err = -EINVAL;                    \
1379                         goto exit;                        \
1380                 }                                         \
1381                 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) {     \
1382                         err = -EFAULT;                                  \
1383                         goto exit;                                      \
1384                 }                                                       \
1385         } while (0)
1386
1387         err = 0;
1388         sk = sock->sk;
1389         vsk = vsock_sk(sk);
1390
1391         lock_sock(sk);
1392
1393         switch (optname) {
1394         case SO_VM_SOCKETS_BUFFER_SIZE:
1395                 COPY_IN(val);
1396                 transport->set_buffer_size(vsk, val);
1397                 break;
1398
1399         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1400                 COPY_IN(val);
1401                 transport->set_max_buffer_size(vsk, val);
1402                 break;
1403
1404         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1405                 COPY_IN(val);
1406                 transport->set_min_buffer_size(vsk, val);
1407                 break;
1408
1409         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1410                 struct __kernel_old_timeval tv;
1411                 COPY_IN(tv);
1412                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1413                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1414                         vsk->connect_timeout = tv.tv_sec * HZ +
1415                             DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ));
1416                         if (vsk->connect_timeout == 0)
1417                                 vsk->connect_timeout =
1418                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1419
1420                 } else {
1421                         err = -ERANGE;
1422                 }
1423                 break;
1424         }
1425
1426         default:
1427                 err = -ENOPROTOOPT;
1428                 break;
1429         }
1430
1431 #undef COPY_IN
1432
1433 exit:
1434         release_sock(sk);
1435         return err;
1436 }
1437
1438 static int vsock_stream_getsockopt(struct socket *sock,
1439                                    int level, int optname,
1440                                    char __user *optval,
1441                                    int __user *optlen)
1442 {
1443         int err;
1444         int len;
1445         struct sock *sk;
1446         struct vsock_sock *vsk;
1447         u64 val;
1448
1449         if (level != AF_VSOCK)
1450                 return -ENOPROTOOPT;
1451
1452         err = get_user(len, optlen);
1453         if (err != 0)
1454                 return err;
1455
1456 #define COPY_OUT(_v)                            \
1457         do {                                    \
1458                 if (len < sizeof(_v))           \
1459                         return -EINVAL;         \
1460                                                 \
1461                 len = sizeof(_v);               \
1462                 if (copy_to_user(optval, &_v, len) != 0)        \
1463                         return -EFAULT;                         \
1464                                                                 \
1465         } while (0)
1466
1467         err = 0;
1468         sk = sock->sk;
1469         vsk = vsock_sk(sk);
1470
1471         switch (optname) {
1472         case SO_VM_SOCKETS_BUFFER_SIZE:
1473                 val = transport->get_buffer_size(vsk);
1474                 COPY_OUT(val);
1475                 break;
1476
1477         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1478                 val = transport->get_max_buffer_size(vsk);
1479                 COPY_OUT(val);
1480                 break;
1481
1482         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1483                 val = transport->get_min_buffer_size(vsk);
1484                 COPY_OUT(val);
1485                 break;
1486
1487         case SO_VM_SOCKETS_CONNECT_TIMEOUT: {
1488                 struct __kernel_old_timeval tv;
1489                 tv.tv_sec = vsk->connect_timeout / HZ;
1490                 tv.tv_usec =
1491                     (vsk->connect_timeout -
1492                      tv.tv_sec * HZ) * (1000000 / HZ);
1493                 COPY_OUT(tv);
1494                 break;
1495         }
1496         default:
1497                 return -ENOPROTOOPT;
1498         }
1499
1500         err = put_user(len, optlen);
1501         if (err != 0)
1502                 return -EFAULT;
1503
1504 #undef COPY_OUT
1505
1506         return 0;
1507 }
1508
1509 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1510                                 size_t len)
1511 {
1512         struct sock *sk;
1513         struct vsock_sock *vsk;
1514         ssize_t total_written;
1515         long timeout;
1516         int err;
1517         struct vsock_transport_send_notify_data send_data;
1518         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1519
1520         sk = sock->sk;
1521         vsk = vsock_sk(sk);
1522         total_written = 0;
1523         err = 0;
1524
1525         if (msg->msg_flags & MSG_OOB)
1526                 return -EOPNOTSUPP;
1527
1528         lock_sock(sk);
1529
1530         /* Callers should not provide a destination with stream sockets. */
1531         if (msg->msg_namelen) {
1532                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1533                 goto out;
1534         }
1535
1536         /* Send data only if both sides are not shutdown in the direction. */
1537         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1538             vsk->peer_shutdown & RCV_SHUTDOWN) {
1539                 err = -EPIPE;
1540                 goto out;
1541         }
1542
1543         if (sk->sk_state != TCP_ESTABLISHED ||
