Pileus Git - ~andy/linux/blob - net/vmw_vsock/af_vsock.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 /* Implementation notes:
  17  *
  18  * - There are two kinds of sockets: those created by user action (such as
  19  * calling socket(2)) and those created by incoming connection request packets.
  20  *
  21  * - There are two "global" tables, one for bound sockets (sockets that have
  22  * specified an address that they are responsible for) and one for connected
  23  * sockets (sockets that have established a connection with another socket).
  24  * These tables are "global" in that all sockets on the system are placed
  25  * within them. - Note, though, that the bound table contains an extra entry
  26  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  27  * that list. The bound table is used solely for lookup of sockets when packets
  28  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  29  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  30  * sockets out of the bound hash buckets will reduce the chance of collisions
  31  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  32  * socket type in the hash table lookups.
  33  *
  34  * - Sockets created by user action will either be "client" sockets that
  35  * initiate a connection or "server" sockets that listen for connections; we do
  36  * not support simultaneous connects (two "client" sockets connecting).
  37  *
  38  * - "Server" sockets are referred to as listener sockets throughout this
  39  * implementation because they are in the SS_LISTEN state.  When a connection
  40  * request is received (the second kind of socket mentioned above), we create a
  41  * new socket and refer to it as a pending socket.  These pending sockets are
  42  * placed on the pending connection list of the listener socket.  When future
  43  * packets are received for the address the listener socket is bound to, we
  44  * check if the source of the packet is from one that has an existing pending
  45  * connection.  If it does, we process the packet for the pending socket.  When
  46  * that socket reaches the connected state, it is removed from the listener
  47  * socket's pending list and enqueued in the listener socket's accept queue.
  48  * Callers of accept(2) will accept connected sockets from the listener socket's
  49  * accept queue.  If the socket cannot be accepted for some reason then it is
  50  * marked rejected.  Once the connection is accepted, it is owned by the user
  51  * process and the responsibility for cleanup falls with that user process.
  52  *
  53  * - It is possible that these pending sockets will never reach the connected
  54  * state; in fact, we may never receive another packet after the connection
  55  * request.  Because of this, we must schedule a cleanup function to run in the
  56  * future, after some amount of time passes where a connection should have been
  57  * established.  This function ensures that the socket is off all lists so it
  58  * cannot be retrieved, then drops all references to the socket so it is cleaned
  59  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  60  * function will also cleanup rejected sockets, those that reach the connected
  61  * state but leave it before they have been accepted.
  62  *
  63  * - Sockets created by user action will be cleaned up when the user process
  64  * calls close(2), causing our release implementation to be called. Our release
  65  * implementation will perform some cleanup then drop the last reference so our
  66  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  67  * perform additional cleanup that's common for both types of sockets.
  68  *
  69  * - A socket's reference count is what ensures that the structure won't be
  70  * freed.  Each entry in a list (such as the "global" bound and connected tables
  71  * and the listener socket's pending list and connected queue) ensures a
  72  * reference.  When we defer work until process context and pass a socket as our
  73  * argument, we must ensure the reference count is increased to ensure the
  74  * socket isn't freed before the function is run; the deferred function will
  75  * then drop the reference.
  76  */
  77
  78 #include <linux/types.h>
  79 #include <linux/bitops.h>
  80 #include <linux/cred.h>
  81 #include <linux/init.h>
  82 #include <linux/io.h>
  83 #include <linux/kernel.h>
  84 #include <linux/kmod.h>
  85 #include <linux/list.h>
  86 #include <linux/miscdevice.h>
  87 #include <linux/module.h>
  88 #include <linux/mutex.h>
  89 #include <linux/net.h>
  90 #include <linux/poll.h>
  91 #include <linux/skbuff.h>
  92 #include <linux/smp.h>
  93 #include <linux/socket.h>
  94 #include <linux/stddef.h>
  95 #include <linux/unistd.h>
  96 #include <linux/wait.h>
  97 #include <linux/workqueue.h>
  98 #include <net/sock.h>
  99
 100 #include "af_vsock.h"
 101
 102 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 103 static void vsock_sk_destruct(struct sock *sk);
 104 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 105
 106 /* Protocol family. */
 107 static struct proto vsock_proto = {
 108         .name = "AF_VSOCK",
 109         .owner = THIS_MODULE,
 110         .obj_size = sizeof(struct vsock_sock),
 111 };
 112
 113 /* The default peer timeout indicates how long we will wait for a peer response
 114  * to a control message.
 115  */
 116 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 117
 118 #define SS_LISTEN 255
 119
 120 static const struct vsock_transport *transport;
 121 static DEFINE_MUTEX(vsock_register_mutex);
 122
 123 /**** EXPORTS ****/
 124
 125 /* Get the ID of the local context.  This is transport dependent. */
 126
 127 int vm_sockets_get_local_cid(void)
 128 {
 129         return transport->get_local_cid();
 130 }
 131 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid);
 132
 133 /**** UTILS ****/
 134
 135 /* Each bound VSocket is stored in the bind hash table and each connected
 136  * VSocket is stored in the connected hash table.
 137  *
 138  * Unbound sockets are all put on the same list attached to the end of the hash
 139  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 140  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 141  * represents the list that addr hashes to).
 142  *
 143  * Specifically, we initialize the vsock_bind_table array to a size of
 144  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 145  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 146  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 147  * mods with VSOCK_HASH_SIZE - 1 to ensure this.
