1/* 2 * VMware vSockets Driver 3 * 4 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License as published by the Free 8 * Software Foundation version 2 and no later version. 9 * 10 * This program is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 */ 15 16/* Implementation notes: 17 * 18 * - There are two kinds of sockets: those created by user action (such as 19 * calling socket(2)) and those created by incoming connection request packets. 20 * 21 * - There are two "global" tables, one for bound sockets (sockets that have 22 * specified an address that they are responsible for) and one for connected 23 * sockets (sockets that have established a connection with another socket). 24 * These tables are "global" in that all sockets on the system are placed 25 * within them. - Note, though, that the bound table contains an extra entry 26 * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in 27 * that list. The bound table is used solely for lookup of sockets when packets 28 * are received and that's not necessary for SOCK_DGRAM sockets since we create 29 * a datagram handle for each and need not perform a lookup. Keeping SOCK_DGRAM 30 * sockets out of the bound hash buckets will reduce the chance of collisions 31 * when looking for SOCK_STREAM sockets and prevents us from having to check the 32 * socket type in the hash table lookups. 33 * 34 * - Sockets created by user action will either be "client" sockets that 35 * initiate a connection or "server" sockets that listen for connections; we do 36 * not support simultaneous connects (two "client" sockets connecting). 37 * 38 * - "Server" sockets are referred to as listener sockets throughout this 39 * implementation because they are in the SS_LISTEN state. When a connection 40 * request is received (the second kind of socket mentioned above), we create a 41 * new socket and refer to it as a pending socket. These pending sockets are 42 * placed on the pending connection list of the listener socket. When future 43 * packets are received for the address the listener socket is bound to, we 44 * check if the source of the packet is from one that has an existing pending 45 * connection. If it does, we process the packet for the pending socket. When 46 * that socket reaches the connected state, it is removed from the listener 47 * socket's pending list and enqueued in the listener socket's accept queue. 48 * Callers of accept(2) will accept connected sockets from the listener socket's 49 * accept queue. If the socket cannot be accepted for some reason then it is 50 * marked rejected. Once the connection is accepted, it is owned by the user 51 * process and the responsibility for cleanup falls with that user process. 52 * 53 * - It is possible that these pending sockets will never reach the connected 54 * state; in fact, we may never receive another packet after the connection 55 * request. Because of this, we must schedule a cleanup function to run in the 56 * future, after some amount of time passes where a connection should have been 57 * established. This function ensures that the socket is off all lists so it 58 * cannot be retrieved, then drops all references to the socket so it is cleaned 59 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this 60 * function will also cleanup rejected sockets, those that reach the connected 61 * state but leave it before they have been accepted. 62 * 63 * - Sockets created by user action will be cleaned up when the user process 64 * calls close(2), causing our release implementation to be called. Our release 65 * implementation will perform some cleanup then drop the last reference so our 66 * sk_destruct implementation is invoked. Our sk_destruct implementation will 67 * perform additional cleanup that's common for both types of sockets. 68 * 69 * - A socket's reference count is what ensures that the structure won't be 70 * freed. Each entry in a list (such as the "global" bound and connected tables 71 * and the listener socket's pending list and connected queue) ensures a 72 * reference. When we defer work until process context and pass a socket as our 73 * argument, we must ensure the reference count is increased to ensure the 74 * socket isn't freed before the function is run; the deferred function will 75 * then drop the reference. 76 */ 77 78#include <linux/types.h> 79#include <linux/bitops.h> 80#include <linux/cred.h> 81#include <linux/init.h> 82#include <linux/io.h> 83#include <linux/kernel.h> 84#include <linux/kmod.h> 85#include <linux/list.h> 86#include <linux/miscdevice.h> 87#include <linux/module.h> 88#include <linux/mutex.h> 89#include <linux/net.h> 90#include <linux/poll.h> 91#include <linux/skbuff.h> 92#include <linux/smp.h> 93#include <linux/socket.h> 94#include <linux/stddef.h> 95#include <linux/unistd.h> 96#include <linux/wait.h> 97#include <linux/workqueue.h> 98#include <net/sock.h> 99#include <net/af_vsock.h> 100 101static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr); 102static void vsock_sk_destruct(struct sock *sk); 103static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 104 105/* Protocol family. */ 106static struct proto vsock_proto = { 107 .name = "AF_VSOCK", 108 .owner = THIS_MODULE, 109 .obj_size = sizeof(struct vsock_sock), 110}; 111 112/* The default peer timeout indicates how long we will wait for a peer response 113 * to a control message. 114 */ 115#define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ) 116 117#define SS_LISTEN 255 118 119static const struct vsock_transport *transport; 120static DEFINE_MUTEX(vsock_register_mutex); 121 122/**** EXPORTS ****/ 123 124/* Get the ID of the local context. This is transport dependent. */ 125 126int vm_sockets_get_local_cid(void) 127{ 128 return transport->get_local_cid(); 129} 130EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid); 131 132/**** UTILS ****/ 133 134/* Each bound VSocket is stored in the bind hash table and each connected 135 * VSocket is stored in the connected hash table. 136 * 137 * Unbound sockets are all put on the same list attached to the end of the hash 138 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in 139 * the bucket that their local address hashes to (vsock_bound_sockets(addr) 140 * represents the list that addr hashes to). 