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/compat.h>
  89 #include <linux/types.h>
  90 #include <linux/bitops.h>
  91 #include <linux/cred.h>
  92 #include <linux/errqueue.h>
  93 #include <linux/init.h>
  94 #include <linux/io.h>
  95 #include <linux/kernel.h>
  96 #include <linux/sched/signal.h>
  97 #include <linux/kmod.h>
  98 #include <linux/list.h>
  99 #include <linux/miscdevice.h>
 100 #include <linux/module.h>
 101 #include <linux/mutex.h>
 102 #include <linux/net.h>
 103 #include <linux/poll.h>
 104 #include <linux/random.h>
 105 #include <linux/skbuff.h>
 106 #include <linux/smp.h>
 107 #include <linux/socket.h>
 108 #include <linux/stddef.h>
 109 #include <linux/unistd.h>
 110 #include <linux/wait.h>
 111 #include <linux/workqueue.h>
 112 #include <net/sock.h>
 113 #include <net/af_vsock.h>
 114 #include <uapi/linux/vm_sockets.h>
 115
 116 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 117 static void vsock_sk_destruct(struct sock *sk);
 118 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 119
 120 /* Protocol family. */
 121 static struct proto vsock_proto = {
 122         .name = "AF_VSOCK",
 123         .owner = THIS_MODULE,
 124         .obj_size = sizeof(struct vsock_sock),
 125 };
 126
 127 /* The default peer timeout indicates how long we will wait for a peer response
 128  * to a control message.
 129  */
 130 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 131
 132 #define VSOCK_DEFAULT_BUFFER_SIZE     (1024 * 256)
 133 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
 134 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
 135
 136 /* Transport used for host->guest communication */
 137 static const struct vsock_transport *transport_h2g;
 138 /* Transport used for guest->host communication */
 139 static const struct vsock_transport *transport_g2h;
 140 /* Transport used for DGRAM communication */
 141 static const struct vsock_transport *transport_dgram;
 142 /* Transport used for local communication */
 143 static const struct vsock_transport *transport_local;
 144 static DEFINE_MUTEX(vsock_register_mutex);
 145
 146 /**** UTILS ****/
 147
 148 /* Each bound VSocket is stored in the bind hash table and each connected
 149  * VSocket is stored in the connected hash table.
 150  *
 151  * Unbound sockets are all put on the same list attached to the end of the hash
 152  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 153  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 154  * represents the list that addr hashes to).
 155  *
 156  * Specifically, we initialize the vsock_bind_table array to a size of
 157  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 158  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 159  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 160  * mods with VSOCK_HASH_SIZE to ensure this.
 161  */
 162 #define MAX_PORT_RETRIES        24
 163
 164 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 165 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 166 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 167
 168 /* XXX This can probably be implemented in a better way. */
 169 #define VSOCK_CONN_HASH(src, dst)                               \
 170         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 171 #define vsock_connected_sockets(src, dst)               \
 172         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 173 #define vsock_connected_sockets_vsk(vsk)                                \
 174         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 175
 176 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 177 EXPORT_SYMBOL_GPL(vsock_bind_table);
 178 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 179 EXPORT_SYMBOL_GPL(vsock_connected_table);
 180 DEFINE_SPINLOCK(vsock_table_lock);
 181 EXPORT_SYMBOL_GPL(vsock_table_lock);
 182
 183 /* Autobind this socket to the local address if necessary. */
 184 static int vsock_auto_bind(struct vsock_sock *vsk)
 185 {
 186         struct sock *sk = sk_vsock(vsk);
 187         struct sockaddr_vm local_addr;
 188
 189         if (vsock_addr_bound(&vsk->local_addr))
 190                 return 0;
 191         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 192         return __vsock_bind(sk, &local_addr);
 193 }
 194
 195 static void vsock_init_tables(void)
 196 {
 197         int i;
 198
 199         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 200                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 201
 202         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 203                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 204 }
 205
 206 static void __vsock_insert_bound(struct list_head *list,
 207                                  struct vsock_sock *vsk)
 208 {
 209         sock_hold(&vsk->sk);
 210         list_add(&vsk->bound_table, list);
 211 }
 212
 213 static void __vsock_insert_connected(struct list_head *list,
 214                                      struct vsock_sock *vsk)
 215 {
 216         sock_hold(&vsk->sk);
 217         list_add(&vsk->connected_table, list);
 218 }
 219
 220 static void __vsock_remove_bound(struct vsock_sock *vsk)
 221 {
 222         list_del_init(&vsk->bound_table);
 223         sock_put(&vsk->sk);
 224 }
 225
 226 static void __vsock_remove_connected(struct vsock_sock *vsk)
 227 {
 228         list_del_init(&vsk->connected_table);
 229         sock_put(&vsk->sk);
 230 }
 231
 232 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 233 {
 234         struct vsock_sock *vsk;
 235
 236         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
 237                 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
 238                         return sk_vsock(vsk);
 239
 240                 if (addr->svm_port == vsk->local_addr.svm_port &&
 241                     (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
 242                      addr->svm_cid == VMADDR_CID_ANY))
 243                         return sk_vsock(vsk);
 244         }
 245
 246         return NULL;
 247 }
 248
 249 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 250                                                   struct sockaddr_vm *dst)
 251 {
 252         struct vsock_sock *vsk;
 253
 254         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 255                             connected_table) {
 256                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 257                     dst->svm_port == vsk->local_addr.svm_port) {
 258                         return sk_vsock(vsk);
 259                 }
 260         }
 261
 262         return NULL;
 263 }
 264
 265 static void vsock_insert_unbound(struct vsock_sock *vsk)
 266 {
 267         spin_lock_bh(&vsock_table_lock);
 268         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 269         spin_unlock_bh(&vsock_table_lock);
 270 }
 271
 272 void vsock_insert_connected(struct vsock_sock *vsk)
 273 {
 274         struct list_head *list = vsock_connected_sockets(
 275                 &vsk->remote_addr, &vsk->local_addr);
 276
 277         spin_lock_bh(&vsock_table_lock);
 278         __vsock_insert_connected(list, vsk);
 279         spin_unlock_bh(&vsock_table_lock);
 280 }
 281 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 282
 283 void vsock_remove_bound(struct vsock_sock *vsk)
 284 {
 285         spin_lock_bh(&vsock_table_lock);
 286         if (__vsock_in_bound_table(vsk))
 287                 __vsock_remove_bound(vsk);
 288         spin_unlock_bh(&vsock_table_lock);
 289 }
 290 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 291
 292 void vsock_remove_connected(struct vsock_sock *vsk)
 293 {
 294         spin_lock_bh(&vsock_table_lock);
 295         if (__vsock_in_connected_table(vsk))
 296                 __vsock_remove_connected(vsk);
 297         spin_unlock_bh(&vsock_table_lock);
 298 }
 299 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 300
 301 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 302 {
 303         struct sock *sk;
 304
 305         spin_lock_bh(&vsock_table_lock);
 306         sk = __vsock_find_bound_socket(addr);
 307         if (sk)
 308                 sock_hold(sk);
 309
 310         spin_unlock_bh(&vsock_table_lock);
 311
 312         return sk;
 313 }
 314 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 315
 316 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 317                                          struct sockaddr_vm *dst)
 318 {
 319         struct sock *sk;
 320
 321         spin_lock_bh(&vsock_table_lock);
 322         sk = __vsock_find_connected_socket(src, dst);
 323         if (sk)
 324                 sock_hold(sk);
 325
 326         spin_unlock_bh(&vsock_table_lock);
 327
 328         return sk;
 329 }
 330 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 331
 332 void vsock_remove_sock(struct vsock_sock *vsk)
 333 {
 334         vsock_remove_bound(vsk);
 335         vsock_remove_connected(vsk);
 336 }
 337 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 338
 339 void vsock_for_each_connected_socket(struct vsock_transport *transport,
 340                                      void (*fn)(struct sock *sk))
 341 {
 342         int i;
 343
 344         spin_lock_bh(&vsock_table_lock);
 345
 346         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 347                 struct vsock_sock *vsk;
 348                 list_for_each_entry(vsk, &vsock_connected_table[i],
 349                                     connected_table) {
 350                         if (vsk->transport != transport)
 351                                 continue;
 352
 353                         fn(sk_vsock(vsk));
 354                 }
 355         }
 356
 357         spin_unlock_bh(&vsock_table_lock);
 358 }
 359 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 360
 361 void vsock_add_pending(struct sock *listener, struct sock *pending)
 362 {
 363         struct vsock_sock *vlistener;
 364         struct vsock_sock *vpending;
 365
 366         vlistener = vsock_sk(listener);
 367         vpending = vsock_sk(pending);
 368
 369         sock_hold(pending);
 370         sock_hold(listener);
 371         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 372 }
 373 EXPORT_SYMBOL_GPL(vsock_add_pending);
 374
 375 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 376 {
 377         struct vsock_sock *vpending = vsock_sk(pending);
 378
 379         list_del_init(&vpending->pending_links);
 380         sock_put(listener);
 381         sock_put(pending);
 382 }
 383 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 384
 385 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 386 {
 387         struct vsock_sock *vlistener;
 388         struct vsock_sock *vconnected;
 389
 390         vlistener = vsock_sk(listener);
 391         vconnected = vsock_sk(connected);
 392
 393         sock_hold(connected);
 394         sock_hold(listener);
 395         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 396 }
 397 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 398
 399 static bool vsock_use_local_transport(unsigned int remote_cid)
 400 {
 401         if (!transport_local)
 402                 return false;
 403
 404         if (remote_cid == VMADDR_CID_LOCAL)
 405                 return true;
 406
 407         if (transport_g2h) {
 408                 return remote_cid == transport_g2h->get_local_cid();
 409         } else {
 410                 return remote_cid == VMADDR_CID_HOST;
 411         }
 412 }
 413
 414 static void vsock_deassign_transport(struct vsock_sock *vsk)
 415 {
 416         if (!vsk->transport)
 417                 return;
 418
 419         vsk->transport->destruct(vsk);
 420         module_put(vsk->transport->module);
 421         vsk->transport = NULL;
 422 }
 423
 424 /* Assign a transport to a socket and call the .init transport callback.
