net/vmw_vsock/af_vsock.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VMware vSockets Driver
   4  *
   5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   6  */
   7
   8 /* Implementation notes:
   9  *
  10  * - There are two kinds of sockets: those created by user action (such as
  11  * calling socket(2)) and those created by incoming connection request packets.
  12  *
  13  * - There are two "global" tables, one for bound sockets (sockets that have
  14  * specified an address that they are responsible for) and one for connected
  15  * sockets (sockets that have established a connection with another socket).
  16  * These tables are "global" in that all sockets on the system are placed
  17  * within them. - Note, though, that the bound table contains an extra entry
  18  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  19  * that list. The bound table is used solely for lookup of sockets when packets
  20  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  21  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  22  * sockets out of the bound hash buckets will reduce the chance of collisions
  23  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  24  * socket type in the hash table lookups.
  25  *
  26  * - Sockets created by user action will either be "client" sockets that
  27  * initiate a connection or "server" sockets that listen for connections; we do
  28  * not support simultaneous connects (two "client" sockets connecting).
  29  *
  30  * - "Server" sockets are referred to as listener sockets throughout this
  31  * implementation because they are in the TCP_LISTEN state.  When a
  32  * connection request is received (the second kind of socket mentioned above),
  33  * we create a new socket and refer to it as a pending socket.  These pending
  34  * sockets are placed on the pending connection list of the listener socket.
  35  * When future packets are received for the address the listener socket is
  36  * bound to, we check if the source of the packet is from one that has an
  37  * existing pending connection.  If it does, we process the packet for the
  38  * pending socket.  When that socket reaches the connected state, it is removed
  39  * from the listener socket's pending list and enqueued in the listener
  40  * socket's accept queue.  Callers of accept(2) will accept connected sockets
  41  * from the listener socket's accept queue.  If the socket cannot be accepted
  42  * for some reason then it is marked rejected.  Once the connection is
  43  * accepted, it is owned by the user process and the responsibility for cleanup
  44  * falls with that user process.
  45  *
  46  * - It is possible that these pending sockets will never reach the connected
  47  * state; in fact, we may never receive another packet after the connection
  48  * request.  Because of this, we must schedule a cleanup function to run in the
  49  * future, after some amount of time passes where a connection should have been
  50  * established.  This function ensures that the socket is off all lists so it
  51  * cannot be retrieved, then drops all references to the socket so it is cleaned
  52  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  53  * function will also cleanup rejected sockets, those that reach the connected
  54  * state but leave it before they have been accepted.
  55  *
  56  * - Lock ordering for pending or accept queue sockets is:
  57  *
  58  *     lock_sock(listener);
  59  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  60  *
  61  * Using explicit nested locking keeps lockdep happy since normally only one
  62  * lock of a given class may be taken at a time.
  63  *
  64  * - Sockets created by user action will be cleaned up when the user process
  65  * calls close(2), causing our release implementation to be called. Our release
  66  * implementation will perform some cleanup then drop the last reference so our
  67  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  68  * perform additional cleanup that's common for both types of sockets.
  69  *
  70  * - A socket's reference count is what ensures that the structure won't be
  71  * freed.  Each entry in a list (such as the "global" bound and connected tables
  72  * and the listener socket's pending list and connected queue) ensures a
  73  * reference.  When we defer work until process context and pass a socket as our
  74  * argument, we must ensure the reference count is increased to ensure the
  75  * socket isn't freed before the function is run; the deferred function will
  76  * then drop the reference.
  77  *
  78  * - sk->sk_state uses the TCP state constants because they are widely used by
  79  * other address families and exposed to userspace tools like ss(8):
  80  *
  81  *   TCP_CLOSE - unconnected
  82  *   TCP_SYN_SENT - connecting
  83  *   TCP_ESTABLISHED - connected
  84  *   TCP_CLOSING - disconnecting
  85  *   TCP_LISTEN - listening
  86  */
  87
  88 #include <linux/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 #include <uapi/asm-generic/ioctls.h>
 116
 117 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 118 static void vsock_sk_destruct(struct sock *sk);
 119 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 120 static void vsock_close(struct sock *sk, long timeout);
 121
 122 /* Protocol family. */
 123 struct proto vsock_proto = {
 124         .name = "AF_VSOCK",
 125         .owner = THIS_MODULE,
 126         .obj_size = sizeof(struct vsock_sock),
 127         .close = vsock_close,
 128 #ifdef CONFIG_BPF_SYSCALL
 129         .psock_update_sk_prot = vsock_bpf_update_proto,
 130 #endif
 131 };
 132
 133 /* The default peer timeout indicates how long we will wait for a peer response
 134  * to a control message.
 135  */
 136 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 137
 138 #define VSOCK_DEFAULT_BUFFER_SIZE     (1024 * 256)
 139 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
 140 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
 141
 142 /* Transport used for host->guest communication */
 143 static const struct vsock_transport *transport_h2g;
 144 /* Transport used for guest->host communication */
 145 static const struct vsock_transport *transport_g2h;
 146 /* Transport used for DGRAM communication */
 147 static const struct vsock_transport *transport_dgram;
 148 /* Transport used for local communication */
 149 static const struct vsock_transport *transport_local;
 150 static DEFINE_MUTEX(vsock_register_mutex);
 151
 152 /**** UTILS ****/
 153
 154 /* Each bound VSocket is stored in the bind hash table and each connected
 155  * VSocket is stored in the connected hash table.
 156  *
 157  * Unbound sockets are all put on the same list attached to the end of the hash
 158  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 159  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 160  * represents the list that addr hashes to).
 161  *
 162  * Specifically, we initialize the vsock_bind_table array to a size of
 163  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 164  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 165  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 166  * mods with VSOCK_HASH_SIZE to ensure this.
 167  */
 168 #define MAX_PORT_RETRIES        24
 169
 170 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 171 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 172 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 173
 174 /* XXX This can probably be implemented in a better way. */
 175 #define VSOCK_CONN_HASH(src, dst)                               \
 176         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 177 #define vsock_connected_sockets(src, dst)               \
 178         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 179 #define vsock_connected_sockets_vsk(vsk)                                \
 180         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 181
 182 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 183 EXPORT_SYMBOL_GPL(vsock_bind_table);
 184 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 185 EXPORT_SYMBOL_GPL(vsock_connected_table);
 186 DEFINE_SPINLOCK(vsock_table_lock);
 187 EXPORT_SYMBOL_GPL(vsock_table_lock);
 188
 189 /* Autobind this socket to the local address if necessary. */
 190 static int vsock_auto_bind(struct vsock_sock *vsk)
 191 {
 192         struct sock *sk = sk_vsock(vsk);
 193         struct sockaddr_vm local_addr;
 194
 195         if (vsock_addr_bound(&vsk->local_addr))
 196                 return 0;
 197         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 198         return __vsock_bind(sk, &local_addr);
 199 }
 200
 201 static void vsock_init_tables(void)
 202 {
 203         int i;
 204
 205         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 206                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 207
 208         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 209                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 210 }
 211
 212 static void __vsock_insert_bound(struct list_head *list,
 213                                  struct vsock_sock *vsk)
 214 {
 215         sock_hold(&vsk->sk);
 216         list_add(&vsk->bound_table, list);
 217 }
 218
 219 static void __vsock_insert_connected(struct list_head *list,
 220                                      struct vsock_sock *vsk)
 221 {
 222         sock_hold(&vsk->sk);
 223         list_add(&vsk->connected_table, list);
 224 }
 225
 226 static void __vsock_remove_bound(struct vsock_sock *vsk)
 227 {
 228         list_del_init(&vsk->bound_table);
 229         sock_put(&vsk->sk);
 230 }
 231
 232 static void __vsock_remove_connected(struct vsock_sock *vsk)
 233 {
 234         list_del_init(&vsk->connected_table);
 235         sock_put(&vsk->sk);
 236 }
 237
 238 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 239 {
 240         struct vsock_sock *vsk;
 241
 242         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
 243                 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
 244                         return sk_vsock(vsk);
 245
 246                 if (addr->svm_port == vsk->local_addr.svm_port &&
 247                     (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
 248                      addr->svm_cid == VMADDR_CID_ANY))
 249                         return sk_vsock(vsk);
 250         }
 251
 252         return NULL;
 253 }
 254
 255 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 256                                                   struct sockaddr_vm *dst)
 257 {
 258         struct vsock_sock *vsk;
 259
 260         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 261                             connected_table) {
 262                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 263                     dst->svm_port == vsk->local_addr.svm_port) {
 264                         return sk_vsock(vsk);
 265                 }
 266         }
 267
 268         return NULL;
 269 }
 270
 271 static void vsock_insert_unbound(struct vsock_sock *vsk)
 272 {
 273         spin_lock_bh(&vsock_table_lock);
 274         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 275         spin_unlock_bh(&vsock_table_lock);
 276 }
 277
 278 void vsock_insert_connected(struct vsock_sock *vsk)
 279 {
 280         struct list_head *list = vsock_connected_sockets(
 281                 &vsk->remote_addr, &vsk->local_addr);
 282
 283         spin_lock_bh(&vsock_table_lock);
 284         __vsock_insert_connected(list, vsk);
 285         spin_unlock_bh(&vsock_table_lock);
 286 }
 287 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 288
 289 void vsock_remove_bound(struct vsock_sock *vsk)
 290 {
 291         spin_lock_bh(&vsock_table_lock);
 292         if (__vsock_in_bound_table(vsk))
 293                 __vsock_remove_bound(vsk);
 294         spin_unlock_bh(&vsock_table_lock);
 295 }
 296 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 297
 298 void vsock_remove_connected(struct vsock_sock *vsk)
 299 {
 300         spin_lock_bh(&vsock_table_lock);
 301         if (__vsock_in_connected_table(vsk))
 302                 __vsock_remove_connected(vsk);
 303         spin_unlock_bh(&vsock_table_lock);
 304 }
 305 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 306
 307 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 308 {
 309         struct sock *sk;
 310
 311         spin_lock_bh(&vsock_table_lock);