1544             !vsock_addr_bound(&vsk->local_addr)) {
1545                 err = -ENOTCONN;
1546                 goto out;
1547         }
1548
1549         if (!vsock_addr_bound(&vsk->remote_addr)) {
1550                 err = -EDESTADDRREQ;
1551                 goto out;
1552         }
1553
1554         /* Wait for room in the produce queue to enqueue our user's data. */
1555         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1556
1557         err = transport->notify_send_init(vsk, &send_data);
1558         if (err < 0)
1559                 goto out;
1560
1561         while (total_written < len) {
1562                 ssize_t written;
1563
1564                 add_wait_queue(sk_sleep(sk), &wait);
1565                 while (vsock_stream_has_space(vsk) == 0 &&
1566                        sk->sk_err == 0 &&
1567                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1568                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1569
1570                         /* Don't wait for non-blocking sockets. */
1571                         if (timeout == 0) {
1572                                 err = -EAGAIN;
1573                                 remove_wait_queue(sk_sleep(sk), &wait);
1574                                 goto out_err;
1575                         }
1576
1577                         err = transport->notify_send_pre_block(vsk, &send_data);
1578                         if (err < 0) {
1579                                 remove_wait_queue(sk_sleep(sk), &wait);
1580                                 goto out_err;
1581                         }
1582
1583                         release_sock(sk);
1584                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1585                         lock_sock(sk);
1586                         if (signal_pending(current)) {
1587                                 err = sock_intr_errno(timeout);
1588                                 remove_wait_queue(sk_sleep(sk), &wait);
1589                                 goto out_err;
1590                         } else if (timeout == 0) {
1591                                 err = -EAGAIN;
1592                                 remove_wait_queue(sk_sleep(sk), &wait);
1593                                 goto out_err;
1594                         }
1595                 }
1596                 remove_wait_queue(sk_sleep(sk), &wait);
1597
1598                 /* These checks occur both as part of and after the loop
1599                  * conditional since we need to check before and after
1600                  * sleeping.
1601                  */
1602                 if (sk->sk_err) {
1603                         err = -sk->sk_err;
1604                         goto out_err;
1605                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1606                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1607                         err = -EPIPE;
1608                         goto out_err;
1609                 }
1610
1611                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1612                 if (err < 0)
1613                         goto out_err;
1614
1615                 /* Note that enqueue will only write as many bytes as are free
1616                  * in the produce queue, so we don't need to ensure len is
1617                  * smaller than the queue size.  It is the caller's
1618                  * responsibility to check how many bytes we were able to send.
1619                  */
1620
1621                 written = transport->stream_enqueue(
1622                                 vsk, msg,
1623                                 len - total_written);
1624                 if (written < 0) {
1625                         err = -ENOMEM;
1626                         goto out_err;
1627                 }
1628
1629                 total_written += written;
1630
1631                 err = transport->notify_send_post_enqueue(
1632                                 vsk, written, &send_data);
1633                 if (err < 0)
1634                         goto out_err;
1635
1636         }
1637
1638 out_err:
1639         if (total_written > 0)
1640                 err = total_written;
1641 out:
1642         release_sock(sk);
1643         return err;
1644 }
1645
1646
1647 static int
1648 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1649                      int flags)
1650 {
1651         struct sock *sk;
1652         struct vsock_sock *vsk;
1653         int err;
1654         size_t target;
1655         ssize_t copied;
1656         long timeout;
1657         struct vsock_transport_recv_notify_data recv_data;
1658
1659         DEFINE_WAIT(wait);
1660
1661         sk = sock->sk;
1662         vsk = vsock_sk(sk);
1663         err = 0;
1664
1665         lock_sock(sk);
1666
1667         if (sk->sk_state != TCP_ESTABLISHED) {
1668                 /* Recvmsg is supposed to return 0 if a peer performs an
1669                  * orderly shutdown. Differentiate between that case and when a
1670                  * peer has not connected or a local shutdown occured with the
1671                  * SOCK_DONE flag.
1672                  */
1673                 if (sock_flag(sk, SOCK_DONE))
1674                         err = 0;
1675                 else
1676                         err = -ENOTCONN;
1677
1678                 goto out;
1679         }
1680
1681         if (flags & MSG_OOB) {
1682                 err = -EOPNOTSUPP;
1683                 goto out;
1684         }
1685
1686         /* We don't check peer_shutdown flag here since peer may actually shut
1687          * down, but there can be data in the queue that a local socket can
1688          * receive.
1689          */
1690         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1691                 err = 0;
1692                 goto out;
1693         }
1694
1695         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1696          * is not an error.  We may as well bail out now.