 425  *
 426  * Note: for connection oriented socket this must be called when vsk->remote_addr
 427  * is set (e.g. during the connect() or when a connection request on a listener
 428  * socket is received).
 429  * The vsk->remote_addr is used to decide which transport to use:
 430  *  - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
 431  *    g2h is not loaded, will use local transport;
 432  *  - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
 433  *    includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
 434  *  - remote CID > VMADDR_CID_HOST will use host->guest transport;
 435  */
 436 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
 437 {
 438         const struct vsock_transport *new_transport;
 439         struct sock *sk = sk_vsock(vsk);
 440         unsigned int remote_cid = vsk->remote_addr.svm_cid;
 441         __u8 remote_flags;
 442         int ret;
 443
 444         /* If the packet is coming with the source and destination CIDs higher
 445          * than VMADDR_CID_HOST, then a vsock channel where all the packets are
 446          * forwarded to the host should be established. Then the host will
 447          * need to forward the packets to the guest.
 448          *
 449          * The flag is set on the (listen) receive path (psk is not NULL). On
 450          * the connect path the flag can be set by the user space application.
 451          */
 452         if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
 453             vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
 454                 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
 455
 456         remote_flags = vsk->remote_addr.svm_flags;
 457
 458         switch (sk->sk_type) {
 459         case SOCK_DGRAM:
 460                 new_transport = transport_dgram;
 461                 break;
 462         case SOCK_STREAM:
 463         case SOCK_SEQPACKET:
 464                 if (vsock_use_local_transport(remote_cid))
 465                         new_transport = transport_local;
 466                 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
 467                          (remote_flags & VMADDR_FLAG_TO_HOST))
 468                         new_transport = transport_g2h;
 469                 else
 470                         new_transport = transport_h2g;
 471                 break;
 472         default:
 473                 return -ESOCKTNOSUPPORT;
 474         }
 475
 476         if (vsk->transport) {
 477                 if (vsk->transport == new_transport)
 478                         return 0;
 479
 480                 /* transport->release() must be called with sock lock acquired.
 481                  * This path can only be taken during vsock_connect(), where we
 482                  * have already held the sock lock. In the other cases, this
 483                  * function is called on a new socket which is not assigned to
 484                  * any transport.
 485                  */
 486                 vsk->transport->release(vsk);
 487                 vsock_deassign_transport(vsk);
 488         }
 489
 490         /* We increase the module refcnt to prevent the transport unloading
 491          * while there are open sockets assigned to it.
 492          */
 493         if (!new_transport || !try_module_get(new_transport->module))
 494                 return -ENODEV;
 495
 496         if (sk->sk_type == SOCK_SEQPACKET) {
 497                 if (!new_transport->seqpacket_allow ||
 498                     !new_transport->seqpacket_allow(remote_cid)) {
 499                         module_put(new_transport->module);
 500                         return -ESOCKTNOSUPPORT;
 501                 }
 502         }
 503
 504         ret = new_transport->init(vsk, psk);
 505         if (ret) {
 506                 module_put(new_transport->module);
 507                 return ret;
 508         }
 509
 510         vsk->transport = new_transport;
 511
 512         return 0;
 513 }
 514 EXPORT_SYMBOL_GPL(vsock_assign_transport);
 515
 516 bool vsock_find_cid(unsigned int cid)
 517 {
 518         if (transport_g2h && cid == transport_g2h->get_local_cid())
 519                 return true;
 520
 521         if (transport_h2g && cid == VMADDR_CID_HOST)
 522                 return true;
 523
 524         if (transport_local && cid == VMADDR_CID_LOCAL)
 525                 return true;
 526
 527         return false;
 528 }
 529 EXPORT_SYMBOL_GPL(vsock_find_cid);
 530
 531 static struct sock *vsock_dequeue_accept(struct sock *listener)
 532 {
 533         struct vsock_sock *vlistener;
 534         struct vsock_sock *vconnected;
 535
 536         vlistener = vsock_sk(listener);
 537
 538         if (list_empty(&vlistener->accept_queue))
 539                 return NULL;
 540
 541         vconnected = list_entry(vlistener->accept_queue.next,
 542                                 struct vsock_sock, accept_queue);
 543
 544         list_del_init(&vconnected->accept_queue);
 545         sock_put(listener);
 546         /* The caller will need a reference on the connected socket so we let
 547          * it call sock_put().
 548          */
 549
 550         return sk_vsock(vconnected);
 551 }
 552
 553 static bool vsock_is_accept_queue_empty(struct sock *sk)
 554 {
 555         struct vsock_sock *vsk = vsock_sk(sk);
 556         return list_empty(&vsk->accept_queue);
 557 }
 558
 559 static bool vsock_is_pending(struct sock *sk)
 560 {
 561         struct vsock_sock *vsk = vsock_sk(sk);
 562         return !list_empty(&vsk->pending_links);
 563 }
 564
 565 static int vsock_send_shutdown(struct sock *sk, int mode)
 566 {
 567         struct vsock_sock *vsk = vsock_sk(sk);
 568
 569         if (!vsk->transport)
 570                 return -ENODEV;
 571
 572         return vsk->transport->shutdown(vsk, mode);
 573 }
 574
 575 static void vsock_pending_work(struct work_struct *work)
 576 {
 577         struct sock *sk;
 578         struct sock *listener;
 579         struct vsock_sock *vsk;
 580         bool cleanup;
 581
 582         vsk = container_of(work, struct vsock_sock, pending_work.work);
 583         sk = sk_vsock(vsk);
 584         listener = vsk->listener;
 585         cleanup = true;
 586
 587         lock_sock(listener);
 588         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 589
 590         if (vsock_is_pending(sk)) {
 591                 vsock_remove_pending(listener, sk);
 592
 593                 sk_acceptq_removed(listener);
 594         } else if (!vsk->rejected) {
 595                 /* We are not on the pending list and accept() did not reject
 596                  * us, so we must have been accepted by our user process.  We
 597                  * just need to drop our references to the sockets and be on
 598                  * our way.
 599                  */
 600                 cleanup = false;
 601                 goto out;
 602         }
 603
 604         /* We need to remove ourself from the global connected sockets list so
 605          * incoming packets can't find this socket, and to reduce the reference
 606          * count.
 607          */
 608         vsock_remove_connected(vsk);
 609
 610         sk->sk_state = TCP_CLOSE;
 611
 612 out:
 613         release_sock(sk);
 614         release_sock(listener);
 615         if (cleanup)
 616                 sock_put(sk);
 617
 618         sock_put(sk);
 619         sock_put(listener);
 620 }
 621
 622 /**** SOCKET OPERATIONS ****/
 623
 624 static int __vsock_bind_connectible(struct vsock_sock *vsk,
 625                                     struct sockaddr_vm *addr)
 626 {
 627         static u32 port;
 628         struct sockaddr_vm new_addr;
 629
 630         if (!port)
 631                 port = LAST_RESERVED_PORT + 1 +
 632                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 633
 634         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 635
 636         if (addr->svm_port == VMADDR_PORT_ANY) {
 637                 bool found = false;
 638                 unsigned int i;
 639
 640                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 641                         if (port <= LAST_RESERVED_PORT)
 642                                 port = LAST_RESERVED_PORT + 1;
 643
 644                         new_addr.svm_port = port++;
 645
 646                         if (!__vsock_find_bound_socket(&new_addr)) {
 647                                 found = true;
 648                                 break;
 649                         }
 650                 }
 651
 652                 if (!found)
 653                         return -EADDRNOTAVAIL;
 654         } else {
 655                 /* If port is in reserved range, ensure caller
 656                  * has necessary privileges.