 312         sk = __vsock_find_bound_socket(addr);
 313         if (sk)
 314                 sock_hold(sk);
 315
 316         spin_unlock_bh(&vsock_table_lock);
 317
 318         return sk;
 319 }
 320 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 321
 322 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 323                                          struct sockaddr_vm *dst)
 324 {
 325         struct sock *sk;
 326
 327         spin_lock_bh(&vsock_table_lock);
 328         sk = __vsock_find_connected_socket(src, dst);
 329         if (sk)
 330                 sock_hold(sk);
 331
 332         spin_unlock_bh(&vsock_table_lock);
 333
 334         return sk;
 335 }
 336 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 337
 338 void vsock_remove_sock(struct vsock_sock *vsk)
 339 {
 340         vsock_remove_bound(vsk);
 341         vsock_remove_connected(vsk);
 342 }
 343 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 344
 345 void vsock_for_each_connected_socket(struct vsock_transport *transport,
 346                                      void (*fn)(struct sock *sk))
 347 {
 348         int i;
 349
 350         spin_lock_bh(&vsock_table_lock);
 351
 352         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 353                 struct vsock_sock *vsk;
 354                 list_for_each_entry(vsk, &vsock_connected_table[i],
 355                                     connected_table) {
 356                         if (vsk->transport != transport)
 357                                 continue;
 358
 359                         fn(sk_vsock(vsk));
 360                 }
 361         }
 362
 363         spin_unlock_bh(&vsock_table_lock);
 364 }
 365 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 366
 367 void vsock_add_pending(struct sock *listener, struct sock *pending)
 368 {
 369         struct vsock_sock *vlistener;
 370         struct vsock_sock *vpending;
 371
 372         vlistener = vsock_sk(listener);
 373         vpending = vsock_sk(pending);
 374
 375         sock_hold(pending);
 376         sock_hold(listener);
 377         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 378 }
 379 EXPORT_SYMBOL_GPL(vsock_add_pending);
 380
 381 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 382 {
 383         struct vsock_sock *vpending = vsock_sk(pending);
 384
 385         list_del_init(&vpending->pending_links);
 386         sock_put(listener);
 387         sock_put(pending);
 388 }
 389 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 390
 391 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 392 {
 393         struct vsock_sock *vlistener;
 394         struct vsock_sock *vconnected;
 395
 396         vlistener = vsock_sk(listener);
 397         vconnected = vsock_sk(connected);
 398
 399         sock_hold(connected);
 400         sock_hold(listener);
 401         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 402 }
 403 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 404
 405 static bool vsock_use_local_transport(unsigned int remote_cid)
 406 {
 407         if (!transport_local)
 408                 return false;
 409
 410         if (remote_cid == VMADDR_CID_LOCAL)
 411                 return true;
 412
 413         if (transport_g2h) {
 414                 return remote_cid == transport_g2h->get_local_cid();
 415         } else {
 416                 return remote_cid == VMADDR_CID_HOST;
 417         }
 418 }
 419
 420 static void vsock_deassign_transport(struct vsock_sock *vsk)
 421 {
 422         if (!vsk->transport)
 423                 return;
 424
 425         vsk->transport->destruct(vsk);
 426         module_put(vsk->transport->module);
 427         vsk->transport = NULL;
 428 }
 429
 430 /* Assign a transport to a socket and call the .init transport callback.
 431  *
 432  * Note: for connection oriented socket this must be called when vsk->remote_addr
 433  * is set (e.g. during the connect() or when a connection request on a listener
 434  * socket is received).
 435  * The vsk->remote_addr is used to decide which transport to use:
 436  *  - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
 437  *    g2h is not loaded, will use local transport;
 438  *  - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
 439  *    includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
 440  *  - remote CID > VMADDR_CID_HOST will use host->guest transport;
 441  */
 442 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
 443 {
 444         const struct vsock_transport *new_transport;
 445         struct sock *sk = sk_vsock(vsk);
 446         unsigned int remote_cid = vsk->remote_addr.svm_cid;
 447         __u8 remote_flags;
 448         int ret;
 449
 450         /* If the packet is coming with the source and destination CIDs higher
 451          * than VMADDR_CID_HOST, then a vsock channel where all the packets are
 452          * forwarded to the host should be established. Then the host will
 453          * need to forward the packets to the guest.
 454          *
 455          * The flag is set on the (listen) receive path (psk is not NULL). On
 456          * the connect path the flag can be set by the user space application.
 457          */
 458         if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
 459             vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
 460                 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
 461
 462         remote_flags = vsk->remote_addr.svm_flags;
 463
 464         switch (sk->sk_type) {
 465         case SOCK_DGRAM:
 466                 new_transport = transport_dgram;
 467                 break;
 468         case SOCK_STREAM:
 469         case SOCK_SEQPACKET:
 470                 if (vsock_use_local_transport(remote_cid))
 471                         new_transport = transport_local;
 472                 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
 473                          (remote_flags & VMADDR_FLAG_TO_HOST))
 474                         new_transport = transport_g2h;
 475                 else
 476                         new_transport = transport_h2g;
 477                 break;
 478         default:
 479                 return -ESOCKTNOSUPPORT;
 480         }
 481
 482         if (vsk->transport) {
 483                 if (vsk->transport == new_transport)
 484                         return 0;
 485
 486                 /* transport->release() must be called with sock lock acquired.
 487                  * This path can only be taken during vsock_connect(), where we
 488                  * have already held the sock lock. In the other cases, this
 489                  * function is called on a new socket which is not assigned to
 490                  * any transport.
 491                  */
 492                 vsk->transport->release(vsk);
 493                 vsock_deassign_transport(vsk);
 494         }
 495
 496         /* We increase the module refcnt to prevent the transport unloading
 497          * while there are open sockets assigned to it.
 498          */
 499         if (!new_transport || !try_module_get(new_transport->module))
 500                 return -ENODEV;
 501
 502         if (sk->sk_type == SOCK_SEQPACKET) {
 503                 if (!new_transport->seqpacket_allow ||
 504                     !new_transport->seqpacket_allow(remote_cid)) {
 505                         module_put(new_transport->module);
 506                         return -ESOCKTNOSUPPORT;
 507                 }
 508         }
 509
 510         ret = new_transport->init(vsk, psk);
 511         if (ret) {
 512                 module_put(new_transport->module);
 513                 return ret;
 514         }
 515
 516         vsk->transport = new_transport;
 517
 518         return 0;
 519 }
 520 EXPORT_SYMBOL_GPL(vsock_assign_transport);
 521
 522 bool vsock_find_cid(unsigned int cid)
 523 {
 524         if (transport_g2h && cid == transport_g2h->get_local_cid())
 525                 return true;
 526
 527         if (transport_h2g && cid == VMADDR_CID_HOST)
 528                 return true;
 529
 530         if (transport_local && cid == VMADDR_CID_LOCAL)
 531                 return true;
 532
 533         return false;
 534 }
 535 EXPORT_SYMBOL_GPL(vsock_find_cid);
 536
 537 static struct sock *vsock_dequeue_accept(struct sock *listener)
 538 {
 539         struct vsock_sock *vlistener;
 540         struct vsock_sock *vconnected;
 541
 542         vlistener = vsock_sk(listener);
 543
 544         if (list_empty(&vlistener->accept_queue))
 545                 return NULL;
 546
 547         vconnected = list_entry(vlistener->accept_queue.next,
 548                                 struct vsock_sock, accept_queue);
 549
 550         list_del_init(&vconnected->accept_queue);
 551         sock_put(listener);
 552         /* The caller will need a reference on the connected socket so we let
 553          * it call sock_put().
 554          */
 555
 556         return sk_vsock(vconnected);
 557 }
 558
 559 static bool vsock_is_accept_queue_empty(struct sock *sk)
 560 {
 561         struct vsock_sock *vsk = vsock_sk(sk);
 562         return list_empty(&vsk->accept_queue);
 563 }
 564
 565 static bool vsock_is_pending(struct sock *sk)
 566 {
 567         struct vsock_sock *vsk = vsock_sk(sk);
 568         return !list_empty(&vsk->pending_links);
 569 }
 570
 571 static int vsock_send_shutdown(struct sock *sk, int mode)
 572 {
 573         struct vsock_sock *vsk = vsock_sk(sk);
 574
 575         if (!vsk->transport)
 576                 return -ENODEV;
 577
 578         return vsk->transport->shutdown(vsk, mode);
 579 }
 580
 581 static void vsock_pending_work(struct work_struct *work)
 582 {
 583         struct sock *sk;
 584         struct sock *listener;
 585         struct vsock_sock *vsk;
 586         bool cleanup;
 587
 588         vsk = container_of(work, struct vsock_sock, pending_work.work);
 589         sk = sk_vsock(vsk);
 590         listener = vsk->listener;
 591         cleanup = true;
 592
 593         lock_sock(listener);
 594         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 595
 596         if (vsock_is_pending(sk)) {
 597                 vsock_remove_pending(listener, sk);
 598
 599                 sk_acceptq_removed(listener);
 600         } else if (!vsk->rejected) {
 601                 /* We are not on the pending list and accept() did not reject
 602                  * us, so we must have been accepted by our user process.  We
 603                  * just need to drop our references to the sockets and be on
 604                  * our way.
 605                  */
 606                 cleanup = false;
 607                 goto out;
 608         }
 609
 610         /* We need to remove ourself from the global connected sockets list so
 611          * incoming packets can't find this socket, and to reduce the reference
 612          * count.
 613          */
 614         vsock_remove_connected(vsk);
 615
 616         sk->sk_state = TCP_CLOSE;
 617
 618 out:
 619         release_sock(sk);
 620         release_sock(listener);
 621         if (cleanup)
 622                 sock_put(sk);
 623
 624         sock_put(sk);
 625         sock_put(listener);
 626 }
 627
 628 /**** SOCKET OPERATIONS ****/
 629
 630 static int __vsock_bind_connectible(struct vsock_sock *vsk,
 631                                     struct sockaddr_vm *addr)
 632 {
 633         static u32 port;
 634         struct sockaddr_vm new_addr;
 635
 636         if (!port)
 637                 port = get_random_u32_above(LAST_RESERVED_PORT);
 638
 639         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 640
 641         if (addr->svm_port == VMADDR_PORT_ANY) {
 642                 bool found = false;
 643                 unsigned int i;
 644
 645                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 646                         if (port <= LAST_RESERVED_PORT)
 647                                 port = LAST_RESERVED_PORT + 1;
 648
 649                         new_addr.svm_port = port++;
 650
 651                         if (!__vsock_find_bound_socket(&new_addr)) {
 652                                 found = true;
 653                                 break;
 654                         }
 655                 }
 656
 657                 if (!found)
 658                         return -EADDRNOTAVAIL;
 659         } else {
 660                 /* If port is in reserved range, ensure caller
 661                  * has necessary privileges.
