1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * VMware vSockets Driver 4 * 5 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved. 6 */ 7 8 /* Implementation notes: 9 * 10 * - There are two kinds of sockets: those created by user action (such as 11 * calling socket(2)) and those created by incoming connection request packets. 12 * 13 * - There are two "global" tables, one for bound sockets (sockets that have 14 * specified an address that they are responsible for) and one for connected 15 * sockets (sockets that have established a connection with another socket). 16 * These tables are "global" in that all sockets on the system are placed 17 * within them. - Note, though, that the bound table contains an extra entry 18 * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in 19 * that list. The bound table is used solely for lookup of sockets when packets 20 * are received and that's not necessary for SOCK_DGRAM sockets since we create 21 * a datagram handle for each and need not perform a lookup. Keeping SOCK_DGRAM 22 * sockets out of the bound hash buckets will reduce the chance of collisions 23 * when looking for SOCK_STREAM sockets and prevents us from having to check the 24 * socket type in the hash table lookups. 25 * 26 * - Sockets created by user action will either be "client" sockets that 27 * initiate a connection or "server" sockets that listen for connections; we do 28 * not support simultaneous connects (two "client" sockets connecting). 29 * 30 * - "Server" sockets are referred to as listener sockets throughout this 31 * implementation because they are in the TCP_LISTEN state. When a 32 * connection request is received (the second kind of socket mentioned above), 33 * we create a new socket and refer to it as a pending socket. These pending 34 * sockets are placed on the pending connection list of the listener socket. 35 * When future packets are received for the address the listener socket is 36 * bound to, we check if the source of the packet is from one that has an 37 * existing pending connection. If it does, we process the packet for the 38 * pending socket. When that socket reaches the connected state, it is removed 39 * from the listener socket's pending list and enqueued in the listener 40 * socket's accept queue. Callers of accept(2) will accept connected sockets 41 * from the listener socket's accept queue. If the socket cannot be accepted 42 * for some reason then it is marked rejected. Once the connection is 43 * accepted, it is owned by the user process and the responsibility for cleanup 44 * falls with that user process. 45 * 46 * - It is possible that these pending sockets will never reach the connected 47 * state; in fact, we may never receive another packet after the connection 48 * request. Because of this, we must schedule a cleanup function to run in the 49 * future, after some amount of time passes where a connection should have been 50 * established. This function ensures that the socket is off all lists so it 51 * cannot be retrieved, then drops all references to the socket so it is cleaned 52 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this 53 * function will also cleanup rejected sockets, those that reach the connected 54 * state but leave it before they have been accepted. 55 * 56 * - Lock ordering for pending or accept queue sockets is: 57 * 58 * lock_sock(listener); 59 * lock_sock_nested(pending, SINGLE_DEPTH_NESTING); 60 * 61 * Using explicit nested locking keeps lockdep happy since normally only one 62 * lock of a given class may be taken at a time. 63 * 64 * - Sockets created by user action will be cleaned up when the user process 65 * calls close(2), causing our release implementation to be called. Our release 66 * implementation will perform some cleanup then drop the last reference so our 67 * sk_destruct implementation is invoked. Our sk_destruct implementation will 68 * perform additional cleanup that's common for both types of sockets. 69 * 70 * - A socket's reference count is what ensures that the structure won't be 71 * freed. Each entry in a list (such as the "global" bound and connected tables 72 * and the listener socket's pending list and connected queue) ensures a 73 * reference. When we defer work until process context and pass a socket as our 74 * argument, we must ensure the reference count is increased to ensure the 75 * socket isn't freed before the function is run; the deferred function will 76 * then drop the reference. 77 * 78 * - sk->sk_state uses the TCP state constants because they are widely used by 79 * other address families and exposed to userspace tools like ss(8): 80 * 81 * TCP_CLOSE - unconnected 82 * TCP_SYN_SENT - connecting 83 * TCP_ESTABLISHED - connected 84 * TCP_CLOSING - disconnecting 85 * TCP_LISTEN - listening 86 */ 87 88 #include <linux/types.h> 89 #include <linux/bitops.h> 90 #include <linux/cred.h> 91 #include <linux/init.h> 92 #include <linux/io.h> 93 #include <linux/kernel.h> 94 #include <linux/sched/signal.h> 95 #include <linux/kmod.h> 96 #include <linux/list.h> 97 #include <linux/miscdevice.h> 98 #include <linux/module.h> 99 #include <linux/mutex.h> 100 #include <linux/net.h> 101 #include <linux/poll.h> 102 #include <linux/random.h> 103 #include <linux/skbuff.h> 104 #include <linux/smp.h> 105 #include <linux/socket.h> 106 #include <linux/stddef.h> 107 #include <linux/unistd.h> 108 #include <linux/wait.h> 109 #include <linux/workqueue.h> 110 #include <net/sock.h> 111 #include <net/af_vsock.h> 112 113 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr); 114 static void vsock_sk_destruct(struct sock *sk); 115 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 116 117 /* Protocol family. */ 118 static struct proto vsock_proto = { 119 .name = "AF_VSOCK", 120 .owner = THIS_MODULE, 121 .obj_size = sizeof(struct vsock_sock), 122 }; 123 124 /* The default peer timeout indicates how long we will wait for a peer response 125 * to a control message. 126 */ 127 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ) 128 129 #define VSOCK_DEFAULT_BUFFER_SIZE (1024 * 256) 130 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256) 131 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128 132 133 static const struct vsock_transport *transport_single; 134 static DEFINE_MUTEX(vsock_register_mutex); 135 136 /**** UTILS ****/ 137 138 /* Each bound VSocket is stored in the bind hash table and each connected 139 * VSocket is stored in the connected hash table. 140 * 141 * Unbound sockets are all put on the same list attached to the end of the hash 142 * table (vsock_unbound_sockets). Bound sockets are added to the hash table in 143 * the bucket that their local address hashes to (vsock_bound_sockets(addr) 144 * represents the list that addr hashes to). 