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 TCP_LISTEN state. When a 40 * connection request is received (the second kind of socket mentioned above), 41 * we create a new socket and refer to it as a pending socket. These pending 42 * sockets are placed on the pending connection list of the listener socket. 43 * When future packets are received for the address the listener socket is 44 * bound to, we check if the source of the packet is from one that has an 45 * existing pending connection. If it does, we process the packet for the 46 * pending socket. When that socket reaches the connected state, it is removed 47 * from the listener socket's pending list and enqueued in the listener 48 * socket's accept queue. Callers of accept(2) will accept connected sockets 49 * from the listener socket's accept queue. If the socket cannot be accepted 50 * for some reason then it is marked rejected. Once the connection is 51 * accepted, it is owned by the user process and the responsibility for cleanup 52 * falls with that user process. 53 * 54 * - It is possible that these pending sockets will never reach the connected 55 * state; in fact, we may never receive another packet after the connection 56 * request. Because of this, we must schedule a cleanup function to run in the 57 * future, after some amount of time passes where a connection should have been 58 * established. This function ensures that the socket is off all lists so it 59 * cannot be retrieved, then drops all references to the socket so it is cleaned 60 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this 61 * function will also cleanup rejected sockets, those that reach the connected 62 * state but leave it before they have been accepted. 63 * 64 * - Lock ordering for pending or accept queue sockets is: 65 * 66 * lock_sock(listener); 67 * lock_sock_nested(pending, SINGLE_DEPTH_NESTING); 68 * 69 * Using explicit nested locking keeps lockdep happy since normally only one 70 * lock of a given class may be taken at a time. 71 * 72 * - Sockets created by user action will be cleaned up when the user process 73 * calls close(2), causing our release implementation to be called. Our release 74 * implementation will perform some cleanup then drop the last reference so our 75 * sk_destruct implementation is invoked. Our sk_destruct implementation will 76 * perform additional cleanup that's common for both types of sockets. 77 * 78 * - A socket's reference count is what ensures that the structure won't be 79 * freed. Each entry in a list (such as the "global" bound and connected tables 80 * and the listener socket's pending list and connected queue) ensures a 81 * reference. When we defer work until process context and pass a socket as our 82 * argument, we must ensure the reference count is increased to ensure the 83 * socket isn't freed before the function is run; the deferred function will 84 * then drop the reference. 85 * 86 * - sk->sk_state uses the TCP state constants because they are widely used by 87 * other address families and exposed to userspace tools like ss(8): 88 * 89 * TCP_CLOSE - unconnected 90 * TCP_SYN_SENT - connecting 91 * TCP_ESTABLISHED - connected 92 * TCP_CLOSING - disconnecting 93 * TCP_LISTEN - listening 94 */ 95 96 #include <linux/types.h> 97 #include <linux/bitops.h> 98 #include <linux/cred.h> 99 #include <linux/init.h> 100 #include <linux/io.h> 101 #include <linux/kernel.h> 102 #include <linux/sched/signal.h> 103 #include <linux/kmod.h> 104 #include <linux/list.h> 105 #include <linux/miscdevice.h> 106 #include <linux/module.h> 107 #include <linux/mutex.h> 108 #include <linux/net.h> 109 #include <linux/poll.h> 110 #include <linux/skbuff.h> 111 #include <linux/smp.h> 112 #include <linux/socket.h> 113 #include <linux/stddef.h> 114 #include <linux/unistd.h> 115 #include <linux/wait.h> 116 #include <linux/workqueue.h> 117 #include <net/sock.h> 118 #include <net/af_vsock.h> 119 120 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr); 121 static void vsock_sk_destruct(struct sock *sk); 122 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 123 124 /* Protocol family. */ 125 static struct proto vsock_proto = { 126 .name = "AF_VSOCK", 127 .owner = THIS_MODULE, 128 .obj_size = sizeof(struct vsock_sock), 129 }; 130 131 /* The default peer timeout indicates how long we will wait for a peer response 132 * to a control message. 133 */ 134 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ) 135 136 static const struct vsock_transport *transport; 137 static DEFINE_MUTEX(vsock_register_mutex); 138 139 /**** EXPORTS ****/ 140 141 /* Get the ID of the local context. This is transport dependent. */ 142 143 int vm_sockets_get_local_cid(void) 144 { 145 return transport->get_local_cid(); 146 } 147 EXPORT_SYMBOL_GPL(vm_sockets_get_local_cid); 148 149 /**** UTILS ****/ 150 151 /* Each bound VSocket is stored in the bind hash table and each connected 152 * VSocket is stored in the connected hash table. 153 * 154 * Unbound sockets are all put on the same list attached to the end of the hash 155 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in 156 * the bucket that their local address hashes to (vsock_bound_sockets(addr) 157 * represents the list that addr hashes to). 158 * 159 * Specifically, we initialize the vsock_bind_table array to a size of 160 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through 161 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and 162 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function 163 * mods with VSOCK_HASH_SIZE to ensure this. 164 */ 165 #define MAX_PORT_RETRIES 24 166 167 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE) 168 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)]) 169 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE]) 170 171 /* XXX This can probably be implemented in a better way. */ 172 #define VSOCK_CONN_HASH(src, dst) \ 173 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE) 174 #define vsock_connected_sockets(src, dst) \ 175 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)]) 176 #define vsock_connected_sockets_vsk(vsk) \ 177 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr) 178 179 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1]; 180 EXPORT_SYMBOL_GPL(vsock_bind_table); 181 struct list_head vsock_connected_table[VSOCK_HASH_SIZE]; 182 EXPORT_SYMBOL_GPL(vsock_connected_table); 183 DEFINE_SPINLOCK(vsock_table_lock); 184 EXPORT_SYMBOL_GPL(vsock_table_lock); 185 186 /* Autobind this socket to the local address if necessary. */ 187 static int vsock_auto_bind(struct vsock_sock *vsk) 