1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * VMware vSockets Driver 4 * 5 * Copyright (C) 2007-2013 VMware, Inc. All rights reserved. 6 */ 7 8 /* Implementation notes: 9 * 10 * - There are two kinds of sockets: those created by user action (such as 11 * calling socket(2)) and those created by incoming connection request packets. 12 * 13 * - There are two "global" tables, one for bound sockets (sockets that have 14 * specified an address that they are responsible for) and one for connected 15 * sockets (sockets that have established a connection with another socket). 16 * These tables are "global" in that all sockets on the system are placed 17 * within them. - Note, though, that the bound table contains an extra entry 18 * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in 19 * that list. The bound table is used solely for lookup of sockets when packets 20 * are received and that's not necessary for SOCK_DGRAM sockets since we create 21 * a datagram handle for each and need not perform a lookup. Keeping SOCK_DGRAM 22 * sockets out of the bound hash buckets will reduce the chance of collisions 23 * when looking for SOCK_STREAM sockets and prevents us from having to check the 24 * socket type in the hash table lookups. 25 * 26 * - Sockets created by user action will either be "client" sockets that 27 * initiate a connection or "server" sockets that listen for connections; we do 28 * not support simultaneous connects (two "client" sockets connecting). 29 * 30 * - "Server" sockets are referred to as listener sockets throughout this 31 * implementation because they are in the TCP_LISTEN state. When a 32 * connection request is received (the second kind of socket mentioned above), 33 * we create a new socket and refer to it as a pending socket. These pending 34 * sockets are placed on the pending connection list of the listener socket. 35 * When future packets are received for the address the listener socket is 36 * bound to, we check if the source of the packet is from one that has an 37 * existing pending connection. If it does, we process the packet for the 38 * pending socket. When that socket reaches the connected state, it is removed 39 * from the listener socket's pending list and enqueued in the listener 40 * socket's accept queue. Callers of accept(2) will accept connected sockets 41 * from the listener socket's accept queue. If the socket cannot be accepted 42 * for some reason then it is marked rejected. Once the connection is 43 * accepted, it is owned by the user process and the responsibility for cleanup 44 * falls with that user process. 45 * 46 * - It is possible that these pending sockets will never reach the connected 47 * state; in fact, we may never receive another packet after the connection 48 * request. Because of this, we must schedule a cleanup function to run in the 49 * future, after some amount of time passes where a connection should have been 50 * established. This function ensures that the socket is off all lists so it 51 * cannot be retrieved, then drops all references to the socket so it is cleaned 52 * up (sock_put() -> sk_free() -> our sk_destruct implementation). Note this 53 * function will also cleanup rejected sockets, those that reach the connected 54 * state but leave it before they have been accepted. 55 * 56 * - Lock ordering for pending or accept queue sockets is: 57 * 58 * lock_sock(listener); 59 * lock_sock_nested(pending, SINGLE_DEPTH_NESTING); 60 * 61 * Using explicit nested locking keeps lockdep happy since normally only one 62 * lock of a given class may be taken at a time. 63 * 64 * - Sockets created by user action will be cleaned up when the user process 65 * calls close(2), causing our release implementation to be called. Our release 66 * implementation will perform some cleanup then drop the last reference so our 67 * sk_destruct implementation is invoked. Our sk_destruct implementation will 68 * perform additional cleanup that's common for both types of sockets. 69 * 70 * - A socket's reference count is what ensures that the structure won't be 71 * freed. Each entry in a list (such as the "global" bound and connected tables 72 * and the listener socket's pending list and connected queue) ensures a 73 * reference. When we defer work until process context and pass a socket as our 74 * argument, we must ensure the reference count is increased to ensure the 75 * socket isn't freed before the function is run; the deferred function will 76 * then drop the reference. 77 * 78 * - sk->sk_state uses the TCP state constants because they are widely used by 79 * other address families and exposed to userspace tools like ss(8): 80 * 81 * TCP_CLOSE - unconnected 82 * TCP_SYN_SENT - connecting 83 * TCP_ESTABLISHED - connected 84 * TCP_CLOSING - disconnecting 85 * TCP_LISTEN - listening 86 */ 87 88 #include <linux/compat.h> 89 #include <linux/types.h> 90 #include <linux/bitops.h> 91 #include <linux/cred.h> 92 #include <linux/errqueue.h> 93 #include <linux/init.h> 94 #include <linux/io.h> 95 #include <linux/kernel.h> 96 #include <linux/sched/signal.h> 97 #include <linux/kmod.h> 98 #include <linux/list.h> 99 #include <linux/miscdevice.h> 100 #include <linux/module.h> 101 #include <linux/mutex.h> 102 #include <linux/net.h> 103 #include <linux/poll.h> 104 #include <linux/random.h> 105 #include <linux/skbuff.h> 106 #include <linux/smp.h> 107 #include <linux/socket.h> 108 #include <linux/stddef.h> 109 #include <linux/unistd.h> 110 #include <linux/wait.h> 111 #include <linux/workqueue.h> 112 #include <net/sock.h> 113 #include <net/af_vsock.h> 114 #include <uapi/linux/vm_sockets.h> 115 116 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr); 117 static void vsock_sk_destruct(struct sock *sk); 118 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 119 120 /* Protocol family. */ 121 struct proto vsock_proto = { 122 .name = "AF_VSOCK", 123 .owner = THIS_MODULE, 124 .obj_size = sizeof(struct vsock_sock), 125 #ifdef CONFIG_BPF_SYSCALL 126 .psock_update_sk_prot = vsock_bpf_update_proto, 127 #endif 128 }; 129 130 /* The default peer timeout indicates how long we will wait for a peer response 131 * to a control message. 132 */ 133 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ) 134 135 #define VSOCK_DEFAULT_BUFFER_SIZE (1024 * 256) 136 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256) 137 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128 138 139 /* Transport used for host->guest communication */ 140 static const struct vsock_transport *transport_h2g; 141 /* Transport used for guest->host communication */ 142 static const struct vsock_transport *transport_g2h; 143 /* Transport used for DGRAM communication */ 144 static const struct vsock_transport *transport_dgram; 145 /* Transport used for local communication */ 146 static const struct vsock_transport *transport_local; 147 static DEFINE_MUTEX(vsock_register_mutex); 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 void 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 } 208 209 static void __vsock_insert_bound(struct list_head *list, 210 struct vsock_sock *vsk) 211 { 212 sock_hold(&vsk->sk); 213 list_add(&vsk->bound_table, list); 214 } 215 216 static void __vsock_insert_connected(struct list_head *list, 217 struct vsock_sock *vsk) 218 { 219 sock_hold(&vsk->sk); 220 list_add(&vsk->connected_table, list); 221 } 222 223 static void __vsock_remove_bound(struct vsock_sock *vsk) 224 { 225 list_del_init(&vsk->bound_table); 226 sock_put(&vsk->sk); 227 } 228 229 static void __vsock_remove_connected(struct vsock_sock *vsk) 230 { 231 list_del_init(&vsk->connected_table); 232 sock_put(&vsk->sk); 233 } 234 235 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr) 236 { 237 struct vsock_sock *vsk; 238 239 list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) { 240 if (vsock_addr_equals_addr(addr, &vsk->local_addr)) 241 return sk_vsock(vsk); 242 243 if (addr->svm_port == vsk->local_addr.svm_port && 244 (vsk->local_addr.svm_cid == VMADDR_CID_ANY || 245 addr->svm_cid == VMADDR_CID_ANY)) 246 return sk_vsock(vsk); 247 } 248 249 return NULL; 250 } 251 252 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src, 253 struct sockaddr_vm *dst) 254 { 255 struct vsock_sock *vsk; 256 257 list_for_each_entry(vsk, vsock_connected_sockets(src, dst), 258 connected_table) { 259 if (vsock_addr_equals_addr(src, &vsk->remote_addr) && 260 dst->svm_port == vsk->local_addr.svm_port) { 261 return sk_vsock(vsk); 262 } 263 } 264 265 return NULL; 266 } 267 268 static void vsock_insert_unbound(struct vsock_sock *vsk) 269 { 270 spin_lock_bh(&vsock_table_lock); 271 __vsock_insert_bound(vsock_unbound_sockets, vsk); 272 spin_unlock_bh(&vsock_table_lock); 273 } 274 275 void vsock_insert_connected(struct vsock_sock *vsk) 276 { 277 struct list_head *list = vsock_connected_sockets( 278 &vsk->remote_addr, &vsk->local_addr); 279 280 spin_lock_bh(&vsock_table_lock); 281 __vsock_insert_connected(list, vsk); 282 spin_unlock_bh(&vsock_table_lock); 283 } 284 EXPORT_SYMBOL_GPL(vsock_insert_connected); 285 286 void vsock_remove_bound(struct vsock_sock *vsk) 287 { 288 spin_lock_bh(&vsock_table_lock); 289 if (__vsock_in_bound_table(vsk)) 290 __vsock_remove_bound(vsk); 291 spin_unlock_bh(&vsock_table_lock); 292 } 293 EXPORT_SYMBOL_GPL(vsock_remove_bound); 294 295 void vsock_remove_connected(struct vsock_sock *vsk) 296 { 297 spin_lock_bh(&vsock_table_lock); 298 if (__vsock_in_connected_table(vsk)) 299 __vsock_remove_connected(vsk); 300 spin_unlock_bh(&vsock_table_lock); 301 } 302 EXPORT_SYMBOL_GPL(vsock_remove_connected); 303 304 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr) 305 { 306 struct sock *sk; 307 308 spin_lock_bh(&vsock_table_lock); 309 sk = __vsock_find_bound_socket(addr); 310 if (sk) 311 sock_hold(sk); 312 313 spin_unlock_bh(&vsock_table_lock); 314 315 return sk; 316 } 317 EXPORT_SYMBOL_GPL(vsock_find_bound_socket); 318 319 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src, 320 struct sockaddr_vm *dst) 321 { 322 struct sock *sk; 323 324 spin_lock_bh(&vsock_table_lock); 325 sk = __vsock_find_connected_socket(src, dst); 326 if (sk) 327 sock_hold(sk); 328 329 spin_unlock_bh(&vsock_table_lock); 330 331 return sk; 332 } 333 EXPORT_SYMBOL_GPL(vsock_find_connected_socket); 334 335 void vsock_remove_sock(struct vsock_sock *vsk) 336 { 337 vsock_remove_bound(vsk); 338 vsock_remove_connected(vsk); 339 } 340 EXPORT_SYMBOL_GPL(vsock_remove_sock); 341 342 void vsock_for_each_connected_socket(struct vsock_transport *transport, 343 void (*fn)(struct sock *sk)) 344 { 345 int i; 346 347 spin_lock_bh(&vsock_table_lock); 348 349 for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) { 350 struct vsock_sock *vsk; 351 list_for_each_entry(vsk, &vsock_connected_table[i], 352 connected_table) { 353 if (vsk->transport != transport) 354 continue; 355 356 fn(sk_vsock(vsk)); 357 } 358 } 359 360 spin_unlock_bh(&vsock_table_lock); 361 } 362 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket); 363 364 void vsock_add_pending(struct sock *listener, struct sock *pending) 365 { 366 struct vsock_sock *vlistener; 367 struct vsock_sock *vpending; 368 369 vlistener = vsock_sk(listener); 370 vpending = vsock_sk(pending); 371 372 sock_hold(pending); 373 sock_hold(listener); 374 list_add_tail(&vpending->pending_links, &vlistener->pending_links); 375 } 376 EXPORT_SYMBOL_GPL(vsock_add_pending); 377 378 void vsock_remove_pending(struct sock *listener, struct sock *pending) 379 { 380 struct vsock_sock *vpending = vsock_sk(pending); 381 382 list_del_init(&vpending->pending_links); 383 sock_put(listener); 384 sock_put(pending); 385 } 386 EXPORT_SYMBOL_GPL(vsock_remove_pending); 387 388 void vsock_enqueue_accept(struct sock *listener, struct sock *connected) 389 { 390 struct vsock_sock *vlistener; 391 struct vsock_sock *vconnected; 392 393 vlistener = vsock_sk(listener); 394 vconnected = vsock_sk(connected); 395 396 sock_hold(connected); 397 sock_hold(listener); 398 list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue); 399 } 400 EXPORT_SYMBOL_GPL(vsock_enqueue_accept); 401 402 static bool vsock_use_local_transport(unsigned int remote_cid) 403 { 404 if (!transport_local) 405 return false; 406 407 if (remote_cid == VMADDR_CID_LOCAL) 408 return true; 409 410 if (transport_g2h) { 411 return remote_cid == transport_g2h->get_local_cid(); 412 } else { 413 return remote_cid == VMADDR_CID_HOST; 414 } 415 } 416 417 static void vsock_deassign_transport(struct vsock_sock *vsk) 418 { 419 if (!vsk->transport) 420 return; 421 422 vsk->transport->destruct(vsk); 423 module_put(vsk->transport->module); 424 vsk->transport = NULL; 425 } 426 427 /* Assign a transport to a socket and call the .init transport callback. 428 * 429 * Note: for connection oriented socket this must be called when vsk->remote_addr 430 * is set (e.g. during the connect() or when a connection request on a listener 431 * socket is received). 432 * The vsk->remote_addr is used to decide which transport to use: 433 * - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if 434 * g2h is not loaded, will use local transport; 435 * - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field 436 * includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport; 437 * - remote CID > VMADDR_CID_HOST will use host->guest transport; 438 */ 439 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk) 440 { 441 const struct vsock_transport *new_transport; 442 struct sock *sk = sk_vsock(vsk); 443 unsigned int remote_cid = vsk->remote_addr.svm_cid; 444 __u8 remote_flags; 445 int ret; 446 447 /* If the packet is coming with the source and destination CIDs higher 448 * than VMADDR_CID_HOST, then a vsock channel where all the packets are 449 * forwarded to the host should be established. Then the host will 450 * need to forward the packets to the guest. 451 * 452 * The flag is set on the (listen) receive path (psk is not NULL). On 453 * the connect path the flag can be set by the user space application. 454 */ 455 if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST && 456 vsk->remote_addr.svm_cid > VMADDR_CID_HOST) 457 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST; 458 459 remote_flags = vsk->remote_addr.svm_flags; 460 461 switch (sk->sk_type) { 462 case SOCK_DGRAM: 463 new_transport = transport_dgram; 464 break; 465 case SOCK_STREAM: 466 case SOCK_SEQPACKET: 467 if (vsock_use_local_transport(remote_cid)) 468 new_transport = transport_local; 469 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g || 470 (remote_flags & VMADDR_FLAG_TO_HOST)) 471 new_transport = transport_g2h; 472 else 473 new_transport = transport_h2g; 474 break; 475 default: 476 return -ESOCKTNOSUPPORT; 477 } 478 479 if (vsk->transport) { 480 if (vsk->transport == new_transport) 481 return 0; 482 483 /* transport->release() must be called with sock lock acquired. 484 * This path can only be taken during vsock_connect(), where we 485 * have already held the sock lock. In the other cases, this 486 * function is called on a new socket which is not assigned to 487 * any transport. 488 */ 489 vsk->transport->release(vsk); 490 vsock_deassign_transport(vsk); 491 } 492 493 /* We increase the module refcnt to prevent the transport unloading 494 * while there are open sockets assigned to it. 495 */ 496 if (!new_transport || !try_module_get(new_transport->module)) 497 return -ENODEV; 498 499 if (sk->sk_type == SOCK_SEQPACKET) { 500 if (!new_transport->seqpacket_allow || 501 !new_transport->seqpacket_allow(remote_cid)) { 502 module_put(new_transport->module); 503 return -ESOCKTNOSUPPORT; 504 } 505 } 506 507 ret = new_transport->init(vsk, psk); 508 if (ret) { 509 module_put(new_transport->module); 510 return ret; 511 } 512 513 vsk->transport = new_transport; 514 515 return 0; 516 } 517 EXPORT_SYMBOL_GPL(vsock_assign_transport); 518 519 bool vsock_find_cid(unsigned int cid) 520 { 521 if (transport_g2h && cid == transport_g2h->get_local_cid()) 522 return true; 523 524 if (transport_h2g && cid == VMADDR_CID_HOST) 525 return true; 526 527 if (transport_local && cid == VMADDR_CID_LOCAL) 528 return true; 529 530 return false; 531 } 532 EXPORT_SYMBOL_GPL(vsock_find_cid); 533 534 static struct sock *vsock_dequeue_accept(struct sock *listener) 535 { 536 struct vsock_sock *vlistener; 537 struct vsock_sock *vconnected; 538 539 vlistener = vsock_sk(listener); 540 541 if (list_empty(&vlistener->accept_queue)) 542 return NULL; 543 544 vconnected = list_entry(vlistener->accept_queue.next, 545 struct vsock_sock, accept_queue); 546 547 list_del_init(&vconnected->accept_queue); 548 sock_put(listener); 549 /* The caller will need a reference on the connected socket so we let 550 * it call sock_put(). 