145 * 146 * Specifically, we initialize the vsock_bind_table array to a size of 147 * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through 148 * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and 149 * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets. The hash function 150 * mods with VSOCK_HASH_SIZE to ensure this. 151 */ 152 #define MAX_PORT_RETRIES 24 153 154 #define VSOCK_HASH(addr) ((addr)->svm_port % VSOCK_HASH_SIZE) 155 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)]) 156 #define vsock_unbound_sockets (&vsock_bind_table[VSOCK_HASH_SIZE]) 157 158 /* XXX This can probably be implemented in a better way. */ 159 #define VSOCK_CONN_HASH(src, dst) \ 160 (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE) 161 #define vsock_connected_sockets(src, dst) \ 162 (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)]) 163 #define vsock_connected_sockets_vsk(vsk) \ 164 vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr) 165 166 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1]; 167 EXPORT_SYMBOL_GPL(vsock_bind_table); 168 struct list_head vsock_connected_table[VSOCK_HASH_SIZE]; 169 EXPORT_SYMBOL_GPL(vsock_connected_table); 170 DEFINE_SPINLOCK(vsock_table_lock); 171 EXPORT_SYMBOL_GPL(vsock_table_lock); 172 173 /* Autobind this socket to the local address if necessary. */ 174 static int vsock_auto_bind(struct vsock_sock *vsk) 175 { 176 struct sock *sk = sk_vsock(vsk); 177 struct sockaddr_vm local_addr; 178 179 if (vsock_addr_bound(&vsk->local_addr)) 180 return 0; 181 vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 182 return __vsock_bind(sk, &local_addr); 183 } 184 185 static int __init vsock_init_tables(void) 186 { 187 int i; 188 189 for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++) 190 INIT_LIST_HEAD(&vsock_bind_table[i]); 191 192 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) 193 INIT_LIST_HEAD(&vsock_connected_table[i]); 194 return 0; 195 } 196 197 static void __vsock_insert_bound(struct list_head *list, 198 struct vsock_sock *vsk) 199 { 200 sock_hold(&vsk->sk); 201 list_add(&vsk->bound_table, list); 202 } 203 204 static void __vsock_insert_connected(struct list_head *list, 205 struct vsock_sock *vsk) 206 { 207 sock_hold(&vsk->sk); 208 list_add(&vsk->connected_table, list); 209 } 210 211 static void __vsock_remove_bound(struct vsock_sock *vsk) 212 { 213 list_del_init(&vsk->bound_table); 214 sock_put(&vsk->sk); 215 } 216 217 static void __vsock_remove_connected(struct vsock_sock *vsk) 218 { 219 list_del_init(&vsk->connected_table); 220 sock_put(&vsk->sk); 221 } 222 223 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr) 224 { 225 struct vsock_sock *vsk; 226 227 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) 228 if (addr->svm_port == vsk->local_addr.svm_port) 229 return sk_vsock(vsk); 230 231 return NULL; 232 } 233 234 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src, 235 struct sockaddr_vm *dst) 236 { 237 struct vsock_sock *vsk; 238 239 list_for_each_entry(vsk, vsock_connected_sockets(src, dst), 240 connected_table) { 241 if (vsock_addr_equals_addr(src, &vsk->remote_addr) && 242 dst->svm_port == vsk->local_addr.svm_port) { 243 return sk_vsock(vsk); 244 } 245 } 246 247 return NULL; 248 } 249 250 static void vsock_insert_unbound(struct vsock_sock *vsk) 251 { 252 spin_lock_bh(&vsock_table_lock); 253 __vsock_insert_bound(vsock_unbound_sockets, vsk); 254 spin_unlock_bh(&vsock_table_lock); 255 } 256 257 void vsock_insert_connected(struct vsock_sock *vsk) 258 { 259 struct list_head *list = vsock_connected_sockets( 260 &vsk->remote_addr, &vsk->local_addr); 261 262 spin_lock_bh(&vsock_table_lock); 263 __vsock_insert_connected(list, vsk); 264 spin_unlock_bh(&vsock_table_lock); 265 } 266 EXPORT_SYMBOL_GPL(vsock_insert_connected); 267 268 void vsock_remove_bound(struct vsock_sock *vsk) 269 { 270 spin_lock_bh(&vsock_table_lock); 271 if (__vsock_in_bound_table(vsk)) 272 __vsock_remove_bound(vsk); 273 spin_unlock_bh(&vsock_table_lock); 274 } 275 EXPORT_SYMBOL_GPL(vsock_remove_bound); 276 277 void vsock_remove_connected(struct vsock_sock *vsk) 278 { 279 spin_lock_bh(&vsock_table_lock); 280 if (__vsock_in_connected_table(vsk)) 281 __vsock_remove_connected(vsk); 282 spin_unlock_bh(&vsock_table_lock); 283 } 284 EXPORT_SYMBOL_GPL(vsock_remove_connected); 285 286 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr) 287 { 288 struct sock *sk; 289 290 spin_lock_bh(&vsock_table_lock); 291 sk = __vsock_find_bound_socket(addr); 292 if (sk) 293 sock_hold(sk); 294 295 spin_unlock_bh(&vsock_table_lock); 296 297 return sk; 298 } 299 EXPORT_SYMBOL_GPL(vsock_find_bound_socket); 300 301 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src, 302 struct sockaddr_vm *dst) 303 { 304 struct sock *sk; 305 306 spin_lock_bh(&vsock_table_lock); 307 sk = __vsock_find_connected_socket(src, dst); 308 if (sk) 309 sock_hold(sk); 310 311 spin_unlock_bh(&vsock_table_lock); 312 313 return sk; 314 } 315 EXPORT_SYMBOL_GPL(vsock_find_connected_socket); 316 317 void vsock_remove_sock(struct vsock_sock *vsk) 318 { 319 vsock_remove_bound(vsk); 320 vsock_remove_connected(vsk); 321 } 322 EXPORT_SYMBOL_GPL(vsock_remove_sock); 323 324 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk)) 325 { 326 int i; 327 328 spin_lock_bh(&vsock_table_lock); 329 330 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) { 331 struct vsock_sock *vsk; 332 list_for_each_entry(vsk, &vsock_connected_table[i], 333 connected_table) 334 fn(sk_vsock(vsk)); 335 } 336 337 spin_unlock_bh(&vsock_table_lock); 338 } 339 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket); 340 341 void vsock_add_pending(struct sock *listener, struct sock *pending) 342 { 343 struct vsock_sock *vlistener; 344 struct vsock_sock *vpending; 345 346 vlistener = vsock_sk(listener); 347 vpending = vsock_sk(pending); 348 349 sock_hold(pending); 350 sock_hold(listener); 351 list_add_tail(&vpending->pending_links, &vlistener->pending_links); 352 } 353 EXPORT_SYMBOL_GPL(vsock_add_pending); 354 355 void vsock_remove_pending(struct sock *listener, struct sock *pending) 356 { 357 struct vsock_sock *vpending = vsock_sk(pending); 358 359 list_del_init(&vpending->pending_links); 360 sock_put(listener); 361 sock_put(pending); 362 } 363 EXPORT_SYMBOL_GPL(vsock_remove_pending); 364 365 void vsock_enqueue_accept(struct sock *listener, struct sock *connected) 366 { 367 struct vsock_sock *vlistener; 368 struct vsock_sock *vconnected; 369 370 vlistener = vsock_sk(listener); 371 vconnected = vsock_sk(connected); 372 373 sock_hold(connected); 374 sock_hold(listener); 375 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue); 376 } 377 EXPORT_SYMBOL_GPL(vsock_enqueue_accept); 378 379 static struct sock *vsock_dequeue_accept(struct sock *listener) 380 { 381 struct vsock_sock *vlistener; 382 struct vsock_sock *vconnected; 383 384 vlistener = vsock_sk(listener); 385 386 if (list_empty(&vlistener->accept_queue)) 387 return NULL; 388 389 vconnected = list_entry(vlistener->accept_queue.next, 390 struct vsock_sock, accept_queue); 391 392 list_del_init(&vconnected->accept_queue); 393 sock_put(listener); 394 /* The caller will need a reference on the connected socket so we let 395 * it call sock_put(). 