 657                  */
 658                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 659                     !capable(CAP_NET_BIND_SERVICE)) {
 660                         return -EACCES;
 661                 }
 662
 663                 if (__vsock_find_bound_socket(&new_addr))
 664                         return -EADDRINUSE;
 665         }
 666
 667         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 668
 669         /* Remove connection oriented sockets from the unbound list and add them
 670          * to the hash table for easy lookup by its address.  The unbound list
 671          * is simply an extra entry at the end of the hash table, a trick used
 672          * by AF_UNIX.
 673          */
 674         __vsock_remove_bound(vsk);
 675         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 676
 677         return 0;
 678 }
 679
 680 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 681                               struct sockaddr_vm *addr)
 682 {
 683         return vsk->transport->dgram_bind(vsk, addr);
 684 }
 685
 686 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 687 {
 688         struct vsock_sock *vsk = vsock_sk(sk);
 689         int retval;
 690
 691         /* First ensure this socket isn't already bound. */
 692         if (vsock_addr_bound(&vsk->local_addr))
 693                 return -EINVAL;
 694
 695         /* Now bind to the provided address or select appropriate values if
 696          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 697          * like AF_INET prevents binding to a non-local IP address (in most
 698          * cases), we only allow binding to a local CID.
 699          */
 700         if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
 701                 return -EADDRNOTAVAIL;
 702
 703         switch (sk->sk_socket->type) {
 704         case SOCK_STREAM:
 705         case SOCK_SEQPACKET:
 706                 spin_lock_bh(&vsock_table_lock);
 707                 retval = __vsock_bind_connectible(vsk, addr);
 708                 spin_unlock_bh(&vsock_table_lock);
 709                 break;
 710
 711         case SOCK_DGRAM:
 712                 retval = __vsock_bind_dgram(vsk, addr);
 713                 break;
 714
 715         default:
 716                 retval = -EINVAL;
 717                 break;
 718         }
 719
 720         return retval;
 721 }
 722
 723 static void vsock_connect_timeout(struct work_struct *work);
 724
 725 static struct sock *__vsock_create(struct net *net,
 726                                    struct socket *sock,
 727                                    struct sock *parent,
 728                                    gfp_t priority,
 729                                    unsigned short type,
 730                                    int kern)
 731 {
 732         struct sock *sk;
 733         struct vsock_sock *psk;
 734         struct vsock_sock *vsk;
 735
 736         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 737         if (!sk)
 738                 return NULL;
 739
 740         sock_init_data(sock, sk);
 741
 742         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 743          * non-NULL. We make sure that our sockets always have a type by
 744          * setting it here if needed.
 745          */
 746         if (!sock)
 747                 sk->sk_type = type;
 748
 749         vsk = vsock_sk(sk);
 750         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 751         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 752
 753         sk->sk_destruct = vsock_sk_destruct;
 754         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 755         sock_reset_flag(sk, SOCK_DONE);
 756
 757         INIT_LIST_HEAD(&vsk->bound_table);
 758         INIT_LIST_HEAD(&vsk->connected_table);
 759         vsk->listener = NULL;
 760         INIT_LIST_HEAD(&vsk->pending_links);
 761         INIT_LIST_HEAD(&vsk->accept_queue);
 762         vsk->rejected = false;
 763         vsk->sent_request = false;
 764         vsk->ignore_connecting_rst = false;
 765         vsk->peer_shutdown = 0;
 766         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 767         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 768
 769         psk = parent ? vsock_sk(parent) : NULL;
 770         if (parent) {
 771                 vsk->trusted = psk->trusted;
 772                 vsk->owner = get_cred(psk->owner);
 773                 vsk->connect_timeout = psk->connect_timeout;
 774                 vsk->buffer_size = psk->buffer_size;
 775                 vsk->buffer_min_size = psk->buffer_min_size;
 776                 vsk->buffer_max_size = psk->buffer_max_size;
 777                 security_sk_clone(parent, sk);
 778         } else {
 779                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 780                 vsk->owner = get_current_cred();
 781                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 782                 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
 783                 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
 784                 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
 785         }
 786
 787         return sk;
 788 }
 789
 790 static bool sock_type_connectible(u16 type)
 791 {
 792         return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
 793 }
 794
 795 static void __vsock_release(struct sock *sk, int level)
 796 {
 797         if (sk) {
 798                 struct sock *pending;
 799                 struct vsock_sock *vsk;
 800
 801                 vsk = vsock_sk(sk);
 802                 pending = NULL; /* Compiler warning. */
 803
 804                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 805                  * version to avoid the warning "possible recursive locking
 806                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 807                  * is the same as lock_sock(sk).
 808                  */
 809                 lock_sock_nested(sk, level);
 810
 811                 if (vsk->transport)
 812                         vsk->transport->release(vsk);
 813                 else if (sock_type_connectible(sk->sk_type))
 814                         vsock_remove_sock(vsk);
 815
 816                 sock_orphan(sk);
 817                 sk->sk_shutdown = SHUTDOWN_MASK;
 818
 819                 skb_queue_purge(&sk->sk_receive_queue);
 820
 821                 /* Clean up any sockets that never were accepted. */
 822                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 823                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 824                         sock_put(pending);
 825                 }
 826
 827                 release_sock(sk);
 828                 sock_put(sk);
 829         }
 830 }
 831
 832 static void vsock_sk_destruct(struct sock *sk)
 833 {
 834         struct vsock_sock *vsk = vsock_sk(sk);
 835
 836         vsock_deassign_transport(vsk);
 837
 838         /* When clearing these addresses, there's no need to set the family and
 839          * possibly register the address family with the kernel.
 840          */
 841         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 842         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 843
 844         put_cred(vsk->owner);
 845 }
 846
 847 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 848 {
 849         int err;
 850
 851         err = sock_queue_rcv_skb(sk, skb);
 852         if (err)
 853                 kfree_skb(skb);
 854
 855         return err;
 856 }
 857
 858 struct sock *vsock_create_connected(struct sock *parent)
 859 {
 860         return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
 861                               parent->sk_type, 0);
 862 }
 863 EXPORT_SYMBOL_GPL(vsock_create_connected);
 864
 865 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 866 {
 867         return vsk->transport->stream_has_data(vsk);
 868 }
 869 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 870
 871 static s64 vsock_connectible_has_data(struct vsock_sock *vsk)
 872 {
 873         struct sock *sk = sk_vsock(vsk);
 874
 875         if (sk->sk_type == SOCK_SEQPACKET)
 876                 return vsk->transport->seqpacket_has_data(vsk);
 877         else
 878                 return vsock_stream_has_data(vsk);
 879 }
 880
 881 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 882 {
 883         return vsk->transport->stream_has_space(vsk);
 884 }
 885 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 886
 887 void vsock_data_ready(struct sock *sk)
 888 {
 889         struct vsock_sock *vsk = vsock_sk(sk);
 890
 891         if (vsock_stream_has_data(vsk) >= sk->sk_rcvlowat ||
 892             sock_flag(sk, SOCK_DONE))
 893                 sk->sk_data_ready(sk);
 894 }
 895 EXPORT_SYMBOL_GPL(vsock_data_ready);
 896
 897 static int vsock_release(struct socket *sock)
 898 {
 899         __vsock_release(sock->sk, 0);
 900         sock->sk = NULL;
 901         sock->state = SS_FREE;
 902
 903         return 0;
 904 }
 905
 906 static int
 907 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 908 {
 909         int err;
 910         struct sock *sk;
 911         struct sockaddr_vm *vm_addr;
 912
 913         sk = sock->sk;
 914
 915         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 916                 return -EINVAL;
 917
 918         lock_sock(sk);
 919         err = __vsock_bind(sk, vm_addr);
 920         release_sock(sk);
 921
 922         return err;
 923 }
 924
 925 static int vsock_getname(struct socket *sock,
 926                          struct sockaddr *addr, int peer)
 927 {
 928         int err;
 929         struct sock *sk;
 930         struct vsock_sock *vsk;
 931         struct sockaddr_vm *vm_addr;
 932
 933         sk = sock->sk;
 934         vsk = vsock_sk(sk);
 935         err = 0;
 936
 937         lock_sock(sk);
 938
 939         if (peer) {
 940                 if (sock->state != SS_CONNECTED) {
 941                         err = -ENOTCONN;
 942                         goto out;
 943                 }
 944                 vm_addr = &vsk->remote_addr;
 945         } else {
 946                 vm_addr = &vsk->local_addr;
 947         }
 948
 949         if (!vm_addr) {
 950                 err = -EINVAL;
 951                 goto out;
 952         }
 953
 954         /* sys_getsockname() and sys_getpeername() pass us a
 955          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 956          * that macro is defined in socket.c instead of .h, so we hardcode its
 957          * value here.
