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