188 { 189 struct sock *sk = sk_vsock(vsk); 190 struct sockaddr_vm local_addr; 191 192 if (vsock_addr_bound(&vsk->local_addr)) 193 return 0; 194 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 195 return __vsock_bind(sk, &local_addr); 196 } 197 198 static int __init vsock_init_tables(void) 199 { 200 int i; 201 202 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++) 203 INIT_LIST_HEAD(&vsock_bind_table[i]); 204 205 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) 206 INIT_LIST_HEAD(&vsock_connected_table[i]); 207 return 0; 208 } 209 210 static void __vsock_insert_bound(struct list_head *list, 211 struct vsock_sock *vsk) 212 { 213 sock_hold(&vsk->sk); 214 list_add(&vsk->bound_table, list); 215 } 216 217 static void __vsock_insert_connected(struct list_head *list, 218 struct vsock_sock *vsk) 219 { 220 sock_hold(&vsk->sk); 221 list_add(&vsk->connected_table, list); 222 } 223 224 static void __vsock_remove_bound(struct vsock_sock *vsk) 225 { 226 list_del_init(&vsk->bound_table); 227 sock_put(&vsk->sk); 228 } 229 230 static void __vsock_remove_connected(struct vsock_sock *vsk) 231 { 232 list_del_init(&vsk->connected_table); 233 sock_put(&vsk->sk); 234 } 235 236 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr) 237 { 238 struct vsock_sock *vsk; 239 240 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) 241 if (addr->svm_port == vsk->local_addr.svm_port) 242 return sk_vsock(vsk); 243 244 return NULL; 245 } 246 247 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src, 248 struct sockaddr_vm *dst) 249 { 250 struct vsock_sock *vsk; 251 252 list_for_each_entry(vsk, vsock_connected_sockets(src, dst), 253 connected_table) { 254 if (vsock_addr_equals_addr(src, &vsk->remote_addr) && 255 dst->svm_port == vsk->local_addr.svm_port) { 256 return sk_vsock(vsk); 257 } 258 } 259 260 return NULL; 261 } 262 263 static void vsock_insert_unbound(struct vsock_sock *vsk) 264 { 265 spin_lock_bh(&vsock_table_lock); 266 __vsock_insert_bound(vsock_unbound_sockets, vsk); 267 spin_unlock_bh(&vsock_table_lock); 268 } 269 270 void vsock_insert_connected(struct vsock_sock *vsk) 271 { 272 struct list_head *list = vsock_connected_sockets( 273 &vsk->remote_addr, &vsk->local_addr); 274 275 spin_lock_bh(&vsock_table_lock); 276 __vsock_insert_connected(list, vsk); 277 spin_unlock_bh(&vsock_table_lock); 278 } 279 EXPORT_SYMBOL_GPL(vsock_insert_connected); 280 281 void vsock_remove_bound(struct vsock_sock *vsk) 282 { 283 spin_lock_bh(&vsock_table_lock); 284 __vsock_remove_bound(vsk); 285 spin_unlock_bh(&vsock_table_lock); 286 } 287 EXPORT_SYMBOL_GPL(vsock_remove_bound); 288 289 void vsock_remove_connected(struct vsock_sock *vsk) 290 { 291 spin_lock_bh(&vsock_table_lock); 292 __vsock_remove_connected(vsk); 293 spin_unlock_bh(&vsock_table_lock); 294 } 295 EXPORT_SYMBOL_GPL(vsock_remove_connected); 296 297 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr) 298 { 299 struct sock *sk; 300 301 spin_lock_bh(&vsock_table_lock); 302 sk = __vsock_find_bound_socket(addr); 303 if (sk) 304 sock_hold(sk); 305 306 spin_unlock_bh(&vsock_table_lock); 307 308 return sk; 309 } 310 EXPORT_SYMBOL_GPL(vsock_find_bound_socket); 311 312 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src, 313 struct sockaddr_vm *dst) 314 { 315 struct sock *sk; 316 317 spin_lock_bh(&vsock_table_lock); 318 sk = __vsock_find_connected_socket(src, dst); 319 if (sk) 320 sock_hold(sk); 321 322 spin_unlock_bh(&vsock_table_lock); 323 324 return sk; 325 } 326 EXPORT_SYMBOL_GPL(vsock_find_connected_socket); 327 328 static bool vsock_in_bound_table(struct vsock_sock *vsk) 329 { 330 bool ret; 331 332 spin_lock_bh(&vsock_table_lock); 333 ret = __vsock_in_bound_table(vsk); 334 spin_unlock_bh(&vsock_table_lock); 335 336 return ret; 337 } 338 339 static bool vsock_in_connected_table(struct vsock_sock *vsk) 340 { 341 bool ret; 342 343 spin_lock_bh(&vsock_table_lock); 344 ret = __vsock_in_connected_table(vsk); 345 spin_unlock_bh(&vsock_table_lock); 346 347 return ret; 348 } 349 350 void vsock_remove_sock(struct vsock_sock *vsk) 351 { 352 if (vsock_in_bound_table(vsk)) 353 vsock_remove_bound(vsk); 354 355 if (vsock_in_connected_table(vsk)) 356 vsock_remove_connected(vsk); 357 } 358 EXPORT_SYMBOL_GPL(vsock_remove_sock); 359 360 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk)) 361 { 362 int i; 363 364 spin_lock_bh(&vsock_table_lock); 365 366 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) { 367 struct vsock_sock *vsk; 368 list_for_each_entry(vsk, &vsock_connected_table[i], 369 connected_table) 370 fn(sk_vsock(vsk)); 371 } 372 373 spin_unlock_bh(&vsock_table_lock); 374 } 375 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket); 376 377 void vsock_add_pending(struct sock *listener, struct sock *pending) 378 { 379 struct vsock_sock *vlistener; 380 struct vsock_sock *vpending; 381 382 vlistener = vsock_sk(listener); 383 vpending = vsock_sk(pending); 384 385 sock_hold(pending); 386 sock_hold(listener); 387 list_add_tail(&vpending->pending_links, &vlistener->pending_links); 388 } 389 EXPORT_SYMBOL_GPL(vsock_add_pending); 390 391 void vsock_remove_pending(struct sock *listener, struct sock *pending) 392 { 393 struct vsock_sock *vpending = vsock_sk(pending); 394 395 list_del_init(&vpending->pending_links); 396 sock_put(listener); 397 sock_put(pending); 398 } 399 EXPORT_SYMBOL_GPL(vsock_remove_pending); 400 401 void vsock_enqueue_accept(struct sock *listener, struct sock *connected) 402 { 403 struct vsock_sock *vlistener; 404 struct vsock_sock *vconnected; 405 406 vlistener = vsock_sk(listener); 407 vconnected = vsock_sk(connected); 408 409 sock_hold(connected); 410 sock_hold(listener); 411 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue); 412 } 413 EXPORT_SYMBOL_GPL(vsock_enqueue_accept); 414 415 static struct sock *vsock_dequeue_accept(struct sock *listener) 416 { 417 struct vsock_sock *vlistener; 418 struct vsock_sock *vconnected; 419 420 vlistener = vsock_sk(listener); 421 422 if (list_empty(&vlistener->accept_queue)) 423 return NULL; 424 425 vconnected = list_entry(vlistener->accept_queue.next, 426 struct vsock_sock, accept_queue); 427 428 list_del_init(&vconnected->accept_queue); 429 sock_put(listener); 430 /* The caller will need a reference on the connected socket so we let 431 * it call sock_put(). 