 958          */
 959         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 960         memcpy(addr, vm_addr, sizeof(*vm_addr));
 961         err = sizeof(*vm_addr);
 962
 963 out:
 964         release_sock(sk);
 965         return err;
 966 }
 967
 968 static int vsock_shutdown(struct socket *sock, int mode)
 969 {
 970         int err;
 971         struct sock *sk;
 972
 973         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 974          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 975          * here like the other address families do.  Note also that the
 976          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 977          * which is what we want.
 978          */
 979         mode++;
 980
 981         if ((mode & ~SHUTDOWN_MASK) || !mode)
 982                 return -EINVAL;
 983
 984         /* If this is a connection oriented socket and it is not connected then
 985          * bail out immediately.  If it is a DGRAM socket then we must first
 986          * kick the socket so that it wakes up from any sleeping calls, for
 987          * example recv(), and then afterwards return the error.
 988          */
 989
 990         sk = sock->sk;
 991
 992         lock_sock(sk);
 993         if (sock->state == SS_UNCONNECTED) {
 994                 err = -ENOTCONN;
 995                 if (sock_type_connectible(sk->sk_type))
 996                         goto out;
 997         } else {
 998                 sock->state = SS_DISCONNECTING;
 999                 err = 0;
1000         }
1001
1002         /* Receive and send shutdowns are treated alike. */
1003         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
1004         if (mode) {
1005                 sk->sk_shutdown |= mode;
1006                 sk->sk_state_change(sk);
1007
1008                 if (sock_type_connectible(sk->sk_type)) {
1009                         sock_reset_flag(sk, SOCK_DONE);
1010                         vsock_send_shutdown(sk, mode);
1011                 }
1012         }
1013
1014 out:
1015         release_sock(sk);
1016         return err;
1017 }
1018
1019 static __poll_t vsock_poll(struct file *file, struct socket *sock,
1020                                poll_table *wait)
1021 {
1022         struct sock *sk;
1023         __poll_t mask;
1024         struct vsock_sock *vsk;
1025
1026         sk = sock->sk;
1027         vsk = vsock_sk(sk);
1028
1029         poll_wait(file, sk_sleep(sk), wait);
1030         mask = 0;
1031
1032         if (sk->sk_err)
1033                 /* Signify that there has been an error on this socket. */
1034                 mask |= EPOLLERR;
1035
1036         /* INET sockets treat local write shutdown and peer write shutdown as a
1037          * case of EPOLLHUP set.
1038          */
1039         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1040             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1041              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1042                 mask |= EPOLLHUP;
1043         }
1044
1045         if (sk->sk_shutdown & RCV_SHUTDOWN ||
1046             vsk->peer_shutdown & SEND_SHUTDOWN) {
1047                 mask |= EPOLLRDHUP;
1048         }
1049
1050         if (sock->type == SOCK_DGRAM) {
1051                 /* For datagram sockets we can read if there is something in
1052                  * the queue and write as long as the socket isn't shutdown for
1053                  * sending.
1054                  */
1055                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1056                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
1057                         mask |= EPOLLIN | EPOLLRDNORM;
1058                 }
1059
1060                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1061                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1062
1063         } else if (sock_type_connectible(sk->sk_type)) {
1064                 const struct vsock_transport *transport;
1065
1066                 lock_sock(sk);
1067
1068                 transport = vsk->transport;
1069
1070                 /* Listening sockets that have connections in their accept
1071                  * queue can be read.
1072                  */
1073                 if (sk->sk_state == TCP_LISTEN
1074                     && !vsock_is_accept_queue_empty(sk))
1075                         mask |= EPOLLIN | EPOLLRDNORM;
1076
1077                 /* If there is something in the queue then we can read. */
1078                 if (transport && transport->stream_is_active(vsk) &&
1079                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1080                         bool data_ready_now = false;
1081                         int target = sock_rcvlowat(sk, 0, INT_MAX);
1082                         int ret = transport->notify_poll_in(
1083                                         vsk, target, &data_ready_now);
1084                         if (ret < 0) {
1085                                 mask |= EPOLLERR;
1086                         } else {
1087                                 if (data_ready_now)
1088                                         mask |= EPOLLIN | EPOLLRDNORM;
1089
1090                         }
1091                 }
1092
1093                 /* Sockets whose connections have been closed, reset, or
1094                  * terminated should also be considered read, and we check the
1095                  * shutdown flag for that.
1096                  */
1097                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1098                     vsk->peer_shutdown & SEND_SHUTDOWN) {
1099                         mask |= EPOLLIN | EPOLLRDNORM;
1100                 }
1101
1102                 /* Connected sockets that can produce data can be written. */
1103                 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1104                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1105                                 bool space_avail_now = false;
1106                                 int ret = transport->notify_poll_out(
1107                                                 vsk, 1, &space_avail_now);
1108                                 if (ret < 0) {
1109                                         mask |= EPOLLERR;
1110                                 } else {
1111                                         if (space_avail_now)
1112                                                 /* Remove EPOLLWRBAND since INET
1113                                                  * sockets are not setting it.
1114                                                  */
1115                                                 mask |= EPOLLOUT | EPOLLWRNORM;
1116
1117                                 }
1118                         }
1119                 }
1120
1121                 /* Simulate INET socket poll behaviors, which sets
1122                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1123                  * but local send is not shutdown.
1124                  */
1125                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1126                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1127                                 mask |= EPOLLOUT | EPOLLWRNORM;
1128
1129                 }
1130
1131                 release_sock(sk);
1132         }
1133
1134         return mask;
1135 }
1136
1137 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1138                                size_t len)
1139 {
1140         int err;
1141         struct sock *sk;
1142         struct vsock_sock *vsk;
1143         struct sockaddr_vm *remote_addr;
1144         const struct vsock_transport *transport;
1145
1146         if (msg->msg_flags & MSG_OOB)
1147                 return -EOPNOTSUPP;
1148
1149         /* For now, MSG_DONTWAIT is always assumed... */
1150         err = 0;
1151         sk = sock->sk;
1152         vsk = vsock_sk(sk);
1153
1154         lock_sock(sk);
1155
1156         transport = vsk->transport;
1157
1158         err = vsock_auto_bind(vsk);
1159         if (err)
1160                 goto out;
1161
1162
1163         /* If the provided message contains an address, use that.  Otherwise
1164          * fall back on the socket's remote handle (if it has been connected).
1165          */
1166         if (msg->msg_name &&
1167             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1168                             &remote_addr) == 0) {
1169                 /* Ensure this address is of the right type and is a valid
1170                  * destination.
1171                  */
1172
1173                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1174                         remote_addr->svm_cid = transport->get_local_cid();
1175
1176                 if (!vsock_addr_bound(remote_addr)) {
1177                         err = -EINVAL;
1178                         goto out;
1179                 }
1180         } else if (sock->state == SS_CONNECTED) {
1181                 remote_addr = &vsk->remote_addr;
1182
1183                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1184                         remote_addr->svm_cid = transport->get_local_cid();
1185
1186                 /* XXX Should connect() or this function ensure remote_addr is
1187                  * bound?