396 */ 397 398 return sk_vsock(vconnected); 399 } 400 401 static bool vsock_is_accept_queue_empty(struct sock *sk) 402 { 403 struct vsock_sock *vsk = vsock_sk(sk); 404 return list_empty(&vsk->accept_queue); 405 } 406 407 static bool vsock_is_pending(struct sock *sk) 408 { 409 struct vsock_sock *vsk = vsock_sk(sk); 410 return !list_empty(&vsk->pending_links); 411 } 412 413 static int vsock_send_shutdown(struct sock *sk, int mode) 414 { 415 struct vsock_sock *vsk = vsock_sk(sk); 416 417 return vsk->transport->shutdown(vsk, mode); 418 } 419 420 static void vsock_pending_work(struct work_struct *work) 421 { 422 struct sock *sk; 423 struct sock *listener; 424 struct vsock_sock *vsk; 425 bool cleanup; 426 427 vsk = container_of(work, struct vsock_sock, pending_work.work); 428 sk = sk_vsock(vsk); 429 listener = vsk->listener; 430 cleanup = true; 431 432 lock_sock(listener); 433 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 434 435 if (vsock_is_pending(sk)) { 436 vsock_remove_pending(listener, sk); 437 438 sk_acceptq_removed(listener); 439 } else if (!vsk->rejected) { 440 /* We are not on the pending list and accept() did not reject 441 * us, so we must have been accepted by our user process. We 442 * just need to drop our references to the sockets and be on 443 * our way. 444 */ 445 cleanup = false; 446 goto out; 447 } 448 449 /* We need to remove ourself from the global connected sockets list so 450 * incoming packets can't find this socket, and to reduce the reference 451 * count. 452 */ 453 vsock_remove_connected(vsk); 454 455 sk->sk_state = TCP_CLOSE; 456 457 out: 458 release_sock(sk); 459 release_sock(listener); 460 if (cleanup) 461 sock_put(sk); 462 463 sock_put(sk); 464 sock_put(listener); 465 } 466 467 /**** SOCKET OPERATIONS ****/ 468 469 static int __vsock_bind_stream(struct vsock_sock *vsk, 470 struct sockaddr_vm *addr) 471 { 472 static u32 port; 473 struct sockaddr_vm new_addr; 474 475 if (!port) 476 port = LAST_RESERVED_PORT + 1 + 477 prandom_u32_max(U32_MAX - LAST_RESERVED_PORT); 478 479 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port); 480 481 if (addr->svm_port == VMADDR_PORT_ANY) { 482 bool found = false; 483 unsigned int i; 484 485 for (i = 0; i < MAX_PORT_RETRIES; i++) { 486 if (port <= LAST_RESERVED_PORT) 487 port = LAST_RESERVED_PORT + 1; 488 489 new_addr.svm_port = port++; 490 491 if (!__vsock_find_bound_socket(&new_addr)) { 492 found = true; 493 break; 494 } 495 } 496 497 if (!found) 498 return -EADDRNOTAVAIL; 499 } else { 500 /* If port is in reserved range, ensure caller 501 * has necessary privileges. 502 */ 503 if (addr->svm_port <= LAST_RESERVED_PORT && 504 !capable(CAP_NET_BIND_SERVICE)) { 505 return -EACCES; 506 } 507 508 if (__vsock_find_bound_socket(&new_addr)) 509 return -EADDRINUSE; 510 } 511 512 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port); 513 514 /* Remove stream sockets from the unbound list and add them to the hash 515 * table for easy lookup by its address. The unbound list is simply an 516 * extra entry at the end of the hash table, a trick used by AF_UNIX. 517 */ 518 __vsock_remove_bound(vsk); 519 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk); 520 521 return 0; 522 } 523 524 static int __vsock_bind_dgram(struct vsock_sock *vsk, 525 struct sockaddr_vm *addr) 526 { 527 return vsk->transport->dgram_bind(vsk, addr); 528 } 529 530 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr) 531 { 532 struct vsock_sock *vsk = vsock_sk(sk); 533 u32 cid; 534 int retval; 535 536 /* First ensure this socket isn't already bound. */ 537 if (vsock_addr_bound(&vsk->local_addr)) 538 return -EINVAL; 539 540 /* Now bind to the provided address or select appropriate values if 541 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that 542 * like AF_INET prevents binding to a non-local IP address (in most 543 * cases), we only allow binding to the local CID. 544 */ 545 cid = vsk->transport->get_local_cid(); 546 if (addr->svm_cid != cid && addr->svm_cid != VMADDR_CID_ANY) 547 return -EADDRNOTAVAIL; 548 549 switch (sk->sk_socket->type) { 550 case SOCK_STREAM: 551 spin_lock_bh(&vsock_table_lock); 552 retval = __vsock_bind_stream(vsk, addr); 553 spin_unlock_bh(&vsock_table_lock); 554 break; 555 556 case SOCK_DGRAM: 557 retval = __vsock_bind_dgram(vsk, addr); 558 break; 559 560 default: 561 retval = -EINVAL; 562 break; 563 } 564 565 return retval; 566 } 567 568 static void vsock_connect_timeout(struct work_struct *work); 569 570 static struct sock *__vsock_create(struct net *net, 571 struct socket *sock, 572 struct sock *parent, 573 gfp_t priority, 574 unsigned short type, 575 int kern) 576 { 577 struct sock *sk; 578 struct vsock_sock *psk; 579 struct vsock_sock *vsk; 580 581 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern); 582 if (!sk) 583 return NULL; 584 585 sock_init_data(sock, sk); 586 587 /* sk->sk_type is normally set in sock_init_data, but only if sock is 588 * non-NULL. We make sure that our sockets always have a type by 589 * setting it here if needed. 590 */ 591 if (!sock) 592 sk->sk_type = type; 593 594 vsk = vsock_sk(sk); 595 vsk->transport = transport_single; 596 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 597 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 598 599 sk->sk_destruct = vsock_sk_destruct; 600 sk->sk_backlog_rcv = vsock_queue_rcv_skb; 601 sock_reset_flag(sk, SOCK_DONE); 602 603 INIT_LIST_HEAD(&vsk->bound_table); 604 INIT_LIST_HEAD(&vsk->connected_table); 605 vsk->listener = NULL; 606 INIT_LIST_HEAD(&vsk->pending_links); 607 INIT_LIST_HEAD(&vsk->accept_queue); 608 vsk->rejected = false; 609 vsk->sent_request = false; 610 vsk->ignore_connecting_rst = false; 611 vsk->peer_shutdown = 0; 612 INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout); 613 INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work); 614 615 psk = parent ? vsock_sk(parent) : NULL; 616 if (parent) { 617 vsk->trusted = psk->trusted; 618 vsk->owner = get_cred(psk->owner); 619 vsk->connect_timeout = psk->connect_timeout; 620 vsk->buffer_size = psk->buffer_size; 621 vsk->buffer_min_size = psk->buffer_min_size; 622 vsk->buffer_max_size = psk->buffer_max_size; 623 } else { 624 vsk->trusted = capable(CAP_NET_ADMIN); 625 vsk->owner = get_current_cred(); 626 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT; 627 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE; 628 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE; 629 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE; 630 } 631 632 if (vsk->transport->init(vsk, psk) < 0) { 633 sk_free(sk); 634 return NULL; 635 } 636 637 return sk; 638 } 639 640 static void __vsock_release(struct sock *sk, int level) 641 { 642 if (sk) { 643 struct sock *pending; 644 struct vsock_sock *vsk; 645 646 vsk = vsock_sk(sk); 647 pending = NULL; /* Compiler warning. */ 648 649 /* The release call is supposed to use lock_sock_nested() 650 * rather than lock_sock(), if a sock lock should be acquired. 