551 */ 552 553 return sk_vsock(vconnected); 554 } 555 556 static bool vsock_is_accept_queue_empty(struct sock *sk) 557 { 558 struct vsock_sock *vsk = vsock_sk(sk); 559 return list_empty(&vsk->accept_queue); 560 } 561 562 static bool vsock_is_pending(struct sock *sk) 563 { 564 struct vsock_sock *vsk = vsock_sk(sk); 565 return !list_empty(&vsk->pending_links); 566 } 567 568 static int vsock_send_shutdown(struct sock *sk, int mode) 569 { 570 struct vsock_sock *vsk = vsock_sk(sk); 571 572 if (!vsk->transport) 573 return -ENODEV; 574 575 return vsk->transport->shutdown(vsk, mode); 576 } 577 578 static void vsock_pending_work(struct work_struct *work) 579 { 580 struct sock *sk; 581 struct sock *listener; 582 struct vsock_sock *vsk; 583 bool cleanup; 584 585 vsk = container_of(work, struct vsock_sock, pending_work.work); 586 sk = sk_vsock(vsk); 587 listener = vsk->listener; 588 cleanup = true; 589 590 lock_sock(listener); 591 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 592 593 if (vsock_is_pending(sk)) { 594 vsock_remove_pending(listener, sk); 595 596 sk_acceptq_removed(listener); 597 } else if (!vsk->rejected) { 598 /* We are not on the pending list and accept() did not reject 599 * us, so we must have been accepted by our user process. We 600 * just need to drop our references to the sockets and be on 601 * our way. 602 */ 603 cleanup = false; 604 goto out; 605 } 606 607 /* We need to remove ourself from the global connected sockets list so 608 * incoming packets can't find this socket, and to reduce the reference 609 * count. 610 */ 611 vsock_remove_connected(vsk); 612 613 sk->sk_state = TCP_CLOSE; 614 615 out: 616 release_sock(sk); 617 release_sock(listener); 618 if (cleanup) 619 sock_put(sk); 620 621 sock_put(sk); 622 sock_put(listener); 623 } 624 625 /**** SOCKET OPERATIONS ****/ 626 627 static int __vsock_bind_connectible(struct vsock_sock *vsk, 628 struct sockaddr_vm *addr) 629 { 630 static u32 port; 631 struct sockaddr_vm new_addr; 632 633 if (!port) 634 port = get_random_u32_above(LAST_RESERVED_PORT); 635 636 vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port); 637 638 if (addr->svm_port == VMADDR_PORT_ANY) { 639 bool found = false; 640 unsigned int i; 641 642 for (i = 0; i < MAX_PORT_RETRIES; i++) { 643 if (port <= LAST_RESERVED_PORT) 644 port = LAST_RESERVED_PORT + 1; 645 646 new_addr.svm_port = port++; 647 648 if (!__vsock_find_bound_socket(&new_addr)) { 649 found = true; 650 break; 651 } 652 } 653 654 if (!found) 655 return -EADDRNOTAVAIL; 656 } else { 657 /* If port is in reserved range, ensure caller 658 * has necessary privileges. 659 */ 660 if (addr->svm_port <= LAST_RESERVED_PORT && 661 !capable(CAP_NET_BIND_SERVICE)) { 662 return -EACCES; 663 } 664 665 if (__vsock_find_bound_socket(&new_addr)) 666 return -EADDRINUSE; 667 } 668 669 vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port); 670 671 /* Remove connection oriented sockets from the unbound list and add them 672 * to the hash table for easy lookup by its address. The unbound list 673 * is simply an extra entry at the end of the hash table, a trick used 674 * by AF_UNIX. 675 */ 676 __vsock_remove_bound(vsk); 677 __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk); 678 679 return 0; 680 } 681 682 static int __vsock_bind_dgram(struct vsock_sock *vsk, 683 struct sockaddr_vm *addr) 684 { 685 return vsk->transport->dgram_bind(vsk, addr); 686 } 687 688 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr) 689 { 690 struct vsock_sock *vsk = vsock_sk(sk); 691 int retval; 692 693 /* First ensure this socket isn't already bound. */ 694 if (vsock_addr_bound(&vsk->local_addr)) 695 return -EINVAL; 696 697 /* Now bind to the provided address or select appropriate values if 698 * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY). Note that 699 * like AF_INET prevents binding to a non-local IP address (in most 700 * cases), we only allow binding to a local CID. 701 */ 702 if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid)) 703 return -EADDRNOTAVAIL; 704 705 switch (sk->sk_socket->type) { 706 case SOCK_STREAM: 707 case SOCK_SEQPACKET: 708 spin_lock_bh(&vsock_table_lock); 709 retval = __vsock_bind_connectible(vsk, addr); 710 spin_unlock_bh(&vsock_table_lock); 711 break; 712 713 case SOCK_DGRAM: 714 retval = __vsock_bind_dgram(vsk, addr); 715 break; 716 717 default: 718 retval = -EINVAL; 719 break; 720 } 721 722 return retval; 723 } 724 725 static void vsock_connect_timeout(struct work_struct *work); 726 727 static struct sock *__vsock_create(struct net *net, 728 struct socket *sock, 729 struct sock *parent, 730 gfp_t priority, 731 unsigned short type, 732 int kern) 733 { 734 struct sock *sk; 735 struct vsock_sock *psk; 736 struct vsock_sock *vsk; 737 738 sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern); 739 if (!sk) 740 return NULL; 741 742 sock_init_data(sock, sk); 743 744 /* sk->sk_type is normally set in sock_init_data, but only if sock is 745 * non-NULL. We make sure that our sockets always have a type by 746 * setting it here if needed. 747 */ 748 if (!sock) 749 sk->sk_type = type; 750 751 vsk = vsock_sk(sk); 752 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 753 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 754 755 sk->sk_destruct = vsock_sk_destruct; 756 sk->sk_backlog_rcv = vsock_queue_rcv_skb; 757 sock_reset_flag(sk, SOCK_DONE); 758 759 INIT_LIST_HEAD(&vsk->bound_table); 760 INIT_LIST_HEAD(&vsk->connected_table); 761 vsk->listener = NULL; 762 INIT_LIST_HEAD(&vsk->pending_links); 763 INIT_LIST_HEAD(&vsk->accept_queue); 764 vsk->rejected = false; 765 vsk->sent_request = false; 766 vsk->ignore_connecting_rst = false; 767 vsk->peer_shutdown = 0; 768 INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout); 769 INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work); 770 771 psk = parent ? vsock_sk(parent) : NULL; 772 if (parent) { 773 vsk->trusted = psk->trusted; 774 vsk->owner = get_cred(psk->owner); 775 vsk->connect_timeout = psk->connect_timeout; 776 vsk->buffer_size = psk->buffer_size; 777 vsk->buffer_min_size = psk->buffer_min_size; 778 vsk->buffer_max_size = psk->buffer_max_size; 779 security_sk_clone(parent, sk); 780 } else { 781 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN); 782 vsk->owner = get_current_cred(); 783 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT; 784 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE; 785 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE; 786 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE; 787 } 788 789 return sk; 790 } 791 792 static bool sock_type_connectible(u16 type) 793 { 794 return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET); 795 } 796 797 static void __vsock_release(struct sock *sk, int level) 798 { 799 if (sk) { 800 struct sock *pending; 801 struct vsock_sock *vsk; 802 803 vsk = vsock_sk(sk); 804 pending = NULL; /* Compiler warning. */ 805 806 /* When "level" is SINGLE_DEPTH_NESTING, use the nested 807 * version to avoid the warning "possible recursive locking 808 * detected". When "level" is 0, lock_sock_nested(sk, level) 809 * is the same as lock_sock(sk). 810 */ 811 lock_sock_nested(sk, level); 812 813 if (vsk->transport) 814 vsk->transport->release(vsk); 815 else if (sock_type_connectible(sk->sk_type)) 816 vsock_remove_sock(vsk); 817 818 sock_orphan(sk); 819 sk->sk_shutdown = SHUTDOWN_MASK; 820 821 skb_queue_purge(&sk->sk_receive_queue); 822 823 /* Clean up any sockets that never were accepted. */ 824 while ((pending = vsock_dequeue_accept(sk)) != NULL) { 825 __vsock_release(pending, SINGLE_DEPTH_NESTING); 826 sock_put(pending); 827 } 828 829 release_sock(sk); 830 sock_put(sk); 831 } 832 } 833 834 static void vsock_sk_destruct(struct sock *sk) 835 { 836 struct vsock_sock *vsk = vsock_sk(sk); 837 838 vsock_deassign_transport(vsk); 839 840 /* When clearing these addresses, there's no need to set the family and 841 * possibly register the address family with the kernel. 