1188                  */
1189                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1190                         err = -EINVAL;
1191                         goto out;
1192                 }
1193         } else {
1194                 err = -EINVAL;
1195                 goto out;
1196         }
1197
1198         if (!transport->dgram_allow(remote_addr->svm_cid,
1199                                     remote_addr->svm_port)) {
1200                 err = -EINVAL;
1201                 goto out;
1202         }
1203
1204         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1205
1206 out:
1207         release_sock(sk);
1208         return err;
1209 }
1210
1211 static int vsock_dgram_connect(struct socket *sock,
1212                                struct sockaddr *addr, int addr_len, int flags)
1213 {
1214         int err;
1215         struct sock *sk;
1216         struct vsock_sock *vsk;
1217         struct sockaddr_vm *remote_addr;
1218
1219         sk = sock->sk;
1220         vsk = vsock_sk(sk);
1221
1222         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1223         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1224                 lock_sock(sk);
1225                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1226                                 VMADDR_PORT_ANY);
1227                 sock->state = SS_UNCONNECTED;
1228                 release_sock(sk);
1229                 return 0;
1230         } else if (err != 0)
1231                 return -EINVAL;
1232
1233         lock_sock(sk);
1234
1235         err = vsock_auto_bind(vsk);
1236         if (err)
1237                 goto out;
1238
1239         if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1240                                          remote_addr->svm_port)) {
1241                 err = -EINVAL;
1242                 goto out;
1243         }
1244
1245         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1246         sock->state = SS_CONNECTED;
1247
1248 out:
1249         release_sock(sk);
1250         return err;
1251 }
1252
1253 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1254                                size_t len, int flags)
1255 {
1256         struct vsock_sock *vsk = vsock_sk(sock->sk);
1257
1258         return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1259 }
1260
1261 static const struct proto_ops vsock_dgram_ops = {
1262         .family = PF_VSOCK,
1263         .owner = THIS_MODULE,
1264         .release = vsock_release,
1265         .bind = vsock_bind,
1266         .connect = vsock_dgram_connect,
1267         .socketpair = sock_no_socketpair,
1268         .accept = sock_no_accept,
1269         .getname = vsock_getname,
1270         .poll = vsock_poll,
1271         .ioctl = sock_no_ioctl,
1272         .listen = sock_no_listen,
1273         .shutdown = vsock_shutdown,
1274         .sendmsg = vsock_dgram_sendmsg,
1275         .recvmsg = vsock_dgram_recvmsg,
1276         .mmap = sock_no_mmap,
1277         .sendpage = sock_no_sendpage,
1278 };
1279
1280 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1281 {
1282         const struct vsock_transport *transport = vsk->transport;
1283
1284         if (!transport || !transport->cancel_pkt)
1285                 return -EOPNOTSUPP;
1286
1287         return transport->cancel_pkt(vsk);
1288 }
1289
1290 static void vsock_connect_timeout(struct work_struct *work)
1291 {
1292         struct sock *sk;
1293         struct vsock_sock *vsk;
1294
1295         vsk = container_of(work, struct vsock_sock, connect_work.work);
1296         sk = sk_vsock(vsk);
1297
1298         lock_sock(sk);
1299         if (sk->sk_state == TCP_SYN_SENT &&
1300             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1301                 sk->sk_state = TCP_CLOSE;
1302                 sk->sk_socket->state = SS_UNCONNECTED;
1303                 sk->sk_err = ETIMEDOUT;
1304                 sk_error_report(sk);
1305                 vsock_transport_cancel_pkt(vsk);
1306         }
1307         release_sock(sk);
1308
1309         sock_put(sk);
1310 }
1311
1312 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1313                          int addr_len, int flags)
1314 {
1315         int err;
1316         struct sock *sk;
1317         struct vsock_sock *vsk;
1318         const struct vsock_transport *transport;
1319         struct sockaddr_vm *remote_addr;
1320         long timeout;
1321         DEFINE_WAIT(wait);
1322
1323         err = 0;
1324         sk = sock->sk;
1325         vsk = vsock_sk(sk);
1326
1327         lock_sock(sk);
1328
1329         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1330         switch (sock->state) {
1331         case SS_CONNECTED:
1332                 err = -EISCONN;
1333                 goto out;
1334         case SS_DISCONNECTING:
1335                 err = -EINVAL;
1336                 goto out;
1337         case SS_CONNECTING:
1338                 /* This continues on so we can move sock into the SS_CONNECTED
1339                  * state once the connection has completed (at which point err
1340                  * will be set to zero also).  Otherwise, we will either wait
1341                  * for the connection or return -EALREADY should this be a
1342                  * non-blocking call.
1343                  */
1344                 err = -EALREADY;
1345                 if (flags & O_NONBLOCK)
1346                         goto out;
1347                 break;
1348         default:
1349                 if ((sk->sk_state == TCP_LISTEN) ||
1350                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1351                         err = -EINVAL;
1352                         goto out;
1353                 }
1354
1355                 /* Set the remote address that we are connecting to. */
1356                 memcpy(&vsk->remote_addr, remote_addr,
1357                        sizeof(vsk->remote_addr));
1358
1359                 err = vsock_assign_transport(vsk, NULL);
1360                 if (err)
1361                         goto out;
1362
1363                 transport = vsk->transport;
1364
1365                 /* The hypervisor and well-known contexts do not have socket
1366                  * endpoints.
1367                  */
1368                 if (!transport ||
1369                     !transport->stream_allow(remote_addr->svm_cid,
1370                                              remote_addr->svm_port)) {
1371                         err = -ENETUNREACH;
1372                         goto out;
1373                 }
1374
1375                 err = vsock_auto_bind(vsk);
1376                 if (err)
1377                         goto out;
1378
1379                 sk->sk_state = TCP_SYN_SENT;
1380
1381                 err = transport->connect(vsk);
1382                 if (err < 0)
1383                         goto out;
1384
1385                 /* Mark sock as connecting and set the error code to in
1386                  * progress in case this is a non-blocking connect.
1387                  */
1388                 sock->state = SS_CONNECTING;
1389                 err = -EINPROGRESS;
1390         }
1391
1392         /* The receive path will handle all communication until we are able to
1393          * enter the connected state.  Here we wait for the connection to be
1394          * completed or a notification of an error.
1395          */
1396         timeout = vsk->connect_timeout;
1397         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1398
1399         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1400                 if (flags & O_NONBLOCK) {
1401                         /* If we're not going to block, we schedule a timeout
1402                          * function to generate a timeout on the connection
1403                          * attempt, in case the peer doesn't respond in a
1404                          * timely manner. We hold on to the socket until the
1405                          * timeout fires.
1406                          */
1407                         sock_hold(sk);
1408
1409                         /* If the timeout function is already scheduled,
1410                          * reschedule it, then ungrab the socket refcount to
1411                          * keep it balanced.
1412                          */
1413                         if (mod_delayed_work(system_wq, &vsk->connect_work,
1414                                              timeout))
1415                                 sock_put(sk);
1416
1417                         /* Skip ahead to preserve error code set above. */
1418                         goto out_wait;
1419                 }
1420
1421                 release_sock(sk);
1422                 timeout = schedule_timeout(timeout);
1423                 lock_sock(sk);
1424
1425                 if (signal_pending(current)) {
1426                         err = sock_intr_errno(timeout);
1427                         sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1428                         sock->state = SS_UNCONNECTED;
1429                         vsock_transport_cancel_pkt(vsk);
1430                         vsock_remove_connected(vsk);
1431                         goto out_wait;
1432                 } else if ((sk->sk_state != TCP_ESTABLISHED) && (timeout == 0)) {
1433                         err = -ETIMEDOUT;
1434                         sk->sk_state = TCP_CLOSE;
1435                         sock->state = SS_UNCONNECTED;
1436                         vsock_transport_cancel_pkt(vsk);
1437                         goto out_wait;
1438                 }
1439
1440                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1441         }
1442
1443         if (sk->sk_err) {
1444                 err = -sk->sk_err;
1445                 sk->sk_state = TCP_CLOSE;
1446                 sock->state = SS_UNCONNECTED;
1447         } else {
1448                 err = 0;
1449         }
1450
1451 out_wait:
1452         finish_wait(sk_sleep(sk), &wait);
1453 out:
1454         release_sock(sk);
1455         return err;
1456 }
1457
1458 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1459                         bool kern)
1460 {
1461         struct sock *listener;
1462         int err;
1463         struct sock *connected;
1464         struct vsock_sock *vconnected;
1465         long timeout;
1466         DEFINE_WAIT(wait);
1467
1468         err = 0;
1469         listener = sock->sk;
1470
1471         lock_sock(listener);
1472
1473         if (!sock_type_connectible(sock->type)) {
1474                 err = -EOPNOTSUPP;
1475                 goto out;
1476         }
1477
1478         if (listener->sk_state != TCP_LISTEN) {
1479                 err = -EINVAL;
1480                 goto out;
1481         }
1482
1483         /* Wait for children sockets to appear; these are the new sockets
1484          * created upon connection establishment.
1485          */
1486         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1487         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1488
1489         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1490                listener->sk_err == 0) {
1491                 release_sock(listener);
1492                 timeout = schedule_timeout(timeout);
1493                 finish_wait(sk_sleep(listener), &wait);
1494                 lock_sock(listener);
1495
1496                 if (signal_pending(current)) {
1497                         err = sock_intr_errno(timeout);
1498                         goto out;
1499                 } else if (timeout == 0) {
1500                         err = -EAGAIN;
1501                         goto out;
1502                 }
1503
1504                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1505         }
1506         finish_wait(sk_sleep(listener), &wait);
1507
1508         if (listener->sk_err)
1509                 err = -listener->sk_err;
1510
1511         if (connected) {
1512                 sk_acceptq_removed(listener);
1513
1514                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1515                 vconnected = vsock_sk(connected);
1516
1517                 /* If the listener socket has received an error, then we should
1518                  * reject this socket and return.  Note that we simply mark the
1519                  * socket rejected, drop our reference, and let the cleanup
1520                  * function handle the cleanup; the fact that we found it in
1521                  * the listener's accept queue guarantees that the cleanup
1522                  * function hasn't run yet.