651 */ 652 vsk->transport->release(vsk); 653 654 /* When "level" is SINGLE_DEPTH_NESTING, use the nested 655 * version to avoid the warning "possible recursive locking 656 * detected". When "level" is 0, lock_sock_nested(sk, level) 657 * is the same as lock_sock(sk). 658 */ 659 lock_sock_nested(sk, level); 660 sock_orphan(sk); 661 sk->sk_shutdown = SHUTDOWN_MASK; 662 663 skb_queue_purge(&sk->sk_receive_queue); 664 665 /* Clean up any sockets that never were accepted. */ 666 while ((pending = vsock_dequeue_accept(sk)) != NULL) { 667 __vsock_release(pending, SINGLE_DEPTH_NESTING); 668 sock_put(pending); 669 } 670 671 release_sock(sk); 672 sock_put(sk); 673 } 674 } 675 676 static void vsock_sk_destruct(struct sock *sk) 677 { 678 struct vsock_sock *vsk = vsock_sk(sk); 679 680 vsk->transport->destruct(vsk); 681 682 /* When clearing these addresses, there's no need to set the family and 683 * possibly register the address family with the kernel. 684 */ 685 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 686 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 687 688 put_cred(vsk->owner); 689 } 690 691 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 692 { 693 int err; 694 695 err = sock_queue_rcv_skb(sk, skb); 696 if (err) 697 kfree_skb(skb); 698 699 return err; 700 } 701 702 struct sock *vsock_create_connected(struct sock *parent) 703 { 704 return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL, 705 parent->sk_type, 0); 706 } 707 EXPORT_SYMBOL_GPL(vsock_create_connected); 708 709 s64 vsock_stream_has_data(struct vsock_sock *vsk) 710 { 711 return vsk->transport->stream_has_data(vsk); 712 } 713 EXPORT_SYMBOL_GPL(vsock_stream_has_data); 714 715 s64 vsock_stream_has_space(struct vsock_sock *vsk) 716 { 717 return vsk->transport->stream_has_space(vsk); 718 } 719 EXPORT_SYMBOL_GPL(vsock_stream_has_space); 720 721 static int vsock_release(struct socket *sock) 722 { 723 __vsock_release(sock->sk, 0); 724 sock->sk = NULL; 725 sock->state = SS_FREE; 726 727 return 0; 728 } 729 730 static int 731 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 732 { 733 int err; 734 struct sock *sk; 735 struct sockaddr_vm *vm_addr; 736 737 sk = sock->sk; 738 739 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0) 740 return -EINVAL; 741 742 lock_sock(sk); 743 err = __vsock_bind(sk, vm_addr); 744 release_sock(sk); 745 746 return err; 747 } 748 749 static int vsock_getname(struct socket *sock, 750 struct sockaddr *addr, int peer) 751 { 752 int err; 753 struct sock *sk; 754 struct vsock_sock *vsk; 755 struct sockaddr_vm *vm_addr; 756 757 sk = sock->sk; 758 vsk = vsock_sk(sk); 759 err = 0; 760 761 lock_sock(sk); 762 763 if (peer) { 764 if (sock->state != SS_CONNECTED) { 765 err = -ENOTCONN; 766 goto out; 767 } 768 vm_addr = &vsk->remote_addr; 769 } else { 770 vm_addr = &vsk->local_addr; 771 } 772 773 if (!vm_addr) { 774 err = -EINVAL; 775 goto out; 776 } 777 778 /* sys_getsockname() and sys_getpeername() pass us a 779 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately 780 * that macro is defined in socket.c instead of .h, so we hardcode its 781 * value here. 782 */ 783 BUILD_BUG_ON(sizeof(*vm_addr) > 128); 784 memcpy(addr, vm_addr, sizeof(*vm_addr)); 785 err = sizeof(*vm_addr); 786 787 out: 788 release_sock(sk); 789 return err; 790 } 791 792 static int vsock_shutdown(struct socket *sock, int mode) 793 { 794 int err; 795 struct sock *sk; 796 797 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses 798 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode 799 * here like the other address families do. Note also that the 800 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3), 801 * which is what we want. 802 */ 803 mode++; 804 805 if ((mode & ~SHUTDOWN_MASK) || !mode) 806 return -EINVAL; 807 808 /* If this is a STREAM socket and it is not connected then bail out 809 * immediately. If it is a DGRAM socket then we must first kick the 810 * socket so that it wakes up from any sleeping calls, for example 811 * recv(), and then afterwards return the error. 812 */ 813 814 sk = sock->sk; 815 if (sock->state == SS_UNCONNECTED) { 816 err = -ENOTCONN; 817 if (sk->sk_type == SOCK_STREAM) 818 return err; 819 } else { 820 sock->state = SS_DISCONNECTING; 821 err = 0; 822 } 823 824 /* Receive and send shutdowns are treated alike. */ 825 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN); 826 if (mode) { 827 lock_sock(sk); 828 sk->sk_shutdown |= mode; 829 sk->sk_state_change(sk); 830 release_sock(sk); 831 832 if (sk->sk_type == SOCK_STREAM) { 833 sock_reset_flag(sk, SOCK_DONE); 834 vsock_send_shutdown(sk, mode); 835 } 836 } 837 838 return err; 839 } 840 841 static __poll_t vsock_poll(struct file *file, struct socket *sock, 842 poll_table *wait) 843 { 844 struct sock *sk; 845 __poll_t mask; 846 struct vsock_sock *vsk; 847 848 sk = sock->sk; 849 vsk = vsock_sk(sk); 850 851 poll_wait(file, sk_sleep(sk), wait); 852 mask = 0; 853 854 if (sk->sk_err) 855 /* Signify that there has been an error on this socket. */ 856 mask |= EPOLLERR; 857 858 /* INET sockets treat local write shutdown and peer write shutdown as a 859 * case of EPOLLHUP set. 860 */ 861 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 862 ((sk->sk_shutdown & SEND_SHUTDOWN) && 863 (vsk->peer_shutdown & SEND_SHUTDOWN))) { 864 mask |= EPOLLHUP; 865 } 866 867 if (sk->sk_shutdown & RCV_SHUTDOWN || 868 vsk->peer_shutdown & SEND_SHUTDOWN) { 869 mask |= EPOLLRDHUP; 870 } 871 872 if (sock->type == SOCK_DGRAM) { 873 /* For datagram sockets we can read if there is something in 874 * the queue and write as long as the socket isn't shutdown for 875 * sending. 876 */ 877 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) || 878 (sk->sk_shutdown & RCV_SHUTDOWN)) { 879 mask |= EPOLLIN | EPOLLRDNORM; 880 } 881 882 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 883 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND; 884 885 } else if (sock->type == SOCK_STREAM) { 886 const struct vsock_transport *transport = vsk->transport; 887 lock_sock(sk); 888 889 /* Listening sockets that have connections in their accept 890 * queue can be read. 891 */ 892 if (sk->sk_state == TCP_LISTEN 893 && !vsock_is_accept_queue_empty(sk)) 894 mask |= EPOLLIN | EPOLLRDNORM; 895 896 /* If there is something in the queue then we can read. */ 897 if (transport->stream_is_active(vsk) && 898 !