141 * 142 * Specifically, we initialize the vsock_bind_table array to a size of 143 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through 144 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and 145 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function 146 * mods with VSOCK_HASH_SIZE to ensure this. 147 */ 148#define VSOCK_HASH_SIZE 251 149#define MAX_PORT_RETRIES 24 150 151#define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE) 152#define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)]) 153#define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE]) 154 155/* XXX This can probably be implemented in a better way. */ 156#define VSOCK_CONN_HASH(src, dst) \ 157 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE) 158#define vsock_connected_sockets(src, dst) \ 159 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)]) 160#define vsock_connected_sockets_vsk(vsk) \ 161 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr) 162 163static struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1]; 164static struct list_head vsock_connected_table[VSOCK_HASH_SIZE]; 165static DEFINE_SPINLOCK(vsock_table_lock); 166 167/* Autobind this socket to the local address if necessary. */ 168static int vsock_auto_bind(struct vsock_sock *vsk) 169{ 170 struct sock *sk = sk_vsock(vsk); 171 struct sockaddr_vm local_addr; 172 173 if (vsock_addr_bound(&vsk->local_addr)) 174 return 0; 175 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 176 return __vsock_bind(sk, &local_addr); 177} 178 179static void vsock_init_tables(void) 180{ 181 int i; 182 183 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++) 184 INIT_LIST_HEAD(&vsock_bind_table[i]); 185 186 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) 187 INIT_LIST_HEAD(&vsock_connected_table[i]); 188} 189 190static void __vsock_insert_bound(struct list_head *list, 191 struct vsock_sock *vsk) 192{ 193 sock_hold(&vsk->sk); 194 list_add(&vsk->bound_table, list); 195} 196 197static void __vsock_insert_connected(struct list_head *list, 198 struct vsock_sock *vsk) 199{ 200 sock_hold(&vsk->sk); 201 list_add(&vsk->connected_table, list); 202} 203 204static void __vsock_remove_bound(struct vsock_sock *vsk) 205{ 206 list_del_init(&vsk->bound_table); 207 sock_put(&vsk->sk); 208} 209 210static void __vsock_remove_connected(struct vsock_sock *vsk) 211{ 212 list_del_init(&vsk->connected_table); 213 sock_put(&vsk->sk); 214} 215 216static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr) 217{ 218 struct vsock_sock *vsk; 219 220 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) 221 if (addr->svm_port == vsk->local_addr.svm_port) 222 return sk_vsock(vsk); 223 224 return NULL; 225} 226 227static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src, 228 struct sockaddr_vm *dst) 229{ 230 struct vsock_sock *vsk; 231 232 list_for_each_entry(vsk, vsock_connected_sockets(src, dst), 233 connected_table) { 234 if (vsock_addr_equals_addr(src, &vsk->remote_addr) && 235 dst->svm_port == vsk->local_addr.svm_port) { 236 return sk_vsock(vsk); 237 } 238 } 239 240 return NULL; 241} 242 243static bool __vsock_in_bound_table(struct vsock_sock *vsk) 244{ 245 return !list_empty(&vsk->bound_table); 246} 247 248static bool __vsock_in_connected_table(struct vsock_sock *vsk) 249{ 250 return !list_empty(&vsk->connected_table); 251} 252 253static void vsock_insert_unbound(struct vsock_sock *vsk) 254{ 255 spin_lock_bh(&vsock_table_lock); 256 __vsock_insert_bound(vsock_unbound_sockets, vsk); 257 spin_unlock_bh(&vsock_table_lock); 258} 259 260void vsock_insert_connected(struct vsock_sock *vsk) 261{ 262 struct list_head *list = vsock_connected_sockets( 263 &vsk->remote_addr, &vsk->local_addr); 264 265 spin_lock_bh(&vsock_table_lock); 266 __vsock_insert_connected(list, vsk); 267 spin_unlock_bh(&vsock_table_lock); 268} 269EXPORT_SYMBOL_GPL(vsock_insert_connected); 270 271void vsock_remove_bound(struct vsock_sock *vsk) 272{ 273 spin_lock_bh(&vsock_table_lock); 274 __vsock_remove_bound(vsk); 275 spin_unlock_bh(&vsock_table_lock); 276} 277EXPORT_SYMBOL_GPL(vsock_remove_bound); 278 279void vsock_remove_connected(struct vsock_sock *vsk) 280{ 281 spin_lock_bh(&vsock_table_lock); 282 __vsock_remove_connected(vsk); 283 spin_unlock_bh(&vsock_table_lock); 284} 285EXPORT_SYMBOL_GPL(vsock_remove_connected); 286 287struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr) 288{ 289 struct sock *sk; 290 291 spin_lock_bh(&vsock_table_lock); 292 sk = __vsock_find_bound_socket(addr); 293 if (sk) 294 sock_hold(sk); 295 296 spin_unlock_bh(&vsock_table_lock); 297 298 return sk; 299} 300EXPORT_SYMBOL_GPL(vsock_find_bound_socket); 301 302struct sock *vsock_find_connected_socket(struct sockaddr_vm *src, 303 struct sockaddr_vm *dst) 304{ 305 struct sock *sk; 306 307 spin_lock_bh(&vsock_table_lock); 308 sk = __vsock_find_connected_socket(src, dst); 309 if (sk) 310 sock_hold(sk); 311 312 spin_unlock_bh(&vsock_table_lock); 313 314 return sk; 315} 316EXPORT_SYMBOL_GPL(vsock_find_connected_socket); 317 318static bool vsock_in_bound_table(struct vsock_sock *vsk) 319{ 320 bool ret; 321 322 spin_lock_bh(&vsock_table_lock); 323 ret = __vsock_in_bound_table(vsk); 324 spin_unlock_bh(&vsock_table_lock); 325 326 return ret; 327} 328 329static bool vsock_in_connected_table(struct vsock_sock *vsk) 330{ 331 bool ret; 332 333 spin_lock_bh(&vsock_table_lock); 334 ret = __vsock_in_connected_table(vsk); 335 spin_unlock_bh(&vsock_table_lock); 336 337 return ret; 338} 339 340void vsock_for_each_connected_socket(void (*fn)(struct sock *sk)) 341{ 342 int i; 343 344 spin_lock_bh(&vsock_table_lock); 345 346 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) { 347 struct vsock_sock *vsk; 348 list_for_each_entry(vsk, &vsock_connected_table[i], 349 connected_table) 350 fn(sk_vsock(vsk)); 351 } 352 353 spin_unlock_bh(&vsock_table_lock); 354} 355EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket); 356 357void vsock_add_pending(struct sock *listener, struct sock *pending) 358{ 359 struct vsock_sock *vlistener; 360 struct vsock_sock *vpending; 361 362 vlistener = vsock_sk(listener); 363 vpending = vsock_sk(pending); 364 365 sock_hold(pending); 366 sock_hold(listener); 367 list_add_tail(&vpending->pending_links, &vlistener->pending_links); 368} 369EXPORT_SYMBOL_GPL(vsock_add_pending); 370 371void vsock_remove_pending(struct sock *listener, struct sock *pending) 372{ 373 struct vsock_sock *vpending = vsock_sk(pending); 374 375 list_del_init(&vpending->pending_links); 376 sock_put(listener); 377 sock_put(pending); 378} 379EXPORT_SYMBOL_GPL(vsock_remove_pending); 380 381void vsock_enqueue_accept(struct sock *listener, struct sock *connected) 382{ 383 struct vsock_sock *vlistener; 384 struct vsock_sock *vconnected; 385 386 vlistener = vsock_sk(listener); 387 vconnected = vsock_sk(connected); 388 389 sock_hold(connected); 390 sock_hold(listener); 391 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue); 392} 393EXPORT_SYMBOL_GPL(vsock_enqueue_accept); 394 395static struct sock *vsock_dequeue_accept(struct sock *listener) 396{ 397 struct vsock_sock *vlistener; 398 struct vsock_sock *vconnected; 399 400 vlistener = vsock_sk(listener); 401 402 if (list_empty(&vlistener->accept_queue)) 403 return NULL; 404 405 vconnected = list_entry(vlistener->accept_queue.next, 406 struct vsock_sock, accept_queue); 407 408 list_del_init(&vconnected->accept_queue); 409 sock_put(listener); 410 /* The caller will need a reference on the connected socket so we let 411 * it call sock_put(). 