432 */ 433 434 return sk_vsock(vconnected); 435 } 436 437 static bool vsock_is_accept_queue_empty(struct sock *sk) 438 { 439 struct vsock_sock *vsk = vsock_sk(sk); 440 return list_empty(&vsk->accept_queue); 441 } 442 443 static bool vsock_is_pending(struct sock *sk) 444 { 445 struct vsock_sock *vsk = vsock_sk(sk); 446 return !list_empty(&vsk->pending_links); 447 } 448 449 static int vsock_send_shutdown(struct sock *sk, int mode) 450 { 451 return transport->shutdown(vsock_sk(sk), mode); 452 } 453 454 void vsock_pending_work(struct work_struct *work) 455 { 456 struct sock *sk; 457 struct sock *listener; 458 struct vsock_sock *vsk; 459 bool cleanup; 460 461 vsk = container_of(work, struct vsock_sock, dwork.work); 462 sk = sk_vsock(vsk); 463 listener = vsk->listener; 464 cleanup = true; 465 466 lock_sock(listener); 467 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 468 469 if (vsock_is_pending(sk)) { 470 vsock_remove_pending(listener, sk); 471 472 listener->sk_ack_backlog--; 473 } else if (!vsk->rejected) { 474 /* We are not on the pending list and accept() did not reject 475 * us, so we must have been accepted by our user process. We 476 * just need to drop our references to the sockets and be on 477 * our way. 478 */ 479 cleanup = false; 480 goto out; 481 } 482 483 /* We need to remove ourself from the global connected sockets list so 484 * incoming packets can't find this socket, and to reduce the reference 485 * count. 486 */ 487 if (vsock_in_connected_table(vsk)) 488 vsock_remove_connected(vsk); 489 490 sk->sk_state = TCP_CLOSE; 491 492 out: 493 release_sock(sk); 494 release_sock(listener); 495 if (cleanup) 496 sock_put(sk); 497 498 sock_put(sk); 499 sock_put(listener); 500 } 501 EXPORT_SYMBOL_GPL(vsock_pending_work); 502 503 /**** SOCKET OPERATIONS ****/ 504 505 static int __vsock_bind_stream(struct vsock_sock *vsk, 506 struct sockaddr_vm *addr) 507 { 508 static u32 port = LAST_RESERVED_PORT + 1; 509 struct sockaddr_vm new_addr; 510 511 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port); 512 513 if (addr->svm_port == VMADDR_PORT_ANY) { 514 bool found = false; 515 unsigned int i; 516 517 for (i = 0; i < MAX_PORT_RETRIES; i++) { 518 if (port <= LAST_RESERVED_PORT) 519 port = LAST_RESERVED_PORT + 1; 520 521 new_addr.svm_port = port++; 522 523 if (!__vsock_find_bound_socket(&new_addr)) { 524 found = true; 525 break; 526 } 527 } 528 529 if (!found) 530 return -EADDRNOTAVAIL; 531 } else { 532 /* If port is in reserved range, ensure caller 533 * has necessary privileges. 534 */ 535 if (addr->svm_port <= LAST_RESERVED_PORT && 536 !capable(CAP_NET_BIND_SERVICE)) { 537 return -EACCES; 538 } 539 540 if (__vsock_find_bound_socket(&new_addr)) 541 return -EADDRINUSE; 542 } 543 544 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port); 545 546 /* Remove stream sockets from the unbound list and add them to the hash 547 * table for easy lookup by its address. The unbound list is simply an 548 * extra entry at the end of the hash table, a trick used by AF_UNIX. 549 */ 550 __vsock_remove_bound(vsk); 551 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk); 552 553 return 0; 554 } 555 556 static int __vsock_bind_dgram(struct vsock_sock *vsk, 557 struct sockaddr_vm *addr) 558 { 559 return transport->dgram_bind(vsk, addr); 560 } 561 562 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr) 563 { 564 struct vsock_sock *vsk = vsock_sk(sk); 565 u32 cid; 566 int retval; 567 568 /* First ensure this socket isn't already bound. */ 569 if (vsock_addr_bound(&vsk->local_addr)) 570 return -EINVAL; 571 572 /* Now bind to the provided address or select appropriate values if 573 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that 574 * like AF_INET prevents binding to a non-local IP address (in most 575 * cases), we only allow binding to the local CID. 576 */ 577 cid = transport->get_local_cid(); 578 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY) 579 return -EADDRNOTAVAIL; 580 581 switch (sk->sk_socket->type) { 582 case SOCK_STREAM: 583 spin_lock_bh(&vsock_table_lock); 584 retval = __vsock_bind_stream(vsk, addr); 585 spin_unlock_bh(&vsock_table_lock); 586 break; 587 588 case SOCK_DGRAM: 589 retval = __vsock_bind_dgram(vsk, addr); 590 break; 591 592 default: 593 retval = -EINVAL; 594 break; 595 } 596 597 return retval; 598 } 599 600 struct sock *__vsock_create(struct net *net, 601 struct socket *sock, 602 struct sock *parent, 603 gfp_t priority, 604 unsigned short type, 605 int kern) 606 { 607 struct sock *sk; 608 struct vsock_sock *psk; 609 struct vsock_sock *vsk; 610 611 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern); 612 if (!sk) 613 return NULL; 614 615 sock_init_data(sock, sk); 616 617 /* sk->sk_type is normally set in sock_init_data, but only if sock is 618 * non-NULL. We make sure that our sockets always have a type by 619 * setting it here if needed. 620 */ 621 if (!sock) 622 sk->sk_type = type; 623 624 vsk = vsock_sk(sk); 625 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 626 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 627 628 sk->sk_destruct = vsock_sk_destruct; 629 sk->sk_backlog_rcv = vsock_queue_rcv_skb; 630 sock_reset_flag(sk, SOCK_DONE); 631 632 INIT_LIST_HEAD(&vsk->bound_table); 633 INIT_LIST_HEAD(&vsk->connected_table); 634 vsk->listener = NULL; 635 INIT_LIST_HEAD(&vsk->pending_links); 636 INIT_LIST_HEAD(&vsk->accept_queue); 637 vsk->rejected = false; 638 vsk->sent_request = false; 639 vsk->ignore_connecting_rst = false; 640 vsk->peer_shutdown = 0; 641 642 psk = parent ? vsock_sk(parent) : NULL; 643 if (parent) { 644 vsk->trusted = psk->trusted; 645 vsk->owner = get_cred(psk->owner); 646 vsk->connect_timeout = psk->connect_timeout; 647 } else { 648 vsk->trusted = capable(CAP_NET_ADMIN); 649 vsk->owner = get_current_cred(); 650 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT; 651 } 652 653 if (transport->init(vsk, psk) < 0) { 654 sk_free(sk); 655 return NULL; 656 } 657 658 if (sock) 659 vsock_insert_unbound(vsk); 660 661 return sk; 662 } 663 EXPORT_SYMBOL_GPL(__vsock_create); 664 665 static void __vsock_release(struct sock *sk) 666 { 667 if (sk) { 668 struct sk_buff *skb; 669 struct sock *pending; 670 struct vsock_sock *vsk; 671 672 vsk = vsock_sk(sk); 673 pending = NULL; /* Compiler warning. */ 674 675 transport->release(vsk); 676 677 lock_sock(sk); 678 sock_orphan(sk); 679 sk->sk_shutdown = SHUTDOWN_MASK; 680 681 while ((skb = skb_dequeue(&sk->sk_receive_queue))) 682 kfree_skb(skb); 683 684 /* Clean up any sockets that never were accepted. */ 685 while ((pending = vsock_dequeue_accept(sk)) != NULL) { 686 __vsock_release(pending); 687 sock_put(pending); 688 } 689 690 release_sock(sk); 691 sock_put(sk); 692 } 693 } 694 695 static void vsock_sk_destruct(struct sock *sk) 696 { 697 struct vsock_sock *vsk = vsock_sk(sk); 698 699 transport->destruct(vsk); 700 701 /* When clearing these addresses, there's no need to set the family and 702 * possibly register the address family with the kernel. 