 662                  */
 663                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 664                     !capable(CAP_NET_BIND_SERVICE)) {
 665                         return -EACCES;
 666                 }
 667
 668                 if (__vsock_find_bound_socket(&new_addr))
 669                         return -EADDRINUSE;
 670         }
 671
 672         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 673
 674         /* Remove connection oriented sockets from the unbound list and add them
 675          * to the hash table for easy lookup by its address.  The unbound list
 676          * is simply an extra entry at the end of the hash table, a trick used
 677          * by AF_UNIX.
 678          */
 679         __vsock_remove_bound(vsk);
 680         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 681
 682         return 0;
 683 }
 684
 685 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 686                               struct sockaddr_vm *addr)
 687 {
 688         return vsk->transport->dgram_bind(vsk, addr);
 689 }
 690
 691 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 692 {
 693         struct vsock_sock *vsk = vsock_sk(sk);
 694         int retval;
 695
 696         /* First ensure this socket isn't already bound. */
 697         if (vsock_addr_bound(&vsk->local_addr))
 698                 return -EINVAL;
 699
 700         /* Now bind to the provided address or select appropriate values if
 701          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 702          * like AF_INET prevents binding to a non-local IP address (in most
 703          * cases), we only allow binding to a local CID.
 704          */
 705         if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
 706                 return -EADDRNOTAVAIL;
 707
 708         switch (sk->sk_socket->type) {
 709         case SOCK_STREAM:
 710         case SOCK_SEQPACKET:
 711                 spin_lock_bh(&vsock_table_lock);
 712                 retval = __vsock_bind_connectible(vsk, addr);
 713                 spin_unlock_bh(&vsock_table_lock);
 714                 break;
 715
 716         case SOCK_DGRAM:
 717                 retval = __vsock_bind_dgram(vsk, addr);
 718                 break;
 719
 720         default:
 721                 retval = -EINVAL;
 722                 break;
 723         }
 724
 725         return retval;
 726 }
 727
 728 static void vsock_connect_timeout(struct work_struct *work);
 729
 730 static struct sock *__vsock_create(struct net *net,
 731                                    struct socket *sock,
 732                                    struct sock *parent,
 733                                    gfp_t priority,
 734                                    unsigned short type,
 735                                    int kern)
 736 {
 737         struct sock *sk;
 738         struct vsock_sock *psk;
 739         struct vsock_sock *vsk;
 740
 741         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 742         if (!sk)
 743                 return NULL;
 744
 745         sock_init_data(sock, sk);
 746
 747         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 748          * non-NULL. We make sure that our sockets always have a type by
 749          * setting it here if needed.
 750          */
 751         if (!sock)
 752                 sk->sk_type = type;
 753
 754         vsk = vsock_sk(sk);
 755         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 756         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 757
 758         sk->sk_destruct = vsock_sk_destruct;
 759         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 760         sock_reset_flag(sk, SOCK_DONE);
 761
 762         INIT_LIST_HEAD(&vsk->bound_table);
 763         INIT_LIST_HEAD(&vsk->connected_table);
 764         vsk->listener = NULL;
 765         INIT_LIST_HEAD(&vsk->pending_links);
 766         INIT_LIST_HEAD(&vsk->accept_queue);
 767         vsk->rejected = false;
 768         vsk->sent_request = false;
 769         vsk->ignore_connecting_rst = false;
 770         vsk->peer_shutdown = 0;
 771         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 772         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 773
 774         psk = parent ? vsock_sk(parent) : NULL;
 775         if (parent) {
 776                 vsk->trusted = psk->trusted;
 777                 vsk->owner = get_cred(psk->owner);
 778                 vsk->connect_timeout = psk->connect_timeout;
 779                 vsk->buffer_size = psk->buffer_size;
 780                 vsk->buffer_min_size = psk->buffer_min_size;
 781                 vsk->buffer_max_size = psk->buffer_max_size;
 782                 security_sk_clone(parent, sk);
 783         } else {
 784                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 785                 vsk->owner = get_current_cred();
 786                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 787                 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
 788                 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
 789                 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
 790         }
 791
 792         return sk;
 793 }
 794
 795 static bool sock_type_connectible(u16 type)
 796 {
 797         return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
 798 }
 799
 800 static void __vsock_release(struct sock *sk, int level)
 801 {
 802         struct vsock_sock *vsk;
 803         struct sock *pending;
 804
 805         vsk = vsock_sk(sk);
 806         pending = NULL; /* Compiler warning. */
 807
 808         /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 809          * version to avoid the warning "possible recursive locking
 810          * detected". When "level" is 0, lock_sock_nested(sk, level)
 811          * is the same as lock_sock(sk).
 812          */
 813         lock_sock_nested(sk, level);
 814
 815         if (vsk->transport)
 816                 vsk->transport->release(vsk);
 817         else if (sock_type_connectible(sk->sk_type))
 818                 vsock_remove_sock(vsk);
 819
 820         sock_orphan(sk);
 821         sk->sk_shutdown = SHUTDOWN_MASK;
 822
 823         skb_queue_purge(&sk->sk_receive_queue);
 824
 825         /* Clean up any sockets that never were accepted. */
 826         while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 827                 __vsock_release(pending, SINGLE_DEPTH_NESTING);
 828                 sock_put(pending);
 829         }
 830
 831         release_sock(sk);
 832         sock_put(sk);
 833 }
 834
 835 static void vsock_sk_destruct(struct sock *sk)
 836 {
 837         struct vsock_sock *vsk = vsock_sk(sk);
 838
 839         /* Flush MSG_ZEROCOPY leftovers. */
 840         __skb_queue_purge(&sk->sk_error_queue);
 841
 842         vsock_deassign_transport(vsk);
 843
 844         /* When clearing these addresses, there's no need to set the family and
 845          * possibly register the address family with the kernel.
 846          */
 847         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 848         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 849
 850         put_cred(vsk->owner);
 851 }
 852
 853 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 854 {
 855         int err;
 856
 857         err = sock_queue_rcv_skb(sk, skb);
 858         if (err)
 859                 kfree_skb(skb);
 860
 861         return err;
 862 }
 863
 864 struct sock *vsock_create_connected(struct sock *parent)
 865 {
 866         return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
 867                               parent->sk_type, 0);
 868 }
 869 EXPORT_SYMBOL_GPL(vsock_create_connected);
 870
 871 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 872 {
 873         return vsk->transport->stream_has_data(vsk);
 874 }
 875 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 876
 877 s64 vsock_connectible_has_data(struct vsock_sock *vsk)
 878 {
 879         struct sock *sk = sk_vsock(vsk);
 880
 881         if (sk->sk_type == SOCK_SEQPACKET)
 882                 return vsk->transport->seqpacket_has_data(vsk);
 883         else
 884                 return vsock_stream_has_data(vsk);
 885 }
 886 EXPORT_SYMBOL_GPL(vsock_connectible_has_data);
 887
 888 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 889 {
 890         return vsk->transport->stream_has_space(vsk);
 891 }
 892 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 893
 894 void vsock_data_ready(struct sock *sk)
 895 {
 896         struct vsock_sock *vsk = vsock_sk(sk);
 897
 898         if (vsock_stream_has_data(vsk) >= sk->sk_rcvlowat ||
 899             sock_flag(sk, SOCK_DONE))
 900                 sk->sk_data_ready(sk);
 901 }
 902 EXPORT_SYMBOL_GPL(vsock_data_ready);
 903
 904 /* Dummy callback required by sockmap.
 905  * See unconditional call of saved_close() in sock_map_close().
 906  */
 907 static void vsock_close(struct sock *sk, long timeout)
 908 {
 909 }
 910
 911 static int vsock_release(struct socket *sock)
 912 {
 913         struct sock *sk = sock->sk;
 914
 915         if (!sk)
 916                 return 0;
 917
 918         sk->sk_prot->close(sk, 0);
 919         __vsock_release(sk, 0);
 920         sock->sk = NULL;
 921         sock->state = SS_FREE;
 922
 923         return 0;
 924 }
 925
 926 static int
 927 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 928 {
 929         int err;
 930         struct sock *sk;
 931         struct sockaddr_vm *vm_addr;
 932
 933         sk = sock->sk;
 934
 935         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 936                 return -EINVAL;
 937
 938         lock_sock(sk);
 939         err = __vsock_bind(sk, vm_addr);
 940         release_sock(sk);
 941
 942         return err;
 943 }
 944
 945 static int vsock_getname(struct socket *sock,
 946                          struct sockaddr *addr, int peer)
 947 {
 948         int err;
 949         struct sock *sk;
 950         struct vsock_sock *vsk;
 951         struct sockaddr_vm *vm_addr;
 952
 953         sk = sock->sk;
 954         vsk = vsock_sk(sk);
 955         err = 0;
 956
 957         lock_sock(sk);
 958
 959         if (peer) {
 960                 if (sock->state != SS_CONNECTED) {
 961                         err = -ENOTCONN;
 962                         goto out;
 963                 }
 964                 vm_addr = &vsk->remote_addr;
 965         } else {
 966                 vm_addr = &vsk->local_addr;
 967         }
 968
 969         if (!vm_addr) {
 970                 err = -EINVAL;
 971                 goto out;
 972         }
 973
 974         /* sys_getsockname() and sys_getpeername() pass us a
 975          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 976          * that macro is defined in socket.c instead of .h, so we hardcode its
 977          * value here.