(sk->sk_shutdown & RCV_SHUTDOWN)) { 899 bool data_ready_now = false; 900 int ret = transport->notify_poll_in( 901 vsk, 1, &data_ready_now); 902 if (ret < 0) { 903 mask |= EPOLLERR; 904 } else { 905 if (data_ready_now) 906 mask |= EPOLLIN | EPOLLRDNORM; 907 908 } 909 } 910 911 /* Sockets whose connections have been closed, reset, or 912 * terminated should also be considered read, and we check the 913 * shutdown flag for that. 914 */ 915 if (sk->sk_shutdown & RCV_SHUTDOWN || 916 vsk->peer_shutdown & SEND_SHUTDOWN) { 917 mask |= EPOLLIN | EPOLLRDNORM; 918 } 919 920 /* Connected sockets that can produce data can be written. */ 921 if (sk->sk_state == TCP_ESTABLISHED) { 922 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 923 bool space_avail_now = false; 924 int ret = transport->notify_poll_out( 925 vsk, 1, &space_avail_now); 926 if (ret < 0) { 927 mask |= EPOLLERR; 928 } else { 929 if (space_avail_now) 930 /* Remove EPOLLWRBAND since INET 931 * sockets are not setting it. 932 */ 933 mask |= EPOLLOUT | EPOLLWRNORM; 934 935 } 936 } 937 } 938 939 /* Simulate INET socket poll behaviors, which sets 940 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read, 941 * but local send is not shutdown. 942 */ 943 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) { 944 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 945 mask |= EPOLLOUT | EPOLLWRNORM; 946 947 } 948 949 release_sock(sk); 950 } 951 952 return mask; 953 } 954 955 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg, 956 size_t len) 957 { 958 int err; 959 struct sock *sk; 960 struct vsock_sock *vsk; 961 struct sockaddr_vm *remote_addr; 962 const struct vsock_transport *transport; 963 964 if (msg->msg_flags & MSG_OOB) 965 return -EOPNOTSUPP; 966 967 /* For now, MSG_DONTWAIT is always assumed... */ 968 err = 0; 969 sk = sock->sk; 970 vsk = vsock_sk(sk); 971 transport = vsk->transport; 972 973 lock_sock(sk); 974 975 err = vsock_auto_bind(vsk); 976 if (err) 977 goto out; 978 979 980 /* If the provided message contains an address, use that. Otherwise 981 * fall back on the socket's remote handle (if it has been connected). 982 */ 983 if (msg->msg_name && 984 vsock_addr_cast(msg->msg_name, msg->msg_namelen, 985 &remote_addr) == 0) { 986 /* Ensure this address is of the right type and is a valid 987 * destination. 988 */ 989 990 if (remote_addr->svm_cid == VMADDR_CID_ANY) 991 remote_addr->svm_cid = transport->get_local_cid(); 992 993 if (!vsock_addr_bound(remote_addr)) { 994 err = -EINVAL; 995 goto out; 996 } 997 } else if (sock->state == SS_CONNECTED) { 998 remote_addr = &vsk->remote_addr; 999 1000 if (remote_addr->svm_cid == VMADDR_CID_ANY) 1001 remote_addr->svm_cid = transport->get_local_cid(); 1002 1003 /* XXX Should connect() or this function ensure remote_addr is 1004 * bound? 1005 */ 1006 if (!vsock_addr_bound(&vsk->remote_addr)) { 1007 err = -EINVAL; 1008 goto out; 1009 } 1010 } else { 1011 err = -EINVAL; 1012 goto out; 1013 } 1014 1015 if (!transport->dgram_allow(remote_addr->svm_cid, 1016 remote_addr->svm_port)) { 1017 err = -EINVAL; 1018 goto out; 1019 } 1020 1021 err = transport->dgram_enqueue(vsk, remote_addr, msg, len); 1022 1023 out: 1024 release_sock(sk); 1025 return err; 1026 } 1027 1028 static int vsock_dgram_connect(struct socket *sock, 1029 struct sockaddr *addr, int addr_len, int flags) 1030 { 1031 int err; 1032 struct sock *sk; 1033 struct vsock_sock *vsk; 1034 struct sockaddr_vm *remote_addr; 1035 1036 sk = sock->sk; 1037 vsk = vsock_sk(sk); 1038 1039 err = vsock_addr_cast(addr, addr_len, &remote_addr); 1040 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) { 1041 lock_sock(sk); 1042 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, 1043 VMADDR_PORT_ANY); 1044 sock->state = SS_UNCONNECTED; 1045 release_sock(sk); 1046 return 0; 1047 } else if (err != 0) 1048 return -EINVAL; 1049 1050 lock_sock(sk); 1051 1052 err = vsock_auto_bind(vsk); 1053 if (err) 1054 goto out; 1055 1056 if (!vsk->transport->dgram_allow(remote_addr->svm_cid, 1057 remote_addr->svm_port)) { 1058 err = -EINVAL; 1059 goto out; 1060 } 1061 1062 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr)); 1063 sock->state = SS_CONNECTED; 1064 1065 out: 1066 release_sock(sk); 1067 return err; 1068 } 1069 1070 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg, 1071 size_t len, int flags) 1072 { 1073 struct vsock_sock *vsk = vsock_sk(sock->sk); 1074 1075 return vsk->transport->dgram_dequeue(vsk, 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 = vsock_poll, 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 const struct vsock_transport *transport = vsk->transport; 1102 1103 if (!transport->cancel_pkt) 1104 return -EOPNOTSUPP; 1105 1106 return transport->cancel_pkt(vsk); 1107 } 1108 1109 static void vsock_connect_timeout(struct work_struct *work) 1110 { 1111 struct sock *sk; 1112 struct vsock_sock *vsk; 1113 int cancel = 0; 1114 1115 vsk = container_of(work, struct vsock_sock, connect_work.work); 1116 sk = sk_vsock(vsk); 1117 1118 lock_sock(sk); 1119 if (sk->sk_state == TCP_SYN_SENT && 1120 (sk->sk_shutdown != SHUTDOWN_MASK)) { 1121 sk->sk_state = TCP_CLOSE; 1122 sk->sk_err = ETIMEDOUT; 1123 sk->sk_error_report(sk); 1124 cancel = 1; 1125 } 1126 release_sock(sk); 1127 if (cancel) 1128 vsock_transport_cancel_pkt(vsk); 1129 1130 sock_put(sk); 1131 } 1132 1133 static int vsock_stream_connect(struct socket *sock, struct sockaddr *addr, 1134 int addr_len, int flags) 1135 { 1136 int err; 1137 struct sock *sk; 1138 struct vsock_sock *vsk; 1139 const struct vsock_transport *transport; 1140 struct sockaddr_vm *remote_addr; 1141 long timeout; 1142 DEFINE_WAIT(wait); 1143 1144 err = 0; 1145 sk = sock->sk; 1146 vsk = vsock_sk(sk); 1147 transport = vsk->transport; 1148 1149 lock_sock(sk); 1150 1151 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */ 1152 switch (sock->state) { 1153 case SS_CONNECTED: 1154 err = -EISCONN; 1155 goto out; 1156 case SS_DISCONNECTING: 1157 err = -EINVAL; 1158 goto out; 1159 case SS_CONNECTING: 1160 /* This continues on so we can move sock into the SS_CONNECTED 1161 * state once the connection has completed (at which point err 1162 * will be set to zero also). Otherwise, we will either wait 1163 * for the connection or return -EALREADY should this be a 1164 * non-blocking call. 1165 */ 1166 err = -EALREADY; 1167 break; 1168 default: 1169 if ((sk->sk_state == TCP_LISTEN) || 1170 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) { 1171 err = -EINVAL; 1172 goto out; 1173 } 1174 1175 /* The hypervisor and well-known contexts do not have socket 1176 * endpoints. 1177 */ 1178 if (!transport->stream_allow(remote_addr->svm_cid, 1179 remote_addr->svm_port)) { 1180 err = -ENETUNREACH; 1181 goto out; 1182 } 1183 1184 /* Set the remote address that we are connecting to. */ 1185 memcpy(&vsk->remote_addr, remote_addr, 1186 sizeof(vsk->remote_addr)); 1187 1188 err = vsock_auto_bind(vsk); 1189 if (err) 1190 goto out; 1191 1192 sk->sk_state = TCP_SYN_SENT; 1193 1194 err = transport->connect(vsk); 1195 if (err < 0) 1196 goto out; 1197 1198 /* Mark sock as connecting and set the error code to in 1199 * progress in case this is a non-blocking connect. 1200 */ 1201 sock->state = SS_CONNECTING; 1202 err = -EINPROGRESS; 1203 } 1204 1205 /* The receive path will handle all communication until we are able to 1206 * enter the connected state. Here we wait for the connection to be 1207 * completed or a notification of an error. 