842 */ 843 vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 844 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY); 845 846 put_cred(vsk->owner); 847 } 848 849 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 850 { 851 int err; 852 853 err = sock_queue_rcv_skb(sk, skb); 854 if (err) 855 kfree_skb(skb); 856 857 return err; 858 } 859 860 struct sock *vsock_create_connected(struct sock *parent) 861 { 862 return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL, 863 parent->sk_type, 0); 864 } 865 EXPORT_SYMBOL_GPL(vsock_create_connected); 866 867 s64 vsock_stream_has_data(struct vsock_sock *vsk) 868 { 869 return vsk->transport->stream_has_data(vsk); 870 } 871 EXPORT_SYMBOL_GPL(vsock_stream_has_data); 872 873 s64 vsock_connectible_has_data(struct vsock_sock *vsk) 874 { 875 struct sock *sk = sk_vsock(vsk); 876 877 if (sk->sk_type == SOCK_SEQPACKET) 878 return vsk->transport->seqpacket_has_data(vsk); 879 else 880 return vsock_stream_has_data(vsk); 881 } 882 EXPORT_SYMBOL_GPL(vsock_connectible_has_data); 883 884 s64 vsock_stream_has_space(struct vsock_sock *vsk) 885 { 886 return vsk->transport->stream_has_space(vsk); 887 } 888 EXPORT_SYMBOL_GPL(vsock_stream_has_space); 889 890 void vsock_data_ready(struct sock *sk) 891 { 892 struct vsock_sock *vsk = vsock_sk(sk); 893 894 if (vsock_stream_has_data(vsk) >= sk->sk_rcvlowat || 895 sock_flag(sk, SOCK_DONE)) 896 sk->sk_data_ready(sk); 897 } 898 EXPORT_SYMBOL_GPL(vsock_data_ready); 899 900 static int vsock_release(struct socket *sock) 901 { 902 __vsock_release(sock->sk, 0); 903 sock->sk = NULL; 904 sock->state = SS_FREE; 905 906 return 0; 907 } 908 909 static int 910 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 911 { 912 int err; 913 struct sock *sk; 914 struct sockaddr_vm *vm_addr; 915 916 sk = sock->sk; 917 918 if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0) 919 return -EINVAL; 920 921 lock_sock(sk); 922 err = __vsock_bind(sk, vm_addr); 923 release_sock(sk); 924 925 return err; 926 } 927 928 static int vsock_getname(struct socket *sock, 929 struct sockaddr *addr, int peer) 930 { 931 int err; 932 struct sock *sk; 933 struct vsock_sock *vsk; 934 struct sockaddr_vm *vm_addr; 935 936 sk = sock->sk; 937 vsk = vsock_sk(sk); 938 err = 0; 939 940 lock_sock(sk); 941 942 if (peer) { 943 if (sock->state != SS_CONNECTED) { 944 err = -ENOTCONN; 945 goto out; 946 } 947 vm_addr = &vsk->remote_addr; 948 } else { 949 vm_addr = &vsk->local_addr; 950 } 951 952 if (!vm_addr) { 953 err = -EINVAL; 954 goto out; 955 } 956 957 /* sys_getsockname() and sys_getpeername() pass us a 958 * MAX_SOCK_ADDR-sized buffer and don't set addr_len. Unfortunately 959 * that macro is defined in socket.c instead of .h, so we hardcode its 960 * value here. 961 */ 962 BUILD_BUG_ON(sizeof(*vm_addr) > 128); 963 memcpy(addr, vm_addr, sizeof(*vm_addr)); 964 err = sizeof(*vm_addr); 965 966 out: 967 release_sock(sk); 968 return err; 969 } 970 971 static int vsock_shutdown(struct socket *sock, int mode) 972 { 973 int err; 974 struct sock *sk; 975 976 /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses 977 * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode 978 * here like the other address families do. Note also that the 979 * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3), 980 * which is what we want. 981 */ 982 mode++; 983 984 if ((mode & ~SHUTDOWN_MASK) || !mode) 985 return -EINVAL; 986 987 /* If this is a connection oriented socket and it is not connected then 988 * bail out immediately. If it is a DGRAM socket then we must first 989 * kick the socket so that it wakes up from any sleeping calls, for 990 * example recv(), and then afterwards return the error. 991 */ 992 993 sk = sock->sk; 994 995 lock_sock(sk); 996 if (sock->state == SS_UNCONNECTED) { 997 err = -ENOTCONN; 998 if (sock_type_connectible(sk->sk_type)) 999 goto out; 1000 } else { 1001 sock->state = SS_DISCONNECTING; 1002 err = 0; 1003 } 1004 1005 /* Receive and send shutdowns are treated alike. */ 1006 mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN); 1007 if (mode) { 1008 sk->sk_shutdown |= mode; 1009 sk->sk_state_change(sk); 1010 1011 if (sock_type_connectible(sk->sk_type)) { 1012 sock_reset_flag(sk, SOCK_DONE); 1013 vsock_send_shutdown(sk, mode); 1014 } 1015 } 1016 1017 out: 1018 release_sock(sk); 1019 return err; 1020 } 1021 1022 static __poll_t vsock_poll(struct file *file, struct socket *sock, 1023 poll_table *wait) 1024 { 1025 struct sock *sk; 1026 __poll_t mask; 1027 struct vsock_sock *vsk; 1028 1029 sk = sock->sk; 1030 vsk = vsock_sk(sk); 1031 1032 poll_wait(file, sk_sleep(sk), wait); 1033 mask = 0; 1034 1035 if (sk->sk_err) 1036 /* Signify that there has been an error on this socket. */ 1037 mask |= EPOLLERR; 1038 1039 /* INET sockets treat local write shutdown and peer write shutdown as a 1040 * case of EPOLLHUP set. 1041 */ 1042 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1043 ((sk->sk_shutdown & SEND_SHUTDOWN) && 1044 (vsk->peer_shutdown & SEND_SHUTDOWN))) { 1045 mask |= EPOLLHUP; 1046 } 1047 1048 if (sk->sk_shutdown & RCV_SHUTDOWN || 1049 vsk->peer_shutdown & SEND_SHUTDOWN) { 1050 mask |= EPOLLRDHUP; 1051 } 1052 1053 if (sk_is_readable(sk)) 1054 mask |= EPOLLIN | EPOLLRDNORM; 1055 1056 if (sock->type == SOCK_DGRAM) { 1057 /* For datagram sockets we can read if there is something in 1058 * the queue and write as long as the socket isn't shutdown for 1059 * sending. 1060 */ 1061 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) || 1062 (sk->sk_shutdown & RCV_SHUTDOWN)) { 1063 mask |= EPOLLIN | EPOLLRDNORM; 1064 } 1065 1066 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 1067 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND; 1068 1069 } else if (sock_type_connectible(sk->sk_type)) { 1070 const struct vsock_transport *transport; 1071 1072 lock_sock(sk); 1073 1074 transport = vsk->transport; 1075 1076 /* Listening sockets that have connections in their accept 1077 * queue can be read. 1078 */ 1079 if (sk->sk_state == TCP_LISTEN 1080 && !vsock_is_accept_queue_empty(sk)) 1081 mask |= EPOLLIN | EPOLLRDNORM; 1082 1083 /* If there is something in the queue then we can read. */ 1084 if (transport && transport->stream_is_active(vsk) && 1085 !(sk->sk_shutdown & RCV_SHUTDOWN)) { 1086 bool data_ready_now = false; 1087 int target = sock_rcvlowat(sk, 0, INT_MAX); 1088 int ret = transport->notify_poll_in( 1089 vsk, target, &data_ready_now); 1090 if (ret < 0) { 1091 mask |= EPOLLERR; 1092 } else { 1093 if (data_ready_now) 1094 mask |= EPOLLIN | EPOLLRDNORM; 1095 1096 } 1097 } 1098 1099 /* Sockets whose connections have been closed, reset, or 1100 * terminated should also be considered read, and we check the 1101 * shutdown flag for that. 1102 */ 1103 if (sk->sk_shutdown & RCV_SHUTDOWN || 1104 vsk->peer_shutdown & SEND_SHUTDOWN) { 1105 mask |= EPOLLIN | EPOLLRDNORM; 1106 } 1107 1108 /* Connected sockets that can produce data can be written. */ 1109 if (transport && sk->sk_state == TCP_ESTABLISHED) { 1110 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1111 bool space_avail_now = false; 1112 int ret = transport->notify_poll_out( 1113 vsk, 1, &space_avail_now); 1114 if (ret < 0) { 1115 mask |= EPOLLERR; 1116 } else { 1117 if (space_avail_now) 1118 /* Remove EPOLLWRBAND since INET 1119 * sockets are not setting it. 1120 */ 1121 mask |= EPOLLOUT | EPOLLWRNORM; 1122 1123 } 1124 } 1125 } 1126 1127 /* Simulate INET socket poll behaviors, which sets 1128 * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read, 1129 * but local send is not shutdown. 1130 */ 1131 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) { 1132 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) 1133 mask |= EPOLLOUT | EPOLLWRNORM; 1134 1135 } 1136 1137 release_sock(sk); 1138 } 1139 1140 return mask; 1141 } 1142 1143 static int vsock_read_skb(struct sock *sk, skb_read_actor_t read_actor) 1144 { 1145 struct vsock_sock *vsk = vsock_sk(sk); 1146 1147 return vsk->transport->read_skb(vsk, read_actor); 1148 } 1149 1150 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg, 1151 size_t len) 1152 { 1153 int err; 1154 struct sock *sk; 1155 struct vsock_sock *vsk; 1156 struct sockaddr_vm *remote_addr; 1157 const struct vsock_transport *transport; 1158 1159 if (msg->msg_flags & MSG_OOB) 1160 return -EOPNOTSUPP; 1161 1162 /* For now, MSG_DONTWAIT is always assumed... */ 1163 err = 0; 1164 sk = sock->sk; 1165 vsk = vsock_sk(sk); 1166 1167 lock_sock(sk); 1168 1169 transport = vsk->transport; 1170 1171 err = vsock_auto_bind(vsk); 1172 if (err) 1173 goto out; 1174 1175 1176 /* If the provided message contains an address, use that. Otherwise 1177 * fall back on the socket's remote handle (if it has been connected). 1178 */ 1179 if (msg->msg_name && 1180 vsock_addr_cast(msg->msg_name, msg->msg_namelen, 1181 &remote_addr) == 0) { 1182 /* Ensure this address is of the right type and is a valid 1183 * destination. 