 148  */
 149 #define VSOCK_HASH_SIZE         251
 150 #define MAX_PORT_RETRIES        24
 151
 152 #define VSOCK_HASH(addr)        ((addr)->svm_port % (VSOCK_HASH_SIZE - 1))
 153 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 154 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 155
 156 /* XXX This can probably be implemented in a better way. */
 157 #define VSOCK_CONN_HASH(src, dst)                               \
 158         (((src)->svm_cid ^ (dst)->svm_port) % (VSOCK_HASH_SIZE - 1))
 159 #define vsock_connected_sockets(src, dst)               \
 160         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 161 #define vsock_connected_sockets_vsk(vsk)                                \
 162         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 163
 164 static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 165 static struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 166 static DEFINE_SPINLOCK(vsock_table_lock);
 167
 168 /* Autobind this socket to the local address if necessary. */
 169 static int vsock_auto_bind(struct vsock_sock *vsk)
 170 {
 171         struct sock *sk = sk_vsock(vsk);
 172         struct sockaddr_vm local_addr;
 173
 174         if (vsock_addr_bound(&vsk->local_addr))
 175                 return 0;
 176         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 177         return __vsock_bind(sk, &local_addr);
 178 }
 179
 180 static void vsock_init_tables(void)
 181 {
 182         int i;
 183
 184         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 185                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 186
 187         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 188                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 189 }
 190
 191 static void __vsock_insert_bound(struct list_head *list,
 192                                  struct vsock_sock *vsk)
 193 {
 194         sock_hold(&vsk->sk);
 195         list_add(&vsk->bound_table, list);
 196 }
 197
 198 static void __vsock_insert_connected(struct list_head *list,
 199                                      struct vsock_sock *vsk)
 200 {
 201         sock_hold(&vsk->sk);
 202         list_add(&vsk->connected_table, list);
 203 }
 204
 205 static void __vsock_remove_bound(struct vsock_sock *vsk)
 206 {
 207         list_del_init(&vsk->bound_table);
 208         sock_put(&vsk->sk);
 209 }
 210
 211 static void __vsock_remove_connected(struct vsock_sock *vsk)
 212 {
 213         list_del_init(&vsk->connected_table);
 214         sock_put(&vsk->sk);
 215 }
 216
 217 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 218 {
 219         struct vsock_sock *vsk;
 220
 221         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table)
 222                 if (addr->svm_port == vsk->local_addr.svm_port)
 223                         return sk_vsock(vsk);
 224
 225         return NULL;
 226 }
 227
 228 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 229                                                   struct sockaddr_vm *dst)
 230 {
 231         struct vsock_sock *vsk;
 232
 233         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 234                             connected_table) {
 235                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 236                     dst->svm_port == vsk->local_addr.svm_port) {
 237                         return sk_vsock(vsk);
 238                 }
 239         }
 240
 241         return NULL;
 242 }
 243
 244 static bool __vsock_in_bound_table(struct vsock_sock *vsk)
 245 {
 246         return !list_empty(&vsk->bound_table);
 247 }
 248
 249 static bool __vsock_in_connected_table(struct vsock_sock *vsk)
 250 {
 251         return !list_empty(&vsk->connected_table);
 252 }
 253
 254 static void vsock_insert_unbound(struct vsock_sock *vsk)
 255 {
 256         spin_lock_bh(&vsock_table_lock);
 257         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 258         spin_unlock_bh(&vsock_table_lock);
 259 }
 260
 261 void vsock_insert_connected(struct vsock_sock *vsk)
 262 {
 263         struct list_head *list = vsock_connected_sockets(
 264                 &vsk->remote_addr, &vsk->local_addr);
 265
 266         spin_lock_bh(&vsock_table_lock);
 267         __vsock_insert_connected(list, vsk);
 268         spin_unlock_bh(&vsock_table_lock);
 269 }
 270 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 271
 272 void vsock_remove_bound(struct vsock_sock *vsk)
 273 {
 274         spin_lock_bh(&vsock_table_lock);
 275         __vsock_remove_bound(vsk);
 276         spin_unlock_bh(&vsock_table_lock);
 277 }
 278 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 279
 280 void vsock_remove_connected(struct vsock_sock *vsk)
 281 {
 282         spin_lock_bh(&vsock_table_lock);
 283         __vsock_remove_connected(vsk);
 284         spin_unlock_bh(&vsock_table_lock);
 285 }
 286 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 287
 288 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 289 {
 290         struct sock *sk;
 291
 292         spin_lock_bh(&vsock_table_lock);
 293         sk = __vsock_find_bound_socket(addr);
 294         if (sk)
 295                 sock_hold(sk);
 296
 297         spin_unlock_bh(&vsock_table_lock);
 298
 299         return sk;
 300 }
 301 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 302
 303 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 304                                          struct sockaddr_vm *dst)
 305 {
 306         struct sock *sk;
 307
 308         spin_lock_bh(&vsock_table_lock);
 309         sk = __vsock_find_connected_socket(src, dst);
 310         if (sk)
 311                 sock_hold(sk);
 312
 313         spin_unlock_bh(&vsock_table_lock);
 314
 315         return sk;
 316 }
 317 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 318
 319 static bool vsock_in_bound_table(struct vsock_sock *vsk)
 320 {
 321         bool ret;
 322
 323         spin_lock_bh(&vsock_table_lock);
 324         ret = __vsock_in_bound_table(vsk);
 325         spin_unlock_bh(&vsock_table_lock);
 326
 327         return ret;
 328 }
 329
 330 static bool vsock_in_connected_table(struct vsock_sock *vsk)
 331 {
 332         bool ret;
 333
 334         spin_lock_bh(&vsock_table_lock);
 335         ret = __vsock_in_connected_table(vsk);
 336         spin_unlock_bh(&vsock_table_lock);
 337
 338         return ret;
 339 }
 340
 341 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 342 {
 343         int i;
 344
 345         spin_lock_bh(&vsock_table_lock);
 346
 347         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 348                 struct vsock_sock *vsk;
 349                 list_for_each_entry(vsk, &vsock_connected_table[i],
 350                                     connected_table);
 351                         fn(sk_vsock(vsk));
 352         }
 353
 354         spin_unlock_bh(&vsock_table_lock);
 355 }
 356 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 357
 358 void vsock_add_pending(struct sock *listener, struct sock *pending)
 359 {
 360         struct vsock_sock *vlistener;
 361         struct vsock_sock *vpending;
 362
 363         vlistener = vsock_sk(listener);
 364         vpending = vsock_sk(pending);
 365
 366         sock_hold(pending);
 367         sock_hold(listener);
 368         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 369 }
 370 EXPORT_SYMBOL_GPL(vsock_add_pending);
 371
 372 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 373 {
 374         struct vsock_sock *vpending = vsock_sk(pending);
 375
 376         list_del_init(&vpending->pending_links);
 377         sock_put(listener);
 378         sock_put(pending);
 379 }
 380 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 381
 382 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 383 {
 384         struct vsock_sock *vlistener;
 385         struct vsock_sock *vconnected;
 386
 387         vlistener = vsock_sk(listener);
 388         vconnected = vsock_sk(connected);
 389
 390         sock_hold(connected);
 391         sock_hold(listener);
 392         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 393 }
 394 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 395
 396 static struct sock *vsock_dequeue_accept(struct sock *listener)
 397 {
 398         struct vsock_sock *vlistener;
 399         struct vsock_sock *vconnected;
 400
 401         vlistener = vsock_sk(listener);
 402
 403         if (list_empty(&vlistener->accept_queue))
 404                 return NULL;
 405
 406         vconnected = list_entry(vlistener->accept_queue.next,
 407                                 struct vsock_sock, accept_queue);
 408
 409         list_del_init(&vconnected->accept_queue);
 410         sock_put(listener);
 411         /* The caller will need a reference on the connected socket so we let
 412          * it call sock_put().