703 */ 704 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 705 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 706 707 put_cred(vsk->owner); 708 } 709 710 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 711 { 712 int err; 713 714 err = sock_queue_rcv_skb(sk, skb); 715 if (err) 716 kfree_skb(skb); 717 718 return err; 719 } 720 721 s64 vsock_stream_has_data(struct vsock_sock *vsk) 722 { 723 return transport->stream_has_data(vsk); 724 } 725 EXPORT_SYMBOL_GPL(vsock_stream_has_data); 726 727 s64 vsock_stream_has_space(struct vsock_sock *vsk) 728 { 729 return transport->stream_has_space(vsk); 730 } 731 EXPORT_SYMBOL_GPL(vsock_stream_has_space); 732 733 static int vsock_release(struct socket *sock) 734 { 735 __vsock_release(sock->sk); 736 sock->sk = NULL; 737 sock->state = SS_FREE; 738 739 return 0; 740 } 741 742 static int 743 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 744 { 745 int err; 746 struct sock *sk; 747 struct sockaddr_vm *vm_addr; 748 749 sk = sock->sk; 750 751 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0) 752 return -EINVAL; 753 754 lock_sock(sk); 755 err = __vsock_bind(sk, vm_addr); 756 release_sock(sk); 757 758 return err; 759 } 760 761 static int vsock_getname(struct socket *sock, 762 struct sockaddr *addr, int peer) 763 { 764 int err; 765 struct sock *sk; 766 struct vsock_sock *vsk; 767 struct sockaddr_vm *vm_addr; 768 769 sk = sock->sk; 770 vsk = vsock_sk(sk); 771 err = 0; 772 773 lock_sock(sk); 774 775 if (peer) { 776 if (sock->state != SS_CONNECTED) { 777 err = -ENOTCONN; 778 goto out; 779 } 780 vm_addr = &vsk->remote_addr; 781 } else { 782 vm_addr = &vsk->local_addr; 783 } 784 785 if (!vm_addr) { 786 err = -EINVAL; 787 goto out; 788 } 789 790 /* sys_getsockname() and sys_getpeername() pass us a 791 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately 792 * that macro is defined in socket.c instead of .h, so we hardcode its 793 * value here. 794 */ 795 BUILD_BUG_ON(sizeof(*vm_addr) > 128); 796 memcpy(addr, vm_addr, sizeof(*vm_addr)); 797 err = sizeof(*vm_addr); 798 799 out: 800 release_sock(sk); 801 return err; 802 } 803 804 static int vsock_shutdown(struct socket *sock, int mode) 805 { 806 int err; 807 struct sock *sk; 808 809 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses 810 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode 811 * here like the other address families do. Note also that the 812 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3), 813 * which is what we want. 814 */ 815 mode++; 816 817 if ((mode & ~SHUTDOWN_MASK) || !mode) 818 return -EINVAL; 819 820 /* If this is a STREAM socket and it is not connected then bail out 821 * immediately. If it is a DGRAM socket then we must first kick the 822 * socket so that it wakes up from any sleeping calls, for example 823 * recv(), and then afterwards return the error. 824 */ 825 826 sk = sock->sk; 827 if (sock->state == SS_UNCONNECTED) { 828 err = -ENOTCONN; 829 if (sk->sk_type == SOCK_STREAM) 830 return err; 831 } else { 832 sock->state = SS_DISCONNECTING; 833 err = 0; 834 } 835 836 /* Receive and send shutdowns are treated alike. */ 837 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN); 838 if (mode) { 839 lock_sock(sk); 840 sk->sk_shutdown |= mode; 841 sk->sk_state_change(sk); 842 release_sock(sk); 843 844 if (sk->sk_type == SOCK_STREAM) { 845 sock_reset_flag(sk, SOCK_DONE); 846 vsock_send_shutdown(sk, mode); 847 } 848 } 849 850 return err; 851 } 852 853 static __poll_t vsock_poll_mask(struct socket *sock, __poll_t events) 854 { 855 struct sock *sk = sock->sk; 856 struct vsock_sock *vsk = vsock_sk(sk); 857 __poll_t mask = 0; 858 859 if (sk->sk_err) 860 /* Signify that there has been an error on this socket. */ 861 mask |= EPOLLERR; 862 863 /* INET sockets treat local write shutdown and peer write shutdown as a 864 * case of EPOLLHUP set. 865 */ 866 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 867 ((sk->sk_shutdown & SEND_SHUTDOWN) && 868 (vsk->peer_shutdown & SEND_SHUTDOWN))) { 869 mask |= EPOLLHUP; 870 } 871 872 if (sk->sk_shutdown & RCV_SHUTDOWN || 873 vsk->peer_shutdown & SEND_SHUTDOWN) { 874 mask |= EPOLLRDHUP; 875 } 876 877 if (sock->type == SOCK_DGRAM) { 878 /* For datagram sockets we can read if there is something in 879 * the queue and write as long as the socket isn't shutdown for 880 * sending. 881 */ 882 if (!skb_queue_empty(&sk->sk_receive_queue) || 883 (sk->sk_shutdown & RCV_SHUTDOWN)) { 884 mask |= EPOLLIN | EPOLLRDNORM; 885 } 886 887 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 888 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND; 889 890 } else if (sock->type == SOCK_STREAM) { 891 lock_sock(sk); 892 893 /* Listening sockets that have connections in their accept 894 * queue can be read. 895 */ 896 if (sk->sk_state == TCP_LISTEN 897 && !vsock_is_accept_queue_empty(sk)) 898 mask |= EPOLLIN | EPOLLRDNORM; 899 900 /* If there is something in the queue then we can read. */ 901 if (transport->stream_is_active(vsk) && 902 !(sk->sk_shutdown & RCV_SHUTDOWN)) { 903 bool data_ready_now = false; 904 int ret = transport->notify_poll_in( 905 vsk, 1, &data_ready_now); 906 if (ret < 0) { 907 mask |= EPOLLERR; 908 } else { 909 if (data_ready_now) 910 mask |= EPOLLIN | EPOLLRDNORM; 911 912 } 913 } 914 915 /* Sockets whose connections have been closed, reset, or 916 * terminated should also be considered read, and we check the 917 * shutdown flag for that. 918 */ 919 if (sk->sk_shutdown & RCV_SHUTDOWN || 920 vsk->peer_shutdown & SEND_SHUTDOWN) { 921 mask |= EPOLLIN | EPOLLRDNORM; 922 } 923 924 /* Connected sockets that can produce data can be written. */ 925 if (sk->sk_state == TCP_ESTABLISHED) { 926 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 927 bool space_avail_now = false; 928 int ret = transport->notify_poll_out( 929 vsk, 1, &space_avail_now); 930 if (ret < 0) { 931 mask |= EPOLLERR; 932 } else { 933 if (space_avail_now) 934 /* Remove EPOLLWRBAND since INET 935 * sockets are not setting it. 936 */ 937 mask |= EPOLLOUT | EPOLLWRNORM; 938 939 } 940 } 941 } 942 943 /* Simulate INET socket poll behaviors, which sets 944 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read, 945 * but local send is not shutdown. 