 978          */
 979         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 980         memcpy(addr, vm_addr, sizeof(*vm_addr));
 981         err = sizeof(*vm_addr);
 982
 983 out:
 984         release_sock(sk);
 985         return err;
 986 }
 987
 988 static int vsock_shutdown(struct socket *sock, int mode)
 989 {
 990         int err;
 991         struct sock *sk;
 992
 993         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 994          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 995          * here like the other address families do.  Note also that the
 996          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 997          * which is what we want.
 998          */
 999         mode++;
1000
1001         if ((mode & ~SHUTDOWN_MASK) || !mode)
1002                 return -EINVAL;
1003
1004         /* If this is a connection oriented socket and it is not connected then
1005          * bail out immediately.  If it is a DGRAM socket then we must first
1006          * kick the socket so that it wakes up from any sleeping calls, for
1007          * example recv(), and then afterwards return the error.
1008          */
1009
1010         sk = sock->sk;
1011
1012         lock_sock(sk);
1013         if (sock->state == SS_UNCONNECTED) {
1014                 err = -ENOTCONN;
1015                 if (sock_type_connectible(sk->sk_type))
1016                         goto out;
1017         } else {
1018                 sock->state = SS_DISCONNECTING;
1019                 err = 0;
1020         }
1021
1022         /* Receive and send shutdowns are treated alike. */
1023         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
1024         if (mode) {
1025                 sk->sk_shutdown |= mode;
1026                 sk->sk_state_change(sk);
1027
1028                 if (sock_type_connectible(sk->sk_type)) {
1029                         sock_reset_flag(sk, SOCK_DONE);
1030                         vsock_send_shutdown(sk, mode);
1031                 }
1032         }
1033
1034 out:
1035         release_sock(sk);
1036         return err;
1037 }
1038
1039 static __poll_t vsock_poll(struct file *file, struct socket *sock,
1040                                poll_table *wait)
1041 {
1042         struct sock *sk;
1043         __poll_t mask;
1044         struct vsock_sock *vsk;
1045
1046         sk = sock->sk;
1047         vsk = vsock_sk(sk);
1048
1049         poll_wait(file, sk_sleep(sk), wait);
1050         mask = 0;
1051
1052         if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
1053                 /* Signify that there has been an error on this socket. */
1054                 mask |= EPOLLERR;
1055
1056         /* INET sockets treat local write shutdown and peer write shutdown as a
1057          * case of EPOLLHUP set.
1058          */
1059         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1060             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1061              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1062                 mask |= EPOLLHUP;
1063         }
1064
1065         if (sk->sk_shutdown & RCV_SHUTDOWN ||
1066             vsk->peer_shutdown & SEND_SHUTDOWN) {
1067                 mask |= EPOLLRDHUP;
1068         }
1069
1070         if (sk_is_readable(sk))
1071                 mask |= EPOLLIN | EPOLLRDNORM;
1072
1073         if (sock->type == SOCK_DGRAM) {
1074                 /* For datagram sockets we can read if there is something in
1075                  * the queue and write as long as the socket isn't shutdown for
1076                  * sending.
1077                  */
1078                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1079                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
1080                         mask |= EPOLLIN | EPOLLRDNORM;
1081                 }
1082
1083                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1084                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1085
1086         } else if (sock_type_connectible(sk->sk_type)) {
1087                 const struct vsock_transport *transport;
1088
1089                 lock_sock(sk);
1090
1091                 transport = vsk->transport;
1092
1093                 /* Listening sockets that have connections in their accept
1094                  * queue can be read.
1095                  */
1096                 if (sk->sk_state == TCP_LISTEN
1097                     && !vsock_is_accept_queue_empty(sk))
1098                         mask |= EPOLLIN | EPOLLRDNORM;
1099
1100                 /* If there is something in the queue then we can read. */
1101                 if (transport && transport->stream_is_active(vsk) &&
1102                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1103                         bool data_ready_now = false;
1104                         int target = sock_rcvlowat(sk, 0, INT_MAX);
1105                         int ret = transport->notify_poll_in(
1106                                         vsk, target, &data_ready_now);
1107                         if (ret < 0) {
1108                                 mask |= EPOLLERR;
1109                         } else {
1110                                 if (data_ready_now)
1111                                         mask |= EPOLLIN | EPOLLRDNORM;
1112
1113                         }
1114                 }
1115
1116                 /* Sockets whose connections have been closed, reset, or
1117                  * terminated should also be considered read, and we check the
1118                  * shutdown flag for that.
1119                  */
1120                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1121                     vsk->peer_shutdown & SEND_SHUTDOWN) {
1122                         mask |= EPOLLIN | EPOLLRDNORM;
1123                 }
1124
1125                 /* Connected sockets that can produce data can be written. */
1126                 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1127                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1128                                 bool space_avail_now = false;
1129                                 int ret = transport->notify_poll_out(
1130                                                 vsk, 1, &space_avail_now);
1131                                 if (ret < 0) {
1132                                         mask |= EPOLLERR;
1133                                 } else {
1134                                         if (space_avail_now)
1135                                                 /* Remove EPOLLWRBAND since INET
1136                                                  * sockets are not setting it.
1137                                                  */
1138                                                 mask |= EPOLLOUT | EPOLLWRNORM;
1139
1140                                 }
1141                         }
1142                 }
1143
1144                 /* Simulate INET socket poll behaviors, which sets
1145                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1146                  * but local send is not shutdown.
1147                  */
1148                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1149                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1150                                 mask |= EPOLLOUT | EPOLLWRNORM;
1151
1152                 }
1153
1154                 release_sock(sk);
1155         }
1156
1157         return mask;
1158 }
1159
1160 static int vsock_read_skb(struct sock *sk, skb_read_actor_t read_actor)
1161 {
1162         struct vsock_sock *vsk = vsock_sk(sk);
1163
1164         return vsk->transport->read_skb(vsk, read_actor);
1165 }
1166
1167 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1168                                size_t len)
1169 {
1170         int err;
1171         struct sock *sk;
1172         struct vsock_sock *vsk;
1173         struct sockaddr_vm *remote_addr;
1174         const struct vsock_transport *transport;
1175
1176         if (msg->msg_flags & MSG_OOB)
1177                 return -EOPNOTSUPP;
1178
1179         /* For now, MSG_DONTWAIT is always assumed... */
1180         err = 0;
1181         sk = sock->sk;
1182         vsk = vsock_sk(sk);
1183
1184         lock_sock(sk);
1185
1186         transport = vsk->transport;
1187
1188         err = vsock_auto_bind(vsk);
1189         if (err)
1190                 goto out;
1191
1192
1193         /* If the provided message contains an address, use that.  Otherwise
1194          * fall back on the socket's remote handle (if it has been connected).
1195          */
1196         if (msg->msg_name &&
1197             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1198                             &remote_addr) == 0) {
1199                 /* Ensure this address is of the right type and is a valid
1200                  * destination.
1201                  */
1202
1203                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1204                         remote_addr->svm_cid = transport->get_local_cid();
1205
1206                 if (!vsock_addr_bound(remote_addr)) {
1207                         err = -EINVAL;
1208                         goto out;
1209                 }
1210         } else if (sock->state == SS_CONNECTED) {
1211                 remote_addr = &vsk->remote_addr;
1212
1213                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1214                         remote_addr->svm_cid = transport->get_local_cid();
1215
1216                 /* XXX Should connect() or this function ensure remote_addr is
1217                  * bound?
1218                  */
1219                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1220                         err = -EINVAL;
1221                         goto out;
1222                 }
1223         } else {
1224                 err = -EINVAL;
1225                 goto out;
1226         }
1227
1228         if (!transport->dgram_allow(remote_addr->svm_cid,
1229                                     remote_addr->svm_port)) {
1230                 err = -EINVAL;
1231                 goto out;
1232         }
1233
1234         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1235
1236 out:
1237         release_sock(sk);
1238         return err;
1239 }
1240
1241 static int vsock_dgram_connect(struct socket *sock,
1242                                struct sockaddr *addr, int addr_len, int flags)
1243 {
1244         int err;
1245         struct sock *sk;
1246         struct vsock_sock *vsk;
1247         struct sockaddr_vm *remote_addr;
1248
1249         sk = sock->sk;
1250         vsk = vsock_sk(sk);
1251
1252         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1253         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1254                 lock_sock(sk);
1255                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1256                                 VMADDR_PORT_ANY);
1257                 sock->state = SS_UNCONNECTED;
1258                 release_sock(sk);
1259                 return 0;
1260         } else if (err != 0)
1261                 return -EINVAL;
1262
1263         lock_sock(sk);
1264
1265         err = vsock_auto_bind(vsk);
1266         if (err)
1267                 goto out;
1268
1269         if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1270                                          remote_addr->svm_port)) {
1271                 err = -EINVAL;
1272                 goto out;
1273         }
1274
1275         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1276         sock->state = SS_CONNECTED;
1277
1278         /* sock map disallows redirection of non-TCP sockets with sk_state !=
1279          * TCP_ESTABLISHED (see sock_map_redirect_allowed()), so we set
1280          * TCP_ESTABLISHED here to allow redirection of connected vsock dgrams.
1281          *
1282          * This doesn't seem to be abnormal state for datagram sockets, as the
1283          * same approach can be see in other datagram socket types as well
1284          * (such as unix sockets).