1523                  */
1524                 if (err) {
1525                         vconnected->rejected = true;
1526                 } else {
1527                         newsock->state = SS_CONNECTED;
1528                         sock_graft(connected, newsock);
1529                 }
1530
1531                 release_sock(connected);
1532                 sock_put(connected);
1533         }
1534
1535 out:
1536         release_sock(listener);
1537         return err;
1538 }
1539
1540 static int vsock_listen(struct socket *sock, int backlog)
1541 {
1542         int err;
1543         struct sock *sk;
1544         struct vsock_sock *vsk;
1545
1546         sk = sock->sk;
1547
1548         lock_sock(sk);
1549
1550         if (!sock_type_connectible(sk->sk_type)) {
1551                 err = -EOPNOTSUPP;
1552                 goto out;
1553         }
1554
1555         if (sock->state != SS_UNCONNECTED) {
1556                 err = -EINVAL;
1557                 goto out;
1558         }
1559
1560         vsk = vsock_sk(sk);
1561
1562         if (!vsock_addr_bound(&vsk->local_addr)) {
1563                 err = -EINVAL;
1564                 goto out;
1565         }
1566
1567         sk->sk_max_ack_backlog = backlog;
1568         sk->sk_state = TCP_LISTEN;
1569
1570         err = 0;
1571
1572 out:
1573         release_sock(sk);
1574         return err;
1575 }
1576
1577 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1578                                      const struct vsock_transport *transport,
1579                                      u64 val)
1580 {
1581         if (val > vsk->buffer_max_size)
1582                 val = vsk->buffer_max_size;
1583
1584         if (val < vsk->buffer_min_size)
1585                 val = vsk->buffer_min_size;
1586
1587         if (val != vsk->buffer_size &&
1588             transport && transport->notify_buffer_size)
1589                 transport->notify_buffer_size(vsk, &val);
1590
1591         vsk->buffer_size = val;
1592 }
1593
1594 static int vsock_connectible_setsockopt(struct socket *sock,
1595                                         int level,
1596                                         int optname,
1597                                         sockptr_t optval,
1598                                         unsigned int optlen)
1599 {
1600         int err;
1601         struct sock *sk;
1602         struct vsock_sock *vsk;
1603         const struct vsock_transport *transport;
1604         u64 val;
1605
1606         if (level != AF_VSOCK)
1607                 return -ENOPROTOOPT;
1608
1609 #define COPY_IN(_v)                                       \
1610         do {                                              \
1611                 if (optlen < sizeof(_v)) {                \
1612                         err = -EINVAL;                    \
1613                         goto exit;                        \
1614                 }                                         \
1615                 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) {  \
1616                         err = -EFAULT;                                  \
1617                         goto exit;                                      \
1618                 }                                                       \
1619         } while (0)
1620
1621         err = 0;
1622         sk = sock->sk;
1623         vsk = vsock_sk(sk);
1624
1625         lock_sock(sk);
1626
1627         transport = vsk->transport;
1628
1629         switch (optname) {
1630         case SO_VM_SOCKETS_BUFFER_SIZE:
1631                 COPY_IN(val);
1632                 vsock_update_buffer_size(vsk, transport, val);
1633                 break;
1634
1635         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1636                 COPY_IN(val);
1637                 vsk->buffer_max_size = val;
1638                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1639                 break;
1640
1641         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1642                 COPY_IN(val);
1643                 vsk->buffer_min_size = val;
1644                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1645                 break;
1646
1647         case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1648         case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: {
1649                 struct __kernel_sock_timeval tv;
1650
1651                 err = sock_copy_user_timeval(&tv, optval, optlen,
1652                                              optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1653                 if (err)
1654                         break;
1655                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1656                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1657                         vsk->connect_timeout = tv.tv_sec * HZ +
1658                                 DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ));
1659                         if (vsk->connect_timeout == 0)
1660                                 vsk->connect_timeout =
1661                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1662
1663                 } else {
1664                         err = -ERANGE;
1665                 }
1666                 break;
1667         }
1668
1669         default:
1670                 err = -ENOPROTOOPT;
1671                 break;
1672         }
1673
1674 #undef COPY_IN
1675
1676 exit:
1677         release_sock(sk);
1678         return err;
1679 }
1680
1681 static int vsock_connectible_getsockopt(struct socket *sock,
1682                                         int level, int optname,
1683                                         char __user *optval,
1684                                         int __user *optlen)
1685 {
1686         struct sock *sk = sock->sk;
1687         struct vsock_sock *vsk = vsock_sk(sk);
1688
1689         union {
1690                 u64 val64;
1691                 struct old_timeval32 tm32;
1692                 struct __kernel_old_timeval tm;
1693                 struct  __kernel_sock_timeval stm;
1694         } v;
1695
1696         int lv = sizeof(v.val64);
1697         int len;
1698
1699         if (level != AF_VSOCK)
1700                 return -ENOPROTOOPT;
1701
1702         if (get_user(len, optlen))
1703                 return -EFAULT;
1704
1705         memset(&v, 0, sizeof(v));
1706
1707         switch (optname) {
1708         case SO_VM_SOCKETS_BUFFER_SIZE:
1709                 v.val64 = vsk->buffer_size;
1710                 break;
1711
1712         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1713                 v.val64 = vsk->buffer_max_size;
1714                 break;
1715
1716         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1717                 v.val64 = vsk->buffer_min_size;
1718                 break;
1719
1720         case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1721         case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD:
1722                 lv = sock_get_timeout(vsk->connect_timeout, &v,
1723                                       optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1724                 break;
1725
1726         default:
1727                 return -ENOPROTOOPT;
1728         }
1729
1730         if (len < lv)
1731                 return -EINVAL;
1732         if (len > lv)
1733                 len = lv;
1734         if (copy_to_user(optval, &v, len))
1735                 return -EFAULT;
1736
1737         if (put_user(len, optlen))
1738                 return -EFAULT;
1739
1740         return 0;
1741 }
1742
1743 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1744                                      size_t len)
1745 {
1746         struct sock *sk;
1747         struct vsock_sock *vsk;
1748         const struct vsock_transport *transport;
1749         ssize_t total_written;
1750         long timeout;
1751         int err;
1752         struct vsock_transport_send_notify_data send_data;
1753         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1754
1755         sk = sock->sk;
1756         vsk = vsock_sk(sk);
1757         total_written = 0;
1758         err = 0;
1759
1760         if (msg->msg_flags & MSG_OOB)
1761                 return -EOPNOTSUPP;
1762
1763         lock_sock(sk);
1764
1765         transport = vsk->transport;
1766
1767         /* Callers should not provide a destination with connection oriented
1768          * sockets.
1769          */
1770         if (msg->msg_namelen) {
1771                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1772                 goto out;
1773         }
1774
1775         /* Send data only if both sides are not shutdown in the direction. */
1776         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1777             vsk->peer_shutdown & RCV_SHUTDOWN) {
1778                 err = -EPIPE;
1779                 goto out;
1780         }
1781
1782         if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1783             !vsock_addr_bound(&vsk->local_addr)) {
1784                 err = -ENOTCONN;
1785                 goto out;
1786         }
1787
1788         if (!vsock_addr_bound(&vsk->remote_addr)) {
1789                 err = -EDESTADDRREQ;
1790                 goto out;
1791         }
1792
1793         /* Wait for room in the produce queue to enqueue our user's data. */
1794         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1795
1796         err = transport->notify_send_init(vsk, &send_data);
1797         if (err < 0)
1798                 goto out;
1799
1800         while (total_written < len) {
1801                 ssize_t written;
1802
1803                 add_wait_queue(sk_sleep(sk), &wait);
1804                 while (vsock_stream_has_space(vsk) == 0 &&
1805                        sk->sk_err == 0 &&
1806                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1807                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1808
1809                         /* Don't wait for non-blocking sockets. */
1810                         if (timeout == 0) {
1811                                 err = -EAGAIN;
1812                                 remove_wait_queue(sk_sleep(sk), &wait);
1813                                 goto out_err;
1814                         }
1815
1816                         err = transport->notify_send_pre_block(vsk, &send_data);
1817                         if (err < 0) {
1818                                 remove_wait_queue(sk_sleep(sk), &wait);
1819                                 goto out_err;
1820                         }
1821
1822                         release_sock(sk);
1823                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1824                         lock_sock(sk);
1825                         if (signal_pending(current)) {
1826                                 err = sock_intr_errno(timeout);
1827                                 remove_wait_queue(sk_sleep(sk), &wait);
1828                                 goto out_err;
1829                         } else if (timeout == 0) {
1830                                 err = -EAGAIN;
1831                                 remove_wait_queue(sk_sleep(sk), &wait);
1832                                 goto out_err;
1833                         }
1834                 }
1835                 remove_wait_queue(sk_sleep(sk), &wait);
1836
1837                 /* These checks occur both as part of and after the loop
1838                  * conditional since we need to check before and after
1839                  * sleeping.
1840                  */
1841                 if (sk->sk_err) {
1842                         err = -sk->sk_err;
1843                         goto out_err;
1844                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1845                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1846                         err = -EPIPE;
1847                         goto out_err;
1848                 }
1849
1850                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1851                 if (err < 0)
1852                         goto out_err;
1853
1854                 /* Note that enqueue will only write as many bytes as are free
1855                  * in the produce queue, so we don't need to ensure len is
1856                  * smaller than the queue size.  It is the caller's
1857                  * responsibility to check how many bytes we were able to send.