1184 */ 1185 1186 if (remote_addr->svm_cid == VMADDR_CID_ANY) 1187 remote_addr->svm_cid = transport->get_local_cid(); 1188 1189 if (!vsock_addr_bound(remote_addr)) { 1190 err = -EINVAL; 1191 goto out; 1192 } 1193 } else if (sock->state == SS_CONNECTED) { 1194 remote_addr = &vsk->remote_addr; 1195 1196 if (remote_addr->svm_cid == VMADDR_CID_ANY) 1197 remote_addr->svm_cid = transport->get_local_cid(); 1198 1199 /* XXX Should connect() or this function ensure remote_addr is 1200 * bound? 1201 */ 1202 if (!vsock_addr_bound(&vsk->remote_addr)) { 1203 err = -EINVAL; 1204 goto out; 1205 } 1206 } else { 1207 err = -EINVAL; 1208 goto out; 1209 } 1210 1211 if (!transport->dgram_allow(remote_addr->svm_cid, 1212 remote_addr->svm_port)) { 1213 err = -EINVAL; 1214 goto out; 1215 } 1216 1217 err = transport->dgram_enqueue(vsk, remote_addr, msg, len); 1218 1219 out: 1220 release_sock(sk); 1221 return err; 1222 } 1223 1224 static int vsock_dgram_connect(struct socket *sock, 1225 struct sockaddr *addr, int addr_len, int flags) 1226 { 1227 int err; 1228 struct sock *sk; 1229 struct vsock_sock *vsk; 1230 struct sockaddr_vm *remote_addr; 1231 1232 sk = sock->sk; 1233 vsk = vsock_sk(sk); 1234 1235 err = vsock_addr_cast(addr, addr_len, &remote_addr); 1236 if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) { 1237 lock_sock(sk); 1238 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, 1239 VMADDR_PORT_ANY); 1240 sock->state = SS_UNCONNECTED; 1241 release_sock(sk); 1242 return 0; 1243 } else if (err != 0) 1244 return -EINVAL; 1245 1246 lock_sock(sk); 1247 1248 err = vsock_auto_bind(vsk); 1249 if (err) 1250 goto out; 1251 1252 if (!vsk->transport->dgram_allow(remote_addr->svm_cid, 1253 remote_addr->svm_port)) { 1254 err = -EINVAL; 1255 goto out; 1256 } 1257 1258 memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr)); 1259 sock->state = SS_CONNECTED; 1260 1261 /* sock map disallows redirection of non-TCP sockets with sk_state != 1262 * TCP_ESTABLISHED (see sock_map_redirect_allowed()), so we set 1263 * TCP_ESTABLISHED here to allow redirection of connected vsock dgrams. 1264 * 1265 * This doesn't seem to be abnormal state for datagram sockets, as the 1266 * same approach can be see in other datagram socket types as well 1267 * (such as unix sockets). 1268 */ 1269 sk->sk_state = TCP_ESTABLISHED; 1270 1271 out: 1272 release_sock(sk); 1273 return err; 1274 } 1275 1276 int __vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg, 1277 size_t len, int flags) 1278 { 1279 struct sock *sk = sock->sk; 1280 struct vsock_sock *vsk = vsock_sk(sk); 1281 1282 return vsk->transport->dgram_dequeue(vsk, msg, len, flags); 1283 } 1284 1285 int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg, 1286 size_t len, int flags) 1287 { 1288 #ifdef CONFIG_BPF_SYSCALL 1289 struct sock *sk = sock->sk; 1290 const struct proto *prot; 1291 1292 prot = READ_ONCE(sk->sk_prot); 1293 if (prot != &vsock_proto) 1294 return prot->recvmsg(sk, msg, len, flags, NULL); 1295 #endif 1296 1297 return __vsock_dgram_recvmsg(sock, msg, len, flags); 1298 } 1299 EXPORT_SYMBOL_GPL(vsock_dgram_recvmsg); 1300 1301 static const struct proto_ops vsock_dgram_ops = { 1302 .family = PF_VSOCK, 1303 .owner = THIS_MODULE, 1304 .release = vsock_release, 1305 .bind = vsock_bind, 1306 .connect = vsock_dgram_connect, 1307 .socketpair = sock_no_socketpair, 1308 .accept = sock_no_accept, 1309 .getname = vsock_getname, 1310 .poll = vsock_poll, 1311 .ioctl = sock_no_ioctl, 1312 .listen = sock_no_listen, 1313 .shutdown = vsock_shutdown, 1314 .sendmsg = vsock_dgram_sendmsg, 1315 .recvmsg = vsock_dgram_recvmsg, 1316 .mmap = sock_no_mmap, 1317 .read_skb = vsock_read_skb, 1318 }; 1319 1320 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk) 1321 { 1322 const struct vsock_transport *transport = vsk->transport; 1323 1324 if (!transport || !transport->cancel_pkt) 1325 return -EOPNOTSUPP; 1326 1327 return transport->cancel_pkt(vsk); 1328 } 1329 1330 static void vsock_connect_timeout(struct work_struct *work) 1331 { 1332 struct sock *sk; 1333 struct vsock_sock *vsk; 1334 1335 vsk = container_of(work, struct vsock_sock, connect_work.work); 1336 sk = sk_vsock(vsk); 1337 1338 lock_sock(sk); 1339 if (sk->sk_state == TCP_SYN_SENT && 1340 (sk->sk_shutdown != SHUTDOWN_MASK)) { 1341 sk->sk_state = TCP_CLOSE; 1342 sk->sk_socket->state = SS_UNCONNECTED; 1343 sk->sk_err = ETIMEDOUT; 1344 sk_error_report(sk); 1345 vsock_transport_cancel_pkt(vsk); 1346 } 1347 release_sock(sk); 1348 1349 sock_put(sk); 1350 } 1351 1352 static int vsock_connect(struct socket *sock, struct sockaddr *addr, 1353 int addr_len, int flags) 1354 { 1355 int err; 1356 struct sock *sk; 1357 struct vsock_sock *vsk; 1358 const struct vsock_transport *transport; 1359 struct sockaddr_vm *remote_addr; 1360 long timeout; 1361 DEFINE_WAIT(wait); 1362 1363 err = 0; 1364 sk = sock->sk; 1365 vsk = vsock_sk(sk); 1366 1367 lock_sock(sk); 1368 1369 /* XXX AF_UNSPEC should make us disconnect like AF_INET. */ 1370 switch (sock->state) { 1371 case SS_CONNECTED: 1372 err = -EISCONN; 1373 goto out; 1374 case SS_DISCONNECTING: 1375 err = -EINVAL; 1376 goto out; 1377 case SS_CONNECTING: 1378 /* This continues on so we can move sock into the SS_CONNECTED 1379 * state once the connection has completed (at which point err 1380 * will be set to zero also). Otherwise, we will either wait 1381 * for the connection or return -EALREADY should this be a 1382 * non-blocking call. 1383 */ 1384 err = -EALREADY; 1385 if (flags & O_NONBLOCK) 1386 goto out; 1387 break; 1388 default: 1389 if ((sk->sk_state == TCP_LISTEN) || 1390 vsock_addr_cast(addr, addr_len, &remote_addr) != 0) { 1391 err = -EINVAL; 1392 goto out; 1393 } 1394 1395 /* Set the remote address that we are connecting to. */ 1396 memcpy(&vsk->remote_addr, remote_addr, 1397 sizeof(vsk->remote_addr)); 1398 1399 err = vsock_assign_transport(vsk, NULL); 1400 if (err) 1401 goto out; 1402 1403 transport = vsk->transport; 1404 1405 /* The hypervisor and well-known contexts do not have socket 1406 * endpoints. 1407 */ 1408 if (!transport || 1409 !transport->stream_allow(remote_addr->svm_cid, 1410 remote_addr->svm_port)) { 1411 err = -ENETUNREACH; 1412 goto out; 1413 } 1414 1415 err = vsock_auto_bind(vsk); 1416 if (err) 1417 goto out; 1418 1419 sk->sk_state = TCP_SYN_SENT; 1420 1421 err = transport->connect(vsk); 1422 if (err < 0) 1423 goto out; 1424 1425 /* Mark sock as connecting and set the error code to in 1426 * progress in case this is a non-blocking connect. 1427 */ 1428 sock->state = SS_CONNECTING; 1429 err = -EINPROGRESS; 1430 } 1431 1432 /* The receive path will handle all communication until we are able to 1433 * enter the connected state. Here we wait for the connection to be 1434 * completed or a notification of an error. 1435 */ 1436 timeout = vsk->connect_timeout; 1437 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1438 1439 while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) { 1440 if (flags & O_NONBLOCK) { 1441 /* If we're not going to block, we schedule a timeout 1442 * function to generate a timeout on the connection 1443 * attempt, in case the peer doesn't respond in a 1444 * timely manner. We hold on to the socket until the 1445 * timeout fires. 1446 */ 1447 sock_hold(sk); 1448 1449 /* If the timeout function is already scheduled, 1450 * reschedule it, then ungrab the socket refcount to 1451 * keep it balanced. 1452 */ 1453 if (mod_delayed_work(system_wq, &vsk->connect_work, 1454 timeout)) 1455 sock_put(sk); 1456 1457 /* Skip ahead to preserve error code set above. */ 1458 goto out_wait; 1459 } 1460 1461 release_sock(sk); 1462 timeout = schedule_timeout(timeout); 1463 lock_sock(sk); 1464 1465 if (signal_pending(current)) { 1466 err = sock_intr_errno(timeout); 1467 sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE; 1468 sock->state = SS_UNCONNECTED; 1469 vsock_transport_cancel_pkt(vsk); 1470 vsock_remove_connected(vsk); 1471 goto out_wait; 1472 } else if ((sk->sk_state != TCP_ESTABLISHED) && (timeout == 0)) { 1473 err = -ETIMEDOUT; 1474 sk->sk_state = TCP_CLOSE; 1475 sock->state = SS_UNCONNECTED; 1476 vsock_transport_cancel_pkt(vsk); 1477 goto out_wait; 1478 } 1479 1480 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1481 } 1482 1483 if (sk->sk_err) { 1484 err = -sk->sk_err; 1485 sk->sk_state = TCP_CLOSE; 1486 sock->state = SS_UNCONNECTED; 1487 } else { 1488 err = 0; 1489 } 1490 1491 out_wait: 1492 finish_wait(sk_sleep(sk), &wait); 1493 out: 1494 release_sock(sk); 1495 return err; 1496 } 1497 1498 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags, 1499 bool kern) 1500 { 1501 struct sock *listener; 1502 int err; 1503 struct sock *connected; 1504 struct vsock_sock *vconnected; 1505 long timeout; 1506 DEFINE_WAIT(wait); 1507 1508 err = 0; 1509 listener = sock->sk; 1510 1511 lock_sock(listener); 1512 1513 if (!sock_type_connectible(sock->type)) { 1514 err = -EOPNOTSUPP; 1515 goto out; 1516 } 1517 1518 if (listener->sk_state != TCP_LISTEN) { 1519 err = -EINVAL; 1520 goto out; 1521 } 1522 1523 /* Wait for children sockets to appear; these are the new sockets 1524 * created upon connection establishment. 