 413          */
 414
 415         return sk_vsock(vconnected);
 416 }
 417
 418 static bool vsock_is_accept_queue_empty(struct sock *sk)
 419 {
 420         struct vsock_sock *vsk = vsock_sk(sk);
 421         return list_empty(&vsk->accept_queue);
 422 }
 423
 424 static bool vsock_is_pending(struct sock *sk)
 425 {
 426         struct vsock_sock *vsk = vsock_sk(sk);
 427         return !list_empty(&vsk->pending_links);
 428 }
 429
 430 static int vsock_send_shutdown(struct sock *sk, int mode)
 431 {
 432         return transport->shutdown(vsock_sk(sk), mode);
 433 }
 434
 435 void vsock_pending_work(struct work_struct *work)
 436 {
 437         struct sock *sk;
 438         struct sock *listener;
 439         struct vsock_sock *vsk;
 440         bool cleanup;
 441
 442         vsk = container_of(work, struct vsock_sock, dwork.work);
 443         sk = sk_vsock(vsk);
 444         listener = vsk->listener;
 445         cleanup = true;
 446
 447         lock_sock(listener);
 448         lock_sock(sk);
 449
 450         if (vsock_is_pending(sk)) {
 451                 vsock_remove_pending(listener, sk);
 452         } else if (!vsk->rejected) {
 453                 /* We are not on the pending list and accept() did not reject
 454                  * us, so we must have been accepted by our user process.  We
 455                  * just need to drop our references to the sockets and be on
 456                  * our way.
 457                  */
 458                 cleanup = false;
 459                 goto out;
 460         }
 461
 462         listener->sk_ack_backlog--;
 463
 464         /* We need to remove ourself from the global connected sockets list so
 465          * incoming packets can't find this socket, and to reduce the reference
 466          * count.
 467          */
 468         if (vsock_in_connected_table(vsk))
 469                 vsock_remove_connected(vsk);
 470
 471         sk->sk_state = SS_FREE;
 472
 473 out:
 474         release_sock(sk);
 475         release_sock(listener);
 476         if (cleanup)
 477                 sock_put(sk);
 478
 479         sock_put(sk);
 480         sock_put(listener);
 481 }
 482 EXPORT_SYMBOL_GPL(vsock_pending_work);
 483
 484 /**** SOCKET OPERATIONS ****/
 485
 486 static int __vsock_bind_stream(struct vsock_sock *vsk,
 487                                struct sockaddr_vm *addr)
 488 {
 489         static u32 port = LAST_RESERVED_PORT + 1;
 490         struct sockaddr_vm new_addr;
 491
 492         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 493
 494         if (addr->svm_port == VMADDR_PORT_ANY) {
 495                 bool found = false;
 496                 unsigned int i;
 497
 498                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 499                         if (port <= LAST_RESERVED_PORT)
 500                                 port = LAST_RESERVED_PORT + 1;
 501
 502                         new_addr.svm_port = port++;
 503
 504                         if (!__vsock_find_bound_socket(&new_addr)) {
 505                                 found = true;
 506                                 break;
 507                         }
 508                 }
 509
 510                 if (!found)
 511                         return -EADDRNOTAVAIL;
 512         } else {
 513                 /* If port is in reserved range, ensure caller
 514                  * has necessary privileges.
 515                  */
 516                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 517                     !capable(CAP_NET_BIND_SERVICE)) {
 518                         return -EACCES;
 519                 }
 520
 521                 if (__vsock_find_bound_socket(&new_addr))
 522                         return -EADDRINUSE;
 523         }
 524
 525         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 526
 527         /* Remove stream sockets from the unbound list and add them to the hash
 528          * table for easy lookup by its address.  The unbound list is simply an
 529          * extra entry at the end of the hash table, a trick used by AF_UNIX.
 530          */
 531         __vsock_remove_bound(vsk);
 532         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 533
 534         return 0;
 535 }
 536
 537 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 538                               struct sockaddr_vm *addr)
 539 {
 540         return transport->dgram_bind(vsk, addr);
 541 }
 542
 543 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 544 {
 545         struct vsock_sock *vsk = vsock_sk(sk);
 546         u32 cid;
 547         int retval;
 548
 549         /* First ensure this socket isn't already bound. */
 550         if (vsock_addr_bound(&vsk->local_addr))
 551                 return -EINVAL;
 552
 553         /* Now bind to the provided address or select appropriate values if
 554          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 555          * like AF_INET prevents binding to a non-local IP address (in most
 556          * cases), we only allow binding to the local CID.