412 */ 413 414 return sk_vsock(vconnected); 415} 416 417static bool vsock_is_accept_queue_empty(struct sock *sk) 418{ 419 struct vsock_sock *vsk = vsock_sk(sk); 420 return list_empty(&vsk->accept_queue); 421} 422 423static bool vsock_is_pending(struct sock *sk) 424{ 425 struct vsock_sock *vsk = vsock_sk(sk); 426 return !list_empty(&vsk->pending_links); 427} 428 429static int vsock_send_shutdown(struct sock *sk, int mode) 430{ 431 return transport->shutdown(vsock_sk(sk), mode); 432} 433 434void vsock_pending_work(struct work_struct *work) 435{ 436 struct sock *sk; 437 struct sock *listener; 438 struct vsock_sock *vsk; 439 bool cleanup; 440 441 vsk = container_of(work, struct vsock_sock, dwork.work); 442 sk = sk_vsock(vsk); 443 listener = vsk->listener; 444 cleanup = true; 445 446 lock_sock(listener); 447 lock_sock(sk); 448 449 if (vsock_is_pending(sk)) { 450 vsock_remove_pending(listener, sk); 451 } else if (!vsk->rejected) { 452 /* We are not on the pending list and accept() did not reject 453 * us, so we must have been accepted by our user process. We 454 * just need to drop our references to the sockets and be on 455 * our way. 456 */ 457 cleanup = false; 458 goto out; 459 } 460 461 listener->sk_ack_backlog--; 462 463 /* We need to remove ourself from the global connected sockets list so 464 * incoming packets can't find this socket, and to reduce the reference 465 * count. 466 */ 467 if (vsock_in_connected_table(vsk)) 468 vsock_remove_connected(vsk); 469 470 sk->sk_state = SS_FREE; 471 472out: 473 release_sock(sk); 474 release_sock(listener); 475 if (cleanup) 476 sock_put(sk); 477 478 sock_put(sk); 479 sock_put(listener); 480} 481EXPORT_SYMBOL_GPL(vsock_pending_work); 482 483/**** SOCKET OPERATIONS ****/ 484 485static int __vsock_bind_stream(struct vsock_sock *vsk, 486 struct sockaddr_vm *addr) 487{ 488 static u32 port = LAST_RESERVED_PORT + 1; 489 struct sockaddr_vm new_addr; 490 491 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port); 492 493 if (addr->svm_port == VMADDR_PORT_ANY) { 494 bool found = false; 495 unsigned int i; 496 497 for (i = 0; i < MAX_PORT_RETRIES; i++) { 498 if (port <= LAST_RESERVED_PORT) 499 port = LAST_RESERVED_PORT + 1; 500 501 new_addr.svm_port = port++; 502 503 if (!__vsock_find_bound_socket(&new_addr)) { 504 found = true; 505 break; 506 } 507 } 508 509 if (!found) 510 return -EADDRNOTAVAIL; 511 } else { 512 /* If port is in reserved range, ensure caller 513 * has necessary privileges. 514 */ 515 if (addr->svm_port <= LAST_RESERVED_PORT && 516 !capable(CAP_NET_BIND_SERVICE)) { 517 return -EACCES; 518 } 519 520 if (__vsock_find_bound_socket(&new_addr)) 521 return -EADDRINUSE; 522 } 523 524 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port); 525 526 /* Remove stream sockets from the unbound list and add them to the hash 527 * table for easy lookup by its address. The unbound list is simply an 528 * extra entry at the end of the hash table, a trick used by AF_UNIX. 529 */ 530 __vsock_remove_bound(vsk); 531 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk); 532 533 return 0; 534} 535 536static int __vsock_bind_dgram(struct vsock_sock *vsk, 537 struct sockaddr_vm *addr) 538{ 539 return transport->dgram_bind(vsk, addr); 540} 541 542static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr) 543{ 544 struct vsock_sock *vsk = vsock_sk(sk); 545 u32 cid; 546 int retval; 547 548 /* First ensure this socket isn't already bound. */ 549 if (vsock_addr_bound(&vsk->local_addr)) 550 return -EINVAL; 551 552 /* Now bind to the provided address or select appropriate values if 553 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that 554 * like AF_INET prevents binding to a non-local IP address (in most 555 * cases), we only allow binding to the local CID. 556 */ 557 cid = transport->get_local_cid(); 558 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY) 559 return -EADDRNOTAVAIL; 560 561 switch (sk->sk_socket->type) { 562 case SOCK_STREAM: 563 spin_lock_bh(&vsock_table_lock); 564 retval = __vsock_bind_stream(vsk, addr); 565 spin_unlock_bh(&vsock_table_lock); 566 break; 567 568 case SOCK_DGRAM: 569 retval = __vsock_bind_dgram(vsk, addr); 570 break; 571 572 default: 573 retval = -EINVAL; 574 break; 575 } 576 577 return retval; 578} 579 580struct sock *__vsock_create(struct net *net, 581 struct socket *sock, 582 struct sock *parent, 583 gfp_t priority, 584 unsigned short type) 585{ 586 struct sock *sk; 587 struct vsock_sock *psk; 588 struct vsock_sock *vsk; 589 590 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto); 591 if (!sk) 592 return NULL; 593 594 sock_init_data(sock, sk); 595 596 /* sk->sk_type is normally set in sock_init_data, but only if sock is 597 * non-NULL. We make sure that our sockets always have a type by 598 * setting it here if needed. 599 */ 600 if (!sock) 601 sk->sk_type = type; 602 603 vsk = vsock_sk(sk); 604 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 605 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 606 607 sk->sk_destruct = vsock_sk_destruct; 608 sk->sk_backlog_rcv = vsock_queue_rcv_skb; 609 sk->sk_state = 0; 610 sock_reset_flag(sk, SOCK_DONE); 611 612 INIT_LIST_HEAD(&vsk->bound_table); 613 INIT_LIST_HEAD(&vsk->connected_table); 614 vsk->listener = NULL; 615 INIT_LIST_HEAD(&vsk->pending_links); 616 INIT_LIST_HEAD(&vsk->accept_queue); 617 vsk->rejected = false; 618 vsk->sent_request = false; 619 vsk->ignore_connecting_rst = false; 620 vsk->peer_shutdown = 0; 621 622 psk = parent ? vsock_sk(parent) : NULL; 623 if (parent) { 624 vsk->trusted = psk->trusted; 625 vsk->owner = get_cred(psk->owner); 626 vsk->connect_timeout = psk->connect_timeout; 627 } else { 628 vsk->trusted = capable(CAP_NET_ADMIN); 629 vsk->owner = get_current_cred(); 630 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT; 631 } 632 633 if (transport->init(vsk, psk) < 0) { 634 sk_free(sk); 635 return NULL; 636 } 637 638 if (sock) 639 vsock_insert_unbound(vsk); 640 641 return sk; 642} 643EXPORT_SYMBOL_GPL(__vsock_create); 644 645static void __vsock_release(struct sock *sk) 646{ 647 if (sk) { 648 struct sk_buff *skb; 649 struct sock *pending; 650 struct vsock_sock *vsk; 651 652 vsk = vsock_sk(sk); 653 pending = NULL; /* Compiler warning. */ 654 655 if (vsock_in_bound_table(vsk)) 656 vsock_remove_bound(vsk); 657 658 if (vsock_in_connected_table(vsk)) 659 vsock_remove_connected(vsk); 660 661 transport->release(vsk); 662 663 lock_sock(sk); 664 sock_orphan(sk); 665 sk->sk_shutdown = SHUTDOWN_MASK; 666 667 while ((skb = skb_dequeue(&sk->sk_receive_queue))) 668 kfree_skb(skb); 669 670 /* Clean up any sockets that never were accepted. */ 671 while ((pending = vsock_dequeue_accept(sk)) != NULL) { 672 __vsock_release(pending); 673 sock_put(pending); 674 } 675 676 release_sock(sk); 677 sock_put(sk); 678 } 679} 680 681static void vsock_sk_destruct(struct sock *sk) 682{ 683 struct vsock_sock *vsk = vsock_sk(sk); 684 685 transport->destruct(vsk); 686 687 /* When clearing these addresses, there's no need to set the family and 688 * possibly register the address family with the kernel. 