1525 */ 1526 timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK); 1527 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1528 1529 while ((connected = vsock_dequeue_accept(listener)) == NULL && 1530 listener->sk_err == 0) { 1531 release_sock(listener); 1532 timeout = schedule_timeout(timeout); 1533 finish_wait(sk_sleep(listener), &wait); 1534 lock_sock(listener); 1535 1536 if (signal_pending(current)) { 1537 err = sock_intr_errno(timeout); 1538 goto out; 1539 } else if (timeout == 0) { 1540 err = -EAGAIN; 1541 goto out; 1542 } 1543 1544 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); 1545 } 1546 finish_wait(sk_sleep(listener), &wait); 1547 1548 if (listener->sk_err) 1549 err = -listener->sk_err; 1550 1551 if (connected) { 1552 sk_acceptq_removed(listener); 1553 1554 lock_sock_nested(connected, SINGLE_DEPTH_NESTING); 1555 vconnected = vsock_sk(connected); 1556 1557 /* If the listener socket has received an error, then we should 1558 * reject this socket and return. Note that we simply mark the 1559 * socket rejected, drop our reference, and let the cleanup 1560 * function handle the cleanup; the fact that we found it in 1561 * the listener's accept queue guarantees that the cleanup 1562 * function hasn't run yet. 1563 */ 1564 if (err) { 1565 vconnected->rejected = true; 1566 } else { 1567 newsock->state = SS_CONNECTED; 1568 sock_graft(connected, newsock); 1569 } 1570 1571 release_sock(connected); 1572 sock_put(connected); 1573 } 1574 1575 out: 1576 release_sock(listener); 1577 return err; 1578 } 1579 1580 static int vsock_listen(struct socket *sock, int backlog) 1581 { 1582 int err; 1583 struct sock *sk; 1584 struct vsock_sock *vsk; 1585 1586 sk = sock->sk; 1587 1588 lock_sock(sk); 1589 1590 if (!sock_type_connectible(sk->sk_type)) { 1591 err = -EOPNOTSUPP; 1592 goto out; 1593 } 1594 1595 if (sock->state != SS_UNCONNECTED) { 1596 err = -EINVAL; 1597 goto out; 1598 } 1599 1600 vsk = vsock_sk(sk); 1601 1602 if (!vsock_addr_bound(&vsk->local_addr)) { 1603 err = -EINVAL; 1604 goto out; 1605 } 1606 1607 sk->sk_max_ack_backlog = backlog; 1608 sk->sk_state = TCP_LISTEN; 1609 1610 err = 0; 1611 1612 out: 1613 release_sock(sk); 1614 return err; 1615 } 1616 1617 static void vsock_update_buffer_size(struct vsock_sock *vsk, 1618 const struct vsock_transport *transport, 1619 u64 val) 1620 { 1621 if (val > vsk->buffer_max_size) 1622 val = vsk->buffer_max_size; 1623 1624 if (val < vsk->buffer_min_size) 1625 val = vsk->buffer_min_size; 1626 1627 if (val != vsk->buffer_size && 1628 transport && transport->notify_buffer_size) 1629 transport->notify_buffer_size(vsk, &val); 1630 1631 vsk->buffer_size = val; 1632 } 1633 1634 static int vsock_connectible_setsockopt(struct socket *sock, 1635 int level, 1636 int optname, 1637 sockptr_t optval, 1638 unsigned int optlen) 1639 { 1640 int err; 1641 struct sock *sk; 1642 struct vsock_sock *vsk; 1643 const struct vsock_transport *transport; 1644 u64 val; 1645 1646 if (level != AF_VSOCK) 1647 return -ENOPROTOOPT; 1648 1649 #define COPY_IN(_v) \ 1650 do { \ 1651 if (optlen < sizeof(_v)) { \ 1652 err = -EINVAL; \ 1653 goto exit; \ 1654 } \ 1655 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) { \ 1656 err = -EFAULT; \ 1657 goto exit; \ 1658 } \ 1659 } while (0) 1660 1661 err = 0; 1662 sk = sock->sk; 1663 vsk = vsock_sk(sk); 1664 1665 lock_sock(sk); 1666 1667 transport = vsk->transport; 1668 1669 switch (optname) { 1670 case SO_VM_SOCKETS_BUFFER_SIZE: 1671 COPY_IN(val); 1672 vsock_update_buffer_size(vsk, transport, val); 1673 break; 1674 1675 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1676 COPY_IN(val); 1677 vsk->buffer_max_size = val; 1678 vsock_update_buffer_size(vsk, transport, vsk->buffer_size); 1679 break; 1680 1681 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1682 COPY_IN(val); 1683 vsk->buffer_min_size = val; 1684 vsock_update_buffer_size(vsk, transport, vsk->buffer_size); 1685 break; 1686 1687 case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW: 1688 case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: { 1689 struct __kernel_sock_timeval tv; 1690 1691 err = sock_copy_user_timeval(&tv, optval, optlen, 1692 optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD); 1693 if (err) 1694 break; 1695 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC && 1696 tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) { 1697 vsk->connect_timeout = tv.tv_sec * HZ + 1698 DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ)); 1699 if (vsk->connect_timeout == 0) 1700 vsk->connect_timeout = 1701 VSOCK_DEFAULT_CONNECT_TIMEOUT; 1702 1703 } else { 1704 err = -ERANGE; 1705 } 1706 break; 1707 } 1708 1709 default: 1710 err = -ENOPROTOOPT; 1711 break; 1712 } 1713 1714 #undef COPY_IN 1715 1716 exit: 1717 release_sock(sk); 1718 return err; 1719 } 1720 1721 static int vsock_connectible_getsockopt(struct socket *sock, 1722 int level, int optname, 1723 char __user *optval, 1724 int __user *optlen) 1725 { 1726 struct sock *sk = sock->sk; 1727 struct vsock_sock *vsk = vsock_sk(sk); 1728 1729 union { 1730 u64 val64; 1731 struct old_timeval32 tm32; 1732 struct __kernel_old_timeval tm; 1733 struct __kernel_sock_timeval stm; 1734 } v; 1735 1736 int lv = sizeof(v.val64); 1737 int len; 1738 1739 if (level != AF_VSOCK) 1740 return -ENOPROTOOPT; 1741 1742 if (get_user(len, optlen)) 1743 return -EFAULT; 1744 1745 memset(&v, 0, sizeof(v)); 1746 1747 switch (optname) { 1748 case SO_VM_SOCKETS_BUFFER_SIZE: 1749 v.val64 = vsk->buffer_size; 1750 break; 1751 1752 case SO_VM_SOCKETS_BUFFER_MAX_SIZE: 1753 v.val64 = vsk->buffer_max_size; 1754 break; 1755 1756 case SO_VM_SOCKETS_BUFFER_MIN_SIZE: 1757 v.val64 = vsk->buffer_min_size; 1758 break; 1759 1760 case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW: 1761 case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: 1762 lv = sock_get_timeout(vsk->connect_timeout, &v, 1763 optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD); 1764 break; 1765 1766 default: 1767 return -ENOPROTOOPT; 1768 } 1769 1770 if (len < lv) 1771 return -EINVAL; 1772 if (len > lv) 1773 len = lv; 1774 if (copy_to_user(optval, &v, len)) 1775 return -EFAULT; 1776 1777 if (put_user(len, optlen)) 1778 return -EFAULT; 1779 1780 return 0; 1781 } 1782 1783 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg, 1784 size_t len) 1785 { 1786 struct sock *sk; 1787 struct vsock_sock *vsk; 1788 const struct vsock_transport *transport; 1789 ssize_t total_written; 1790 long timeout; 1791 int err; 1792 struct vsock_transport_send_notify_data send_data; 1793 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1794 1795 sk = sock->sk; 1796 vsk = vsock_sk(sk); 1797 total_written = 0; 1798 err = 0; 1799 1800 if (msg->msg_flags & MSG_OOB) 1801 return -EOPNOTSUPP; 1802 1803 lock_sock(sk); 1804 1805 transport = vsk->transport; 1806 1807 /* Callers should not provide a destination with connection oriented 1808 * sockets. 1809 */ 1810 if (msg->msg_namelen) { 1811 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1812 goto out; 1813 } 1814 1815 /* Send data only if both sides are not shutdown in the direction. */ 1816 if (sk->sk_shutdown & SEND_SHUTDOWN || 1817 vsk->peer_shutdown & RCV_SHUTDOWN) { 1818 err = -EPIPE; 1819 goto out; 1820 } 1821 1822 if (!transport || sk->sk_state != TCP_ESTABLISHED || 1823 !vsock_addr_bound(&vsk->local_addr)) { 1824 err = -ENOTCONN; 1825 goto out; 1826 } 1827 1828 if (!vsock_addr_bound(&vsk->remote_addr)) { 1829 err = -EDESTADDRREQ; 1830 goto out; 1831 } 1832 1833 /* Wait for room in the produce queue to enqueue our user's data. */ 1834 timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1835 1836 err = transport->notify_send_init(vsk, &send_data); 1837 if (err < 0) 1838 goto out; 1839 1840 while (total_written < len) { 1841 ssize_t written; 1842 1843 add_wait_queue(sk_sleep(sk), &wait); 1844 while (vsock_stream_has_space(vsk) == 0 && 1845 sk->sk_err == 0 && 1846 !