 557          */
 558         cid = transport->get_local_cid();
 559         if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY)
 560                 return -EADDRNOTAVAIL;
 561
 562         switch (sk->sk_socket->type) {
 563         case SOCK_STREAM:
 564                 spin_lock_bh(&vsock_table_lock);
 565                 retval = __vsock_bind_stream(vsk, addr);
 566                 spin_unlock_bh(&vsock_table_lock);
 567                 break;
 568
 569         case SOCK_DGRAM:
 570                 retval = __vsock_bind_dgram(vsk, addr);
 571                 break;
 572
 573         default:
 574                 retval = -EINVAL;
 575                 break;
 576         }
 577
 578         return retval;
 579 }
 580
 581 struct sock *__vsock_create(struct net *net,
 582                             struct socket *sock,
 583                             struct sock *parent,
 584                             gfp_t priority,
 585                             unsigned short type)
 586 {
 587         struct sock *sk;
 588         struct vsock_sock *psk;
 589         struct vsock_sock *vsk;
 590
 591         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto);
 592         if (!sk)
 593                 return NULL;
 594
 595         sock_init_data(sock, sk);
 596
 597         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 598          * non-NULL. We make sure that our sockets always have a type by
 599          * setting it here if needed.
 600          */
 601         if (!sock)
 602                 sk->sk_type = type;
 603
 604         vsk = vsock_sk(sk);
 605         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 606         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 607
 608         sk->sk_destruct = vsock_sk_destruct;
 609         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 610         sk->sk_state = 0;
 611         sock_reset_flag(sk, SOCK_DONE);
 612
 613         INIT_LIST_HEAD(&vsk->bound_table);
 614         INIT_LIST_HEAD(&vsk->connected_table);
 615         vsk->listener = NULL;
 616         INIT_LIST_HEAD(&vsk->pending_links);
 617         INIT_LIST_HEAD(&vsk->accept_queue);
 618         vsk->rejected = false;
 619         vsk->sent_request = false;
 620         vsk->ignore_connecting_rst = false;
 621         vsk->peer_shutdown = 0;
 622
 623         psk = parent ? vsock_sk(parent) : NULL;
 624         if (parent) {
 625                 vsk->trusted = psk->trusted;
 626                 vsk->owner = get_cred(psk->owner);
 627                 vsk->connect_timeout = psk->connect_timeout;
 628         } else {
 629                 vsk->trusted = capable(CAP_NET_ADMIN);
 630                 vsk->owner = get_current_cred();
 631                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 632         }
 633
 634         if (transport->init(vsk, psk) < 0) {
 635                 sk_free(sk);
 636                 return NULL;
 637         }
 638
 639         if (sock)
 640                 vsock_insert_unbound(vsk);
 641
 642         return sk;
 643 }
 644 EXPORT_SYMBOL_GPL(__vsock_create);
 645
 646 static void __vsock_release(struct sock *sk)
 647 {
 648         if (sk) {
 649                 struct sk_buff *skb;
 650                 struct sock *pending;
 651                 struct vsock_sock *vsk;
 652
 653                 vsk = vsock_sk(sk);
 654                 pending = NULL; /* Compiler warning. */
 655
 656                 if (vsock_in_bound_table(vsk))
 657                         vsock_remove_bound(vsk);
 658
 659                 if (vsock_in_connected_table(vsk))
 660                         vsock_remove_connected(vsk);
 661
 662                 transport->release(vsk);
 663
 664                 lock_sock(sk);
 665                 sock_orphan(sk);
 666                 sk->sk_shutdown = SHUTDOWN_MASK;
 667
 668                 while ((skb = skb_dequeue(&sk->sk_receive_queue)))
 669                         kfree_skb(skb);
 670
 671                 /* Clean up any sockets that never were accepted. */
 672                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 673                         __vsock_release(pending);
 674                         sock_put(pending);
 675                 }
 676
 677                 release_sock(sk);
 678                 sock_put(sk);
 679         }
 680 }
 681
 682 static void vsock_sk_destruct(struct sock *sk)
 683 {
 684         struct vsock_sock *vsk = vsock_sk(sk);
 685
 686         transport->destruct(vsk);
 687
 688         /* When clearing these addresses, there's no need to set the family and
 689          * possibly register the address family with the kernel.
 690          */
 691         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 692         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 693
 694         put_cred(vsk->owner);
 695 }
 696
 697 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 698 {
 699         int err;
 700
 701         err = sock_queue_rcv_skb(sk, skb);
 702         if (err)
 703                 kfree_skb(skb);
 704
 705         return err;
 706 }
 707
 708 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 709 {
 710         return transport->stream_has_data(vsk);
 711 }
 712 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 713
 714 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 715 {
 716         return transport->stream_has_space(vsk);
 717 }
 718 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 719
 720 static int vsock_release(struct socket *sock)
 721 {
 722         __vsock_release(sock->sk);
 723         sock->sk = NULL;
 724         sock->state = SS_FREE;
 725
 726         return 0;
 727 }
 728
 729 static int
 730 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 731 {
 732         int err;
 733         struct sock *sk;
 734         struct sockaddr_vm *vm_addr;
 735
 736         sk = sock->sk;
 737
 738         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 739                 return -EINVAL;
 740
 741         lock_sock(sk);
 742         err = __vsock_bind(sk, vm_addr);
 743         release_sock(sk);
 744
 745         return err;
 746 }
 747
 748 static int vsock_getname(struct socket *sock,
 749                          struct sockaddr *addr, int *addr_len, int peer)
 750 {
 751         int err;
 752         struct sock *sk;
 753         struct vsock_sock *vsk;
 754         struct sockaddr_vm *vm_addr;
 755
 756         sk = sock->sk;
 757         vsk = vsock_sk(sk);
 758         err = 0;
 759
 760         lock_sock(sk);
 761
 762         if (peer) {
 763                 if (sock->state != SS_CONNECTED) {
 764                         err = -ENOTCONN;
 765                         goto out;
 766                 }
 767                 vm_addr = &vsk->remote_addr;
 768         } else {
 769                 vm_addr = &vsk->local_addr;
 770         }
 771
 772         if (!vm_addr) {
 773                 err = -EINVAL;
 774                 goto out;
 775         }
 776
 777         /* sys_getsockname() and sys_getpeername() pass us a
 778          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 779          * that macro is defined in socket.c instead of .h, so we hardcode its
 780          * value here.