1285          */
1286         sk->sk_state = TCP_ESTABLISHED;
1287
1288 out:
1289         release_sock(sk);
1290         return err;
1291 }
1292
1293 int __vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1294                           size_t len, int flags)
1295 {
1296         struct sock *sk = sock->sk;
1297         struct vsock_sock *vsk = vsock_sk(sk);
1298
1299         return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1300 }
1301
1302 int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1303                         size_t len, int flags)
1304 {
1305 #ifdef CONFIG_BPF_SYSCALL
1306         struct sock *sk = sock->sk;
1307         const struct proto *prot;
1308
1309         prot = READ_ONCE(sk->sk_prot);
1310         if (prot != &vsock_proto)
1311                 return prot->recvmsg(sk, msg, len, flags, NULL);
1312 #endif
1313
1314         return __vsock_dgram_recvmsg(sock, msg, len, flags);
1315 }
1316 EXPORT_SYMBOL_GPL(vsock_dgram_recvmsg);
1317
1318 static int vsock_do_ioctl(struct socket *sock, unsigned int cmd,
1319                           int __user *arg)
1320 {
1321         struct sock *sk = sock->sk;
1322         struct vsock_sock *vsk;
1323         int ret;
1324
1325         vsk = vsock_sk(sk);
1326
1327         switch (cmd) {
1328         case SIOCOUTQ: {
1329                 ssize_t n_bytes;
1330
1331                 if (!vsk->transport || !vsk->transport->unsent_bytes) {
1332                         ret = -EOPNOTSUPP;
1333                         break;
1334                 }
1335
1336                 if (sock_type_connectible(sk->sk_type) && sk->sk_state == TCP_LISTEN) {
1337                         ret = -EINVAL;
1338                         break;
1339                 }
1340
1341                 n_bytes = vsk->transport->unsent_bytes(vsk);
1342                 if (n_bytes < 0) {
1343                         ret = n_bytes;
1344                         break;
1345                 }
1346
1347                 ret = put_user(n_bytes, arg);
1348                 break;
1349         }
1350         default:
1351                 ret = -ENOIOCTLCMD;
1352         }
1353
1354         return ret;
1355 }
1356
1357 static int vsock_ioctl(struct socket *sock, unsigned int cmd,
1358                        unsigned long arg)
1359 {
1360         int ret;
1361
1362         lock_sock(sock->sk);
1363         ret = vsock_do_ioctl(sock, cmd, (int __user *)arg);
1364         release_sock(sock->sk);
1365
1366         return ret;
1367 }
1368
1369 static const struct proto_ops vsock_dgram_ops = {
1370         .family = PF_VSOCK,
1371         .owner = THIS_MODULE,
1372         .release = vsock_release,
1373         .bind = vsock_bind,
1374         .connect = vsock_dgram_connect,
1375         .socketpair = sock_no_socketpair,
1376         .accept = sock_no_accept,
1377         .getname = vsock_getname,
1378         .poll = vsock_poll,
1379         .ioctl = vsock_ioctl,
1380         .listen = sock_no_listen,
1381         .shutdown = vsock_shutdown,
1382         .sendmsg = vsock_dgram_sendmsg,
1383         .recvmsg = vsock_dgram_recvmsg,
1384         .mmap = sock_no_mmap,
1385         .read_skb = vsock_read_skb,
1386 };
1387
1388 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1389 {
1390         const struct vsock_transport *transport = vsk->transport;
1391
1392         if (!transport || !transport->cancel_pkt)
1393                 return -EOPNOTSUPP;
1394
1395         return transport->cancel_pkt(vsk);
1396 }
1397
1398 static void vsock_connect_timeout(struct work_struct *work)
1399 {
1400         struct sock *sk;
1401         struct vsock_sock *vsk;
1402
1403         vsk = container_of(work, struct vsock_sock, connect_work.work);
1404         sk = sk_vsock(vsk);
1405
1406         lock_sock(sk);
1407         if (sk->sk_state == TCP_SYN_SENT &&
1408             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1409                 sk->sk_state = TCP_CLOSE;
1410                 sk->sk_socket->state = SS_UNCONNECTED;
1411                 sk->sk_err = ETIMEDOUT;
1412                 sk_error_report(sk);
1413                 vsock_transport_cancel_pkt(vsk);
1414         }
1415         release_sock(sk);
1416
1417         sock_put(sk);
1418 }
1419
1420 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1421                          int addr_len, int flags)
1422 {
1423         int err;
1424         struct sock *sk;
1425         struct vsock_sock *vsk;
1426         const struct vsock_transport *transport;
1427         struct sockaddr_vm *remote_addr;
1428         long timeout;
1429         DEFINE_WAIT(wait);
1430
1431         err = 0;
1432         sk = sock->sk;
1433         vsk = vsock_sk(sk);
1434
1435         lock_sock(sk);
1436
1437         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1438         switch (sock->state) {
1439         case SS_CONNECTED:
1440                 err = -EISCONN;
1441                 goto out;
1442         case SS_DISCONNECTING:
1443                 err = -EINVAL;
1444                 goto out;
1445         case SS_CONNECTING:
1446                 /* This continues on so we can move sock into the SS_CONNECTED
1447                  * state once the connection has completed (at which point err
1448                  * will be set to zero also).  Otherwise, we will either wait
1449                  * for the connection or return -EALREADY should this be a
1450                  * non-blocking call.
1451                  */
1452                 err = -EALREADY;
1453                 if (flags & O_NONBLOCK)
1454                         goto out;
1455                 break;
1456         default:
1457                 if ((sk->sk_state == TCP_LISTEN) ||
1458                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1459                         err = -EINVAL;
1460                         goto out;
1461                 }
1462
1463                 /* Set the remote address that we are connecting to. */
1464                 memcpy(&vsk->remote_addr, remote_addr,
1465                        sizeof(vsk->remote_addr));
1466
1467                 err = vsock_assign_transport(vsk, NULL);
1468                 if (err)
1469                         goto out;
1470
1471                 transport = vsk->transport;
1472
1473                 /* The hypervisor and well-known contexts do not have socket
1474                  * endpoints.
1475                  */
1476                 if (!transport ||
1477                     !transport->stream_allow(remote_addr->svm_cid,
1478                                              remote_addr->svm_port)) {
1479                         err = -ENETUNREACH;
1480                         goto out;
1481                 }
1482
1483                 if (vsock_msgzerocopy_allow(transport)) {
1484                         set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1485                 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1486                         /* If this option was set before 'connect()',
1487                          * when transport was unknown, check that this
1488                          * feature is supported here.
1489                          */
1490                         err = -EOPNOTSUPP;
1491                         goto out;
1492                 }
1493
1494                 err = vsock_auto_bind(vsk);
1495                 if (err)
1496                         goto out;
1497
1498                 sk->sk_state = TCP_SYN_SENT;
1499
1500                 err = transport->connect(vsk);
1501                 if (err < 0)
1502                         goto out;
1503
1504                 /* Mark sock as connecting and set the error code to in
1505                  * progress in case this is a non-blocking connect.
1506                  */
1507                 sock->state = SS_CONNECTING;
1508                 err = -EINPROGRESS;
1509         }
1510
1511         /* The receive path will handle all communication until we are able to
1512          * enter the connected state.  Here we wait for the connection to be
1513          * completed or a notification of an error.
1514          */
1515         timeout = vsk->connect_timeout;
1516         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1517
1518         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1519                 if (flags & O_NONBLOCK) {
1520                         /* If we're not going to block, we schedule a timeout
1521                          * function to generate a timeout on the connection
1522                          * attempt, in case the peer doesn't respond in a
1523                          * timely manner. We hold on to the socket until the
1524                          * timeout fires.
1525                          */
1526                         sock_hold(sk);
1527
1528                         /* If the timeout function is already scheduled,
1529                          * reschedule it, then ungrab the socket refcount to
1530                          * keep it balanced.
1531                          */
1532                         if (mod_delayed_work(system_wq, &vsk->connect_work,
1533                                              timeout))
1534                                 sock_put(sk);
1535
1536                         /* Skip ahead to preserve error code set above. */
1537                         goto out_wait;
1538                 }
1539
1540                 release_sock(sk);
1541                 timeout = schedule_timeout(timeout);
1542                 lock_sock(sk);
1543
1544                 if (signal_pending(current)) {
1545                         err = sock_intr_errno(timeout);
1546                         sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1547                         sock->state = SS_UNCONNECTED;
1548                         vsock_transport_cancel_pkt(vsk);
1549                         vsock_remove_connected(vsk);
1550                         goto out_wait;
1551                 } else if ((sk->sk_state != TCP_ESTABLISHED) && (timeout == 0)) {
1552                         err = -ETIMEDOUT;
1553                         sk->sk_state = TCP_CLOSE;
1554                         sock->state = SS_UNCONNECTED;
1555                         vsock_transport_cancel_pkt(vsk);
1556                         goto out_wait;
1557                 }
1558
1559                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1560         }
1561
1562         if (sk->sk_err) {
1563                 err = -sk->sk_err;
1564                 sk->sk_state = TCP_CLOSE;
1565                 sock->state = SS_UNCONNECTED;
1566         } else {
1567                 err = 0;
1568         }
1569
1570 out_wait:
1571         finish_wait(sk_sleep(sk), &wait);
1572 out:
1573         release_sock(sk);
1574         return err;
1575 }
1576
1577 static int vsock_accept(struct socket *sock, struct socket *newsock,
1578                         struct proto_accept_arg *arg)
1579 {
1580         struct sock *listener;
1581         int err;
1582         struct sock *connected;
1583         struct vsock_sock *vconnected;
1584         long timeout;
1585         DEFINE_WAIT(wait);
1586
1587         err = 0;
1588         listener = sock->sk;
1589
1590         lock_sock(listener);
1591
1592         if (!sock_type_connectible(sock->type)) {
1593                 err = -EOPNOTSUPP;
1594                 goto out;
1595         }
1596
1597         if (listener->sk_state != TCP_LISTEN) {
1598                 err = -EINVAL;
1599                 goto out;
1600         }
1601
1602         /* Wait for children sockets to appear; these are the new sockets
1603          * created upon connection establishment.
1604          */
1605         timeout = sock_rcvtimeo(listener, arg->flags & O_NONBLOCK);
1606         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1607
1608         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1609                listener->sk_err == 0) {
1610                 release_sock(listener);
1611                 timeout = schedule_timeout(timeout);
1612                 finish_wait(sk_sleep(listener), &wait);
1613                 lock_sock(listener);
1614
1615                 if (signal_pending(current)) {
1616                         err = sock_intr_errno(timeout);
1617                         goto out;
1618                 } else if (timeout == 0) {
1619                         err = -EAGAIN;
1620                         goto out;
1621                 }
1622
1623                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1624         }
1625         finish_wait(sk_sleep(listener), &wait);
1626
1627         if (listener->sk_err)
1628                 err = -listener->sk_err;
1629
1630         if (connected) {
1631                 sk_acceptq_removed(listener);
1632
1633                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1634                 vconnected = vsock_sk(connected);
1635
1636                 /* If the listener socket has received an error, then we should
1637                  * reject this socket and return.  Note that we simply mark the
1638                  * socket rejected, drop our reference, and let the cleanup
1639                  * function handle the cleanup; the fact that we found it in
1640                  * the listener's accept queue guarantees that the cleanup
1641                  * function hasn't run yet.
