1 /* 2 * net/tipc/socket.c: TIPC socket API 3 * 4 * Copyright (c) 2001-2007, Ericsson AB 5 * Copyright (c) 2004-2008, 2010-2011, Wind River Systems 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the names of the copyright holders nor the names of its 17 * contributors may be used to endorse or promote products derived from 18 * this software without specific prior written permission. 19 * 20 * Alternatively, this software may be distributed under the terms of the 21 * GNU General Public License ("GPL") version 2 as published by the Free 22 * Software Foundation. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 25 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 28 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 31 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 32 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 33 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 34 * POSSIBILITY OF SUCH DAMAGE. 35 */ 36 37 #include "core.h" 38 #include "port.h" 39 40 #include <linux/export.h> 41 #include <net/sock.h> 42 43 #define SS_LISTENING -1 /* socket is listening */ 44 #define SS_READY -2 /* socket is connectionless */ 45 46 #define OVERLOAD_LIMIT_BASE 5000 47 #define CONN_TIMEOUT_DEFAULT 8000 /* default connect timeout = 8s */ 48 49 struct tipc_sock { 50 struct sock sk; 51 struct tipc_port *p; 52 struct tipc_portid peer_name; 53 unsigned int conn_timeout; 54 }; 55 56 #define tipc_sk(sk) ((struct tipc_sock *)(sk)) 57 #define tipc_sk_port(sk) (tipc_sk(sk)->p) 58 59 #define tipc_rx_ready(sock) (!skb_queue_empty(&sock->sk->sk_receive_queue) || \ 60 (sock->state == SS_DISCONNECTING)) 61 62 static int backlog_rcv(struct sock *sk, struct sk_buff *skb); 63 static u32 dispatch(struct tipc_port *tport, struct sk_buff *buf); 64 static void wakeupdispatch(struct tipc_port *tport); 65 66 static const struct proto_ops packet_ops; 67 static const struct proto_ops stream_ops; 68 static const struct proto_ops msg_ops; 69 70 static struct proto tipc_proto; 71 72 static int sockets_enabled; 73 74 static atomic_t tipc_queue_size = ATOMIC_INIT(0); 75 76 /* 77 * Revised TIPC socket locking policy: 78 * 79 * Most socket operations take the standard socket lock when they start 80 * and hold it until they finish (or until they need to sleep). Acquiring 81 * this lock grants the owner exclusive access to the fields of the socket 82 * data structures, with the exception of the backlog queue. A few socket 83 * operations can be done without taking the socket lock because they only 84 * read socket information that never changes during the life of the socket. 85 * 86 * Socket operations may acquire the lock for the associated TIPC port if they 87 * need to perform an operation on the port. If any routine needs to acquire 88 * both the socket lock and the port lock it must take the socket lock first 89 * to avoid the risk of deadlock. 90 * 91 * The dispatcher handling incoming messages cannot grab the socket lock in 92 * the standard fashion, since invoked it runs at the BH level and cannot block. 93 * Instead, it checks to see if the socket lock is currently owned by someone, 94 * and either handles the message itself or adds it to the socket's backlog 95 * queue; in the latter case the queued message is processed once the process 96 * owning the socket lock releases it. 97 * 98 * NOTE: Releasing the socket lock while an operation is sleeping overcomes 99 * the problem of a blocked socket operation preventing any other operations 100 * from occurring. However, applications must be careful if they have 101 * multiple threads trying to send (or receive) on the same socket, as these 102 * operations might interfere with each other. For example, doing a connect 103 * and a receive at the same time might allow the receive to consume the 104 * ACK message meant for the connect. While additional work could be done 105 * to try and overcome this, it doesn't seem to be worthwhile at the present. 