1208 */ 1209 timeout = vsk->connect_timeout; 1210 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1211 1212 while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) { 1213 if (flags & O_NONBLOCK) { 1214 /* If we're not going to block, we schedule a timeout 1215 * function to generate a timeout on the connection 1216 * attempt, in case the peer doesn't respond in a 1217 * timely manner. We hold on to the socket until the 1218 * timeout fires. 1219 */ 1220 sock_hold(sk); 1221 schedule_delayed_work(&vsk->connect_work, timeout); 1222 1223 /* Skip ahead to preserve error code set above. */ 1224 goto out_wait; 1225 } 1226 1227 release_sock(sk); 1228 timeout = schedule_timeout(timeout); 1229 lock_sock(sk); 1230 1231 if (signal_pending(current)) { 1232 err = sock_intr_errno(timeout); 1233 sk->sk_state = TCP_CLOSE; 1234 sock->state = SS_UNCONNECTED; 1235 vsock_transport_cancel_pkt(vsk); 1236 goto out_wait; 1237 } else if (timeout == 0) { 1238 err = -ETIMEDOUT; 1239 sk->sk_state = TCP_CLOSE; 1240 sock->state = SS_UNCONNECTED; 1241 vsock_transport_cancel_pkt(vsk); 1242 goto out_wait; 1243 } 1244 1245 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1246 } 1247 1248 if (sk->sk_err) { 1249 err = -sk->sk_err; 1250 sk->sk_state = TCP_CLOSE; 1251 sock->state = SS_UNCONNECTED; 1252 } else { 1253 err = 0; 1254 } 1255 1256 out_wait: 1257 finish_wait(sk_sleep(sk), &wait); 1258 out: 1259 release_sock(sk); 1260 return err; 1261 } 1262 1263 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags, 1264 bool kern) 1265 { 1266 struct sock *listener; 1267 int err; 1268 struct sock *connected; 1269 struct vsock_sock *vconnected; 1270 long timeout; 1271 DEFINE_WAIT(wait); 1272 1273 err = 0; 1274 listener = sock->sk; 1275 1276 lock_sock(listener); 1277 1278 if (sock->type != SOCK_STREAM) { 1279 err = -EOPNOTSUPP; 1280 goto out; 1281 } 1282 1283 if (listener->sk_state != TCP_LISTEN) { 1284 err = -EINVAL; 1285 goto out; 1286 } 1287 1288 /* Wait for children sockets to appear; these are the new sockets 1289 * created upon connection establishment. 1290 */ 1291 timeout = sock_sndtimeo(listener, flags & O_NONBLOCK); 1292 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1293 1294 while ((connected = vsock_dequeue_accept(listener)) == NULL && 1295 listener->sk_err == 0) { 1296 release_sock(listener); 1297 timeout = schedule_timeout(timeout); 1298 finish_wait(sk_sleep(listener), &wait); 1299 lock_sock(listener); 1300 1301 if (signal_pending(current)) { 1302 err = sock_intr_errno(timeout); 1303 goto out; 1304 } else if (timeout == 0) { 1305 err = -EAGAIN; 1306 goto out; 1307 } 1308 1309 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1310 } 1311 finish_wait(sk_sleep(listener), &wait); 1312 1313 if (listener->sk_err) 1314 err = -listener->sk_err; 1315 1316 if (connected) { 1317 sk_acceptq_removed(listener); 1318 1319 lock_sock_nested(connected, SINGLE_DEPTH_NESTING); 1320 vconnected = vsock_sk(connected); 1321 1322 /* If the listener socket has received an error, then we should 1323 * reject this socket and return. Note that we simply mark the 1324 * socket rejected, drop our reference, and let the cleanup 1325 * function handle the cleanup; the fact that we found it in 1326 * the listener's accept queue guarantees that the cleanup 1327 * function hasn't run yet. 1328 */ 1329 if (err) { 1330 vconnected->rejected = true; 1331 } else { 1332 newsock->state = SS_CONNECTED; 1333 sock_graft(connected, newsock); 1334 } 1335 1336 release_sock(connected); 1337 sock_put(connected); 1338 } 1339 1340 out: 1341 release_sock(listener); 1342 return err; 1343 } 1344 1345 static int vsock_listen(struct socket *sock, int backlog) 1346 { 1347 int err; 1348 struct sock *sk; 1349 struct vsock_sock *vsk; 1350 1351 sk = sock->sk; 1352 1353 lock_sock(sk); 1354 1355 if (sock->type != SOCK_STREAM) { 1356 err = -EOPNOTSUPP; 1357 goto out; 1358 } 1359 1360 if (sock->state != SS_UNCONNECTED) { 1361 err = -EINVAL; 1362 goto out; 1363 } 1364 1365 vsk = vsock_sk(sk); 1366 1367 if (!vsock_addr_bound(&vsk->local_addr)) { 1368 err = -EINVAL; 1369 goto out; 1370 } 1371 1372 sk->sk_max_ack_backlog = backlog; 1373 sk->sk_state = TCP_LISTEN; 1374 1375 err = 0; 1376 1377 out: 1378 release_sock(sk); 1379 return err; 1380 } 1381 1382 static void vsock_update_buffer_size(struct vsock_sock *vsk, 1383 const struct vsock_transport *transport, 1384 u64 val) 1385 { 1386 if (val > vsk->buffer_max_size) 1387 val = vsk->buffer_max_size; 1388 1389 if (val < vsk->buffer_min_size) 1390 val = vsk->buffer_min_size; 1391 1392 if (val != vsk->buffer_size && 1393 transport && transport->notify_buffer_size) 1394 transport->notify_buffer_size(vsk, &val); 1395 1396 vsk->buffer_size = val; 1397 } 1398 1399 static int vsock_stream_setsockopt(struct socket *sock, 1400 int level, 1401 int optname, 1402 char __user *optval, 1403 unsigned int optlen) 1404 { 1405 int err; 1406 struct sock *sk; 1407 struct vsock_sock *vsk; 1408 const struct vsock_transport *transport; 1409 u64 val; 1410 1411 if (level != AF_VSOCK) 1412 return -ENOPROTOOPT; 1413 1414 #define COPY_IN(_v) \ 1415 do { \ 1416 if (optlen < sizeof(_v)) { \ 1417 err = -EINVAL; \ 1418 goto exit; \ 1419 } \ 1420 if (copy_from_user(&_v, optval, sizeof(_v)) != 0) { \ 1421 err = -EFAULT; \ 1422 goto exit; \ 1423 } \ 1424 } while (0) 1425 1426 err = 0; 1427 sk = sock->sk; 1428 vsk = vsock_sk(sk); 1429 transport = vsk->transport; 1430 1431 lock_sock(sk); 1432 1433 switch (optname) { 1434 case SO_VM_SOCKETS_BUFFER_SIZE: 1435 COPY_IN(val); 1436 vsock_update_buffer_size(vsk, transport, val); 1437 break; 1438 1439 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1440 COPY_IN(val); 1441 vsk->buffer_max_size = val; 1442 vsock_update_buffer_size(vsk, transport, vsk->buffer_size); 1443 break; 1444 1445 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1446 COPY_IN(val); 1447 vsk->buffer_min_size = val; 1448 vsock_update_buffer_size(vsk, transport, vsk->buffer_size); 1449 break; 1450 1451 case SO_VM_SOCKETS_CONNECT_TIMEOUT: { 1452 struct __kernel_old_timeval tv; 1453 COPY_IN(tv); 1454 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC && 1455 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) { 1456 vsk->connect_timeout = tv.tv_sec * HZ + 1457 DIV_ROUND_UP(tv.tv_usec, (1000000 / HZ)); 1458 if (vsk->connect_timeout == 0) 1459 vsk->connect_timeout = 1460 VSOCK_DEFAULT_CONNECT_TIMEOUT; 1461 1462 } else { 1463 err = -ERANGE; 1464 } 1465 break; 1466 } 1467 1468 default: 1469 err = -ENOPROTOOPT; 1470 break; 1471 } 1472 1473 #undef COPY_IN 1474 1475 exit: 1476 release_sock(sk); 1477 return err; 1478 } 1479 1480 static int vsock_stream_getsockopt(struct socket *sock, 1481 int level, int optname, 1482 char __user *optval, 1483 int __user *optlen) 1484 { 1485 int err; 1486 int len; 1487 struct sock *sk; 1488 struct vsock_sock *vsk; 1489 u64 val; 1490 1491 if (level != AF_VSOCK) 1492 return -ENOPROTOOPT; 1493 1494 err = get_user(len, optlen); 1495 if (err != 0) 1496 return err; 