689 */ 690 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 691 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 692 693 put_cred(vsk->owner); 694} 695 696static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 697{ 698 int err; 699 700 err = sock_queue_rcv_skb(sk, skb); 701 if (err) 702 kfree_skb(skb); 703 704 return err; 705} 706 707s64 vsock_stream_has_data(struct vsock_sock *vsk) 708{ 709 return transport->stream_has_data(vsk); 710} 711EXPORT_SYMBOL_GPL(vsock_stream_has_data); 712 713s64 vsock_stream_has_space(struct vsock_sock *vsk) 714{ 715 return transport->stream_has_space(vsk); 716} 717EXPORT_SYMBOL_GPL(vsock_stream_has_space); 718 719static int vsock_release(struct socket *sock) 720{ 721 __vsock_release(sock->sk); 722 sock->sk = NULL; 723 sock->state = SS_FREE; 724 725 return 0; 726} 727 728static int 729vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 730{ 731 int err; 732 struct sock *sk; 733 struct sockaddr_vm *vm_addr; 734 735 sk = sock->sk; 736 737 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0) 738 return -EINVAL; 739 740 lock_sock(sk); 741 err = __vsock_bind(sk, vm_addr); 742 release_sock(sk); 743 744 return err; 745} 746 747static int vsock_getname(struct socket *sock, 748 struct sockaddr *addr, int *addr_len, int peer) 749{ 750 int err; 751 struct sock *sk; 752 struct vsock_sock *vsk; 753 struct sockaddr_vm *vm_addr; 754 755 sk = sock->sk; 756 vsk = vsock_sk(sk); 757 err = 0; 758 759 lock_sock(sk); 760 761 if (peer) { 762 if (sock->state != SS_CONNECTED) { 763 err = -ENOTCONN; 764 goto out; 765 } 766 vm_addr = &vsk->remote_addr; 767 } else { 768 vm_addr = &vsk->local_addr; 769 } 770 771 if (!vm_addr) { 772 err = -EINVAL; 773 goto out; 774 } 775 776 /* sys_getsockname() and sys_getpeername() pass us a 777 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately 778 * that macro is defined in socket.c instead of .h, so we hardcode its 779 * value here. 780 */ 781 BUILD_BUG_ON(sizeof(*vm_addr) > 128); 782 memcpy(addr, vm_addr, sizeof(*vm_addr)); 783 *addr_len = sizeof(*vm_addr); 784 785out: 786 release_sock(sk); 787 return err; 788} 789 790static int vsock_shutdown(struct socket *sock, int mode) 791{ 792 int err; 793 struct sock *sk; 794 795 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses 796 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode 797 * here like the other address families do. Note also that the 798 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3), 799 * which is what we want. 800 */ 801 mode++; 802 803 if ((mode & ~SHUTDOWN_MASK) || !mode) 804 return -EINVAL; 805 806 /* If this is a STREAM socket and it is not connected then bail out 807 * immediately. If it is a DGRAM socket then we must first kick the 808 * socket so that it wakes up from any sleeping calls, for example 809 * recv(), and then afterwards return the error. 810 */ 811 812 sk = sock->sk; 813 if (sock->state == SS_UNCONNECTED) { 814 err = -ENOTCONN; 815 if (sk->sk_type == SOCK_STREAM) 816 return err; 817 } else { 818 sock->state = SS_DISCONNECTING; 819 err = 0; 820 } 821 822 /* Receive and send shutdowns are treated alike. */ 823 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN); 824 if (mode) { 825 lock_sock(sk); 826 sk->sk_shutdown |= mode; 827 sk->sk_state_change(sk); 828 release_sock(sk); 829 830 if (sk->sk_type == SOCK_STREAM) { 831 sock_reset_flag(sk, SOCK_DONE); 832 vsock_send_shutdown(sk, mode); 833 } 834 } 835 836 return err; 837} 838 839static unsigned int vsock_poll(struct file *file, struct socket *sock, 840 poll_table *wait) 841{ 842 struct sock *sk; 843 unsigned int mask; 844 struct vsock_sock *vsk; 845 846 sk = sock->sk; 847 vsk = vsock_sk(sk); 848 849 poll_wait(file, sk_sleep(sk), wait); 850 mask = 0; 851 852 if (sk->sk_err) 853 /* Signify that there has been an error on this socket. */ 854 mask |= POLLERR; 855 856 /* INET sockets treat local write shutdown and peer write shutdown as a 857 * case of POLLHUP set. 858 */ 859 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 860 ((sk->sk_shutdown & SEND_SHUTDOWN) && 861 (vsk->peer_shutdown & SEND_SHUTDOWN))) { 862 mask |= POLLHUP; 863 } 864 865 if (sk->sk_shutdown & RCV_SHUTDOWN || 866 vsk->peer_shutdown & SEND_SHUTDOWN) { 867 mask |= POLLRDHUP; 868 } 869 870 if (sock->type == SOCK_DGRAM) { 871 /* For datagram sockets we can read if there is something in 872 * the queue and write as long as the socket isn't shutdown for 873 * sending. 874 */ 875 if (!skb_queue_empty(&sk->sk_receive_queue) || 876 (sk->sk_shutdown & RCV_SHUTDOWN)) { 877 mask |= POLLIN | POLLRDNORM; 878 } 879 880 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 881 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 882 883 } else if (sock->type == SOCK_STREAM) { 884 lock_sock(sk); 885 886 /* Listening sockets that have connections in their accept 887 * queue can be read. 888 */ 889 if (sk->sk_state == SS_LISTEN 890 && !vsock_is_accept_queue_empty(sk)) 891 mask |= POLLIN | POLLRDNORM; 892 893 /* If there is something in the queue then we can read. */ 894 if (transport->stream_is_active(vsk) && 895 !(sk->sk_shutdown & RCV_SHUTDOWN)) { 896 bool data_ready_now = false; 897 int ret = transport->notify_poll_in( 898 vsk, 1, &data_ready_now); 899 if (ret < 0) { 900 mask |= POLLERR; 901 } else { 902 if (data_ready_now) 903 mask |= POLLIN | POLLRDNORM; 904 905 } 906 } 907 908 /* Sockets whose connections have been closed, reset, or 909 * terminated should also be considered read, and we check the 910 * shutdown flag for that. 911 */ 912 if (sk->sk_shutdown & RCV_SHUTDOWN || 913 vsk->peer_shutdown & SEND_SHUTDOWN) { 914 mask |= POLLIN | POLLRDNORM; 915 } 916 917 /* Connected sockets that can produce data can be written. */ 918 if (sk->sk_state == SS_CONNECTED) { 919 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 920 bool space_avail_now = false; 921 int ret = transport->notify_poll_out( 922 vsk, 1, &space_avail_now); 923 if (ret < 0) { 924 mask |= POLLERR; 925 } else { 926 if (space_avail_now) 927 /* Remove POLLWRBAND since INET 928 * sockets are not setting it. 929 */ 930 mask |= POLLOUT | POLLWRNORM; 931 932 } 933 } 934 } 935 936 /* Simulate INET socket poll behaviors, which sets 937 * POLLOUT|POLLWRNORM when peer is closed and nothing to read, 938 * but local send is not shutdown. 