(sk->sk_shutdown & SEND_SHUTDOWN) && 1847 !(vsk->peer_shutdown & RCV_SHUTDOWN)) { 1848 1849 /* Don't wait for non-blocking sockets. */ 1850 if (timeout == 0) { 1851 err = -EAGAIN; 1852 remove_wait_queue(sk_sleep(sk), &wait); 1853 goto out_err; 1854 } 1855 1856 err = transport->notify_send_pre_block(vsk, &send_data); 1857 if (err < 0) { 1858 remove_wait_queue(sk_sleep(sk), &wait); 1859 goto out_err; 1860 } 1861 1862 release_sock(sk); 1863 timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout); 1864 lock_sock(sk); 1865 if (signal_pending(current)) { 1866 err = sock_intr_errno(timeout); 1867 remove_wait_queue(sk_sleep(sk), &wait); 1868 goto out_err; 1869 } else if (timeout == 0) { 1870 err = -EAGAIN; 1871 remove_wait_queue(sk_sleep(sk), &wait); 1872 goto out_err; 1873 } 1874 } 1875 remove_wait_queue(sk_sleep(sk), &wait); 1876 1877 /* These checks occur both as part of and after the loop 1878 * conditional since we need to check before and after 1879 * sleeping. 1880 */ 1881 if (sk->sk_err) { 1882 err = -sk->sk_err; 1883 goto out_err; 1884 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) || 1885 (vsk->peer_shutdown & RCV_SHUTDOWN)) { 1886 err = -EPIPE; 1887 goto out_err; 1888 } 1889 1890 err = transport->notify_send_pre_enqueue(vsk, &send_data); 1891 if (err < 0) 1892 goto out_err; 1893 1894 /* Note that enqueue will only write as many bytes as are free 1895 * in the produce queue, so we don't need to ensure len is 1896 * smaller than the queue size. It is the caller's 1897 * responsibility to check how many bytes we were able to send. 1898 */ 1899 1900 if (sk->sk_type == SOCK_SEQPACKET) { 1901 written = transport->seqpacket_enqueue(vsk, 1902 msg, len - total_written); 1903 } else { 1904 written = transport->stream_enqueue(vsk, 1905 msg, len - total_written); 1906 } 1907 1908 if (written < 0) { 1909 err = written; 1910 goto out_err; 1911 } 1912 1913 total_written += written; 1914 1915 err = transport->notify_send_post_enqueue( 1916 vsk, written, &send_data); 1917 if (err < 0) 1918 goto out_err; 1919 1920 } 1921 1922 out_err: 1923 if (total_written > 0) { 1924 /* Return number of written bytes only if: 1925 * 1) SOCK_STREAM socket. 1926 * 2) SOCK_SEQPACKET socket when whole buffer is sent. 1927 */ 1928 if (sk->sk_type == SOCK_STREAM || total_written == len) 1929 err = total_written; 1930 } 1931 out: 1932 release_sock(sk); 1933 return err; 1934 } 1935 1936 static int vsock_connectible_wait_data(struct sock *sk, 1937 struct wait_queue_entry *wait, 1938 long timeout, 1939 struct vsock_transport_recv_notify_data *recv_data, 1940 size_t target) 1941 { 1942 const struct vsock_transport *transport; 1943 struct vsock_sock *vsk; 1944 s64 data; 1945 int err; 1946 1947 vsk = vsock_sk(sk); 1948 err = 0; 1949 transport = vsk->transport; 1950 1951 while (1) { 1952 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE); 1953 data = vsock_connectible_has_data(vsk); 1954 if (data != 0) 1955 break; 1956 1957 if (sk->sk_err != 0 || 1958 (sk->sk_shutdown & RCV_SHUTDOWN) || 1959 (vsk->peer_shutdown & SEND_SHUTDOWN)) { 1960 break; 1961 } 1962 1963 /* Don't wait for non-blocking sockets. */ 1964 if (timeout == 0) { 1965 err = -EAGAIN; 1966 break; 1967 } 1968 1969 if (recv_data) { 1970 err = transport->notify_recv_pre_block(vsk, target, recv_data); 1971 if (err < 0) 1972 break; 1973 } 1974 1975 release_sock(sk); 1976 timeout = schedule_timeout(timeout); 1977 lock_sock(sk); 1978 1979 if (signal_pending(current)) { 1980 err = sock_intr_errno(timeout); 1981 break; 1982 } else if (timeout == 0) { 1983 err = -EAGAIN; 1984 break; 1985 } 1986 } 1987 1988 finish_wait(sk_sleep(sk), wait); 1989 1990 if (err) 1991 return err; 1992 1993 /* Internal transport error when checking for available 1994 * data. XXX This should be changed to a connection 1995 * reset in a later change. 1996 */ 1997 if (data < 0) 1998 return -ENOMEM; 1999 2000 return data; 2001 } 2002 2003 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg, 2004 size_t len, int flags) 2005 { 2006 struct vsock_transport_recv_notify_data recv_data; 2007 const struct vsock_transport *transport; 2008 struct vsock_sock *vsk; 2009 ssize_t copied; 2010 size_t target; 2011 long timeout; 2012 int err; 2013 2014 DEFINE_WAIT(wait); 2015 2016 vsk = vsock_sk(sk); 2017 transport = vsk->transport; 2018 2019 /* We must not copy less than target bytes into the user's buffer 2020 * before returning successfully, so we wait for the consume queue to 2021 * have that much data to consume before dequeueing. Note that this 2022 * makes it impossible to handle cases where target is greater than the 2023 * queue size. 2024 */ 2025 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); 2026 if (target >= transport->stream_rcvhiwat(vsk)) { 2027 err = -ENOMEM; 2028 goto out; 2029 } 2030 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 2031 copied = 0; 2032 2033 err = transport->notify_recv_init(vsk, target, &recv_data); 2034 if (err < 0) 2035 goto out; 2036 2037 2038 while (1) { 2039 ssize_t read; 2040 2041 err = vsock_connectible_wait_data(sk, &wait, timeout, 2042 &recv_data, target); 2043 if (err <= 0) 2044 break; 2045 2046 err = transport->notify_recv_pre_dequeue(vsk, target, 2047 &recv_data); 2048 if (err < 0) 2049 break; 2050 2051 read = transport->stream_dequeue(vsk, msg, len - copied, flags); 2052 if (read < 0) { 2053 err = read; 2054 break; 2055 } 2056 2057 copied += read; 2058 2059 err = transport->notify_recv_post_dequeue(vsk, target, read, 2060 !(flags & MSG_PEEK), &recv_data); 2061 if (err < 0) 2062 goto out; 2063 2064 if (read >= target || flags & MSG_PEEK) 2065 break; 2066 2067 target -= read; 2068 } 2069 2070 if (sk->sk_err) 2071 err = -sk->sk_err; 2072 else if (sk->sk_shutdown & RCV_SHUTDOWN) 2073 err = 0; 2074 2075 if (copied > 0) 2076 err = copied; 2077 2078 out: 2079 return err; 2080 } 2081 2082 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg, 2083 size_t len, int flags) 2084 { 2085 const struct vsock_transport *transport; 2086 struct vsock_sock *vsk; 2087 ssize_t msg_len; 2088 long timeout; 2089 int err = 0; 2090 DEFINE_WAIT(wait); 2091 2092 vsk = vsock_sk(sk); 2093 transport = vsk->transport; 2094 2095 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 2096 2097 err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0); 2098 if (err <= 0) 2099 goto out; 2100 2101 msg_len = transport->seqpacket_dequeue(vsk, msg, flags); 2102 2103 if (msg_len < 0) { 2104 err = msg_len; 2105 goto out; 2106 } 2107 2108 if (sk->sk_err) { 2109 err = -sk->sk_err; 2110 } else if (sk->sk_shutdown & RCV_SHUTDOWN) { 2111 err = 0; 2112 } else { 2113 /* User sets MSG_TRUNC, so return real length of 2114 * packet. 2115 */ 2116 if (flags & MSG_TRUNC) 2117 err = msg_len; 2118 else 2119 err = len - msg_data_left(msg); 2120 2121 /* Always set MSG_TRUNC if real length of packet is 2122 * bigger than user's buffer. 2123 */ 2124 if (msg_len > len) 2125 msg->msg_flags |= MSG_TRUNC; 2126 } 2127 2128 out: 2129 return err; 2130 } 2131 2132 int 2133 __vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2134 int flags) 2135 { 2136 struct sock *sk; 2137 struct vsock_sock *vsk; 2138 const struct vsock_transport *transport; 2139 int err; 2140 2141 sk = sock->sk; 2142 2143 if (unlikely(flags & MSG_ERRQUEUE)) 2144 return sock_recv_errqueue(sk, msg, len, SOL_VSOCK, VSOCK_RECVERR); 2145 2146 vsk = vsock_sk(sk); 2147 err = 0; 2148 2149 lock_sock(sk); 2150 2151 transport = vsk->transport; 2152 2153 if (!transport || sk->sk_state != TCP_ESTABLISHED) { 2154 /* Recvmsg is supposed to return 0 if a peer performs an 2155 * orderly shutdown. Differentiate between that case and when a 2156 * peer has not connected or a local shutdown occurred with the 2157 * SOCK_DONE flag. 2158 */ 2159 if (sock_flag(sk, SOCK_DONE)) 2160 err = 0; 2161 else 2162 err = -ENOTCONN; 2163 2164 goto out; 2165 } 2166 2167 if (flags & MSG_OOB) { 2168 err = -EOPNOTSUPP; 2169 goto out; 2170 } 2171 2172 /* We don't check peer_shutdown flag here since peer may actually shut 2173 * down, but there can be data in the queue that a local socket can 2174 * receive. 