1497 1498 #define COPY_OUT(_v) \ 1499 do { \ 1500 if (len < sizeof(_v)) \ 1501 return -EINVAL; \ 1502 \ 1503 len = sizeof(_v); \ 1504 if (copy_to_user(optval, &_v, len) != 0) \ 1505 return -EFAULT; \ 1506 \ 1507 } while (0) 1508 1509 err = 0; 1510 sk = sock->sk; 1511 vsk = vsock_sk(sk); 1512 1513 switch (optname) { 1514 case SO_VM_SOCKETS_BUFFER_SIZE: 1515 val = vsk->buffer_size; 1516 COPY_OUT(val); 1517 break; 1518 1519 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1520 val = vsk->buffer_max_size; 1521 COPY_OUT(val); 1522 break; 1523 1524 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1525 val = vsk->buffer_min_size; 1526 COPY_OUT(val); 1527 break; 1528 1529 case SO_VM_SOCKETS_CONNECT_TIMEOUT: { 1530 struct __kernel_old_timeval tv; 1531 tv.tv_sec = vsk->connect_timeout / HZ; 1532 tv.tv_usec = 1533 (vsk->connect_timeout - 1534 tv.tv_sec * HZ) * (1000000 / HZ); 1535 COPY_OUT(tv); 1536 break; 1537 } 1538 default: 1539 return -ENOPROTOOPT; 1540 } 1541 1542 err = put_user(len, optlen); 1543 if (err != 0) 1544 return -EFAULT; 1545 1546 #undef COPY_OUT 1547 1548 return 0; 1549 } 1550 1551 static int vsock_stream_sendmsg(struct socket *sock, struct msghdr *msg, 1552 size_t len) 1553 { 1554 struct sock *sk; 1555 struct vsock_sock *vsk; 1556 const struct vsock_transport *transport; 1557 ssize_t total_written; 1558 long timeout; 1559 int err; 1560 struct vsock_transport_send_notify_data send_data; 1561 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1562 1563 sk = sock->sk; 1564 vsk = vsock_sk(sk); 1565 transport = vsk->transport; 1566 total_written = 0; 1567 err = 0; 1568 1569 if (msg->msg_flags & MSG_OOB) 1570 return -EOPNOTSUPP; 1571 1572 lock_sock(sk); 1573 1574 /* Callers should not provide a destination with stream sockets. */ 1575 if (msg->msg_namelen) { 1576 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1577 goto out; 1578 } 1579 1580 /* Send data only if both sides are not shutdown in the direction. */ 1581 if (sk->sk_shutdown & SEND_SHUTDOWN || 1582 vsk->peer_shutdown & RCV_SHUTDOWN) { 1583 err = -EPIPE; 1584 goto out; 1585 } 1586 1587 if (sk->sk_state != TCP_ESTABLISHED || 1588 !vsock_addr_bound(&vsk->local_addr)) { 1589 err = -ENOTCONN; 1590 goto out; 1591 } 1592 1593 if (!vsock_addr_bound(&vsk->remote_addr)) { 1594 err = -EDESTADDRREQ; 1595 goto out; 1596 } 1597 1598 /* Wait for room in the produce queue to enqueue our user's data. */ 1599 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1600 1601 err = transport->notify_send_init(vsk, &send_data); 1602 if (err < 0) 1603 goto out; 1604 1605 while (total_written < len) { 1606 ssize_t written; 1607 1608 add_wait_queue(sk_sleep(sk), &wait); 1609 while (vsock_stream_has_space(vsk) == 0 && 1610 sk->sk_err == 0 && 1611 !(sk->sk_shutdown & SEND_SHUTDOWN) && 1612 !(vsk->peer_shutdown & RCV_SHUTDOWN)) { 1613 1614 /* Don't wait for non-blocking sockets. */ 1615 if (timeout == 0) { 1616 err = -EAGAIN; 1617 remove_wait_queue(sk_sleep(sk), &wait); 1618 goto out_err; 1619 } 1620 1621 err = transport->notify_send_pre_block(vsk, &send_data); 1622 if (err < 0) { 1623 remove_wait_queue(sk_sleep(sk), &wait); 1624 goto out_err; 1625 } 1626 1627 release_sock(sk); 1628 timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout); 1629 lock_sock(sk); 1630 if (signal_pending(current)) { 1631 err = sock_intr_errno(timeout); 1632 remove_wait_queue(sk_sleep(sk), &wait); 1633 goto out_err; 1634 } else if (timeout == 0) { 1635 err = -EAGAIN; 1636 remove_wait_queue(sk_sleep(sk), &wait); 1637 goto out_err; 1638 } 1639 } 1640 remove_wait_queue(sk_sleep(sk), &wait); 1641 1642 /* These checks occur both as part of and after the loop 1643 * conditional since we need to check before and after 1644 * sleeping. 1645 */ 1646 if (sk->sk_err) { 1647 err = -sk->sk_err; 1648 goto out_err; 1649 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) || 1650 (vsk->peer_shutdown & RCV_SHUTDOWN)) { 1651 err = -EPIPE; 1652 goto out_err; 1653 } 1654 1655 err = transport->notify_send_pre_enqueue(vsk, &send_data); 1656 if (err < 0) 1657 goto out_err; 1658 1659 /* Note that enqueue will only write as many bytes as are free 1660 * in the produce queue, so we don't need to ensure len is 1661 * smaller than the queue size. It is the caller's 1662 * responsibility to check how many bytes we were able to send. 1663 */ 1664 1665 written = transport->stream_enqueue( 1666 vsk, msg, 1667 len - total_written); 1668 if (written < 0) { 1669 err = -ENOMEM; 1670 goto out_err; 1671 } 1672 1673 total_written += written; 1674 1675 err = transport->notify_send_post_enqueue( 1676 vsk, written, &send_data); 1677 if (err < 0) 1678 goto out_err; 1679 1680 } 1681 1682 out_err: 1683 if (total_written > 0) 1684 err = total_written; 1685 out: 1686 release_sock(sk); 1687 return err; 1688 } 1689 1690 1691 static int 1692 vsock_stream_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1693 int flags) 1694 { 1695 struct sock *sk; 1696 struct vsock_sock *vsk; 1697 const struct vsock_transport *transport; 1698 int err; 1699 size_t target; 1700 ssize_t copied; 1701 long timeout; 1702 struct vsock_transport_recv_notify_data recv_data; 1703 1704 DEFINE_WAIT(wait); 1705 1706 sk = sock->sk; 1707 vsk = vsock_sk(sk); 1708 transport = vsk->transport; 1709 err = 0; 1710 1711 lock_sock(sk); 1712 1713 if (sk->sk_state != TCP_ESTABLISHED) { 1714 /* Recvmsg is supposed to return 0 if a peer performs an 1715 * orderly shutdown. Differentiate between that case and when a 1716 * peer has not connected or a local shutdown occured with the 1717 * SOCK_DONE flag. 1718 */ 1719 if (sock_flag(sk, SOCK_DONE)) 1720 err = 0; 1721 else 1722 err = -ENOTCONN; 1723 1724 goto out; 1725 } 1726 1727 if (flags & MSG_OOB) { 1728 err = -EOPNOTSUPP; 1729 goto out; 1730 } 1731 1732 /* We don't check peer_shutdown flag here since peer may actually shut 1733 * down, but there can be data in the queue that a local socket can 1734 * receive. 