1642                  */
1643                 if (err) {
1644                         vconnected->rejected = true;
1645                 } else {
1646                         newsock->state = SS_CONNECTED;
1647                         sock_graft(connected, newsock);
1648                         if (vsock_msgzerocopy_allow(vconnected->transport))
1649                                 set_bit(SOCK_SUPPORT_ZC,
1650                                         &connected->sk_socket->flags);
1651                 }
1652
1653                 release_sock(connected);
1654                 sock_put(connected);
1655         }
1656
1657 out:
1658         release_sock(listener);
1659         return err;
1660 }
1661
1662 static int vsock_listen(struct socket *sock, int backlog)
1663 {
1664         int err;
1665         struct sock *sk;
1666         struct vsock_sock *vsk;
1667
1668         sk = sock->sk;
1669
1670         lock_sock(sk);
1671
1672         if (!sock_type_connectible(sk->sk_type)) {
1673                 err = -EOPNOTSUPP;
1674                 goto out;
1675         }
1676
1677         if (sock->state != SS_UNCONNECTED) {
1678                 err = -EINVAL;
1679                 goto out;
1680         }
1681
1682         vsk = vsock_sk(sk);
1683
1684         if (!vsock_addr_bound(&vsk->local_addr)) {
1685                 err = -EINVAL;
1686                 goto out;
1687         }
1688
1689         sk->sk_max_ack_backlog = backlog;
1690         sk->sk_state = TCP_LISTEN;
1691
1692         err = 0;
1693
1694 out:
1695         release_sock(sk);
1696         return err;
1697 }
1698
1699 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1700                                      const struct vsock_transport *transport,
1701                                      u64 val)
1702 {
1703         if (val > vsk->buffer_max_size)
1704                 val = vsk->buffer_max_size;
1705
1706         if (val < vsk->buffer_min_size)
1707                 val = vsk->buffer_min_size;
1708
1709         if (val != vsk->buffer_size &&
1710             transport && transport->notify_buffer_size)
1711                 transport->notify_buffer_size(vsk, &val);
1712
1713         vsk->buffer_size = val;
1714 }
1715
1716 static int vsock_connectible_setsockopt(struct socket *sock,
1717                                         int level,
1718                                         int optname,
1719                                         sockptr_t optval,
1720                                         unsigned int optlen)
1721 {
1722         int err;
1723         struct sock *sk;
1724         struct vsock_sock *vsk;
1725         const struct vsock_transport *transport;
1726         u64 val;
1727
1728         if (level != AF_VSOCK && level != SOL_SOCKET)
1729                 return -ENOPROTOOPT;
1730
1731 #define COPY_IN(_v)                                       \
1732         do {                                              \
1733                 if (optlen < sizeof(_v)) {                \
1734                         err = -EINVAL;                    \
1735                         goto exit;                        \
1736                 }                                         \
1737                 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) {  \
1738                         err = -EFAULT;                                  \
1739                         goto exit;                                      \
1740                 }                                                       \
1741         } while (0)
1742
1743         err = 0;
1744         sk = sock->sk;
1745         vsk = vsock_sk(sk);
1746
1747         lock_sock(sk);
1748
1749         transport = vsk->transport;
1750
1751         if (level == SOL_SOCKET) {
1752                 int zerocopy;
1753
1754                 if (optname != SO_ZEROCOPY) {
1755                         release_sock(sk);
1756                         return sock_setsockopt(sock, level, optname, optval, optlen);
1757                 }
1758
1759                 /* Use 'int' type here, because variable to
1760                  * set this option usually has this type.
1761                  */
1762                 COPY_IN(zerocopy);
1763
1764                 if (zerocopy < 0 || zerocopy > 1) {
1765                         err = -EINVAL;
1766                         goto exit;
1767                 }
1768
1769                 if (transport && !vsock_msgzerocopy_allow(transport)) {
1770                         err = -EOPNOTSUPP;
1771                         goto exit;
1772                 }
1773
1774                 sock_valbool_flag(sk, SOCK_ZEROCOPY, zerocopy);
1775                 goto exit;
1776         }
1777
1778         switch (optname) {
1779         case SO_VM_SOCKETS_BUFFER_SIZE:
1780                 COPY_IN(val);
1781                 vsock_update_buffer_size(vsk, transport, val);
1782                 break;
1783
1784         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1785                 COPY_IN(val);
1786                 vsk->buffer_max_size = val;
1787                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1788                 break;
1789
1790         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1791                 COPY_IN(val);
1792                 vsk->buffer_min_size = val;
1793                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1794                 break;
1795
1796         case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1797         case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: {
1798                 struct __kernel_sock_timeval tv;
1799
1800                 err = sock_copy_user_timeval(&tv, optval, optlen,
1801                                              optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1802                 if (err)
1803                         break;
1804                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1805                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1806                         vsk->connect_timeout = tv.tv_sec * HZ +
1807                                 DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ));
1808                         if (vsk->connect_timeout == 0)
1809                                 vsk->connect_timeout =
1810                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1811
1812                 } else {
1813                         err = -ERANGE;
1814                 }
1815                 break;
1816         }
1817
1818         default:
1819                 err = -ENOPROTOOPT;
1820                 break;
1821         }
1822
1823 #undef COPY_IN
1824
1825 exit:
1826         release_sock(sk);
1827         return err;
1828 }
1829
1830 static int vsock_connectible_getsockopt(struct socket *sock,
1831                                         int level, int optname,
1832                                         char __user *optval,
1833                                         int __user *optlen)
1834 {
1835         struct sock *sk = sock->sk;
1836         struct vsock_sock *vsk = vsock_sk(sk);
1837
1838         union {
1839                 u64 val64;
1840                 struct old_timeval32 tm32;
1841                 struct __kernel_old_timeval tm;
1842                 struct  __kernel_sock_timeval stm;
1843         } v;
1844
1845         int lv = sizeof(v.val64);
1846         int len;
1847
1848         if (level != AF_VSOCK)
1849                 return -ENOPROTOOPT;
1850
1851         if (get_user(len, optlen))
1852                 return -EFAULT;
1853
1854         memset(&v, 0, sizeof(v));
1855
1856         switch (optname) {
1857         case SO_VM_SOCKETS_BUFFER_SIZE:
1858                 v.val64 = vsk->buffer_size;
1859                 break;
1860
1861         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1862                 v.val64 = vsk->buffer_max_size;
1863                 break;
1864
1865         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1866                 v.val64 = vsk->buffer_min_size;
1867                 break;
1868
1869         case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1870         case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD:
1871                 lv = sock_get_timeout(vsk->connect_timeout, &v,
1872                                       optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1873                 break;
1874
1875         default:
1876                 return -ENOPROTOOPT;
1877         }
1878
1879         if (len < lv)
1880                 return -EINVAL;
1881         if (len > lv)
1882                 len = lv;
1883         if (copy_to_user(optval, &v, len))
1884                 return -EFAULT;
1885
1886         if (put_user(len, optlen))
1887                 return -EFAULT;
1888
1889         return 0;
1890 }
1891
1892 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1893                                      size_t len)
1894 {
1895         struct sock *sk;
1896         struct vsock_sock *vsk;
1897         const struct vsock_transport *transport;
1898         ssize_t total_written;
1899         long timeout;
1900         int err;
1901         struct vsock_transport_send_notify_data send_data;
1902         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1903
1904         sk = sock->sk;
1905         vsk = vsock_sk(sk);
1906         total_written = 0;
1907         err = 0;
1908
1909         if (msg->msg_flags & MSG_OOB)
1910                 return -EOPNOTSUPP;
1911
1912         lock_sock(sk);
1913
1914         transport = vsk->transport;
1915
1916         /* Callers should not provide a destination with connection oriented
1917          * sockets.
1918          */
1919         if (msg->msg_namelen) {
1920                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1921                 goto out;
1922         }
1923
1924         /* Send data only if both sides are not shutdown in the direction. */
1925         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1926             vsk->peer_shutdown & RCV_SHUTDOWN) {
1927                 err = -EPIPE;
1928                 goto out;
1929         }
1930
1931         if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1932             !vsock_addr_bound(&vsk->local_addr)) {
1933                 err = -ENOTCONN;
1934                 goto out;
1935         }
1936
1937         if (!vsock_addr_bound(&vsk->remote_addr)) {
1938                 err = -EDESTADDRREQ;
1939                 goto out;
1940         }
1941
1942         if (msg->msg_flags & MSG_ZEROCOPY &&
1943             !vsock_msgzerocopy_allow(transport)) {
1944                 err = -EOPNOTSUPP;
1945                 goto out;
1946         }
1947
1948         /* Wait for room in the produce queue to enqueue our user's data. */
1949         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1950
1951         err = transport->notify_send_init(vsk, &send_data);
1952         if (err < 0)
1953                 goto out;
1954
1955         while (total_written < len) {
1956                 ssize_t written;
1957
1958                 add_wait_queue(sk_sleep(sk), &wait);
1959                 while (vsock_stream_has_space(vsk) == 0 &&
1960                        sk->sk_err == 0 &&
1961                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1962                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1963
1964                         /* Don't wait for non-blocking sockets. */
1965                         if (timeout == 0) {
1966                                 err = -EAGAIN;
1967                                 remove_wait_queue(sk_sleep(sk), &wait);
1968                                 goto out_err;
1969                         }
1970
1971                         err = transport->notify_send_pre_block(vsk, &send_data);
1972                         if (err < 0) {
1973                                 remove_wait_queue(sk_sleep(sk), &wait);
1974                                 goto out_err;
1975                         }
1976
1977                         release_sock(sk);
1978                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1979                         lock_sock(sk);
1980                         if (signal_pending(current)) {
1981                                 err = sock_intr_errno(timeout);
1982                                 remove_wait_queue(sk_sleep(sk), &wait);
1983                                 goto out_err;
1984                         } else if (timeout == 0) {
1985                                 err = -EAGAIN;
1986                                 remove_wait_queue(sk_sleep(sk), &wait);
1987                                 goto out_err;
1988                         }
1989                 }
1990                 remove_wait_queue(sk_sleep(sk), &wait);
1991
1992                 /* These checks occur both as part of and after the loop
1993                  * conditional since we need to check before and after
1994                  * sleeping.