106 * 107 * NOTE: Releasing the socket lock while an operation is sleeping also ensures 108 * that another operation that must be performed in a non-blocking manner is 109 * not delayed for very long because the lock has already been taken. 110 * 111 * NOTE: This code assumes that certain fields of a port/socket pair are 112 * constant over its lifetime; such fields can be examined without taking 113 * the socket lock and/or port lock, and do not need to be re-read even 114 * after resuming processing after waiting. These fields include: 115 * - socket type 116 * - pointer to socket sk structure (aka tipc_sock structure) 117 * - pointer to port structure 118 * - port reference 119 */ 120 121 /** 122 * advance_rx_queue - discard first buffer in socket receive queue 123 * 124 * Caller must hold socket lock 125 */ 126 static void advance_rx_queue(struct sock *sk) 127 { 128 kfree_skb(__skb_dequeue(&sk->sk_receive_queue)); 129 atomic_dec(&tipc_queue_size); 130 } 131 132 /** 133 * discard_rx_queue - discard all buffers in socket receive queue 134 * 135 * Caller must hold socket lock 136 */ 137 static void discard_rx_queue(struct sock *sk) 138 { 139 struct sk_buff *buf; 140 141 while ((buf = __skb_dequeue(&sk->sk_receive_queue))) { 142 atomic_dec(&tipc_queue_size); 143 kfree_skb(buf); 144 } 145 } 146 147 /** 148 * reject_rx_queue - reject all buffers in socket receive queue 149 * 150 * Caller must hold socket lock 151 */ 152 static void reject_rx_queue(struct sock *sk) 153 { 154 struct sk_buff *buf; 155 156 while ((buf = __skb_dequeue(&sk->sk_receive_queue))) { 157 tipc_reject_msg(buf, TIPC_ERR_NO_PORT); 158 atomic_dec(&tipc_queue_size); 159 } 160 } 161 162 /** 163 * tipc_create - create a TIPC socket 164 * @net: network namespace (must be default network) 165 * @sock: pre-allocated socket structure 166 * @protocol: protocol indicator (must be 0) 167 * @kern: caused by kernel or by userspace? 168 * 169 * This routine creates additional data structures used by the TIPC socket, 170 * initializes them, and links them together. 171 * 172 * Returns 0 on success, errno otherwise 173 */ 174 static int tipc_create(struct net *net, struct socket *sock, int protocol, 175 int kern) 176 { 177 const struct proto_ops *ops; 178 socket_state state; 179 struct sock *sk; 180 struct tipc_port *tp_ptr; 181 182 /* Validate arguments */ 183 if (unlikely(protocol != 0)) 184 return -EPROTONOSUPPORT; 185 186 switch (sock->type) { 187 case SOCK_STREAM: 188 ops = &stream_ops; 189 state = SS_UNCONNECTED; 190 break; 191 case SOCK_SEQPACKET: 192 ops = &packet_ops; 193 state = SS_UNCONNECTED; 194 break; 195 case SOCK_DGRAM: 196 case SOCK_RDM: 197 ops = &msg_ops; 198 state = SS_READY; 199 break; 200 default: 201 return -EPROTOTYPE; 202 } 203 204 /* Allocate socket's protocol area */ 205 sk = sk_alloc(net, AF_TIPC, GFP_KERNEL, &tipc_proto); 206 if (sk == NULL) 207 return -ENOMEM; 208 209 /* Allocate TIPC port for socket to use */ 210 tp_ptr = tipc_createport_raw(sk, &dispatch, &wakeupdispatch, 211 TIPC_LOW_IMPORTANCE); 212 if (unlikely(!tp_ptr)) { 213 sk_free(sk); 214 return -ENOMEM; 215 } 216 217 /* Finish initializing socket data structures */ 218 sock->ops = ops; 219 sock->state = state; 220 221 sock_init_data(sock, sk); 222 sk->sk_backlog_rcv = backlog_rcv; 223 tipc_sk(sk)->p = tp_ptr; 224 tipc_sk(sk)->conn_timeout = CONN_TIMEOUT_DEFAULT; 225 226 spin_unlock_bh(tp_ptr->lock); 227 228 if (sock->state == SS_READY) { 229 tipc_set_portunreturnable(tp_ptr->ref, 1); 230 if (sock->type == SOCK_DGRAM) 231 tipc_set_portunreliable(tp_ptr->ref, 1); 232 } 233 234 return 0; 235 } 236 237 /** 238 * release - destroy a TIPC socket 239 * @sock: socket to destroy 240 * 241 * This routine cleans up any messages that are still queued on the socket. 