939 */ 940 if (sk->sk_state == SS_UNCONNECTED) { 941 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 942 mask |= POLLOUT | POLLWRNORM; 943 944 } 945 946 release_sock(sk); 947 } 948 949 return mask; 950} 951 952static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg, 953 size_t len) 954{ 955 int err; 956 struct sock *sk; 957 struct vsock_sock *vsk; 958 struct sockaddr_vm *remote_addr; 959 960 if (msg->msg_flags & MSG_OOB) 961 return -EOPNOTSUPP; 962 963 /* For now, MSG_DONTWAIT is always assumed... */ 964 err = 0; 965 sk = sock->sk; 966 vsk = vsock_sk(sk); 967 968 lock_sock(sk); 969 970 err = vsock_auto_bind(vsk); 971 if (err) 972 goto out; 973 974 975 /* If the provided message contains an address, use that. Otherwise 976 * fall back on the socket's remote handle (if it has been connected). 977 */ 978 if (msg->msg_name && 979 vsock_addr_cast(msg->msg_name, msg->msg_namelen, 980 &remote_addr) == 0) { 981 /* Ensure this address is of the right type and is a valid 982 * destination. 983 */ 984 985 if (remote_addr->svm_cid == VMADDR_CID_ANY) 986 remote_addr->svm_cid = transport->get_local_cid(); 987 988 if (!vsock_addr_bound(remote_addr)) { 989 err = -EINVAL; 990 goto out; 991 } 992 } else if (sock->state == SS_CONNECTED) { 993 remote_addr = &vsk->remote_addr; 994 995 if (remote_addr->svm_cid == VMADDR_CID_ANY) 996 remote_addr->svm_cid = transport->get_local_cid(); 997 998 /* XXX Should connect() or this function ensure remote_addr is 999 * bound? 1000 */ 1001 if (!vsock_addr_bound(&vsk->remote_addr)) { 1002 err = -EINVAL; 1003 goto out; 1004 } 1005 } else { 1006 err = -EINVAL; 1007 goto out; 1008 } 1009 1010 if (!transport->dgram_allow(remote_addr->svm_cid, 1011 remote_addr->svm_port)) { 1012 err = -EINVAL; 1013 goto out; 1014 } 1015 1016 err = transport->dgram_enqueue(vsk, remote_addr, msg, len); 1017 1018out: 1019 release_sock(sk); 1020 return err; 1021} 1022 1023static int vsock_dgram_connect(struct socket *sock, 1024 struct sockaddr *addr, int addr_len, int flags) 1025{ 1026 int err; 1027 struct sock *sk; 1028 struct vsock_sock *vsk; 1029 struct sockaddr_vm *remote_addr; 1030 1031 sk = sock->sk; 1032 vsk = vsock_sk(sk); 1033 1034 err = vsock_addr_cast(addr, addr_len, &remote_addr); 1035 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) { 1036 lock_sock(sk); 1037 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, 1038 VMADDR_PORT_ANY); 1039 sock->state = SS_UNCONNECTED; 1040 release_sock(sk); 1041 return 0; 1042 } else if (err != 0) 1043 return -EINVAL; 1044 1045 lock_sock(sk); 1046 1047 err = vsock_auto_bind(vsk); 1048 if (err) 1049 goto out; 1050 1051 if (!transport->dgram_allow(remote_addr->svm_cid, 1052 remote_addr->svm_port)) { 1053 err = -EINVAL; 1054 goto out; 1055 } 1056 1057 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr)); 1058 sock->state = SS_CONNECTED; 1059 1060out: 1061 release_sock(sk); 1062 return err; 1063} 1064 1065static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg, 1066 size_t len, int flags) 1067{ 1068 return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags); 1069} 1070 1071static const struct proto_ops vsock_dgram_ops = { 1072 .family = PF_VSOCK, 1073 .owner = THIS_MODULE, 1074 .release = vsock_release, 1075 .bind = vsock_bind, 1076 .connect = vsock_dgram_connect, 1077 .socketpair = sock_no_socketpair, 1078 .accept = sock_no_accept, 1079 .getname = vsock_getname, 1080 .poll = vsock_poll, 1081 .ioctl = sock_no_ioctl, 1082 .listen = sock_no_listen, 1083 .shutdown = vsock_shutdown, 1084 .setsockopt = sock_no_setsockopt, 1085 .getsockopt = sock_no_getsockopt, 1086 .sendmsg = vsock_dgram_sendmsg, 1087 .recvmsg = vsock_dgram_recvmsg, 1088 .mmap = sock_no_mmap, 1089 .sendpage = sock_no_sendpage, 1090}; 1091 1092static void vsock_connect_timeout(struct work_struct *work) 1093{ 1094 struct sock *sk; 1095 struct vsock_sock *vsk; 1096 1097 vsk = container_of(work, struct vsock_sock, dwork.work); 1098 sk = sk_vsock(vsk); 1099 1100 lock_sock(sk); 1101 if (sk->sk_state == SS_CONNECTING && 1102 (sk->sk_shutdown != SHUTDOWN_MASK)) { 1103 sk->sk_state = SS_UNCONNECTED; 1104 sk->sk_err = ETIMEDOUT; 1105 sk->sk_error_report(sk); 1106 } 1107 release_sock(sk); 1108 1109 sock_put(sk); 1110} 1111 1112static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr, 1113 int addr_len, int flags) 1114{ 1115 int err; 1116 struct sock *sk; 1117 struct vsock_sock *vsk; 1118 struct sockaddr_vm *remote_addr; 1119 long timeout; 1120 DEFINE_WAIT(wait); 1121 1122 err = 0; 1123 sk = sock->sk; 1124 vsk = vsock_sk(sk); 1125 1126 lock_sock(sk); 1127 1128 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */ 1129 switch (sock->state) { 1130 case SS_CONNECTED: 1131 err = -EISCONN; 1132 goto out; 1133 case SS_DISCONNECTING: 1134 err = -EINVAL; 1135 goto out; 1136 case SS_CONNECTING: 1137 /* This continues on so we can move sock into the SS_CONNECTED 1138 * state once the connection has completed (at which point err 1139 * will be set to zero also). Otherwise, we will either wait 1140 * for the connection or return -EALREADY should this be a 1141 * non-blocking call. 1142 */ 1143 err = -EALREADY; 1144 break; 1145 default: 1146 if ((sk->sk_state == SS_LISTEN) || 1147 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) { 1148 err = -EINVAL; 1149 goto out; 1150 } 1151 1152 /* The hypervisor and well-known contexts do not have socket 1153 * endpoints. 1154 */ 1155 if (!transport->stream_allow(remote_addr->svm_cid, 1156 remote_addr->svm_port)) { 1157 err = -ENETUNREACH; 1158 goto out; 1159 } 1160 1161 /* Set the remote address that we are connecting to. */ 1162 memcpy(&vsk->remote_addr, remote_addr, 1163 sizeof(vsk->remote_addr)); 1164 1165 err = vsock_auto_bind(vsk); 1166 if (err) 1167 goto out; 1168 1169 sk->sk_state = SS_CONNECTING; 1170 1171 err = transport->connect(vsk); 1172 if (err < 0) 1173 goto out; 1174 1175 /* Mark sock as connecting and set the error code to in 1176 * progress in case this is a non-blocking connect. 