1858                  */
1859
1860                 if (sk->sk_type == SOCK_SEQPACKET) {
1861                         written = transport->seqpacket_enqueue(vsk,
1862                                                 msg, len - total_written);
1863                 } else {
1864                         written = transport->stream_enqueue(vsk,
1865                                         msg, len - total_written);
1866                 }
1867                 if (written < 0) {
1868                         err = -ENOMEM;
1869                         goto out_err;
1870                 }
1871
1872                 total_written += written;
1873
1874                 err = transport->notify_send_post_enqueue(
1875                                 vsk, written, &send_data);
1876                 if (err < 0)
1877                         goto out_err;
1878
1879         }
1880
1881 out_err:
1882         if (total_written > 0) {
1883                 /* Return number of written bytes only if:
1884                  * 1) SOCK_STREAM socket.
1885                  * 2) SOCK_SEQPACKET socket when whole buffer is sent.
1886                  */
1887                 if (sk->sk_type == SOCK_STREAM || total_written == len)
1888                         err = total_written;
1889         }
1890 out:
1891         release_sock(sk);
1892         return err;
1893 }
1894
1895 static int vsock_connectible_wait_data(struct sock *sk,
1896                                        struct wait_queue_entry *wait,
1897                                        long timeout,
1898                                        struct vsock_transport_recv_notify_data *recv_data,
1899                                        size_t target)
1900 {
1901         const struct vsock_transport *transport;
1902         struct vsock_sock *vsk;
1903         s64 data;
1904         int err;
1905
1906         vsk = vsock_sk(sk);
1907         err = 0;
1908         transport = vsk->transport;
1909
1910         while (1) {
1911                 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
1912                 data = vsock_connectible_has_data(vsk);
1913                 if (data != 0)
1914                         break;
1915
1916                 if (sk->sk_err != 0 ||
1917                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1918                     (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1919                         break;
1920                 }
1921
1922                 /* Don't wait for non-blocking sockets. */
1923                 if (timeout == 0) {
1924                         err = -EAGAIN;
1925                         break;
1926                 }
1927
1928                 if (recv_data) {
1929                         err = transport->notify_recv_pre_block(vsk, target, recv_data);
1930                         if (err < 0)
1931                                 break;
1932                 }
1933
1934                 release_sock(sk);
1935                 timeout = schedule_timeout(timeout);
1936                 lock_sock(sk);
1937
1938                 if (signal_pending(current)) {
1939                         err = sock_intr_errno(timeout);
1940                         break;
1941                 } else if (timeout == 0) {
1942                         err = -EAGAIN;
1943                         break;
1944                 }
1945         }
1946
1947         finish_wait(sk_sleep(sk), wait);
1948
1949         if (err)
1950                 return err;
1951
1952         /* Internal transport error when checking for available
1953          * data. XXX This should be changed to a connection
1954          * reset in a later change.
1955          */
1956         if (data < 0)
1957                 return -ENOMEM;
1958
1959         return data;
1960 }
1961
1962 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
1963                                   size_t len, int flags)
1964 {
1965         struct vsock_transport_recv_notify_data recv_data;
1966         const struct vsock_transport *transport;
1967         struct vsock_sock *vsk;
1968         ssize_t copied;
1969         size_t target;
1970         long timeout;
1971         int err;
1972
1973         DEFINE_WAIT(wait);
1974
1975         vsk = vsock_sk(sk);
1976         transport = vsk->transport;
1977
1978         /* We must not copy less than target bytes into the user's buffer
1979          * before returning successfully, so we wait for the consume queue to
1980          * have that much data to consume before dequeueing.  Note that this
1981          * makes it impossible to handle cases where target is greater than the
1982          * queue size.
1983          */
1984         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1985         if (target >= transport->stream_rcvhiwat(vsk)) {
1986                 err = -ENOMEM;
1987                 goto out;
1988         }
1989         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1990         copied = 0;
1991
1992         err = transport->notify_recv_init(vsk, target, &recv_data);
1993         if (err < 0)
1994                 goto out;
1995
1996
1997         while (1) {
1998                 ssize_t read;
1999
2000                 err = vsock_connectible_wait_data(sk, &wait, timeout,
2001                                                   &recv_data, target);
2002                 if (err <= 0)
2003                         break;
2004
2005                 err = transport->notify_recv_pre_dequeue(vsk, target,
2006                                                          &recv_data);
2007                 if (err < 0)
2008                         break;
2009
2010                 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
2011                 if (read < 0) {
2012                         err = -ENOMEM;
2013                         break;
2014                 }
2015
2016                 copied += read;
2017
2018                 err = transport->notify_recv_post_dequeue(vsk, target, read,
2019                                                 !(flags & MSG_PEEK), &recv_data);
2020                 if (err < 0)
2021                         goto out;
2022
2023                 if (read >= target || flags & MSG_PEEK)
2024                         break;
2025
2026                 target -= read;
2027         }
2028
2029         if (sk->sk_err)
2030                 err = -sk->sk_err;
2031         else if (sk->sk_shutdown & RCV_SHUTDOWN)
2032                 err = 0;
2033
2034         if (copied > 0)
2035                 err = copied;
2036
2037 out:
2038         return err;
2039 }
2040
2041 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2042                                      size_t len, int flags)
2043 {
2044         const struct vsock_transport *transport;
2045         struct vsock_sock *vsk;
2046         ssize_t msg_len;
2047         long timeout;
2048         int err = 0;
2049         DEFINE_WAIT(wait);
2050
2051         vsk = vsock_sk(sk);
2052         transport = vsk->transport;
2053
2054         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2055
2056         err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0);
2057         if (err <= 0)
2058                 goto out;
2059
2060         msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2061
2062         if (msg_len < 0) {
2063                 err = -ENOMEM;
2064                 goto out;
2065         }
2066
2067         if (sk->sk_err) {
2068                 err = -sk->sk_err;
2069         } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2070                 err = 0;
2071         } else {
2072                 /* User sets MSG_TRUNC, so return real length of
2073                  * packet.
2074                  */
2075                 if (flags & MSG_TRUNC)
2076                         err = msg_len;
2077                 else
2078                         err = len - msg_data_left(msg);
2079
2080                 /* Always set MSG_TRUNC if real length of packet is
2081                  * bigger than user's buffer.
2082                  */
2083                 if (msg_len > len)
2084                         msg->msg_flags |= MSG_TRUNC;
2085         }
2086
2087 out:
2088         return err;
2089 }
2090
2091 static int
2092 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2093                           int flags)
2094 {
2095         struct sock *sk;
2096         struct vsock_sock *vsk;
2097         const struct vsock_transport *transport;
2098         int err;
2099
2100         sk = sock->sk;
2101
2102         if (unlikely(flags & MSG_ERRQUEUE))
2103                 return sock_recv_errqueue(sk, msg, len, SOL_VSOCK, VSOCK_RECVERR);
2104
2105         vsk = vsock_sk(sk);
2106         err = 0;
2107
2108         lock_sock(sk);
2109
2110         transport = vsk->transport;
2111
2112         if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2113                 /* Recvmsg is supposed to return 0 if a peer performs an
2114                  * orderly shutdown. Differentiate between that case and when a
2115                  * peer has not connected or a local shutdown occurred with the
2116                  * SOCK_DONE flag.
2117                  */
2118                 if (sock_flag(sk, SOCK_DONE))
2119                         err = 0;
2120                 else
2121                         err = -ENOTCONN;
2122
2123                 goto out;
2124         }
2125
2126         if (flags & MSG_OOB) {
2127                 err = -EOPNOTSUPP;
2128                 goto out;
2129         }
2130
2131         /* We don't check peer_shutdown flag here since peer may actually shut
2132          * down, but there can be data in the queue that a local socket can
2133          * receive.
2134          */
2135         if (sk->sk_shutdown & RCV_SHUTDOWN) {
2136                 err = 0;
2137                 goto out;
2138         }
2139
2140         /* It is valid on Linux to pass in a zero-length receive buffer.  This
2141          * is not an error.  We may as well bail out now.