242 * For DGRAM and RDM socket types, all queued messages are rejected. 243 * For SEQPACKET and STREAM socket types, the first message is rejected 244 * and any others are discarded. (If the first message on a STREAM socket 245 * is partially-read, it is discarded and the next one is rejected instead.) 246 * 247 * NOTE: Rejected messages are not necessarily returned to the sender! They 248 * are returned or discarded according to the "destination droppable" setting 249 * specified for the message by the sender. 250 * 251 * Returns 0 on success, errno otherwise 252 */ 253 static int release(struct socket *sock) 254 { 255 struct sock *sk = sock->sk; 256 struct tipc_port *tport; 257 struct sk_buff *buf; 258 int res; 259 260 /* 261 * Exit if socket isn't fully initialized (occurs when a failed accept() 262 * releases a pre-allocated child socket that was never used) 263 */ 264 if (sk == NULL) 265 return 0; 266 267 tport = tipc_sk_port(sk); 268 lock_sock(sk); 269 270 /* 271 * Reject all unreceived messages, except on an active connection 272 * (which disconnects locally & sends a 'FIN+' to peer) 273 */ 274 while (sock->state != SS_DISCONNECTING) { 275 buf = __skb_dequeue(&sk->sk_receive_queue); 276 if (buf == NULL) 277 break; 278 atomic_dec(&tipc_queue_size); 279 if (TIPC_SKB_CB(buf)->handle != 0) 280 kfree_skb(buf); 281 else { 282 if ((sock->state == SS_CONNECTING) || 283 (sock->state == SS_CONNECTED)) { 284 sock->state = SS_DISCONNECTING; 285 tipc_disconnect(tport->ref); 286 } 287 tipc_reject_msg(buf, TIPC_ERR_NO_PORT); 288 } 289 } 290 291 /* 292 * Delete TIPC port; this ensures no more messages are queued 293 * (also disconnects an active connection & sends a 'FIN-' to peer) 294 */ 295 res = tipc_deleteport(tport->ref); 296 297 /* Discard any remaining (connection-based) messages in receive queue */ 298 discard_rx_queue(sk); 299 300 /* Reject any messages that accumulated in backlog queue */ 301 sock->state = SS_DISCONNECTING; 302 release_sock(sk); 303 304 sock_put(sk); 305 sock->sk = NULL; 306 307 return res; 308 } 309 310 /** 311 * bind - associate or disassocate TIPC name(s) with a socket 312 * @sock: socket structure 313 * @uaddr: socket address describing name(s) and desired operation 314 * @uaddr_len: size of socket address data structure 315 * 316 * Name and name sequence binding is indicated using a positive scope value; 317 * a negative scope value unbinds the specified name. Specifying no name 318 * (i.e. a socket address length of 0) unbinds all names from the socket. 319 * 320 * Returns 0 on success, errno otherwise 321 * 322 * NOTE: This routine doesn't need to take the socket lock since it doesn't 323 * access any non-constant socket information. 324 */ 325 static int bind(struct socket *sock, struct sockaddr *uaddr, int uaddr_len) 326 { 327 struct sockaddr_tipc *addr = (struct sockaddr_tipc *)uaddr; 328 u32 portref = tipc_sk_port(sock->sk)->ref; 329 330 if (unlikely(!uaddr_len)) 331 return tipc_withdraw(portref, 0, NULL); 332 333 if (uaddr_len < sizeof(struct sockaddr_tipc)) 334 return -EINVAL; 335 if (addr->family != AF_TIPC) 336 return -EAFNOSUPPORT; 337 338 if (addr->addrtype == TIPC_ADDR_NAME) 339 addr->addr.nameseq.upper = addr->addr.nameseq.lower; 340 else if (addr->addrtype != TIPC_ADDR_NAMESEQ) 341 return -EAFNOSUPPORT; 342 343 if (addr->addr.nameseq.type < TIPC_RESERVED_TYPES) 344 return -EACCES; 345 346 return (addr->scope > 0) ? 