1177 */ 1178 sock->state = SS_CONNECTING; 1179 err = -EINPROGRESS; 1180 } 1181 1182 /* The receive path will handle all communication until we are able to 1183 * enter the connected state. Here we wait for the connection to be 1184 * completed or a notification of an error. 1185 */ 1186 timeout = vsk->connect_timeout; 1187 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1188 1189 while (sk->sk_state != SS_CONNECTED && sk->sk_err == 0) { 1190 if (flags & O_NONBLOCK) { 1191 /* If we're not going to block, we schedule a timeout 1192 * function to generate a timeout on the connection 1193 * attempt, in case the peer doesn't respond in a 1194 * timely manner. We hold on to the socket until the 1195 * timeout fires. 1196 */ 1197 sock_hold(sk); 1198 INIT_DELAYED_WORK(&vsk->dwork, 1199 vsock_connect_timeout); 1200 schedule_delayed_work(&vsk->dwork, timeout); 1201 1202 /* Skip ahead to preserve error code set above. */ 1203 goto out_wait; 1204 } 1205 1206 release_sock(sk); 1207 timeout = schedule_timeout(timeout); 1208 lock_sock(sk); 1209 1210 if (signal_pending(current)) { 1211 err = sock_intr_errno(timeout); 1212 goto out_wait_error; 1213 } else if (timeout == 0) { 1214 err = -ETIMEDOUT; 1215 goto out_wait_error; 1216 } 1217 1218 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1219 } 1220 1221 if (sk->sk_err) { 1222 err = -sk->sk_err; 1223 goto out_wait_error; 1224 } else 1225 err = 0; 1226 1227out_wait: 1228 finish_wait(sk_sleep(sk), &wait); 1229out: 1230 release_sock(sk); 1231 return err; 1232 1233out_wait_error: 1234 sk->sk_state = SS_UNCONNECTED; 1235 sock->state = SS_UNCONNECTED; 1236 goto out_wait; 1237} 1238 1239static int vsock_accept(struct socket *sock, struct socket *newsock, int flags) 1240{ 1241 struct sock *listener; 1242 int err; 1243 struct sock *connected; 1244 struct vsock_sock *vconnected; 1245 long timeout; 1246 DEFINE_WAIT(wait); 1247 1248 err = 0; 1249 listener = sock->sk; 1250 1251 lock_sock(listener); 1252 1253 if (sock->type != SOCK_STREAM) { 1254 err = -EOPNOTSUPP; 1255 goto out; 1256 } 1257 1258 if (listener->sk_state != SS_LISTEN) { 1259 err = -EINVAL; 1260 goto out; 1261 } 1262 1263 /* Wait for children sockets to appear; these are the new sockets 1264 * created upon connection establishment. 1265 */ 1266 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK); 1267 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1268 1269 while ((connected = vsock_dequeue_accept(listener)) == NULL && 1270 listener->sk_err == 0) { 1271 release_sock(listener); 1272 timeout = schedule_timeout(timeout); 1273 lock_sock(listener); 1274 1275 if (signal_pending(current)) { 1276 err = sock_intr_errno(timeout); 1277 goto out_wait; 1278 } else if (timeout == 0) { 1279 err = -EAGAIN; 1280 goto out_wait; 1281 } 1282 1283 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1284 } 1285 1286 if (listener->sk_err) 1287 err = -listener->sk_err; 1288 1289 if (connected) { 1290 listener->sk_ack_backlog--; 1291 1292 lock_sock(connected); 1293 vconnected = vsock_sk(connected); 1294 1295 /* If the listener socket has received an error, then we should 1296 * reject this socket and return. Note that we simply mark the 1297 * socket rejected, drop our reference, and let the cleanup 1298 * function handle the cleanup; the fact that we found it in 1299 * the listener's accept queue guarantees that the cleanup 1300 * function hasn't run yet. 1301 */ 1302 if (err) { 1303 vconnected->rejected = true; 1304 release_sock(connected); 1305 sock_put(connected); 1306 goto out_wait; 1307 } 1308 1309 newsock->state = SS_CONNECTED; 1310 sock_graft(connected, newsock); 1311 release_sock(connected); 1312 sock_put(connected); 1313 } 1314 1315out_wait: 1316 finish_wait(sk_sleep(listener), &wait); 1317out: 1318 release_sock(listener); 1319 return err; 1320} 1321 1322static int vsock_listen(struct socket *sock, int backlog) 1323{ 1324 int err; 1325 struct sock *sk; 1326 struct vsock_sock *vsk; 1327 1328 sk = sock->sk; 1329 1330 lock_sock(sk); 1331 1332 if (sock->type != SOCK_STREAM) { 1333 err = -EOPNOTSUPP; 1334 goto out; 1335 } 1336 1337 if (sock->state != SS_UNCONNECTED) { 1338 err = -EINVAL; 1339 goto out; 1340 } 1341 1342 vsk = vsock_sk(sk); 1343 1344 if (!vsock_addr_bound(&vsk->local_addr)) { 1345 err = -EINVAL; 1346 goto out; 1347 } 1348 1349 sk->sk_max_ack_backlog = backlog; 1350 sk->sk_state = SS_LISTEN; 1351 1352 err = 0; 1353 1354out: 1355 release_sock(sk); 1356 return err; 1357} 1358 1359static int vsock_stream_setsockopt(struct socket *sock, 1360 int level, 1361 int optname, 1362 char __user *optval, 1363 unsigned int optlen) 1364{ 1365 int err; 1366 struct sock *sk; 1367 struct vsock_sock *vsk; 1368 u64 val; 1369 1370 if (level != AF_VSOCK) 1371 return -ENOPROTOOPT; 1372 1373#define COPY_IN(_v) \ 1374 do { \ 1375 if (optlen < sizeof(_v)) { \ 1376 err = -EINVAL; \ 1377 goto exit; \ 1378 } \ 1379 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \ 1380 err = -EFAULT; \ 1381 goto exit; \ 1382 } \ 1383 } while (0) 1384 1385 err = 0; 1386 sk = sock->sk; 1387 vsk = vsock_sk(sk); 1388 1389 lock_sock(sk); 1390 1391 switch (optname) { 1392 case SO_VM_SOCKETS_BUFFER_SIZE: 1393 COPY_IN(val); 1394 transport->set_buffer_size(vsk, val); 1395 break; 1396 1397 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1398 COPY_IN(val); 1399 transport->set_max_buffer_size(vsk, val); 1400 break; 1401 1402 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1403 COPY_IN(val); 1404 transport->set_min_buffer_size(vsk, val); 1405 break; 1406 1407 case SO_VM_SOCKETS_CONNECT_TIMEOUT: { 1408 struct timeval tv; 1409 COPY_IN(tv); 1410 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC && 1411 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) { 1412 vsk->connect_timeout = tv.tv_sec * HZ + 1413 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ)); 1414 if (vsk->connect_timeout == 0) 1415 vsk->connect_timeout = 1416 VSOCK_DEFAULT_CONNECT_TIMEOUT; 1417 1418 } else { 1419 err = -ERANGE; 1420 } 1421 break; 1422 } 1423 1424 default: 1425 err = -ENOPROTOOPT; 1426 break; 1427 } 1428 1429#undef COPY_IN 1430 1431exit: 1432 release_sock(sk); 1433 return err; 1434} 1435 1436static int vsock_stream_getsockopt(struct socket *sock, 1437 int level, int optname, 1438 char __user *optval, 1439 int __user *optlen) 1440{ 1441 int err; 1442 int len; 1443 struct sock *sk; 1444 struct vsock_sock *vsk; 1445 u64 val; 1446 1447 if (level != AF_VSOCK) 1448 return -ENOPROTOOPT; 1449 1450 err = get_user(len, optlen); 1451 if (err != 0) 1452 return err; 1453 1454#define COPY_OUT(_v) \ 1455 do { \ 1456 if (len < sizeof(_v)) \ 1457 return -EINVAL; \ 1458 \ 1459 len = sizeof(_v); \ 1460 if (copy_to_user(optval, &_v, len) != 0) \ 1461 return -EFAULT; \ 