1735 */ 1736 if (sk->sk_shutdown & RCV_SHUTDOWN) { 1737 err = 0; 1738 goto out; 1739 } 1740 1741 /* It is valid on Linux to pass in a zero-length receive buffer. This 1742 * is not an error. We may as well bail out now. 1743 */ 1744 if (!len) { 1745 err = 0; 1746 goto out; 1747 } 1748 1749 /* We must not copy less than target bytes into the user's buffer 1750 * before returning successfully, so we wait for the consume queue to 1751 * have that much data to consume before dequeueing. Note that this 1752 * makes it impossible to handle cases where target is greater than the 1753 * queue size. 1754 */ 1755 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 1756 if (target >= transport->stream_rcvhiwat(vsk)) { 1757 err = -ENOMEM; 1758 goto out; 1759 } 1760 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1761 copied = 0; 1762 1763 err = transport->notify_recv_init(vsk, target, &recv_data); 1764 if (err < 0) 1765 goto out; 1766 1767 1768 while (1) { 1769 s64 ready; 1770 1771 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1772 ready = vsock_stream_has_data(vsk); 1773 1774 if (ready == 0) { 1775 if (sk->sk_err != 0 || 1776 (sk->sk_shutdown & RCV_SHUTDOWN) || 1777 (vsk->peer_shutdown & SEND_SHUTDOWN)) { 1778 finish_wait(sk_sleep(sk), &wait); 1779 break; 1780 } 1781 /* Don't wait for non-blocking sockets. */ 1782 if (timeout == 0) { 1783 err = -EAGAIN; 1784 finish_wait(sk_sleep(sk), &wait); 1785 break; 1786 } 1787 1788 err = transport->notify_recv_pre_block( 1789 vsk, target, &recv_data); 1790 if (err < 0) { 1791 finish_wait(sk_sleep(sk), &wait); 1792 break; 1793 } 1794 release_sock(sk); 1795 timeout = schedule_timeout(timeout); 1796 lock_sock(sk); 1797 1798 if (signal_pending(current)) { 1799 err = sock_intr_errno(timeout); 1800 finish_wait(sk_sleep(sk), &wait); 1801 break; 1802 } else if (timeout == 0) { 1803 err = -EAGAIN; 1804 finish_wait(sk_sleep(sk), &wait); 1805 break; 1806 } 1807 } else { 1808 ssize_t read; 1809 1810 finish_wait(sk_sleep(sk), &wait); 1811 1812 if (ready < 0) { 1813 /* Invalid queue pair content. XXX This should 1814 * be changed to a connection reset in a later 1815 * change. 1816 */ 1817 1818 err = -ENOMEM; 1819 goto out; 1820 } 1821 1822 err = transport->notify_recv_pre_dequeue( 1823 vsk, target, &recv_data); 1824 if (err < 0) 1825 break; 1826 1827 read = transport->stream_dequeue( 1828 vsk, msg, 1829 len - copied, flags); 1830 if (read < 0) { 1831 err = -ENOMEM; 1832 break; 1833 } 1834 1835 copied += read; 1836 1837 err = transport->notify_recv_post_dequeue( 1838 vsk, target, read, 1839 !(flags & MSG_PEEK), &recv_data); 1840 if (err < 0) 1841 goto out; 1842 1843 if (read >= target || flags & MSG_PEEK) 1844 break; 1845 1846 target -= read; 1847 } 1848 } 1849 1850 if (sk->sk_err) 1851 err = -sk->sk_err; 1852 else if (sk->sk_shutdown & RCV_SHUTDOWN) 1853 err = 0; 1854 1855 if (copied > 0) 1856 err = copied; 1857 1858 out: 1859 release_sock(sk); 1860 return err; 1861 } 1862 1863 static const struct proto_ops vsock_stream_ops = { 1864 .family = PF_VSOCK, 1865 .owner = THIS_MODULE, 1866 .release = vsock_release, 1867 .bind = vsock_bind, 1868 .connect = vsock_stream_connect, 1869 .socketpair = sock_no_socketpair, 1870 .accept = vsock_accept, 1871 .getname = vsock_getname, 1872 .poll = vsock_poll, 1873 .ioctl = sock_no_ioctl, 1874 .listen = vsock_listen, 1875 .shutdown = vsock_shutdown, 1876 .setsockopt = vsock_stream_setsockopt, 1877 .getsockopt = vsock_stream_getsockopt, 1878 .sendmsg = vsock_stream_sendmsg, 1879 .recvmsg = vsock_stream_recvmsg, 1880 .mmap = sock_no_mmap, 1881 .sendpage = sock_no_sendpage, 1882 }; 1883 1884 static int vsock_create(struct net *net, struct socket *sock, 1885 int protocol, int kern) 1886 { 1887 struct sock *sk; 1888 1889 if (!sock) 1890 return -EINVAL; 1891 1892 if (protocol && protocol != PF_VSOCK) 1893 return -EPROTONOSUPPORT; 1894 1895 switch (sock->type) { 1896 case SOCK_DGRAM: 1897 sock->ops = &vsock_dgram_ops; 1898 break; 1899 case SOCK_STREAM: 1900 sock->ops = &vsock_stream_ops; 1901 break; 1902 default: 1903 return -ESOCKTNOSUPPORT; 1904 } 1905 1906 sock->state = SS_UNCONNECTED; 1907 1908 sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern); 1909 if (!sk) 1910 return -ENOMEM; 1911 1912 vsock_insert_unbound(vsock_sk(sk)); 1913 1914 return 0; 1915 } 1916 1917 static const struct net_proto_family vsock_family_ops = { 1918 .family = AF_VSOCK, 1919 .create = vsock_create, 1920 .owner = THIS_MODULE, 1921 }; 1922 1923 static long vsock_dev_do_ioctl(struct file *filp, 1924 unsigned int cmd, void __user *ptr) 1925 { 1926 u32 __user *p = ptr; 1927 int retval = 0; 1928 1929 switch (cmd) { 1930 case IOCTL_VM_SOCKETS_GET_LOCAL_CID: 1931 if (put_user(transport_single->get_local_cid(), p) != 0) 1932 retval = -EFAULT; 1933 break; 1934 1935 default: 1936 pr_err("Unknown ioctl %d\n", cmd); 1937 retval = -EINVAL; 1938 } 1939 1940 return retval; 1941 } 1942 1943 static long vsock_dev_ioctl(struct file *filp, 1944 unsigned int cmd, unsigned long arg) 1945 { 1946 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg); 1947 } 1948 1949 #ifdef CONFIG_COMPAT 1950 static long vsock_dev_compat_ioctl(struct file *filp, 1951 unsigned int cmd, unsigned long arg) 1952 { 1953 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg)); 1954 } 1955 #endif 1956 1957 static const struct file_operations vsock_device_ops = { 1958 .owner = THIS_MODULE, 1959 .unlocked_ioctl = vsock_dev_ioctl, 1960 #ifdef CONFIG_COMPAT 1961 .compat_ioctl = vsock_dev_compat_ioctl, 1962 #endif 1963 .open = nonseekable_open, 1964 }; 1965 1966 static struct miscdevice vsock_device = { 1967 .name = "vsock", 1968 .fops = &vsock_device_ops, 1969 }; 1970 1971 int __vsock_core_init(const struct vsock_transport *t, struct module *owner) 1972 { 1973 int err = mutex_lock_interruptible(&vsock_register_mutex); 1974 1975 if (err) 1976 return err; 1977 1978 if (transport_single) { 1979 err = -EBUSY; 1980 goto err_busy; 1981 } 1982 1983 /* Transport must be the owner of the protocol so that it can't 1984 * unload while there are open sockets. 1985 */ 1986 vsock_proto.owner = owner; 1987 transport_single = t; 1988 1989 vsock_device.minor = MISC_DYNAMIC_MINOR; 1990 err = misc_register(&vsock_device); 1991 if (err) { 1992 pr_err("Failed to register misc device\n"); 1993 goto err_reset_transport; 1994 } 1995 1996 err = proto_register(&vsock_proto, 1); /* we want our slab */ 1997 if (err) { 1998 pr_err("Cannot register vsock protocol\n"); 1999 goto err_deregister_misc; 2000 } 2001 2002 err = sock_register(&vsock_family_ops); 2003 if (err) { 2004 pr_err("could not register af_vsock (%d) address family: %d\n", 2005 AF_VSOCK, err); 2006 goto err_unregister_proto; 2007 } 2008 2009 mutex_unlock(&vsock_register_mutex); 2010 return 0; 2011 2012 err_unregister_proto: 2013 proto_unregister(&vsock_proto); 2014 err_deregister_misc: 2015 misc_deregister(&vsock_device); 2016 err_reset_transport: 2017 transport_single = NULL; 2018 err_busy: 2019 mutex_unlock(&vsock_register_mutex); 2020 return err; 2021 } 2022 EXPORT_SYMBOL_GPL(__vsock_core_init); 2023 2024 void vsock_core_exit(void) 2025 { 2026 mutex_lock(&vsock_register_mutex); 2027 2028 misc_deregister(&vsock_device); 2029 sock_unregister(AF_VSOCK); 2030 proto_unregister(&vsock_proto); 2031 2032 /* We do not want the assignment below re-ordered. */ 2033 mb(); 2034 transport_single = NULL; 2035 2036 mutex_unlock(&vsock_register_mutex); 2037 } 2038 EXPORT_SYMBOL_GPL(vsock_core_exit); 2039 2040 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk) 2041 { 2042 return vsk->transport; 2043 } 2044 EXPORT_SYMBOL_GPL(vsock_core_get_transport); 2045 2046 static void __exit vsock_exit(void) 2047 { 2048 /* Do nothing. This function makes this module removable. */ 2049 } 2050 2051 module_init(vsock_init_tables); 2052 module_exit(vsock_exit); 2053 2054 MODULE_AUTHOR("VMware, Inc."); 2055 MODULE_DESCRIPTION("VMware Virtual Socket Family"); 2056 MODULE_VERSION("1.0.2.0-k"); 2057 MODULE_LICENSE("GPL v2"); 2058