 781          */
 782         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 783         memcpy(addr, vm_addr, sizeof(*vm_addr));
 784         *addr_len = sizeof(*vm_addr);
 785
 786 out:
 787         release_sock(sk);
 788         return err;
 789 }
 790
 791 static int vsock_shutdown(struct socket *sock, int mode)
 792 {
 793         int err;
 794         struct sock *sk;
 795
 796         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 797          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 798          * here like the other address families do.  Note also that the
 799          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 800          * which is what we want.
 801          */
 802         mode++;
 803
 804         if ((mode & ~SHUTDOWN_MASK) || !mode)
 805                 return -EINVAL;
 806
 807         /* If this is a STREAM socket and it is not connected then bail out
 808          * immediately.  If it is a DGRAM socket then we must first kick the
 809          * socket so that it wakes up from any sleeping calls, for example
 810          * recv(), and then afterwards return the error.
 811          */
 812
 813         sk = sock->sk;
 814         if (sock->state == SS_UNCONNECTED) {
 815                 err = -ENOTCONN;
 816                 if (sk->sk_type == SOCK_STREAM)
 817                         return err;
 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                 lock_sock(sk);
 827                 sk->sk_shutdown |= mode;
 828                 sk->sk_state_change(sk);
 829                 release_sock(sk);
 830
 831                 if (sk->sk_type == SOCK_STREAM) {
 832                         sock_reset_flag(sk, SOCK_DONE);
 833                         vsock_send_shutdown(sk, mode);
 834                 }
 835         }
 836
 837         return err;
 838 }
 839
 840 static unsigned int vsock_poll(struct file *file, struct socket *sock,
 841                                poll_table *wait)
 842 {
 843         struct sock *sk;
 844         unsigned int 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 |= POLLERR;
 856
 857         /* INET sockets treat local write shutdown and peer write shutdown as a
 858          * case of POLLHUP set.
 859          */
 860         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
 861             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
 862              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
 863                 mask |= POLLHUP;
 864         }
 865
 866         if (sk->sk_shutdown & RCV_SHUTDOWN ||
 867             vsk->peer_shutdown & SEND_SHUTDOWN) {
 868                 mask |= POLLRDHUP;
 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(&sk->sk_receive_queue) ||
 877                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
 878                         mask |= POLLIN | POLLRDNORM;
 879                 }
 880
 881                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 882                         mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 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 == SS_LISTEN
 891                     && !vsock_is_accept_queue_empty(sk))
 892                         mask |= POLLIN | POLLRDNORM;
 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 |= POLLERR;
 902                         } else {
 903                                 if (data_ready_now)
 904                                         mask |= POLLIN | POLLRDNORM;
 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 |= POLLIN | POLLRDNORM;
 916                 }
 917
 918                 /* Connected sockets that can produce data can be written. */
 919                 if (sk->sk_state == SS_CONNECTED) {
 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 |= POLLERR;
 926                                 } else {
 927                                         if (space_avail_now)
 928                                                 /* Remove POLLWRBAND since INET
 929                                                  * sockets are not setting it.
 930                                                  */
 931                                                 mask |= POLLOUT | POLLWRNORM;
 932
 933                                 }
 934                         }
 935                 }
 936
 937                 /* Simulate INET socket poll behaviors, which sets
 938                  * POLLOUT|POLLWRNORM when peer is closed and nothing to read,
 939                  * but local send is not shutdown.
 940                  */
 941                 if (sk->sk_state == SS_UNCONNECTED) {
 942                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
 943                                 mask |= POLLOUT | POLLWRNORM;
 944
 945                 }
 946
 947                 release_sock(sk);
 948         }
 949
 950         return mask;
 951 }
 952
 953 static int vsock_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
 954                                struct msghdr *msg, 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->msg_iov, 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 kiocb *kiocb, struct socket *sock,
1067                                struct msghdr *msg, size_t len, int flags)
1068 {
1069         return transport->dgram_dequeue(kiocb, vsock_sk(sock->sk), msg, len,
1070                                         flags);
1071 }
1072
1073 static const struct proto_ops vsock_dgram_ops = {
1074         .family = PF_VSOCK,
1075         .owner = THIS_MODULE,
1076         .release = vsock_release,
1077         .bind = vsock_bind,
1078         .connect = vsock_dgram_connect,
1079         .socketpair = sock_no_socketpair,
1080         .accept = sock_no_accept,
1081         .getname = vsock_getname,
1082         .poll = vsock_poll,
1083         .ioctl = sock_no_ioctl,
1084         .listen = sock_no_listen,
1085         .shutdown = vsock_shutdown,
1086         .setsockopt = sock_no_setsockopt,
1087         .getsockopt = sock_no_getsockopt,
1088         .sendmsg = vsock_dgram_sendmsg,
1089         .recvmsg = vsock_dgram_recvmsg,
1090         .mmap = sock_no_mmap,
1091         .sendpage = sock_no_sendpage,
1092 };
1093
1094 static void vsock_connect_timeout(struct work_struct *work)
1095 {
1096         struct sock *sk;
1097         struct vsock_sock *vsk;
1098
1099         vsk = container_of(work, struct vsock_sock, dwork.work);
1100         sk = sk_vsock(vsk);
1101
1102         lock_sock(sk);
1103         if (sk->sk_state == SS_CONNECTING &&
1104             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1105                 sk->sk_state = SS_UNCONNECTED;
1106                 sk->sk_err = ETIMEDOUT;
1107                 sk->sk_error_report(sk);
1108         }
1109         release_sock(sk);
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 == SS_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 = SS_CONNECTING;
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 != SS_CONNECTED && 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                         INIT_DELAYED_WORK(&vsk->dwork,
1201                                           vsock_connect_timeout);
1202                         schedule_delayed_work(&vsk->dwork, timeout);
1203
1204                         /* Skip ahead to preserve error code set above. */
1205                         goto out_wait;
1206                 }
1207
1208                 release_sock(sk);
1209                 timeout = schedule_timeout(timeout);
1210                 lock_sock(sk);
1211
1212                 if (signal_pending(current)) {
1213                         err = sock_intr_errno(timeout);
1214                         goto out_wait_error;
1215                 } else if (timeout == 0) {
1216                         err = -ETIMEDOUT;
1217                         goto out_wait_error;
1218                 }
1219
1220                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1221         }
1222
1223         if (sk->sk_err) {
1224                 err = -sk->sk_err;
1225                 goto out_wait_error;
1226         } else
1227                 err = 0;
1228
1229 out_wait:
1230         finish_wait(sk_sleep(sk), &wait);
1231 out:
1232         release_sock(sk);
1233         return err;
1234
1235 out_wait_error:
1236         sk->sk_state = SS_UNCONNECTED;
1237         sock->state = SS_UNCONNECTED;
1238         goto out_wait;
1239 }
1240
1241 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags)
1242 {
1243         struct sock *listener;
1244         int err;
1245         struct sock *connected;
1246         struct vsock_sock *vconnected;
1247         long timeout;
1248         DEFINE_WAIT(wait);
1249
1250         err = 0;
1251         listener = sock->sk;
1252
1253         lock_sock(listener);
1254
1255         if (sock->type != SOCK_STREAM) {
1256                 err = -EOPNOTSUPP;
1257                 goto out;
1258         }
1259
1260         if (listener->sk_state != SS_LISTEN) {
1261                 err = -EINVAL;
1262                 goto out;
1263         }
1264
1265         /* Wait for children sockets to appear; these are the new sockets
1266          * created upon connection establishment.