1462 \ 1463 } while (0) 1464 1465 err = 0; 1466 sk = sock->sk; 1467 vsk = vsock_sk(sk); 1468 1469 switch (optname) { 1470 case SO_VM_SOCKETS_BUFFER_SIZE: 1471 val = transport->get_buffer_size(vsk); 1472 COPY_OUT(val); 1473 break; 1474 1475 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1476 val = transport->get_max_buffer_size(vsk); 1477 COPY_OUT(val); 1478 break; 1479 1480 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1481 val = transport->get_min_buffer_size(vsk); 1482 COPY_OUT(val); 1483 break; 1484 1485 case SO_VM_SOCKETS_CONNECT_TIMEOUT: { 1486 struct timeval tv; 1487 tv.tv_sec = vsk->connect_timeout / HZ; 1488 tv.tv_usec = 1489 (vsk->connect_timeout - 1490 tv.tv_sec * HZ) * (1000000 / HZ); 1491 COPY_OUT(tv); 1492 break; 1493 } 1494 default: 1495 return -ENOPROTOOPT; 1496 } 1497 1498 err = put_user(len, optlen); 1499 if (err != 0) 1500 return -EFAULT; 1501 1502#undef COPY_OUT 1503 1504 return 0; 1505} 1506 1507static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg, 1508 size_t len) 1509{ 1510 struct sock *sk; 1511 struct vsock_sock *vsk; 1512 ssize_t total_written; 1513 long timeout; 1514 int err; 1515 struct vsock_transport_send_notify_data send_data; 1516 1517 DEFINE_WAIT(wait); 1518 1519 sk = sock->sk; 1520 vsk = vsock_sk(sk); 1521 total_written = 0; 1522 err = 0; 1523 1524 if (msg->msg_flags & MSG_OOB) 1525 return -EOPNOTSUPP; 1526 1527 lock_sock(sk); 1528 1529 /* Callers should not provide a destination with stream sockets. */ 1530 if (msg->msg_namelen) { 1531 err = sk->sk_state == SS_CONNECTED ? -EISCONN : -EOPNOTSUPP; 1532 goto out; 1533 } 1534 1535 /* Send data only if both sides are not shutdown in the direction. */ 1536 if (sk->sk_shutdown & SEND_SHUTDOWN || 1537 vsk->peer_shutdown & RCV_SHUTDOWN) { 1538 err = -EPIPE; 1539 goto out; 1540 } 1541 1542 if (sk->sk_state != SS_CONNECTED || 1543 !vsock_addr_bound(&vsk->local_addr)) { 1544 err = -ENOTCONN; 1545 goto out; 1546 } 1547 1548 if (!vsock_addr_bound(&vsk->remote_addr)) { 1549 err = -EDESTADDRREQ; 1550 goto out; 1551 } 1552 1553 /* Wait for room in the produce queue to enqueue our user's data. */ 1554 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1555 1556 err = transport->notify_send_init(vsk, &send_data); 1557 if (err < 0) 1558 goto out; 1559 1560 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1561 1562 while (total_written < len) { 1563 ssize_t written; 1564 1565 while (vsock_stream_has_space(vsk) == 0 && 1566 sk->sk_err == 0 && 1567 !(sk->sk_shutdown & SEND_SHUTDOWN) && 1568 !(vsk->peer_shutdown & RCV_SHUTDOWN)) { 1569 1570 /* Don't wait for non-blocking sockets. */ 1571 if (timeout == 0) { 1572 err = -EAGAIN; 1573 goto out_wait; 1574 } 1575 1576 err = transport->notify_send_pre_block(vsk, &send_data); 1577 if (err < 0) 1578 goto out_wait; 1579 1580 release_sock(sk); 1581 timeout = schedule_timeout(timeout); 1582 lock_sock(sk); 1583 if (signal_pending(current)) { 1584 err = sock_intr_errno(timeout); 1585 goto out_wait; 1586 } else if (timeout == 0) { 1587 err = -EAGAIN; 1588 goto out_wait; 1589 } 1590 1591 prepare_to_wait(sk_sleep(sk), &wait, 1592 TASK_INTERRUPTIBLE); 1593 } 1594 1595 /* These checks occur both as part of and after the loop 1596 * conditional since we need to check before and after 1597 * sleeping. 1598 */ 1599 if (sk->sk_err) { 1600 err = -sk->sk_err; 1601 goto out_wait; 1602 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) || 1603 (vsk->peer_shutdown & RCV_SHUTDOWN)) { 1604 err = -EPIPE; 1605 goto out_wait; 1606 } 1607 1608 err = transport->notify_send_pre_enqueue(vsk, &send_data); 1609 if (err < 0) 1610 goto out_wait; 1611 1612 /* Note that enqueue will only write as many bytes as are free 1613 * in the produce queue, so we don't need to ensure len is 1614 * smaller than the queue size. It is the caller's 1615 * responsibility to check how many bytes we were able to send. 1616 */ 1617 1618 written = transport->stream_enqueue( 1619 vsk, msg, 1620 len - total_written); 1621 if (written < 0) { 1622 err = -ENOMEM; 1623 goto out_wait; 1624 } 1625 1626 total_written += written; 1627 1628 err = transport->notify_send_post_enqueue( 1629 vsk, written, &send_data); 1630 if (err < 0) 1631 goto out_wait; 1632 1633 } 1634 1635out_wait: 1636 if (total_written > 0) 1637 err = total_written; 1638 finish_wait(sk_sleep(sk), &wait); 1639out: 1640 release_sock(sk); 1641 return err; 1642} 1643 1644 1645static int 1646vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1647 int flags) 1648{ 1649 struct sock *sk; 1650 struct vsock_sock *vsk; 1651 int err; 1652 size_t target; 1653 ssize_t copied; 1654 long timeout; 1655 struct vsock_transport_recv_notify_data recv_data; 1656 1657 DEFINE_WAIT(wait); 1658 1659 sk = sock->sk; 1660 vsk = vsock_sk(sk); 1661 err = 0; 1662 1663 lock_sock(sk); 1664 1665 if (sk->sk_state != SS_CONNECTED) { 1666 /* Recvmsg is supposed to return 0 if a peer performs an 1667 * orderly shutdown. Differentiate between that case and when a 1668 * peer has not connected or a local shutdown occured with the 1669 * SOCK_DONE flag. 1670 */ 1671 if (sock_flag(sk, SOCK_DONE)) 1672 err = 0; 1673 else 1674 err = -ENOTCONN; 1675 1676 goto out; 1677 } 1678 1679 if (flags & MSG_OOB) { 1680 err = -EOPNOTSUPP; 1681 goto out; 1682 } 1683 1684 /* We don't check peer_shutdown flag here since peer may actually shut 1685 * down, but there can be data in the queue that a local socket can 1686 * receive. 1687 */ 1688 if (sk->sk_shutdown & RCV_SHUTDOWN) { 1689 err = 0; 1690 goto out; 1691 } 1692 1693 /* It is valid on Linux to pass in a zero-length receive buffer. This 1694 * is not an error. We may as well bail out now. 1695 */ 1696 if (!len) { 1697 err = 0; 1698 goto out; 1699 } 1700 1701 /* We must not copy less than target bytes into the user's buffer 1702 * before returning successfully, so we wait for the consume queue to 1703 * have that much data to consume before dequeueing. Note that this 1704 * makes it impossible to handle cases where target is greater than the 1705 * queue size. 1706 */ 1707 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 1708 if (target >= transport->stream_rcvhiwat(vsk)) { 1709 err = -ENOMEM; 1710 goto out; 1711 } 1712 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1713 copied = 0; 1714 1715 err = transport->notify_recv_init(vsk, target, &recv_data); 1716 if (err < 0) 1717 goto out; 1718 1719 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1720 1721 while (1) { 1722 s64 ready = vsock_stream_has_data(vsk); 1723 1724 if (ready < 0) { 1725 /* Invalid queue pair content. XXX This should be 1726 * changed to a connection reset in a later change. 