2142          */
2143         if (!len) {
2144                 err = 0;
2145                 goto out;
2146         }
2147
2148         if (sk->sk_type == SOCK_STREAM)
2149                 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2150         else
2151                 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2152
2153 out:
2154         release_sock(sk);
2155         return err;
2156 }
2157
2158 static int vsock_set_rcvlowat(struct sock *sk, int val)
2159 {
2160         const struct vsock_transport *transport;
2161         struct vsock_sock *vsk;
2162
2163         vsk = vsock_sk(sk);
2164
2165         if (val > vsk->buffer_size)
2166                 return -EINVAL;
2167
2168         transport = vsk->transport;
2169
2170         if (transport && transport->set_rcvlowat)
2171                 return transport->set_rcvlowat(vsk, val);
2172
2173         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
2174         return 0;
2175 }
2176
2177 static const struct proto_ops vsock_stream_ops = {
2178         .family = PF_VSOCK,
2179         .owner = THIS_MODULE,
2180         .release = vsock_release,
2181         .bind = vsock_bind,
2182         .connect = vsock_connect,
2183         .socketpair = sock_no_socketpair,
2184         .accept = vsock_accept,
2185         .getname = vsock_getname,
2186         .poll = vsock_poll,
2187         .ioctl = sock_no_ioctl,
2188         .listen = vsock_listen,
2189         .shutdown = vsock_shutdown,
2190         .setsockopt = vsock_connectible_setsockopt,
2191         .getsockopt = vsock_connectible_getsockopt,
2192         .sendmsg = vsock_connectible_sendmsg,
2193         .recvmsg = vsock_connectible_recvmsg,
2194         .mmap = sock_no_mmap,
2195         .sendpage = sock_no_sendpage,
2196         .set_rcvlowat = vsock_set_rcvlowat,
2197 };
2198
2199 static const struct proto_ops vsock_seqpacket_ops = {
2200         .family = PF_VSOCK,
2201         .owner = THIS_MODULE,
2202         .release = vsock_release,
2203         .bind = vsock_bind,
2204         .connect = vsock_connect,
2205         .socketpair = sock_no_socketpair,
2206         .accept = vsock_accept,
2207         .getname = vsock_getname,
2208         .poll = vsock_poll,
2209         .ioctl = sock_no_ioctl,
2210         .listen = vsock_listen,
2211         .shutdown = vsock_shutdown,
2212         .setsockopt = vsock_connectible_setsockopt,
2213         .getsockopt = vsock_connectible_getsockopt,
2214         .sendmsg = vsock_connectible_sendmsg,
2215         .recvmsg = vsock_connectible_recvmsg,
2216         .mmap = sock_no_mmap,
2217         .sendpage = sock_no_sendpage,
2218 };
2219
2220 static int vsock_create(struct net *net, struct socket *sock,
2221                         int protocol, int kern)
2222 {
2223         struct vsock_sock *vsk;
2224         struct sock *sk;
2225         int ret;
2226
2227         if (!sock)
2228                 return -EINVAL;
2229
2230         if (protocol && protocol != PF_VSOCK)
2231                 return -EPROTONOSUPPORT;
2232
2233         switch (sock->type) {
2234         case SOCK_DGRAM:
2235                 sock->ops = &vsock_dgram_ops;
2236                 break;
2237         case SOCK_STREAM:
2238                 sock->ops = &vsock_stream_ops;
2239                 break;
2240         case SOCK_SEQPACKET:
2241                 sock->ops = &vsock_seqpacket_ops;
2242                 break;
2243         default:
2244                 return -ESOCKTNOSUPPORT;
2245         }
2246
2247         sock->state = SS_UNCONNECTED;
2248
2249         sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2250         if (!sk)
2251                 return -ENOMEM;
2252
2253         vsk = vsock_sk(sk);
2254
2255         if (sock->type == SOCK_DGRAM) {
2256                 ret = vsock_assign_transport(vsk, NULL);
2257                 if (ret < 0) {
2258                         sock_put(sk);
2259                         return ret;
2260                 }
2261         }
2262
2263         vsock_insert_unbound(vsk);
2264
2265         return 0;
2266 }
2267
2268 static const struct net_proto_family vsock_family_ops = {
2269         .family = AF_VSOCK,
2270         .create = vsock_create,
2271         .owner = THIS_MODULE,
2272 };
2273
2274 static long vsock_dev_do_ioctl(struct file *filp,
2275                                unsigned int cmd, void __user *ptr)
2276 {
2277         u32 __user *p = ptr;
2278         u32 cid = VMADDR_CID_ANY;
2279         int retval = 0;
2280
2281         switch (cmd) {
2282         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2283                 /* To be compatible with the VMCI behavior, we prioritize the
2284                  * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2285                  */
2286                 if (transport_g2h)
2287                         cid = transport_g2h->get_local_cid();
2288                 else if (transport_h2g)
2289                         cid = transport_h2g->get_local_cid();
2290
2291                 if (put_user(cid, p) != 0)
2292                         retval = -EFAULT;
2293                 break;
2294
2295         default:
2296                 retval = -ENOIOCTLCMD;
2297         }
2298
2299         return retval;
2300 }
2301
2302 static long vsock_dev_ioctl(struct file *filp,
2303                             unsigned int cmd, unsigned long arg)
2304 {
2305         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2306 }
2307
2308 #ifdef CONFIG_COMPAT
2309 static long vsock_dev_compat_ioctl(struct file *filp,
2310                                    unsigned int cmd, unsigned long arg)
2311 {
2312         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2313 }
2314 #endif
2315
2316 static const struct file_operations vsock_device_ops = {
2317         .owner          = THIS_MODULE,
2318         .unlocked_ioctl = vsock_dev_ioctl,
2319 #ifdef CONFIG_COMPAT
2320         .compat_ioctl   = vsock_dev_compat_ioctl,
2321 #endif
2322         .open           = nonseekable_open,
2323 };
2324
2325 static struct miscdevice vsock_device = {
2326         .name           = "vsock",
2327         .fops           = &vsock_device_ops,
2328 };
2329
2330 static int __init vsock_init(void)
2331 {
2332         int err = 0;
2333
2334         vsock_init_tables();
2335
2336         vsock_proto.owner = THIS_MODULE;
2337         vsock_device.minor = MISC_DYNAMIC_MINOR;
2338         err = misc_register(&vsock_device);
2339         if (err) {
2340                 pr_err("Failed to register misc device\n");
2341                 goto err_reset_transport;
2342         }
2343
2344         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2345         if (err) {
2346                 pr_err("Cannot register vsock protocol\n");
2347                 goto err_deregister_misc;
2348         }
2349
2350         err = sock_register(&vsock_family_ops);
2351         if (err) {
2352                 pr_err("could not register af_vsock (%d) address family: %d\n",
2353                        AF_VSOCK, err);
2354                 goto err_unregister_proto;
2355         }
2356
2357         return 0;
2358
2359 err_unregister_proto:
2360         proto_unregister(&vsock_proto);
2361 err_deregister_misc:
2362         misc_deregister(&vsock_device);
2363 err_reset_transport:
2364         return err;
2365 }
2366
2367 static void __exit vsock_exit(void)
2368 {
2369         misc_deregister(&vsock_device);
2370         sock_unregister(AF_VSOCK);
2371         proto_unregister(&vsock_proto);
2372 }
2373
2374 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2375 {
2376         return vsk->transport;
2377 }
2378 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2379
2380 int vsock_core_register(const struct vsock_transport *t, int features)
2381 {
2382         const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2383         int err = mutex_lock_interruptible(&vsock_register_mutex);
2384
2385         if (err)
2386                 return err;
2387
2388         t_h2g = transport_h2g;
2389         t_g2h = transport_g2h;
2390         t_dgram = transport_dgram;
2391         t_local = transport_local;
2392
2393         if (features & VSOCK_TRANSPORT_F_H2G) {
2394                 if (t_h2g) {
2395                         err = -EBUSY;
2396                         goto err_busy;
2397                 }
2398                 t_h2g = t;
2399         }
2400
2401         if (features & VSOCK_TRANSPORT_F_G2H) {
2402                 if (t_g2h) {
2403                         err = -EBUSY;
2404                         goto err_busy;
2405                 }
2406                 t_g2h = t;
2407         }
2408
2409         if (features & VSOCK_TRANSPORT_F_DGRAM) {
2410                 if (t_dgram) {
2411                         err = -EBUSY;
2412                         goto err_busy;
2413                 }
2414                 t_dgram = t;
2415         }
2416
2417         if (features & VSOCK_TRANSPORT_F_LOCAL) {
2418                 if (t_local) {
2419                         err = -EBUSY;
2420                         goto err_busy;
2421                 }
2422                 t_local = t;
2423         }
2424
2425         transport_h2g = t_h2g;
2426         transport_g2h = t_g2h;
2427         transport_dgram = t_dgram;
2428         transport_local = t_local;
2429
2430 err_busy:
2431         mutex_unlock(&vsock_register_mutex);
2432         return err;
2433 }
2434 EXPORT_SYMBOL_GPL(vsock_core_register);
2435
2436 void vsock_core_unregister(const struct vsock_transport *t)
2437 {
2438         mutex_lock(&vsock_register_mutex);
2439
2440         if (transport_h2g == t)
2441                 transport_h2g = NULL;
2442
2443         if (transport_g2h == t)
2444                 transport_g2h = NULL;
2445
2446         if (transport_dgram == t)
2447                 transport_dgram = NULL;
2448
2449         if (transport_local == t)
2450                 transport_local = NULL;
2451
2452         mutex_unlock(&vsock_register_mutex);
2453 }
2454 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2455
2456 module_init(vsock_init);
2457 module_exit(vsock_exit);
2458
2459 MODULE_AUTHOR("VMware, Inc.");
2460 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2461 MODULE_VERSION("1.0.2.0-k");
2462 MODULE_LICENSE("GPL v2");