1267          */
1268         timeout = sock_sndtimeo(listener, flags & O_NONBLOCK);
1269         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1270
1271         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1272                listener->sk_err == 0) {
1273                 release_sock(listener);
1274                 timeout = schedule_timeout(timeout);
1275                 lock_sock(listener);
1276
1277                 if (signal_pending(current)) {
1278                         err = sock_intr_errno(timeout);
1279                         goto out_wait;
1280                 } else if (timeout == 0) {
1281                         err = -EAGAIN;
1282                         goto out_wait;
1283                 }
1284
1285                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1286         }
1287
1288         if (listener->sk_err)
1289                 err = -listener->sk_err;
1290
1291         if (connected) {
1292                 listener->sk_ack_backlog--;
1293
1294                 lock_sock(connected);
1295                 vconnected = vsock_sk(connected);
1296
1297                 /* If the listener socket has received an error, then we should
1298                  * reject this socket and return.  Note that we simply mark the
1299                  * socket rejected, drop our reference, and let the cleanup
1300                  * function handle the cleanup; the fact that we found it in
1301                  * the listener's accept queue guarantees that the cleanup
1302                  * function hasn't run yet.
1303                  */
1304                 if (err) {
1305                         vconnected->rejected = true;
1306                         release_sock(connected);
1307                         sock_put(connected);
1308                         goto out_wait;
1309                 }
1310
1311                 newsock->state = SS_CONNECTED;
1312                 sock_graft(connected, newsock);
1313                 release_sock(connected);
1314                 sock_put(connected);
1315         }
1316
1317 out_wait:
1318         finish_wait(sk_sleep(listener), &wait);
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 = SS_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 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 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 kiocb *kiocb, struct socket *sock,
1510                                 struct msghdr *msg, 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
1519         DEFINE_WAIT(wait);
1520
1521         sk = sock->sk;
1522         vsk = vsock_sk(sk);
1523         total_written = 0;
1524         err = 0;
1525
1526         if (msg->msg_flags & MSG_OOB)
1527                 return -EOPNOTSUPP;
1528
1529         lock_sock(sk);
1530
1531         /* Callers should not provide a destination with stream sockets. */
1532         if (msg->msg_namelen) {
1533                 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP;
1534                 goto out;
1535         }
1536
1537         /* Send data only if both sides are not shutdown in the direction. */
1538         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1539             vsk->peer_shutdown & RCV_SHUTDOWN) {
1540                 err = -EPIPE;
1541                 goto out;
1542         }
1543
1544         if (sk->sk_state != SS_CONNECTED ||
1545             !vsock_addr_bound(&vsk->local_addr)) {
1546                 err = -ENOTCONN;
1547                 goto out;
1548         }
1549
1550         if (!vsock_addr_bound(&vsk->remote_addr)) {
1551                 err = -EDESTADDRREQ;
1552                 goto out;
1553         }
1554
1555         /* Wait for room in the produce queue to enqueue our user's data. */
1556         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1557
1558         err = transport->notify_send_init(vsk, &send_data);
1559         if (err < 0)
1560                 goto out;
1561
1562         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1563
1564         while (total_written < len) {
1565                 ssize_t written;
1566
1567                 while (vsock_stream_has_space(vsk) == 0 &&
1568                        sk->sk_err == 0 &&
1569                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1570                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1571
1572                         /* Don't wait for non-blocking sockets. */
1573                         if (timeout == 0) {
1574                                 err = -EAGAIN;
1575                                 goto out_wait;
1576                         }
1577
1578                         err = transport->notify_send_pre_block(vsk, &send_data);
1579                         if (err < 0)
1580                                 goto out_wait;
1581
1582                         release_sock(sk);
1583                         timeout = schedule_timeout(timeout);
1584                         lock_sock(sk);
1585                         if (signal_pending(current)) {
1586                                 err = sock_intr_errno(timeout);
1587                                 goto out_wait;
1588                         } else if (timeout == 0) {
1589                                 err = -EAGAIN;
1590                                 goto out_wait;
1591                         }
1592
1593                         prepare_to_wait(sk_sleep(sk), &wait,
1594                                         TASK_INTERRUPTIBLE);
1595                 }
1596
1597                 /* These checks occur both as part of and after the loop
1598                  * conditional since we need to check before and after
1599                  * sleeping.
1600                  */
1601                 if (sk->sk_err) {
1602                         err = -sk->sk_err;
1603                         goto out_wait;
1604                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1605                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1606                         err = -EPIPE;
1607                         goto out_wait;
1608                 }
1609
1610                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1611                 if (err < 0)
1612                         goto out_wait;
1613
1614                 /* Note that enqueue will only write as many bytes as are free
1615                  * in the produce queue, so we don't need to ensure len is
1616                  * smaller than the queue size.  It is the caller's
1617                  * responsibility to check how many bytes we were able to send.