2175 */ 2176 if (sk->sk_shutdown & RCV_SHUTDOWN) { 2177 err = 0; 2178 goto out; 2179 } 2180 2181 /* It is valid on Linux to pass in a zero-length receive buffer. This 2182 * is not an error. We may as well bail out now. 2183 */ 2184 if (!len) { 2185 err = 0; 2186 goto out; 2187 } 2188 2189 if (sk->sk_type == SOCK_STREAM) 2190 err = __vsock_stream_recvmsg(sk, msg, len, flags); 2191 else 2192 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags); 2193 2194 out: 2195 release_sock(sk); 2196 return err; 2197 } 2198 2199 int 2200 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2201 int flags) 2202 { 2203 #ifdef CONFIG_BPF_SYSCALL 2204 struct sock *sk = sock->sk; 2205 const struct proto *prot; 2206 2207 prot = READ_ONCE(sk->sk_prot); 2208 if (prot != &vsock_proto) 2209 return prot->recvmsg(sk, msg, len, flags, NULL); 2210 #endif 2211 2212 return __vsock_connectible_recvmsg(sock, msg, len, flags); 2213 } 2214 EXPORT_SYMBOL_GPL(vsock_connectible_recvmsg); 2215 2216 static int vsock_set_rcvlowat(struct sock *sk, int val) 2217 { 2218 const struct vsock_transport *transport; 2219 struct vsock_sock *vsk; 2220 2221 vsk = vsock_sk(sk); 2222 2223 if (val > vsk->buffer_size) 2224 return -EINVAL; 2225 2226 transport = vsk->transport; 2227 2228 if (transport && transport->notify_set_rcvlowat) { 2229 int err; 2230 2231 err = transport->notify_set_rcvlowat(vsk, val); 2232 if (err) 2233 return err; 2234 } 2235 2236 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 2237 return 0; 2238 } 2239 2240 static const struct proto_ops vsock_stream_ops = { 2241 .family = PF_VSOCK, 2242 .owner = THIS_MODULE, 2243 .release = vsock_release, 2244 .bind = vsock_bind, 2245 .connect = vsock_connect, 2246 .socketpair = sock_no_socketpair, 2247 .accept = vsock_accept, 2248 .getname = vsock_getname, 2249 .poll = vsock_poll, 2250 .ioctl = sock_no_ioctl, 2251 .listen = vsock_listen, 2252 .shutdown = vsock_shutdown, 2253 .setsockopt = vsock_connectible_setsockopt, 2254 .getsockopt = vsock_connectible_getsockopt, 2255 .sendmsg = vsock_connectible_sendmsg, 2256 .recvmsg = vsock_connectible_recvmsg, 2257 .mmap = sock_no_mmap, 2258 .set_rcvlowat = vsock_set_rcvlowat, 2259 .read_skb = vsock_read_skb, 2260 }; 2261 2262 static const struct proto_ops vsock_seqpacket_ops = { 2263 .family = PF_VSOCK, 2264 .owner = THIS_MODULE, 2265 .release = vsock_release, 2266 .bind = vsock_bind, 2267 .connect = vsock_connect, 2268 .socketpair = sock_no_socketpair, 2269 .accept = vsock_accept, 2270 .getname = vsock_getname, 2271 .poll = vsock_poll, 2272 .ioctl = sock_no_ioctl, 2273 .listen = vsock_listen, 2274 .shutdown = vsock_shutdown, 2275 .setsockopt = vsock_connectible_setsockopt, 2276 .getsockopt = vsock_connectible_getsockopt, 2277 .sendmsg = vsock_connectible_sendmsg, 2278 .recvmsg = vsock_connectible_recvmsg, 2279 .mmap = sock_no_mmap, 2280 .read_skb = vsock_read_skb, 2281 }; 2282 2283 static int vsock_create(struct net *net, struct socket *sock, 2284 int protocol, int kern) 2285 { 2286 struct vsock_sock *vsk; 2287 struct sock *sk; 2288 int ret; 2289 2290 if (!sock) 2291 return -EINVAL; 2292 2293 if (protocol && protocol != PF_VSOCK) 2294 return -EPROTONOSUPPORT; 2295 2296 switch (sock->type) { 2297 case SOCK_DGRAM: 2298 sock->ops = &vsock_dgram_ops; 2299 break; 2300 case SOCK_STREAM: 2301 sock->ops = &vsock_stream_ops; 2302 break; 2303 case SOCK_SEQPACKET: 2304 sock->ops = &vsock_seqpacket_ops; 2305 break; 2306 default: 2307 return -ESOCKTNOSUPPORT; 2308 } 2309 2310 sock->state = SS_UNCONNECTED; 2311 2312 sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern); 2313 if (!sk) 2314 return -ENOMEM; 2315 2316 vsk = vsock_sk(sk); 2317 2318 if (sock->type == SOCK_DGRAM) { 2319 ret = vsock_assign_transport(vsk, NULL); 2320 if (ret < 0) { 2321 sock_put(sk); 2322 return ret; 2323 } 2324 } 2325 2326 vsock_insert_unbound(vsk); 2327 2328 return 0; 2329 } 2330 2331 static const struct net_proto_family vsock_family_ops = { 2332 .family = AF_VSOCK, 2333 .create = vsock_create, 2334 .owner = THIS_MODULE, 2335 }; 2336 2337 static long vsock_dev_do_ioctl(struct file *filp, 2338 unsigned int cmd, void __user *ptr) 2339 { 2340 u32 __user *p = ptr; 2341 u32 cid = VMADDR_CID_ANY; 2342 int retval = 0; 2343 2344 switch (cmd) { 2345 case IOCTL_VM_SOCKETS_GET_LOCAL_CID: 2346 /* To be compatible with the VMCI behavior, we prioritize the 2347 * guest CID instead of well-know host CID (VMADDR_CID_HOST). 2348 */ 2349 if (transport_g2h) 2350 cid = transport_g2h->get_local_cid(); 2351 else if (transport_h2g) 2352 cid = transport_h2g->get_local_cid(); 2353 2354 if (put_user(cid, p) != 0) 2355 retval = -EFAULT; 2356 break; 2357 2358 default: 2359 retval = -ENOIOCTLCMD; 2360 } 2361 2362 return retval; 2363 } 2364 2365 static long vsock_dev_ioctl(struct file *filp, 2366 unsigned int cmd, unsigned long arg) 2367 { 2368 return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg); 2369 } 2370 2371 #ifdef CONFIG_COMPAT 2372 static long vsock_dev_compat_ioctl(struct file *filp, 2373 unsigned int cmd, unsigned long arg) 2374 { 2375 return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg)); 2376 } 2377 #endif 2378 2379 static const struct file_operations vsock_device_ops = { 2380 .owner = THIS_MODULE, 2381 .unlocked_ioctl = vsock_dev_ioctl, 2382 #ifdef CONFIG_COMPAT 2383 .compat_ioctl = vsock_dev_compat_ioctl, 2384 #endif 2385 .open = nonseekable_open, 2386 }; 2387 2388 static struct miscdevice vsock_device = { 2389 .name = "vsock", 2390 .fops = &vsock_device_ops, 2391 }; 2392 2393 static int __init vsock_init(void) 2394 { 2395 int err = 0; 2396 2397 vsock_init_tables(); 2398 2399 vsock_proto.owner = THIS_MODULE; 2400 vsock_device.minor = MISC_DYNAMIC_MINOR; 2401 err = misc_register(&vsock_device); 2402 if (err) { 2403 pr_err("Failed to register misc device\n"); 2404 goto err_reset_transport; 2405 } 2406 2407 err = proto_register(&vsock_proto, 1); /* we want our slab */ 2408 if (err) { 2409 pr_err("Cannot register vsock protocol\n"); 2410 goto err_deregister_misc; 2411 } 2412 2413 err = sock_register(&vsock_family_ops); 2414 if (err) { 2415 pr_err("could not register af_vsock (%d) address family: %d\n", 2416 AF_VSOCK, err); 2417 goto err_unregister_proto; 2418 } 2419 2420 vsock_bpf_build_proto(); 2421 2422 return 0; 2423 2424 err_unregister_proto: 2425 proto_unregister(&vsock_proto); 2426 err_deregister_misc: 2427 misc_deregister(&vsock_device); 2428 err_reset_transport: 2429 return err; 2430 } 2431 2432 static void __exit vsock_exit(void) 2433 { 2434 misc_deregister(&vsock_device); 2435 sock_unregister(AF_VSOCK); 2436 proto_unregister(&vsock_proto); 2437 } 2438 2439 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk) 2440 { 2441 return vsk->transport; 2442 } 2443 EXPORT_SYMBOL_GPL(vsock_core_get_transport); 2444 2445 int vsock_core_register(const struct vsock_transport *t, int features) 2446 { 2447 const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local; 2448 int err = mutex_lock_interruptible(&vsock_register_mutex); 2449 2450 if (err) 2451 return err; 2452 2453 t_h2g = transport_h2g; 2454 t_g2h = transport_g2h; 2455 t_dgram = transport_dgram; 2456 t_local = transport_local; 2457 2458 if (features & VSOCK_TRANSPORT_F_H2G) { 2459 if (t_h2g) { 2460 err = -EBUSY; 2461 goto err_busy; 2462 } 2463 t_h2g = t; 2464 } 2465 2466 if (features & VSOCK_TRANSPORT_F_G2H) { 2467 if (t_g2h) { 2468 err = -EBUSY; 2469 goto err_busy; 2470 } 2471 t_g2h = t; 2472 } 2473 2474 if (features & VSOCK_TRANSPORT_F_DGRAM) { 2475 if (t_dgram) { 2476 err = -EBUSY; 2477 goto err_busy; 2478 } 2479 t_dgram = t; 2480 } 2481 2482 if (features & VSOCK_TRANSPORT_F_LOCAL) { 2483 if (t_local) { 2484 err = -EBUSY; 2485 goto err_busy; 2486 } 2487 t_local = t; 2488 } 2489 2490 transport_h2g = t_h2g; 2491 transport_g2h = t_g2h; 2492 transport_dgram = t_dgram; 2493 transport_local = t_local; 2494 2495 err_busy: 2496 mutex_unlock(&vsock_register_mutex); 2497 return err; 2498 } 2499 EXPORT_SYMBOL_GPL(vsock_core_register); 2500 2501 void vsock_core_unregister(const struct vsock_transport *t) 2502 { 2503 mutex_lock(&vsock_register_mutex); 2504 2505 if (transport_h2g == t) 2506 transport_h2g = NULL; 2507 2508 if (transport_g2h == t) 2509 transport_g2h = NULL; 2510 2511 if (transport_dgram == t) 2512 transport_dgram = NULL; 2513 2514 if (transport_local == t) 2515 transport_local = NULL; 2516 2517 mutex_unlock(&vsock_register_mutex); 2518 } 2519 EXPORT_SYMBOL_GPL(vsock_core_unregister); 2520 2521 module_init(vsock_init); 2522 module_exit(vsock_exit); 2523 2524 MODULE_AUTHOR("VMware, Inc."); 2525 MODULE_DESCRIPTION("VMware Virtual Socket Family"); 2526 MODULE_VERSION("1.0.2.0-k"); 2527 MODULE_LICENSE("GPL v2"); 2528