1995                  */
1996                 if (sk->sk_err) {
1997                         err = -sk->sk_err;
1998                         goto out_err;
1999                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
2000                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
2001                         err = -EPIPE;
2002                         goto out_err;
2003                 }
2004
2005                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
2006                 if (err < 0)
2007                         goto out_err;
2008
2009                 /* Note that enqueue will only write as many bytes as are free
2010                  * in the produce queue, so we don't need to ensure len is
2011                  * smaller than the queue size.  It is the caller's
2012                  * responsibility to check how many bytes we were able to send.
2013                  */
2014
2015                 if (sk->sk_type == SOCK_SEQPACKET) {
2016                         written = transport->seqpacket_enqueue(vsk,
2017                                                 msg, len - total_written);
2018                 } else {
2019                         written = transport->stream_enqueue(vsk,
2020                                         msg, len - total_written);
2021                 }
2022
2023                 if (written < 0) {
2024                         err = written;
2025                         goto out_err;
2026                 }
2027
2028                 total_written += written;
2029
2030                 err = transport->notify_send_post_enqueue(
2031                                 vsk, written, &send_data);
2032                 if (err < 0)
2033                         goto out_err;
2034
2035         }
2036
2037 out_err:
2038         if (total_written > 0) {
2039                 /* Return number of written bytes only if:
2040                  * 1) SOCK_STREAM socket.
2041                  * 2) SOCK_SEQPACKET socket when whole buffer is sent.
2042                  */
2043                 if (sk->sk_type == SOCK_STREAM || total_written == len)
2044                         err = total_written;
2045         }
2046 out:
2047         if (sk->sk_type == SOCK_STREAM)
2048                 err = sk_stream_error(sk, msg->msg_flags, err);
2049
2050         release_sock(sk);
2051         return err;
2052 }
2053
2054 static int vsock_connectible_wait_data(struct sock *sk,
2055                                        struct wait_queue_entry *wait,
2056                                        long timeout,
2057                                        struct vsock_transport_recv_notify_data *recv_data,
2058                                        size_t target)
2059 {
2060         const struct vsock_transport *transport;
2061         struct vsock_sock *vsk;
2062         s64 data;
2063         int err;
2064
2065         vsk = vsock_sk(sk);
2066         err = 0;
2067         transport = vsk->transport;
2068
2069         while (1) {
2070                 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
2071                 data = vsock_connectible_has_data(vsk);
2072                 if (data != 0)
2073                         break;
2074
2075                 if (sk->sk_err != 0 ||
2076                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
2077                     (vsk->peer_shutdown & SEND_SHUTDOWN)) {
2078                         break;
2079                 }
2080
2081                 /* Don't wait for non-blocking sockets. */
2082                 if (timeout == 0) {
2083                         err = -EAGAIN;
2084                         break;
2085                 }
2086
2087                 if (recv_data) {
2088                         err = transport->notify_recv_pre_block(vsk, target, recv_data);
2089                         if (err < 0)
2090                                 break;
2091                 }
2092
2093                 release_sock(sk);
2094                 timeout = schedule_timeout(timeout);
2095                 lock_sock(sk);
2096
2097                 if (signal_pending(current)) {
2098                         err = sock_intr_errno(timeout);
2099                         break;
2100                 } else if (timeout == 0) {
2101                         err = -EAGAIN;
2102                         break;
2103                 }
2104         }
2105
2106         finish_wait(sk_sleep(sk), wait);
2107
2108         if (err)
2109                 return err;
2110
2111         /* Internal transport error when checking for available
2112          * data. XXX This should be changed to a connection
2113          * reset in a later change.
2114          */
2115         if (data < 0)
2116                 return -ENOMEM;
2117
2118         return data;
2119 }
2120
2121 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2122                                   size_t len, int flags)
2123 {
2124         struct vsock_transport_recv_notify_data recv_data;
2125         const struct vsock_transport *transport;
2126         struct vsock_sock *vsk;
2127         ssize_t copied;
2128         size_t target;
2129         long timeout;
2130         int err;
2131
2132         DEFINE_WAIT(wait);
2133
2134         vsk = vsock_sk(sk);
2135         transport = vsk->transport;
2136
2137         /* We must not copy less than target bytes into the user's buffer
2138          * before returning successfully, so we wait for the consume queue to
2139          * have that much data to consume before dequeueing.  Note that this
2140          * makes it impossible to handle cases where target is greater than the
2141          * queue size.
2142          */
2143         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2144         if (target >= transport->stream_rcvhiwat(vsk)) {
2145                 err = -ENOMEM;
2146                 goto out;
2147         }
2148         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2149         copied = 0;
2150
2151         err = transport->notify_recv_init(vsk, target, &recv_data);
2152         if (err < 0)
2153                 goto out;
2154
2155
2156         while (1) {
2157                 ssize_t read;
2158
2159                 err = vsock_connectible_wait_data(sk, &wait, timeout,
2160                                                   &recv_data, target);
2161                 if (err <= 0)
2162                         break;
2163
2164                 err = transport->notify_recv_pre_dequeue(vsk, target,
2165                                                          &recv_data);
2166                 if (err < 0)
2167                         break;
2168
2169                 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
2170                 if (read < 0) {
2171                         err = read;
2172                         break;
2173                 }
2174
2175                 copied += read;
2176
2177                 err = transport->notify_recv_post_dequeue(vsk, target, read,
2178                                                 !(flags & MSG_PEEK), &recv_data);
2179                 if (err < 0)
2180                         goto out;
2181
2182                 if (read >= target || flags & MSG_PEEK)
2183                         break;
2184
2185                 target -= read;
2186         }
2187
2188         if (sk->sk_err)
2189                 err = -sk->sk_err;
2190         else if (sk->sk_shutdown & RCV_SHUTDOWN)
2191                 err = 0;
2192
2193         if (copied > 0)
2194                 err = copied;
2195
2196 out:
2197         return err;
2198 }
2199
2200 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2201                                      size_t len, int flags)
2202 {
2203         const struct vsock_transport *transport;
2204         struct vsock_sock *vsk;
2205         ssize_t msg_len;
2206         long timeout;
2207         int err = 0;
2208         DEFINE_WAIT(wait);
2209
2210         vsk = vsock_sk(sk);
2211         transport = vsk->transport;
2212
2213         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2214
2215         err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0);
2216         if (err <= 0)
2217                 goto out;
2218
2219         msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2220
2221         if (msg_len < 0) {
2222                 err = msg_len;
2223                 goto out;
2224         }
2225
2226         if (sk->sk_err) {
2227                 err = -sk->sk_err;
2228         } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2229                 err = 0;
2230         } else {
2231                 /* User sets MSG_TRUNC, so return real length of
2232                  * packet.
2233                  */
2234                 if (flags & MSG_TRUNC)
2235                         err = msg_len;
2236                 else
2237                         err = len - msg_data_left(msg);
2238
2239                 /* Always set MSG_TRUNC if real length of packet is
2240                  * bigger than user's buffer.
2241                  */
2242                 if (msg_len > len)
2243                         msg->msg_flags |= MSG_TRUNC;
2244         }
2245
2246 out:
2247         return err;
2248 }
2249
2250 int
2251 __vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2252                             int flags)
2253 {
2254         struct sock *sk;
2255         struct vsock_sock *vsk;
2256         const struct vsock_transport *transport;
2257         int err;
2258
2259         sk = sock->sk;
2260
2261         if (unlikely(flags & MSG_ERRQUEUE))
2262                 return sock_recv_errqueue(sk, msg, len, SOL_VSOCK, VSOCK_RECVERR);
2263
2264         vsk = vsock_sk(sk);
2265         err = 0;
2266
2267         lock_sock(sk);
2268
2269         transport = vsk->transport;
2270
2271         if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2272                 /* Recvmsg is supposed to return 0 if a peer performs an
2273                  * orderly shutdown. Differentiate between that case and when a
2274                  * peer has not connected or a local shutdown occurred with the
2275                  * SOCK_DONE flag.
2276                  */
2277                 if (sock_flag(sk, SOCK_DONE))
2278                         err = 0;
2279                 else
2280                         err = -ENOTCONN;
2281
2282                 goto out;
2283         }
2284
2285         if (flags & MSG_OOB) {
2286                 err = -EOPNOTSUPP;
2287                 goto out;
2288         }
2289
2290         /* We don't check peer_shutdown flag here since peer may actually shut
2291          * down, but there can be data in the queue that a local socket can
2292          * receive.
2293          */
2294         if (sk->sk_shutdown & RCV_SHUTDOWN) {
2295                 err = 0;
2296                 goto out;
2297         }
2298
2299         /* It is valid on Linux to pass in a zero-length receive buffer.  This
2300          * is not an error.  We may as well bail out now.