1618                  */
1619
1620                 written = transport->stream_enqueue(
1621                                 vsk, msg->msg_iov,
1622                                 len - total_written);
1623                 if (written < 0) {
1624                         err = -ENOMEM;
1625                         goto out_wait;
1626                 }
1627
1628                 total_written += written;
1629
1630                 err = transport->notify_send_post_enqueue(
1631                                 vsk, written, &send_data);
1632                 if (err < 0)
1633                         goto out_wait;
1634
1635         }
1636
1637 out_wait:
1638         if (total_written > 0)
1639                 err = total_written;
1640         finish_wait(sk_sleep(sk), &wait);
1641 out:
1642         release_sock(sk);
1643         return err;
1644 }
1645
1646
1647 static int
1648 vsock_stream_recvmsg(struct kiocb *kiocb,
1649                      struct socket *sock,
1650                      struct msghdr *msg, size_t len, int flags)
1651 {
1652         struct sock *sk;
1653         struct vsock_sock *vsk;
1654         int err;
1655         size_t target;
1656         ssize_t copied;
1657         long timeout;
1658         struct vsock_transport_recv_notify_data recv_data;
1659
1660         DEFINE_WAIT(wait);
1661
1662         sk = sock->sk;
1663         vsk = vsock_sk(sk);
1664         err = 0;
1665
1666         msg->msg_namelen = 0;
1667
1668         lock_sock(sk);
1669
1670         if (sk->sk_state != SS_CONNECTED) {
1671                 /* Recvmsg is supposed to return 0 if a peer performs an
1672                  * orderly shutdown. Differentiate between that case and when a
1673                  * peer has not connected or a local shutdown occured with the
1674                  * SOCK_DONE flag.
1675                  */
1676                 if (sock_flag(sk, SOCK_DONE))
1677                         err = 0;
1678                 else
1679                         err = -ENOTCONN;
1680
1681                 goto out;
1682         }
1683
1684         if (flags & MSG_OOB) {
1685                 err = -EOPNOTSUPP;
1686                 goto out;
1687         }
1688
1689         /* We don't check peer_shutdown flag here since peer may actually shut
1690          * down, but there can be data in the queue that a local socket can
1691          * receive.
1692          */
1693         if (sk->sk_shutdown & RCV_SHUTDOWN) {
1694                 err = 0;
1695                 goto out;
1696         }
1697
1698         /* It is valid on Linux to pass in a zero-length receive buffer.  This
1699          * is not an error.  We may as well bail out now.
1700          */
1701         if (!len) {
1702                 err = 0;
1703                 goto out;
1704         }
1705
1706         /* We must not copy less than target bytes into the user's buffer
1707          * before returning successfully, so we wait for the consume queue to
1708          * have that much data to consume before dequeueing.  Note that this
1709          * makes it impossible to handle cases where target is greater than the
1710          * queue size.
1711          */
1712         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1713         if (target >= transport->stream_rcvhiwat(vsk)) {
1714                 err = -ENOMEM;
1715                 goto out;
1716         }
1717         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1718         copied = 0;
1719
1720         err = transport->notify_recv_init(vsk, target, &recv_data);
1721         if (err < 0)
1722                 goto out;
1723
1724         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1725
1726         while (1) {
1727                 s64 ready = vsock_stream_has_data(vsk);
1728
1729                 if (ready < 0) {
1730                         /* Invalid queue pair content. XXX This should be
1731                          * changed to a connection reset in a later change.
1732                          */
1733
1734                         err = -ENOMEM;
1735                         goto out_wait;
1736                 } else if (ready > 0) {
1737                         ssize_t read;
1738
1739                         err = transport->notify_recv_pre_dequeue(
1740                                         vsk, target, &recv_data);
1741                         if (err < 0)
1742                                 break;
1743
1744                         read = transport->stream_dequeue(
1745                                         vsk, msg->msg_iov,
1746                                         len - copied, flags);
1747                         if (read < 0) {
1748                                 err = -ENOMEM;
1749                                 break;
1750                         }
1751
1752                         copied += read;
1753
1754                         err = transport->notify_recv_post_dequeue(
1755                                         vsk, target, read,
1756                                         !(flags & MSG_PEEK), &recv_data);
1757                         if (err < 0)
1758                                 goto out_wait;
1759
1760                         if (read >= target || flags & MSG_PEEK)
1761                                 break;
1762
1763                         target -= read;
1764                 } else {
1765                         if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN)
1766                             || (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1767                                 break;
1768                         }
1769                         /* Don't wait for non-blocking sockets. */
1770                         if (timeout == 0) {
1771                                 err = -EAGAIN;
1772                                 break;
1773                         }
1774
1775                         err = transport->notify_recv_pre_block(
1776                                         vsk, target, &recv_data);
1777                         if (err < 0)
1778                                 break;
1779
1780                         release_sock(sk);
1781                         timeout = schedule_timeout(timeout);
1782                         lock_sock(sk);
1783
1784                         if (signal_pending(current)) {
1785                                 err = sock_intr_errno(timeout);
1786                                 break;
1787                         } else if (timeout == 0) {
1788                                 err = -EAGAIN;
1789                                 break;
1790                         }
1791
1792                         prepare_to_wait(sk_sleep(sk), &wait,
1793                                         TASK_INTERRUPTIBLE);
1794                 }
1795         }
1796
1797         if (sk->sk_err)
1798                 err = -sk->sk_err;
1799         else if (sk->sk_shutdown & RCV_SHUTDOWN)
1800                 err = 0;
1801
1802         if (copied > 0) {
1803                 /* We only do these additional bookkeeping/notification steps
1804                  * if we actually copied something out of the queue pair
1805                  * instead of just peeking ahead.
1806                  */
1807
1808                 if (!(flags & MSG_PEEK)) {
1809                         /* If the other side has shutdown for sending and there
1810                          * is nothing more to read, then modify the socket
1811                          * state.