1697          */
1698         if (!len) {
1699                 err = 0;
1700                 goto out;
1701         }
1702
1703         /* We must not copy less than target bytes into the user's buffer
1704          * before returning successfully, so we wait for the consume queue to
1705          * have that much data to consume before dequeueing.  Note that this
1706          * makes it impossible to handle cases where target is greater than the
1707          * queue size.
1708          */
1709         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1710         if (target >= transport->stream_rcvhiwat(vsk)) {
1711                 err = -ENOMEM;
1712                 goto out;
1713         }
1714         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1715         copied = 0;
1716
1717         err = transport->notify_recv_init(vsk, target, &recv_data);
1718         if (err < 0)
1719                 goto out;
1720
1721
1722         while (1) {
1723                 s64 ready;
1724
1725                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1726                 ready = vsock_stream_has_data(vsk);
1727
1728                 if (ready == 0) {
1729                         if (sk->sk_err != 0 ||
1730                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1731                             (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1732                                 finish_wait(sk_sleep(sk), &wait);
1733                                 break;
1734                         }
1735                         /* Don't wait for non-blocking sockets. */
1736                         if (timeout == 0) {
1737                                 err = -EAGAIN;
1738                                 finish_wait(sk_sleep(sk), &wait);
1739                                 break;
1740                         }
1741
1742                         err = transport->notify_recv_pre_block(
1743                                         vsk, target, &recv_data);
1744                         if (err < 0) {
1745                                 finish_wait(sk_sleep(sk), &wait);
1746                                 break;
1747                         }
1748                         release_sock(sk);
1749                         timeout = schedule_timeout(timeout);
1750                         lock_sock(sk);
1751
1752                         if (signal_pending(current)) {
1753                                 err = sock_intr_errno(timeout);
1754                                 finish_wait(sk_sleep(sk), &wait);
1755                                 break;
1756                         } else if (timeout == 0) {
1757                                 err = -EAGAIN;
1758                                 finish_wait(sk_sleep(sk), &wait);
1759                                 break;
1760                         }
1761                 } else {
1762                         ssize_t read;
1763
1764                         finish_wait(sk_sleep(sk), &wait);
1765
1766                         if (ready < 0) {
1767                                 /* Invalid queue pair content. XXX This should
1768                                 * be changed to a connection reset in a later
1769                                 * change.
1770                                 */
1771
1772                                 err = -ENOMEM;
1773                                 goto out;
1774                         }
1775
1776                         err = transport->notify_recv_pre_dequeue(
1777                                         vsk, target, &recv_data);
1778                         if (err < 0)
1779                                 break;
1780
1781                         read = transport->stream_dequeue(
1782                                         vsk, msg,
1783                                         len - copied, flags);
1784                         if (read < 0) {
1785                                 err = -ENOMEM;
1786                                 break;
1787                         }
1788
1789                         copied += read;
1790
1791                         err = transport->notify_recv_post_dequeue(
1792                                         vsk, target, read,
1793                                         !(flags & MSG_PEEK), &recv_data);
1794                         if (err < 0)
1795                                 goto out;
1796
1797                         if (read >= target || flags & MSG_PEEK)
1798                                 break;
1799
1800                         target -= read;
1801                 }
1802         }
1803
1804         if (sk->sk_err)
1805                 err = -sk->sk_err;
1806         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1807                 err = 0;
1808
1809         if (copied > 0)
1810                 err = copied;
1811
1812 out:
1813         release_sock(sk);
1814         return err;
1815 }
1816
1817 static const struct proto_ops vsock_stream_ops = {
1818         .family = PF_VSOCK,
1819         .owner = THIS_MODULE,
1820         .release = vsock_release,
1821         .bind = vsock_bind,
1822         .connect = vsock_stream_connect,
1823         .socketpair = sock_no_socketpair,
1824         .accept = vsock_accept,
1825         .getname = vsock_getname,
1826         .poll = vsock_poll,
1827         .ioctl = sock_no_ioctl,
1828         .listen = vsock_listen,
1829         .shutdown = vsock_shutdown,
1830         .setsockopt = vsock_stream_setsockopt,
1831         .getsockopt = vsock_stream_getsockopt,