2301          */
2302         if (!len) {
2303                 err = 0;
2304                 goto out;
2305         }
2306
2307         if (sk->sk_type == SOCK_STREAM)
2308                 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2309         else
2310                 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2311
2312 out:
2313         release_sock(sk);
2314         return err;
2315 }
2316
2317 int
2318 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2319                           int flags)
2320 {
2321 #ifdef CONFIG_BPF_SYSCALL
2322         struct sock *sk = sock->sk;
2323         const struct proto *prot;
2324
2325         prot = READ_ONCE(sk->sk_prot);
2326         if (prot != &vsock_proto)
2327                 return prot->recvmsg(sk, msg, len, flags, NULL);
2328 #endif
2329
2330         return __vsock_connectible_recvmsg(sock, msg, len, flags);
2331 }
2332 EXPORT_SYMBOL_GPL(vsock_connectible_recvmsg);
2333
2334 static int vsock_set_rcvlowat(struct sock *sk, int val)
2335 {
2336         const struct vsock_transport *transport;
2337         struct vsock_sock *vsk;
2338
2339         vsk = vsock_sk(sk);
2340
2341         if (val > vsk->buffer_size)
2342                 return -EINVAL;
2343
2344         transport = vsk->transport;
2345
2346         if (transport && transport->notify_set_rcvlowat) {
2347                 int err;
2348
2349                 err = transport->notify_set_rcvlowat(vsk, val);
2350                 if (err)
2351                         return err;
2352         }
2353
2354         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
2355         return 0;
2356 }
2357
2358 static const struct proto_ops vsock_stream_ops = {
2359         .family = PF_VSOCK,
2360         .owner = THIS_MODULE,
2361         .release = vsock_release,
2362         .bind = vsock_bind,
2363         .connect = vsock_connect,
2364         .socketpair = sock_no_socketpair,
2365         .accept = vsock_accept,
2366         .getname = vsock_getname,
2367         .poll = vsock_poll,
2368         .ioctl = vsock_ioctl,
2369         .listen = vsock_listen,
2370         .shutdown = vsock_shutdown,
2371         .setsockopt = vsock_connectible_setsockopt,
2372         .getsockopt = vsock_connectible_getsockopt,
2373         .sendmsg = vsock_connectible_sendmsg,
2374         .recvmsg = vsock_connectible_recvmsg,
2375         .mmap = sock_no_mmap,
2376         .set_rcvlowat = vsock_set_rcvlowat,
2377         .read_skb = vsock_read_skb,
2378 };
2379
2380 static const struct proto_ops vsock_seqpacket_ops = {
2381         .family = PF_VSOCK,
2382         .owner = THIS_MODULE,
2383         .release = vsock_release,
2384         .bind = vsock_bind,
2385         .connect = vsock_connect,
2386         .socketpair = sock_no_socketpair,
2387         .accept = vsock_accept,
2388         .getname = vsock_getname,
2389         .poll = vsock_poll,
2390         .ioctl = vsock_ioctl,
2391         .listen = vsock_listen,
2392         .shutdown = vsock_shutdown,
2393         .setsockopt = vsock_connectible_setsockopt,
2394         .getsockopt = vsock_connectible_getsockopt,
2395         .sendmsg = vsock_connectible_sendmsg,
2396         .recvmsg = vsock_connectible_recvmsg,
2397         .mmap = sock_no_mmap,
2398         .read_skb = vsock_read_skb,
2399 };
2400
2401 static int vsock_create(struct net *net, struct socket *sock,
2402                         int protocol, int kern)
2403 {
2404         struct vsock_sock *vsk;
2405         struct sock *sk;
2406         int ret;
2407
2408         if (!sock)
2409                 return -EINVAL;
2410
2411         if (protocol && protocol != PF_VSOCK)
2412                 return -EPROTONOSUPPORT;
2413
2414         switch (sock->type) {
2415         case SOCK_DGRAM:
2416                 sock->ops = &vsock_dgram_ops;
2417                 break;
2418         case SOCK_STREAM:
2419                 sock->ops = &vsock_stream_ops;
2420                 break;
2421         case SOCK_SEQPACKET:
2422                 sock->ops = &vsock_seqpacket_ops;
2423                 break;
2424         default:
2425                 return -ESOCKTNOSUPPORT;
2426         }
2427
2428         sock->state = SS_UNCONNECTED;
2429
2430         sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2431         if (!sk)
2432                 return -ENOMEM;
2433
2434         vsk = vsock_sk(sk);
2435
2436         if (sock->type == SOCK_DGRAM) {
2437                 ret = vsock_assign_transport(vsk, NULL);
2438                 if (ret < 0) {
2439                         sock->sk = NULL;
2440                         sock_put(sk);
2441                         return ret;
2442                 }
2443         }
2444
2445         /* SOCK_DGRAM doesn't have 'setsockopt' callback set in its
2446          * proto_ops, so there is no handler for custom logic.
2447          */
2448         if (sock_type_connectible(sock->type))
2449                 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2450
2451         vsock_insert_unbound(vsk);
2452
2453         return 0;
2454 }
2455
2456 static const struct net_proto_family vsock_family_ops = {
2457         .family = AF_VSOCK,
2458         .create = vsock_create,
2459         .owner = THIS_MODULE,
2460 };
2461
2462 static long vsock_dev_do_ioctl(struct file *filp,
2463                                unsigned int cmd, void __user *ptr)
2464 {
2465         u32 __user *p = ptr;
2466         u32 cid = VMADDR_CID_ANY;
2467         int retval = 0;
2468
2469         switch (cmd) {
2470         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2471                 /* To be compatible with the VMCI behavior, we prioritize the
2472                  * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2473                  */
2474                 if (transport_g2h)
2475                         cid = transport_g2h->get_local_cid();
2476                 else if (transport_h2g)
2477                         cid = transport_h2g->get_local_cid();
2478
2479                 if (put_user(cid, p) != 0)
2480                         retval = -EFAULT;
2481                 break;
2482
2483         default:
2484                 retval = -ENOIOCTLCMD;
2485         }
2486
2487         return retval;
2488 }
2489
2490 static long vsock_dev_ioctl(struct file *filp,
2491                             unsigned int cmd, unsigned long arg)
2492 {
2493         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2494 }
2495
2496 #ifdef CONFIG_COMPAT
2497 static long vsock_dev_compat_ioctl(struct file *filp,
2498                                    unsigned int cmd, unsigned long arg)
2499 {
2500         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2501 }
2502 #endif
2503
2504 static const struct file_operations vsock_device_ops = {
2505         .owner          = THIS_MODULE,
2506         .unlocked_ioctl = vsock_dev_ioctl,
2507 #ifdef CONFIG_COMPAT
2508         .compat_ioctl   = vsock_dev_compat_ioctl,
2509 #endif
2510         .open           = nonseekable_open,
2511 };
2512
2513 static struct miscdevice vsock_device = {
2514         .name           = "vsock",
2515         .fops           = &vsock_device_ops,
2516 };
2517
2518 static int __init vsock_init(void)
2519 {
2520         int err = 0;
2521
2522         vsock_init_tables();
2523
2524         vsock_proto.owner = THIS_MODULE;
2525         vsock_device.minor = MISC_DYNAMIC_MINOR;
2526         err = misc_register(&vsock_device);
2527         if (err) {
2528                 pr_err("Failed to register misc device\n");
2529                 goto err_reset_transport;
2530         }
2531
2532         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2533         if (err) {
2534                 pr_err("Cannot register vsock protocol\n");
2535                 goto err_deregister_misc;
2536         }
2537
2538         err = sock_register(&vsock_family_ops);
2539         if (err) {
2540                 pr_err("could not register af_vsock (%d) address family: %d\n",
2541                        AF_VSOCK, err);
2542                 goto err_unregister_proto;
2543         }
2544
2545         vsock_bpf_build_proto();
2546
2547         return 0;
2548
2549 err_unregister_proto:
2550         proto_unregister(&vsock_proto);
2551 err_deregister_misc:
2552         misc_deregister(&vsock_device);
2553 err_reset_transport:
2554         return err;
2555 }
2556
2557 static void __exit vsock_exit(void)
2558 {
2559         misc_deregister(&vsock_device);
2560         sock_unregister(AF_VSOCK);
2561         proto_unregister(&vsock_proto);
2562 }
2563
2564 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2565 {
2566         return vsk->transport;
2567 }
2568 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2569
2570 int vsock_core_register(const struct vsock_transport *t, int features)
2571 {
2572         const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2573         int err = mutex_lock_interruptible(&vsock_register_mutex);
2574
2575         if (err)
2576                 return err;
2577
2578         t_h2g = transport_h2g;
2579         t_g2h = transport_g2h;
2580         t_dgram = transport_dgram;
2581         t_local = transport_local;
2582
2583         if (features & VSOCK_TRANSPORT_F_H2G) {
2584                 if (t_h2g) {
2585                         err = -EBUSY;
2586                         goto err_busy;
2587                 }
2588                 t_h2g = t;
2589         }
2590
2591         if (features & VSOCK_TRANSPORT_F_G2H) {
2592                 if (t_g2h) {
2593                         err = -EBUSY;
2594                         goto err_busy;
2595                 }
2596                 t_g2h = t;
2597         }
2598
2599         if (features & VSOCK_TRANSPORT_F_DGRAM) {
2600                 if (t_dgram) {
2601                         err = -EBUSY;
2602                         goto err_busy;
2603                 }
2604                 t_dgram = t;
2605         }
2606
2607         if (features & VSOCK_TRANSPORT_F_LOCAL) {
2608                 if (t_local) {
2609                         err = -EBUSY;
2610                         goto err_busy;
2611                 }
2612                 t_local = t;
2613         }
2614
2615         transport_h2g = t_h2g;
2616         transport_g2h = t_g2h;
2617         transport_dgram = t_dgram;
2618         transport_local = t_local;
2619
2620 err_busy:
2621         mutex_unlock(&vsock_register_mutex);
2622         return err;
2623 }
2624 EXPORT_SYMBOL_GPL(vsock_core_register);
2625
2626 void vsock_core_unregister(const struct vsock_transport *t)
2627 {
2628         mutex_lock(&vsock_register_mutex);
2629
2630         if (transport_h2g == t)
2631                 transport_h2g = NULL;
2632
2633         if (transport_g2h == t)
2634                 transport_g2h = NULL;
2635
2636         if (transport_dgram == t)
2637                 transport_dgram = NULL;
2638
2639         if (transport_local == t)
2640                 transport_local = NULL;
2641
2642         mutex_unlock(&vsock_register_mutex);
2643 }
2644 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2645
2646 module_init(vsock_init);
2647 module_exit(vsock_exit);
2648
2649 MODULE_AUTHOR("VMware, Inc.");
2650 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2651 MODULE_VERSION("1.0.2.0-k");
2652 MODULE_LICENSE("GPL v2");