1812                          */
1813                         if (vsk->peer_shutdown & SEND_SHUTDOWN) {
1814                                 if (vsock_stream_has_data(vsk) <= 0) {
1815                                         sk->sk_state = SS_UNCONNECTED;
1816                                         sock_set_flag(sk, SOCK_DONE);
1817                                         sk->sk_state_change(sk);
1818                                 }
1819                         }
1820                 }
1821                 err = copied;
1822         }
1823
1824 out_wait:
1825         finish_wait(sk_sleep(sk), &wait);
1826 out:
1827         release_sock(sk);
1828         return err;
1829 }
1830
1831 static const struct proto_ops vsock_stream_ops = {
1832         .family = PF_VSOCK,
1833         .owner = THIS_MODULE,
1834         .release = vsock_release,
1835         .bind = vsock_bind,
1836         .connect = vsock_stream_connect,
1837         .socketpair = sock_no_socketpair,
1838         .accept = vsock_accept,
1839         .getname = vsock_getname,
1840         .poll = vsock_poll,
1841         .ioctl = sock_no_ioctl,
1842         .listen = vsock_listen,
1843         .shutdown = vsock_shutdown,
1844         .setsockopt = vsock_stream_setsockopt,
1845         .getsockopt = vsock_stream_getsockopt,
1846         .sendmsg = vsock_stream_sendmsg,
1847         .recvmsg = vsock_stream_recvmsg,
1848         .mmap = sock_no_mmap,
1849         .sendpage = sock_no_sendpage,
1850 };
1851
1852 static int vsock_create(struct net *net, struct socket *sock,
1853                         int protocol, int kern)
1854 {
1855         if (!sock)
1856                 return -EINVAL;
1857
1858         if (protocol && protocol != PF_VSOCK)
1859                 return -EPROTONOSUPPORT;
1860
1861         switch (sock->type) {
1862         case SOCK_DGRAM:
1863                 sock->ops = &vsock_dgram_ops;
1864                 break;
1865         case SOCK_STREAM:
1866                 sock->ops = &vsock_stream_ops;
1867                 break;
1868         default:
1869                 return -ESOCKTNOSUPPORT;
1870         }
1871
1872         sock->state = SS_UNCONNECTED;
1873
1874         return __vsock_create(net, sock, NULL, GFP_KERNEL, 0) ? 0 : -ENOMEM;
1875 }
1876
1877 static const struct net_proto_family vsock_family_ops = {
1878         .family = AF_VSOCK,
1879         .create = vsock_create,
1880         .owner = THIS_MODULE,
1881 };
1882
1883 static long vsock_dev_do_ioctl(struct file *filp,
1884                                unsigned int cmd, void __user *ptr)
1885 {
1886         u32 __user *p = ptr;
1887         int retval = 0;
1888
1889         switch (cmd) {
1890         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
1891                 if (put_user(transport->get_local_cid(), p) != 0)
1892                         retval = -EFAULT;
1893                 break;
1894
1895         default:
1896                 pr_err("Unknown ioctl %d\n", cmd);
1897                 retval = -EINVAL;
1898         }
1899
1900         return retval;
1901 }
1902
1903 static long vsock_dev_ioctl(struct file *filp,
1904                             unsigned int cmd, unsigned long arg)
1905 {
1906         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
1907 }
1908
1909 #ifdef CONFIG_COMPAT
1910 static long vsock_dev_compat_ioctl(struct file *filp,
1911                                    unsigned int cmd, unsigned long arg)
1912 {
1913         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
1914 }
1915 #endif
1916
1917 static const struct file_operations vsock_device_ops = {
1918         .owner          = THIS_MODULE,
1919         .unlocked_ioctl = vsock_dev_ioctl,
1920 #ifdef CONFIG_COMPAT
1921         .compat_ioctl   = vsock_dev_compat_ioctl,
1922 #endif
1923         .open           = nonseekable_open,
1924 };
1925
1926 static struct miscdevice vsock_device = {
1927         .name           = "vsock",
1928         .fops           = &vsock_device_ops,
1929 };
1930
1931 static int __vsock_core_init(void)
1932 {
1933         int err;
1934
1935         vsock_init_tables();
1936
1937         vsock_device.minor = MISC_DYNAMIC_MINOR;
1938         err = misc_register(&vsock_device);
1939         if (err) {
1940                 pr_err("Failed to register misc device\n");
1941                 return -ENOENT;
1942         }
1943
1944         err = proto_register(&vsock_proto, 1);  /* we want our slab */
1945         if (err) {
1946                 pr_err("Cannot register vsock protocol\n");
1947                 goto err_misc_deregister;
1948         }
1949
1950         err = sock_register(&vsock_family_ops);
1951         if (err) {
1952                 pr_err("could not register af_vsock (%d) address family: %d\n",
1953                        AF_VSOCK, err);
1954                 goto err_unregister_proto;
1955         }
1956
1957         return 0;
1958
1959 err_unregister_proto:
1960         proto_unregister(&vsock_proto);
1961 err_misc_deregister:
1962         misc_deregister(&vsock_device);
1963         return err;
1964 }
1965
1966 int vsock_core_init(const struct vsock_transport *t)
1967 {
1968         int retval = mutex_lock_interruptible(&vsock_register_mutex);
1969         if (retval)
1970                 return retval;
1971
1972         if (transport) {
1973                 retval = -EBUSY;
1974                 goto out;
1975         }
1976
1977         transport = t;
1978         retval = __vsock_core_init();
1979         if (retval)
1980                 transport = NULL;
1981
1982 out:
1983         mutex_unlock(&vsock_register_mutex);
1984         return retval;
1985 }
1986 EXPORT_SYMBOL_GPL(vsock_core_init);
1987
1988 void vsock_core_exit(void)
1989 {
1990         mutex_lock(&vsock_register_mutex);
1991
1992         misc_deregister(&vsock_device);
1993         sock_unregister(AF_VSOCK);
1994         proto_unregister(&vsock_proto);
1995
1996         /* We do not want the assignment below re-ordered. */
1997         mb();
1998         transport = NULL;
1999
2000         mutex_unlock(&vsock_register_mutex);
2001 }
2002 EXPORT_SYMBOL_GPL(vsock_core_exit);
2003
2004 MODULE_AUTHOR("VMware, Inc.");
2005 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2006 MODULE_VERSION("1.0.0.0-k");
2007 MODULE_LICENSE("GPL v2");