1727 */ 1728 1729 err = -ENOMEM; 1730 goto out_wait; 1731 } else if (ready > 0) { 1732 ssize_t read; 1733 1734 err = transport->notify_recv_pre_dequeue( 1735 vsk, target, &recv_data); 1736 if (err < 0) 1737 break; 1738 1739 read = transport->stream_dequeue( 1740 vsk, msg, 1741 len - copied, flags); 1742 if (read < 0) { 1743 err = -ENOMEM; 1744 break; 1745 } 1746 1747 copied += read; 1748 1749 err = transport->notify_recv_post_dequeue( 1750 vsk, target, read, 1751 !(flags & MSG_PEEK), &recv_data); 1752 if (err < 0) 1753 goto out_wait; 1754 1755 if (read >= target || flags & MSG_PEEK) 1756 break; 1757 1758 target -= read; 1759 } else { 1760 if (sk->sk_err != 0 || (sk->sk_shutdown & RCV_SHUTDOWN) 1761 || (vsk->peer_shutdown & SEND_SHUTDOWN)) { 1762 break; 1763 } 1764 /* Don't wait for non-blocking sockets. */ 1765 if (timeout == 0) { 1766 err = -EAGAIN; 1767 break; 1768 } 1769 1770 err = transport->notify_recv_pre_block( 1771 vsk, target, &recv_data); 1772 if (err < 0) 1773 break; 1774 1775 release_sock(sk); 1776 timeout = schedule_timeout(timeout); 1777 lock_sock(sk); 1778 1779 if (signal_pending(current)) { 1780 err = sock_intr_errno(timeout); 1781 break; 1782 } else if (timeout == 0) { 1783 err = -EAGAIN; 1784 break; 1785 } 1786 1787 prepare_to_wait(sk_sleep(sk), &wait, 1788 TASK_INTERRUPTIBLE); 1789 } 1790 } 1791 1792 if (sk->sk_err) 1793 err = -sk->sk_err; 1794 else if (sk->sk_shutdown & RCV_SHUTDOWN) 1795 err = 0; 1796 1797 if (copied > 0) { 1798 /* We only do these additional bookkeeping/notification steps 1799 * if we actually copied something out of the queue pair 1800 * instead of just peeking ahead. 1801 */ 1802 1803 if (!(flags & MSG_PEEK)) { 1804 /* If the other side has shutdown for sending and there 1805 * is nothing more to read, then modify the socket 1806 * state. 1807 */ 1808 if (vsk->peer_shutdown & SEND_SHUTDOWN) { 1809 if (vsock_stream_has_data(vsk) <= 0) { 1810 sk->sk_state = SS_UNCONNECTED; 1811 sock_set_flag(sk, SOCK_DONE); 1812 sk->sk_state_change(sk); 1813 } 1814 } 1815 } 1816 err = copied; 1817 } 1818 1819out_wait: 1820 finish_wait(sk_sleep(sk), &wait); 1821out: 1822 release_sock(sk); 1823 return err; 1824} 1825 1826static const struct proto_ops vsock_stream_ops = { 1827 .family = PF_VSOCK, 1828 .owner = THIS_MODULE, 1829 .release = vsock_release, 1830 .bind = vsock_bind, 1831 .connect = vsock_stream_connect, 1832 .socketpair = sock_no_socketpair, 1833 .accept = vsock_accept, 1834 .getname = vsock_getname, 1835 .poll = vsock_poll, 1836 .ioctl = sock_no_ioctl, 1837 .listen = vsock_listen, 1838 .shutdown = vsock_shutdown, 1839 .setsockopt = vsock_stream_setsockopt, 1840 .getsockopt = vsock_stream_getsockopt, 1841 .sendmsg = vsock_stream_sendmsg, 1842 .recvmsg = vsock_stream_recvmsg, 1843 .mmap = sock_no_mmap, 1844 .sendpage = sock_no_sendpage, 1845}; 1846 1847static int vsock_create(struct net *net, struct socket *sock, 1848 int protocol, int kern) 1849{ 1850 if (!sock) 1851 return -EINVAL; 1852 1853 if (protocol && protocol != PF_VSOCK) 1854 return -EPROTONOSUPPORT; 1855 1856 switch (sock->type) { 1857 case SOCK_DGRAM: 1858 sock->ops = &vsock_dgram_ops; 1859 break; 1860 case SOCK_STREAM: 1861 sock->ops = &vsock_stream_ops; 1862 break; 1863 default: 1864 return -ESOCKTNOSUPPORT; 1865 } 1866 1867 sock->state = SS_UNCONNECTED; 1868 1869 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0) ? 0 : -ENOMEM; 1870} 1871 1872static const struct net_proto_family vsock_family_ops = { 1873 .family = AF_VSOCK, 1874 .create = vsock_create, 1875 .owner = THIS_MODULE, 1876}; 1877 1878static long vsock_dev_do_ioctl(struct file *filp, 1879 unsigned int cmd, void __user *ptr) 1880{ 1881 u32 __user *p = ptr; 1882 int retval = 0; 1883 1884 switch (cmd) { 1885 case IOCTL_VM_SOCKETS_GET_LOCAL_CID: 1886 if (put_user(transport->get_local_cid(), p) != 0) 1887 retval = -EFAULT; 1888 break; 1889 1890 default: 1891 pr_err("Unknown ioctl %d\n", cmd); 1892 retval = -EINVAL; 1893 } 1894 1895 return retval; 1896} 1897 1898static long vsock_dev_ioctl(struct file *filp, 1899 unsigned int cmd, unsigned long arg) 1900{ 1901 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg); 1902} 1903 1904#ifdef CONFIG_COMPAT 1905static long vsock_dev_compat_ioctl(struct file *filp, 1906 unsigned int cmd, unsigned long arg) 1907{ 1908 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg)); 1909} 1910#endif 1911 1912static const struct file_operations vsock_device_ops = { 1913 .owner = THIS_MODULE, 1914 .unlocked_ioctl = vsock_dev_ioctl, 1915#ifdef CONFIG_COMPAT 1916 .compat_ioctl = vsock_dev_compat_ioctl, 1917#endif 1918 .open = nonseekable_open, 1919}; 1920 1921static struct miscdevice vsock_device = { 1922 .name = "vsock", 1923 .fops = &vsock_device_ops, 1924}; 1925 1926int __vsock_core_init(const struct vsock_transport *t, struct module *owner) 1927{ 1928 int err = mutex_lock_interruptible(&vsock_register_mutex); 1929 1930 if (err) 1931 return err; 1932 1933 if (transport) { 1934 err = -EBUSY; 1935 goto err_busy; 1936 } 1937 1938 /* Transport must be the owner of the protocol so that it can't 1939 * unload while there are open sockets. 1940 */ 1941 vsock_proto.owner = owner; 1942 transport = t; 1943 1944 vsock_init_tables(); 1945 1946 vsock_device.minor = MISC_DYNAMIC_MINOR; 1947 err = misc_register(&vsock_device); 1948 if (err) { 1949 pr_err("Failed to register misc device\n"); 1950 return -ENOENT; 1951 } 1952 1953 err = proto_register(&vsock_proto, 1); /* we want our slab */ 1954 if (err) { 1955 pr_err("Cannot register vsock protocol\n"); 1956 goto err_misc_deregister; 1957 } 1958 1959 err = sock_register(&vsock_family_ops); 1960 if (err) { 1961 pr_err("could not register af_vsock (%d) address family: %d\n", 1962 AF_VSOCK, err); 1963 goto err_unregister_proto; 1964 } 1965 1966 mutex_unlock(&vsock_register_mutex); 1967 return 0; 1968 1969err_unregister_proto: 1970 proto_unregister(&vsock_proto); 1971err_misc_deregister: 1972 misc_deregister(&vsock_device); 1973 transport = NULL; 1974err_busy: 1975 mutex_unlock(&vsock_register_mutex); 1976 return err; 1977} 1978EXPORT_SYMBOL_GPL(__vsock_core_init); 1979 1980void vsock_core_exit(void) 1981{ 1982 mutex_lock(&vsock_register_mutex); 1983 1984 misc_deregister(&vsock_device); 1985 sock_unregister(AF_VSOCK); 1986 proto_unregister(&vsock_proto); 1987 1988 /* We do not want the assignment below re-ordered. */ 1989 mb(); 1990 transport = NULL; 1991 1992 mutex_unlock(&vsock_register_mutex); 1993} 1994EXPORT_SYMBOL_GPL(vsock_core_exit); 1995 1996MODULE_AUTHOR("VMware, Inc."); 1997MODULE_DESCRIPTION("VMware Virtual Socket Family"); 1998MODULE_VERSION("1.0.1.0-k"); 1999MODULE_LICENSE("GPL v2"); 2000