946 */ 947 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) { 948 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 949 mask |= EPOLLOUT | EPOLLWRNORM; 950 951 } 952 953 release_sock(sk); 954 } 955 956 return mask; 957 } 958 959 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg, 960 size_t len) 961 { 962 int err; 963 struct sock *sk; 964 struct vsock_sock *vsk; 965 struct sockaddr_vm *remote_addr; 966 967 if (msg->msg_flags & MSG_OOB) 968 return -EOPNOTSUPP; 969 970 /* For now, MSG_DONTWAIT is always assumed... */ 971 err = 0; 972 sk = sock->sk; 973 vsk = vsock_sk(sk); 974 975 lock_sock(sk); 976 977 err = vsock_auto_bind(vsk); 978 if (err) 979 goto out; 980 981 982 /* If the provided message contains an address, use that. Otherwise 983 * fall back on the socket's remote handle (if it has been connected). 984 */ 985 if (msg->msg_name && 986 vsock_addr_cast(msg->msg_name, msg->msg_namelen, 987 &remote_addr) == 0) { 988 /* Ensure this address is of the right type and is a valid 989 * destination. 990 */ 991 992 if (remote_addr->svm_cid == VMADDR_CID_ANY) 993 remote_addr->svm_cid = transport->get_local_cid(); 994 995 if (!vsock_addr_bound(remote_addr)) { 996 err = -EINVAL; 997 goto out; 998 } 999 } else if (sock->state == SS_CONNECTED) { 1000 remote_addr = &vsk->remote_addr; 1001 1002 if (remote_addr->svm_cid == VMADDR_CID_ANY) 1003 remote_addr->svm_cid = transport->get_local_cid(); 1004 1005 /* XXX Should connect() or this function ensure remote_addr is 1006 * bound? 1007 */ 1008 if (!vsock_addr_bound(&vsk->remote_addr)) { 1009 err = -EINVAL; 1010 goto out; 1011 } 1012 } else { 1013 err = -EINVAL; 1014 goto out; 1015 } 1016 1017 if (!transport->dgram_allow(remote_addr->svm_cid, 1018 remote_addr->svm_port)) { 1019 err = -EINVAL; 1020 goto out; 1021 } 1022 1023 err = transport->dgram_enqueue(vsk, remote_addr, msg, len); 1024 1025 out: 1026 release_sock(sk); 1027 return err; 1028 } 1029 1030 static int vsock_dgram_connect(struct socket *sock, 1031 struct sockaddr *addr, int addr_len, int flags) 1032 { 1033 int err; 1034 struct sock *sk; 1035 struct vsock_sock *vsk; 1036 struct sockaddr_vm *remote_addr; 1037 1038 sk = sock->sk; 1039 vsk = vsock_sk(sk); 1040 1041 err = vsock_addr_cast(addr, addr_len, &remote_addr); 1042 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) { 1043 lock_sock(sk); 1044 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, 1045 VMADDR_PORT_ANY); 1046 sock->state = SS_UNCONNECTED; 1047 release_sock(sk); 1048 return 0; 1049 } else if (err != 0) 1050 return -EINVAL; 1051 1052 lock_sock(sk); 1053 1054 err = vsock_auto_bind(vsk); 1055 if (err) 1056 goto out; 1057 1058 if (!transport->dgram_allow(remote_addr->svm_cid, 1059 remote_addr->svm_port)) { 1060 err = -EINVAL; 1061 goto out; 1062 } 1063 1064 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr)); 1065 sock->state = SS_CONNECTED; 1066 1067 out: 1068 release_sock(sk); 1069 return err; 1070 } 1071 1072 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg, 1073 size_t len, int flags) 1074 { 1075 return transport->dgram_dequeue(vsock_sk(sock->sk), msg, len, flags); 1076 } 1077 1078 static const struct proto_ops vsock_dgram_ops = { 1079 .family = PF_VSOCK, 1080 .owner = THIS_MODULE, 1081 .release = vsock_release, 1082 .bind = vsock_bind, 1083 .connect = vsock_dgram_connect, 1084 .socketpair = sock_no_socketpair, 1085 .accept = sock_no_accept, 1086 .getname = vsock_getname, 1087 .poll_mask = vsock_poll_mask, 1088 .ioctl = sock_no_ioctl, 1089 .listen = sock_no_listen, 1090 .shutdown = vsock_shutdown, 1091 .setsockopt = sock_no_setsockopt, 1092 .getsockopt = sock_no_getsockopt, 1093 .sendmsg = vsock_dgram_sendmsg, 1094 .recvmsg = vsock_dgram_recvmsg, 1095 .mmap = sock_no_mmap, 1096 .sendpage = sock_no_sendpage, 1097 }; 1098 1099 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk) 1100 { 1101 if (!transport->cancel_pkt) 1102 return -EOPNOTSUPP; 1103 1104 return transport->cancel_pkt(vsk); 1105 } 1106 1107 static void vsock_connect_timeout(struct work_struct *work) 1108 { 1109 struct sock *sk; 1110 struct vsock_sock *vsk; 1111 int cancel = 0; 1112 1113 vsk = container_of(work, struct vsock_sock, dwork.work); 1114 sk = sk_vsock(vsk); 1115 1116 lock_sock(sk); 1117 if (sk->sk_state == TCP_SYN_SENT && 1118 (sk->sk_shutdown != SHUTDOWN_MASK)) { 1119 sk->sk_state = TCP_CLOSE; 1120 sk->sk_err = ETIMEDOUT; 1121 sk->sk_error_report(sk); 1122 cancel = 1; 1123 } 1124 release_sock(sk); 1125 if (cancel) 1126 vsock_transport_cancel_pkt(vsk); 1127 1128 sock_put(sk); 1129 } 1130 1131 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr, 1132 int addr_len, int flags) 1133 { 1134 int err; 1135 struct sock *sk; 1136 struct vsock_sock *vsk; 1137 struct sockaddr_vm *remote_addr; 1138 long timeout; 1139 DEFINE_WAIT(wait); 1140 1141 err = 0; 1142 sk = sock->sk; 1143 vsk = vsock_sk(sk); 1144 1145 lock_sock(sk); 1146 1147 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */ 1148 switch (sock->state) { 1149 case SS_CONNECTED: 1150 err = -EISCONN; 1151 goto out; 1152 case SS_DISCONNECTING: 1153 err = -EINVAL; 1154 goto out; 1155 case SS_CONNECTING: 1156 /* This continues on so we can move sock into the SS_CONNECTED 1157 * state once the connection has completed (at which point err 1158 * will be set to zero also). Otherwise, we will either wait 1159 * for the connection or return -EALREADY should this be a 1160 * non-blocking call. 1161 */ 1162 err = -EALREADY; 1163 break; 1164 default: 1165 if ((sk->sk_state == TCP_LISTEN) || 1166 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) { 1167 err = -EINVAL; 1168 goto out; 1169 } 1170 1171 /* The hypervisor and well-known contexts do not have socket 1172 * endpoints. 1173 */ 1174 if (!transport->stream_allow(remote_addr->svm_cid, 1175 remote_addr->svm_port)) { 1176 err = -ENETUNREACH; 1177 goto out; 1178 } 1179 1180 /* Set the remote address that we are connecting to. */ 1181 memcpy(&vsk->remote_addr, remote_addr, 1182 sizeof(vsk->remote_addr)); 1183 1184 err = vsock_auto_bind(vsk); 1185 if (err) 1186 goto out; 1187 1188 sk->sk_state = TCP_SYN_SENT; 1189 1190 err = transport->connect(vsk); 1191 if (err < 0) 1192 goto out; 1193 1194 /* Mark sock as connecting and set the error code to in 1195 * progress in case this is a non-blocking connect. 1196 */ 1197 sock->state = SS_CONNECTING; 1198 err = -EINPROGRESS; 1199 } 1200 1201 /* The receive path will handle all communication until we are able to 1202 * enter the connected state. Here we wait for the connection to be 1203 * completed or a notification of an error. 1204 */ 1205 timeout = vsk->connect_timeout; 1206 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1207 1208 while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) { 1209 if (flags & O_NONBLOCK) { 1210 /* If we're not going to block, we schedule a timeout 1211 * function to generate a timeout on the connection 1212 * attempt, in case the peer doesn't respond in a 1213 * timely manner. We hold on to the socket until the 1214 * timeout fires. 