1832         .sendmsg = vsock_stream_sendmsg,
1833         .recvmsg = vsock_stream_recvmsg,
1834         .mmap = sock_no_mmap,
1835         .sendpage = sock_no_sendpage,
1836 };
1837
1838 static int vsock_create(struct net *net, struct socket *sock,
1839                         int protocol, int kern)
1840 {
1841         if (!sock)
1842                 return -EINVAL;
1843
1844         if (protocol && protocol != PF_VSOCK)
1845                 return -EPROTONOSUPPORT;
1846
1847         switch (sock->type) {
1848         case SOCK_DGRAM:
1849                 sock->ops = &vsock_dgram_ops;
1850                 break;
1851         case SOCK_STREAM:
1852                 sock->ops = &vsock_stream_ops;
1853                 break;
1854         default:
1855                 return -ESOCKTNOSUPPORT;
1856         }
1857
1858         sock->state = SS_UNCONNECTED;
1859
1860         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM;
1861 }
1862
1863 static const struct net_proto_family vsock_family_ops = {
1864         .family = AF_VSOCK,
1865         .create = vsock_create,
1866         .owner = THIS_MODULE,
1867 };
1868
1869 static long vsock_dev_do_ioctl(struct file *filp,
1870                                unsigned int cmd, void __user *ptr)
1871 {
1872         u32 __user *p = ptr;
1873         int retval = 0;
1874
1875         switch (cmd) {
1876         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1877                 if (put_user(transport->get_local_cid(), p) != 0)
1878                         retval = -EFAULT;
1879                 break;
1880
1881         default:
1882                 pr_err("Unknown ioctl %d\n", cmd);
1883                 retval = -EINVAL;
1884         }
1885
1886         return retval;
1887 }
1888
1889 static long vsock_dev_ioctl(struct file *filp,
1890                             unsigned int cmd, unsigned long arg)
1891 {
1892         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1893 }
1894
1895 #ifdef CONFIG_COMPAT
1896 static long vsock_dev_compat_ioctl(struct file *filp,
1897                                    unsigned int cmd, unsigned long arg)
1898 {
1899         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1900 }
1901 #endif
1902
1903 static const struct file_operations vsock_device_ops = {
1904         .owner          = THIS_MODULE,
1905         .unlocked_ioctl = vsock_dev_ioctl,
1906 #ifdef CONFIG_COMPAT
1907         .compat_ioctl   = vsock_dev_compat_ioctl,
1908 #endif
1909         .open           = nonseekable_open,
1910 };
1911
1912 static struct miscdevice vsock_device = {
1913         .name           = "vsock",
1914         .fops           = &vsock_device_ops,
1915 };
1916
1917 int __vsock_core_init(const struct vsock_transport *t, struct module *owner)
1918 {
1919         int err = mutex_lock_interruptible(&vsock_register_mutex);
1920
1921         if (err)
1922                 return err;
1923
1924         if (transport) {
1925                 err = -EBUSY;
1926                 goto err_busy;
1927         }
1928
1929         /* Transport must be the owner of the protocol so that it can't
1930          * unload while there are open sockets.
1931          */
1932         vsock_proto.owner = owner;
1933         transport = t;
1934
1935         vsock_device.minor = MISC_DYNAMIC_MINOR;
1936         err = misc_register(&vsock_device);
1937         if (err) {
1938                 pr_err("Failed to register misc device\n");
1939                 goto err_reset_transport;
1940         }
1941
1942         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1943         if (err) {
1944                 pr_err("Cannot register vsock protocol\n");
1945                 goto err_deregister_misc;
1946         }
1947
1948         err = sock_register(&vsock_family_ops);
1949         if (err) {
1950                 pr_err("could not register af_vsock (%d) address family: %d\n",
1951                        AF_VSOCK, err);
1952                 goto err_unregister_proto;
1953         }
1954
1955         mutex_unlock(&vsock_register_mutex);
1956         return 0;
1957
1958 err_unregister_proto:
1959         proto_unregister(&vsock_proto);
1960 err_deregister_misc:
1961         misc_deregister(&vsock_device);
1962 err_reset_transport:
1963         transport = NULL;
1964 err_busy:
1965         mutex_unlock(&vsock_register_mutex);
1966         return err;
1967 }
1968 EXPORT_SYMBOL_GPL(__vsock_core_init);
1969
1970 void vsock_core_exit(void)
1971 {
1972         mutex_lock(&vsock_register_mutex);
1973
1974         misc_deregister(&vsock_device);
1975         sock_unregister(AF_VSOCK);
1976         proto_unregister(&vsock_proto);
1977
1978         /* We do not want the assignment below re-ordered. */
1979         mb();
1980         transport = NULL;
1981
1982         mutex_unlock(&vsock_register_mutex);
1983 }
1984 EXPORT_SYMBOL_GPL(vsock_core_exit);
1985
1986 const struct vsock_transport *vsock_core_get_transport(void)
1987 {
1988         /* vsock_register_mutex not taken since only the transport uses this
1989          * function and only while registered.
1990          */
1991         return transport;
1992 }
1993 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
1994
1995 static void __exit vsock_exit(void)
1996 {
1997         /* Do nothing.  This function makes this module removable. */
1998 }
1999
2000 module_init(vsock_init_tables);
2001 module_exit(vsock_exit);
2002
2003 MODULE_AUTHOR("VMware, Inc.");
2004 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2005 MODULE_VERSION("1.0.2.0-k");
2006 MODULE_LICENSE("GPL v2");