1215 */ 1216 sock_hold(sk); 1217 INIT_DELAYED_WORK(&vsk->dwork, 1218 vsock_connect_timeout); 1219 schedule_delayed_work(&vsk->dwork, timeout); 1220 1221 /* Skip ahead to preserve error code set above. */ 1222 goto out_wait; 1223 } 1224 1225 release_sock(sk); 1226 timeout = schedule_timeout(timeout); 1227 lock_sock(sk); 1228 1229 if (signal_pending(current)) { 1230 err = sock_intr_errno(timeout); 1231 sk->sk_state = TCP_CLOSE; 1232 sock->state = SS_UNCONNECTED; 1233 vsock_transport_cancel_pkt(vsk); 1234 goto out_wait; 1235 } else if (timeout == 0) { 1236 err = -ETIMEDOUT; 1237 sk->sk_state = TCP_CLOSE; 1238 sock->state = SS_UNCONNECTED; 1239 vsock_transport_cancel_pkt(vsk); 1240 goto out_wait; 1241 } 1242 1243 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1244 } 1245 1246 if (sk->sk_err) { 1247 err = -sk->sk_err; 1248 sk->sk_state = TCP_CLOSE; 1249 sock->state = SS_UNCONNECTED; 1250 } else { 1251 err = 0; 1252 } 1253 1254 out_wait: 1255 finish_wait(sk_sleep(sk), &wait); 1256 out: 1257 release_sock(sk); 1258 return err; 1259 } 1260 1261 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags, 1262 bool kern) 1263 { 1264 struct sock *listener; 1265 int err; 1266 struct sock *connected; 1267 struct vsock_sock *vconnected; 1268 long timeout; 1269 DEFINE_WAIT(wait); 1270 1271 err = 0; 1272 listener = sock->sk; 1273 1274 lock_sock(listener); 1275 1276 if (sock->type != SOCK_STREAM) { 1277 err = -EOPNOTSUPP; 1278 goto out; 1279 } 1280 1281 if (listener->sk_state != TCP_LISTEN) { 1282 err = -EINVAL; 1283 goto out; 1284 } 1285 1286 /* Wait for children sockets to appear; these are the new sockets 1287 * created upon connection establishment. 1288 */ 1289 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK); 1290 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1291 1292 while ((connected = vsock_dequeue_accept(listener)) == NULL && 1293 listener->sk_err == 0) { 1294 release_sock(listener); 1295 timeout = schedule_timeout(timeout); 1296 finish_wait(sk_sleep(listener), &wait); 1297 lock_sock(listener); 1298 1299 if (signal_pending(current)) { 1300 err = sock_intr_errno(timeout); 1301 goto out; 1302 } else if (timeout == 0) { 1303 err = -EAGAIN; 1304 goto out; 1305 } 1306 1307 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1308 } 1309 finish_wait(sk_sleep(listener), &wait); 1310 1311 if (listener->sk_err) 1312 err = -listener->sk_err; 1313 1314 if (connected) { 1315 listener->sk_ack_backlog--; 1316 1317 lock_sock_nested(connected, SINGLE_DEPTH_NESTING); 1318 vconnected = vsock_sk(connected); 1319 1320 /* If the listener socket has received an error, then we should 1321 * reject this socket and return. Note that we simply mark the 1322 * socket rejected, drop our reference, and let the cleanup 1323 * function handle the cleanup; the fact that we found it in 1324 * the listener's accept queue guarantees that the cleanup 1325 * function hasn't run yet. 1326 */ 1327 if (err) { 1328 vconnected->rejected = true; 1329 } else { 1330 newsock->state = SS_CONNECTED; 1331 sock_graft(connected, newsock); 1332 } 1333 1334 release_sock(connected); 1335 sock_put(connected); 1336 } 1337 1338 out: 1339 release_sock(listener); 1340 return err; 1341 } 1342 1343 static int vsock_listen(struct socket *sock, int backlog) 1344 { 1345 int err; 1346 struct sock *sk; 1347 struct vsock_sock *vsk; 1348 1349 sk = sock->sk; 1350 1351 lock_sock(sk); 1352 1353 if (sock->type != SOCK_STREAM) { 1354 err = -EOPNOTSUPP; 1355 goto out; 1356 } 1357 1358 if (sock->state != SS_UNCONNECTED) { 1359 err = -EINVAL; 1360 goto out; 1361 } 1362 1363 vsk = vsock_sk(sk); 1364 1365 if (!vsock_addr_bound(&vsk->local_addr)) { 1366 err = -EINVAL; 1367 goto out; 1368 } 1369 1370 sk->sk_max_ack_backlog = backlog; 1371 sk->sk_state = TCP_LISTEN; 1372 1373 err = 0; 1374 1375 out: 1376 release_sock(sk); 1377 return err; 1378 } 1379 1380 static int vsock_stream_setsockopt(struct socket *sock, 1381 int level, 1382 int optname, 1383 char __user *optval, 1384 unsigned int optlen) 1385 { 1386 int err; 1387 struct sock *sk; 1388 struct vsock_sock *vsk; 1389 u64 val; 1390 1391 if (level != AF_VSOCK) 1392 return -ENOPROTOOPT; 1393 1394 #define COPY_IN(_v) \ 1395 do { \ 1396 if (optlen < sizeof(_v)) { \ 1397 err = -EINVAL; \ 1398 goto exit; \ 1399 } \ 1400 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \ 1401 err = -EFAULT; \ 1402 goto exit; \ 1403 } \ 1404 } while (0) 1405 1406 err = 0; 1407 sk = sock->sk; 1408 vsk = vsock_sk(sk); 1409 1410 lock_sock(sk); 1411 1412 switch (optname) { 1413 case SO_VM_SOCKETS_BUFFER_SIZE: 1414 COPY_IN(val); 1415 transport->set_buffer_size(vsk, val); 1416 break; 1417 1418 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1419 COPY_IN(val); 1420 transport->set_max_buffer_size(vsk, val); 1421 break; 1422 1423 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1424 COPY_IN(val); 1425 transport->set_min_buffer_size(vsk, val); 1426 break; 1427 1428 case SO_VM_SOCKETS_CONNECT_TIMEOUT: { 1429 struct timeval tv; 1430 COPY_IN(tv); 1431 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC && 1432 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) { 1433 vsk->connect_timeout = tv.tv_sec * HZ + 1434 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ)); 1435 if (vsk->connect_timeout == 0) 1436 vsk->connect_timeout = 1437 VSOCK_DEFAULT_CONNECT_TIMEOUT; 1438 1439 } else { 1440 err = -ERANGE; 1441 } 1442 break; 1443 } 1444 1445 default: 1446 err = -ENOPROTOOPT; 1447 break; 1448 } 1449 1450 #undef COPY_IN 1451 1452 exit: 1453 release_sock(sk); 1454 return err; 1455 } 1456 1457 static int vsock_stream_getsockopt(struct socket *sock, 1458 int level, int optname, 1459 char __user *optval, 1460 int __user *optlen) 1461 { 1462 int err; 1463 int len; 1464 struct sock *sk; 1465 struct vsock_sock *vsk; 1466 u64 val; 1467 1468 if (level != AF_VSOCK) 1469 return -ENOPROTOOPT; 1470 1471 err = get_user(len, optlen); 1472 if (err != 0) 1473 return err; 1474 1475 #define COPY_OUT(_v) \ 1476 do { \ 1477 if (len < sizeof(_v)) \ 1478 return -EINVAL; \ 1479 \ 1480 len = sizeof(_v); \ 1481 if (copy_to_user(optval, &_v, len) != 0) \ 1482 return -EFAULT; \ 1483 \ 1484 } while (0) 1485 1486 err = 0; 1487 sk = sock->sk; 1488 vsk = vsock_sk(sk); 1489 1490 switch (optname) { 1491 case SO_VM_SOCKETS_BUFFER_SIZE: 1492 val = transport->get_buffer_size(vsk); 1493 COPY_OUT(val); 1494 break; 1495 1496 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1497 val = transport->get_max_buffer_size(vsk); 1498 COPY_OUT(val); 1499 break; 1500 1501 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1502 val = transport->get_min_buffer_size(vsk); 1503 COPY_OUT(val); 1504 break; 1505 1506 case SO_VM_SOCKETS_CONNECT_TIMEOUT: { 1507 struct timeval tv; 1508 tv.tv_sec = vsk->connect_timeout / HZ; 1509 tv.tv_usec = 1510 (vsk->connect_timeout - 1511 tv.tv_sec * HZ) * (1000000 / HZ); 1512 COPY_OUT(tv); 1513 break; 1514 } 1515 default: 1516 return -ENOPROTOOPT; 1517 } 1518 1519 err = put_user(len, optlen); 1520 if (err != 0) 1521 return -EFAULT; 1522 1523 #undef COPY_OUT 1524 1525 return 0; 1526 } 1527 1528 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg, 1529 size_t len) 1530 { 1531 struct sock *sk; 1532 struct vsock_sock *vsk; 1533 ssize_t total_written; 1534 long timeout; 1535 int err; 1536 struct vsock_transport_send_notify_data send_data; 1537 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1538 1539 sk = sock->sk; 1540 vsk = vsock_sk(sk); 1541 total_written = 0; 1542 err = 0; 1543 1544 if (msg->msg_flags & MSG_OOB) 1545 return -EOPNOTSUPP; 1546 1547 lock_sock(sk); 1548 1549 /* Callers should not provide a destination with stream sockets. */ 1550 if (msg->msg_namelen) { 1551 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1552 goto out; 1553 } 1554 1555 /* Send data only if both sides are not shutdown in the direction. */ 1556 if (sk->sk_shutdown & SEND_SHUTDOWN || 1557 vsk->peer_shutdown & RCV_SHUTDOWN) { 1558 err = -EPIPE; 1559 goto out; 1560 } 1561 1562 if (sk->sk_state != TCP_ESTABLISHED || 1563 !vsock_addr_bound(&vsk->local_addr)) { 1564 err = -ENOTCONN; 1565 goto out; 1566 } 1567 1568 if (!vsock_addr_bound(&vsk->remote_addr)) { 1569 err = -EDESTADDRREQ; 1570 goto out; 1571 } 1572 1573 /* Wait for room in the produce queue to enqueue our user's data. */ 1574 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1575 1576 err = transport->notify_send_init(vsk, &send_data); 1577 if (err < 0) 1578 goto out; 1579 1580 while (total_written < len) { 1581 ssize_t written; 1582 1583 add_wait_queue(sk_sleep(sk), &wait); 1584 while (vsock_stream_has_space(vsk) == 0 && 1585 sk->sk_err == 0 && 1586 !(sk->sk_shutdown & SEND_SHUTDOWN) && 1587 !(vsk->peer_shutdown & RCV_SHUTDOWN)) { 1588 1589 /* Don't wait for non-blocking sockets. */ 1590 if (timeout == 0) { 1591 err = -EAGAIN; 1592 remove_wait_queue(sk_sleep(sk), &wait); 1593 goto out_err; 1594 } 1595 1596 err = transport->notify_send_pre_block(vsk, &send_data); 1597 if (err < 0) { 1598 remove_wait_queue(sk_sleep(sk), &wait); 1599 goto out_err; 1600 } 1601 1602 release_sock(sk); 1603 timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout); 1604 lock_sock(sk); 1605 if (signal_pending(current)) { 1606 err = sock_intr_errno(timeout); 1607 remove_wait_queue(sk_sleep(sk), &wait); 1608 goto out_err; 1609 } else if (timeout == 0) { 1610 err = -EAGAIN; 1611 remove_wait_queue(sk_sleep(sk), &wait); 1612 goto out_err; 1613 } 1614 } 1615 remove_wait_queue(sk_sleep(sk), &wait); 1616 1617 /* These checks occur both as part of and after the loop 1618 * conditional since we need to check before and after 1619 * sleeping. 1620 */ 1621 if (sk->sk_err) { 1622 err = -sk->sk_err; 1623 goto out_err; 1624 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) || 1625 (vsk->peer_shutdown & RCV_SHUTDOWN)) { 1626 err = -EPIPE; 1627 goto out_err; 1628 } 1629 1630 err = transport->notify_send_pre_enqueue(vsk, &send_data); 1631 if (err < 0) 1632 goto out_err; 1633 1634 /* Note that enqueue will only write as many bytes as are free 1635 * in the produce queue, so we don't need to ensure len is 1636 * smaller than the queue size. It is the caller's 1637 * responsibility to check how many bytes we were able to send. 1638 */ 1639 1640 written = transport->stream_enqueue( 1641 vsk, msg, 1642 len - total_written); 1643 if (written < 0) { 1644 err = -ENOMEM; 1645 goto out_err; 1646 } 1647 1648 total_written += written; 1649 1650 err = transport->notify_send_post_enqueue( 1651 vsk, written, &send_data); 1652 if (err < 0) 1653 goto out_err; 1654 1655 } 1656 1657 out_err: 1658 if (total_written > 0) 1659 err = total_written; 1660 out: 1661 release_sock(sk); 1662 return err; 1663 } 1664 1665 1666 static int 1667 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1668 int flags) 1669 { 1670 struct sock *sk; 1671 struct vsock_sock *vsk; 1672 int err; 1673 size_t target; 1674 ssize_t copied; 1675 long timeout; 1676 struct vsock_transport_recv_notify_data recv_data; 1677 1678 DEFINE_WAIT(wait); 1679 1680 sk = sock->sk; 1681 vsk = vsock_sk(sk); 1682 err = 0; 1683 1684 lock_sock(sk); 1685 1686 if (sk->sk_state != TCP_ESTABLISHED) { 1687 /* Recvmsg is supposed to return 0 if a peer performs an 1688 * orderly shutdown. Differentiate between that case and when a 1689 * peer has not connected or a local shutdown occured with the 1690 * SOCK_DONE flag. 1691 */ 1692 if (sock_flag(sk, SOCK_DONE)) 1693 err = 0; 1694 else 1695 err = -ENOTCONN; 1696 1697 goto out; 1698 } 1699 1700 if (flags & MSG_OOB) { 1701 err = -EOPNOTSUPP; 1702 goto out; 1703 } 1704 1705 /* We don't check peer_shutdown flag here since peer may actually shut 1706 * down, but there can be data in the queue that a local socket can 1707 * receive. 1708 */ 1709 if (sk->sk_shutdown & RCV_SHUTDOWN) { 1710 err = 0; 1711 goto out; 1712 } 1713 1714 /* It is valid on Linux to pass in a zero-length receive buffer. This 1715 * is not an error. We may as well bail out now. 1716 */ 1717 if (!len) { 1718 err = 0; 1719 goto out; 1720 } 1721 1722 /* We must not copy less than target bytes into the user's buffer 1723 * before returning successfully, so we wait for the consume queue to 1724 * have that much data to consume before dequeueing. Note that this 1725 * makes it impossible to handle cases where target is greater than the 1726 * queue size. 1727 */ 1728 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 1729 if (target >= transport->stream_rcvhiwat(vsk)) { 1730 err = -ENOMEM; 1731 goto out; 1732 } 1733 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1734 copied = 0; 1735 1736 err = transport->notify_recv_init(vsk, target, &recv_data); 1737 if (err < 0) 1738 goto out; 1739 1740 1741 while (1) { 1742 s64 ready; 1743 1744 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1745 ready = vsock_stream_has_data(vsk); 1746 1747 if (ready == 0) { 1748 if (sk->sk_err != 0 || 1749 (sk->sk_shutdown & RCV_SHUTDOWN) || 1750 (vsk->peer_shutdown & SEND_SHUTDOWN)) { 1751 finish_wait(sk_sleep(sk), &wait); 1752 break; 1753 } 1754 /* Don't wait for non-blocking sockets. */ 1755 if (timeout == 0) { 1756 err = -EAGAIN; 1757 finish_wait(sk_sleep(sk), &wait); 1758 break; 1759 } 1760 1761 err = transport->notify_recv_pre_block( 1762 vsk, target, &recv_data); 1763 if (err < 0) { 1764 finish_wait(sk_sleep(sk), &wait); 1765 break; 1766 } 1767 release_sock(sk); 1768 timeout = schedule_timeout(timeout); 1769 lock_sock(sk); 1770 1771 if (signal_pending(current)) { 1772 err = sock_intr_errno(timeout); 1773 finish_wait(sk_sleep(sk), &wait); 1774 break; 1775 } else if (timeout == 0) { 1776 err = -EAGAIN; 1777 finish_wait(sk_sleep(sk), &wait); 1778 break; 1779 } 1780 } else { 1781 ssize_t read; 1782 1783 finish_wait(sk_sleep(sk), &wait); 1784 1785 if (ready < 0) { 1786 /* Invalid queue pair content. XXX This should 1787 * be changed to a connection reset in a later 1788 * change. 1789 */ 1790 1791 err = -ENOMEM; 1792 goto out; 1793 } 1794 1795 err = transport->notify_recv_pre_dequeue( 1796 vsk, target, &recv_data); 1797 if (err < 0) 1798 break; 1799 1800 read = transport->stream_dequeue( 1801 vsk, msg, 1802 len - copied, flags); 1803 if (read < 0) { 1804 err = -ENOMEM; 1805 break; 1806 } 1807 1808 copied += read; 1809 1810 err = transport->notify_recv_post_dequeue( 1811 vsk, target, read, 1812 !(flags & MSG_PEEK), &recv_data); 1813 if (err < 0) 1814 goto out; 1815 1816 if (read >= target || flags & MSG_PEEK) 1817 break; 1818 1819 target -= read; 1820 } 1821 } 1822 1823 if (sk->sk_err) 1824 err = -sk->sk_err; 1825 else if (sk->sk_shutdown & RCV_SHUTDOWN) 1826 err = 0; 1827 1828 if (copied > 0) 1829 err = copied; 1830 1831 out: 1832 release_sock(sk); 1833 return err; 1834 } 1835 1836 static const struct proto_ops vsock_stream_ops = { 1837 .family = PF_VSOCK, 1838 .owner = THIS_MODULE, 1839 .release = vsock_release, 1840 .bind = vsock_bind, 1841 .connect = vsock_stream_connect, 1842 .socketpair = sock_no_socketpair, 1843 .accept = vsock_accept, 1844 .getname = vsock_getname, 1845 .poll_mask = vsock_poll_mask, 1846 .ioctl = sock_no_ioctl, 1847 .listen = vsock_listen, 1848 .shutdown = vsock_shutdown, 1849 .setsockopt = vsock_stream_setsockopt, 1850 .getsockopt = vsock_stream_getsockopt, 1851 .sendmsg = vsock_stream_sendmsg, 1852 .recvmsg = vsock_stream_recvmsg, 1853 .mmap = sock_no_mmap, 1854 .sendpage = sock_no_sendpage, 1855 }; 1856 1857 static int vsock_create(struct net *net, struct socket *sock, 1858 int protocol, int kern) 1859 { 1860 if (!sock) 1861 return -EINVAL; 1862 1863 if (protocol && protocol != PF_VSOCK) 1864 return -EPROTONOSUPPORT; 1865 1866 switch (sock->type) { 1867 case SOCK_DGRAM: 1868 sock->ops = &vsock_dgram_ops; 1869 break; 1870 case SOCK_STREAM: 1871 sock->ops = &vsock_stream_ops; 1872 break; 1873 default: 1874 return -ESOCKTNOSUPPORT; 1875 } 1876 1877 sock->state = SS_UNCONNECTED; 1878 1879 return __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern) ? 0 : -ENOMEM; 1880 } 1881 1882 static const struct net_proto_family vsock_family_ops = { 1883 .family = AF_VSOCK, 1884 .create = vsock_create, 1885 .owner = THIS_MODULE, 1886 }; 1887 1888 static long vsock_dev_do_ioctl(struct file *filp, 1889 unsigned int cmd, void __user *ptr) 1890 { 1891 u32 __user *p = ptr; 1892 int retval = 0; 1893 1894 switch (cmd) { 1895 case IOCTL_VM_SOCKETS_GET_LOCAL_CID: 1896 if (put_user(transport->get_local_cid(), p) != 0) 1897 retval = -EFAULT; 1898 break; 1899 1900 default: 1901 pr_err("Unknown ioctl %d\n", cmd); 1902 retval = -EINVAL; 1903 } 1904 1905 return retval; 1906 } 1907 1908 static long vsock_dev_ioctl(struct file *filp, 1909 unsigned int cmd, unsigned long arg) 1910 { 1911 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg); 1912 } 1913 1914 #ifdef CONFIG_COMPAT 1915 static long vsock_dev_compat_ioctl(struct file *filp, 1916 unsigned int cmd, unsigned long arg) 1917 { 1918 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg)); 1919 } 1920 #endif 1921 1922 static const struct file_operations vsock_device_ops = { 1923 .owner = THIS_MODULE, 1924 .unlocked_ioctl = vsock_dev_ioctl, 1925 #ifdef CONFIG_COMPAT 1926 .compat_ioctl = vsock_dev_compat_ioctl, 1927 #endif 1928 .open = nonseekable_open, 1929 }; 1930 1931 static struct miscdevice vsock_device = { 1932 .name = "vsock", 1933 .fops = &vsock_device_ops, 1934 }; 1935 1936 int __vsock_core_init(const struct vsock_transport *t, struct module *owner) 1937 { 1938 int err = mutex_lock_interruptible(&vsock_register_mutex); 1939 1940 if (err) 1941 return err; 1942 1943 if (transport) { 1944 err = -EBUSY; 1945 goto err_busy; 1946 } 1947 1948 /* Transport must be the owner of the protocol so that it can't 1949 * unload while there are open sockets. 1950 */ 1951 vsock_proto.owner = owner; 1952 transport = t; 1953 1954 vsock_device.minor = MISC_DYNAMIC_MINOR; 1955 err = misc_register(&vsock_device); 1956 if (err) { 1957 pr_err("Failed to register misc device\n"); 1958 goto err_reset_transport; 1959 } 1960 1961 err = proto_register(&vsock_proto, 1); /* we want our slab */ 1962 if (err) { 1963 pr_err("Cannot register vsock protocol\n"); 1964 goto err_deregister_misc; 1965 } 1966 1967 err = sock_register(&vsock_family_ops); 1968 if (err) { 1969 pr_err("could not register af_vsock (%d) address family: %d\n", 1970 AF_VSOCK, err); 1971 goto err_unregister_proto; 1972 } 1973 1974 mutex_unlock(&vsock_register_mutex); 1975 return 0; 1976 1977 err_unregister_proto: 1978 proto_unregister(&vsock_proto); 1979 err_deregister_misc: 1980 misc_deregister(&vsock_device); 1981 err_reset_transport: 1982 transport = NULL; 1983 err_busy: 1984 mutex_unlock(&vsock_register_mutex); 1985 return err; 1986 } 1987 EXPORT_SYMBOL_GPL(__vsock_core_init); 1988 1989 void vsock_core_exit(void) 1990 { 1991 mutex_lock(&vsock_register_mutex); 1992 1993 misc_deregister(&vsock_device); 1994 sock_unregister(AF_VSOCK); 1995 proto_unregister(&vsock_proto); 1996 1997 /* We do not want the assignment below re-ordered. */ 1998 mb(); 1999 transport = NULL; 2000 2001 mutex_unlock(&vsock_register_mutex); 2002 } 2003 EXPORT_SYMBOL_GPL(vsock_core_exit); 2004 2005 const struct vsock_transport *vsock_core_get_transport(void) 2006 { 2007 /* vsock_register_mutex not taken since only the transport uses this 2008 * function and only while registered. 2009 */ 2010 return transport; 2011 } 2012 EXPORT_SYMBOL_GPL(vsock_core_get_transport); 2013 2014 static void __exit vsock_exit(void) 2015 { 2016 /* Do nothing. This function makes this module removable. */ 2017 } 2018 2019 module_init(vsock_init_tables); 2020 module_exit(vsock_exit); 2021 2022 MODULE_AUTHOR("VMware, Inc."); 2023 MODULE_DESCRIPTION("VMware Virtual Socket Family"); 2024 MODULE_VERSION("1.0.2.0-k"); 2025 MODULE_LICENSE("GPL v2"); 2026