347 tipc_publish(portref, addr->scope, &addr->addr.nameseq) : 348 tipc_withdraw(portref, -addr->scope, &addr->addr.nameseq); 349 } 350 351 /** 352 * get_name - get port ID of socket or peer socket 353 * @sock: socket structure 354 * @uaddr: area for returned socket address 355 * @uaddr_len: area for returned length of socket address 356 * @peer: 0 = own ID, 1 = current peer ID, 2 = current/former peer ID 357 * 358 * Returns 0 on success, errno otherwise 359 * 360 * NOTE: This routine doesn't need to take the socket lock since it only 361 * accesses socket information that is unchanging (or which changes in 362 * a completely predictable manner). 363 */ 364 static int get_name(struct socket *sock, struct sockaddr *uaddr, 365 int *uaddr_len, int peer) 366 { 367 struct sockaddr_tipc *addr = (struct sockaddr_tipc *)uaddr; 368 struct tipc_sock *tsock = tipc_sk(sock->sk); 369 370 memset(addr, 0, sizeof(*addr)); 371 if (peer) { 372 if ((sock->state != SS_CONNECTED) && 373 ((peer != 2) || (sock->state != SS_DISCONNECTING))) 374 return -ENOTCONN; 375 addr->addr.id.ref = tsock->peer_name.ref; 376 addr->addr.id.node = tsock->peer_name.node; 377 } else { 378 addr->addr.id.ref = tsock->p->ref; 379 addr->addr.id.node = tipc_own_addr; 380 } 381 382 *uaddr_len = sizeof(*addr); 383 addr->addrtype = TIPC_ADDR_ID; 384 addr->family = AF_TIPC; 385 addr->scope = 0; 386 addr->addr.name.domain = 0; 387 388 return 0; 389 } 390 391 /** 392 * poll - read and possibly block on pollmask 393 * @file: file structure associated with the socket 394 * @sock: socket for which to calculate the poll bits 395 * @wait: ??? 396 * 397 * Returns pollmask value 398 * 399 * COMMENTARY: 400 * It appears that the usual socket locking mechanisms are not useful here 401 * since the pollmask info is potentially out-of-date the moment this routine 402 * exits. TCP and other protocols seem to rely on higher level poll routines 403 * to handle any preventable race conditions, so TIPC will do the same ... 404 * 405 * TIPC sets the returned events as follows: 406 * 407 * socket state flags set 408 * ------------ --------- 409 * unconnected no read flags 410 * no write flags 411 * 412 * connecting POLLIN/POLLRDNORM if ACK/NACK in rx queue 413 * no write flags 414 * 415 * connected POLLIN/POLLRDNORM if data in rx queue 416 * POLLOUT if port is not congested 417 * 418 * disconnecting POLLIN/POLLRDNORM/POLLHUP 419 * no write flags 420 * 421 * listening POLLIN if SYN in rx queue 422 * no write flags 423 * 424 * ready POLLIN/POLLRDNORM if data in rx queue 425 * [connectionless] POLLOUT (since port cannot be congested) 426 * 427 * IMPORTANT: The fact that a read or write operation is indicated does NOT 428 * imply that the operation will succeed, merely that it should be performed 429 * and will not block. 430 */ 431 static unsigned int poll(struct file *file, struct socket *sock, 432 poll_table *wait) 433 { 434 struct sock *sk = sock->sk; 435 u32 mask = 0; 436 437 poll_wait(file, sk_sleep(sk), wait); 438 439 switch ((int)sock->state) { 440 case SS_READY: 441 case SS_CONNECTED: 442 if (!tipc_sk_port(sk)->congested) 443 mask |= POLLOUT; 444 /* fall thru' */ 445 case SS_CONNECTING: 446 case SS_LISTENING: 447 if (!skb_queue_empty(&sk->sk_receive_queue)) 448 mask |= (POLLIN | POLLRDNORM); 449 break; 450 case SS_DISCONNECTING: 451 mask = (POLLIN | POLLRDNORM | POLLHUP); 452 break; 453 } 454 455 return mask; 456 } 457 458 /** 459 * dest_name_check - verify user is permitted to send to specified port name 460 * @dest: destination address 461 * @m: descriptor for message to be sent 462 * 463 * Prevents restricted configuration commands from being issued by 464 * unauthorized users. 465 * 466 * Returns 0 if permission is granted, otherwise errno 467 */ 468 static int dest_name_check(struct sockaddr_tipc *dest, struct msghdr *m) 469 { 470 struct tipc_cfg_msg_hdr hdr; 471 472 if (likely(dest->addr.name.name.type >= TIPC_RESERVED_TYPES)) 473 return 0; 474 if (likely(dest->addr.name.name.type == TIPC_TOP_SRV)) 475 return 0; 476 if (likely(dest->addr.name.name.type != TIPC_CFG_SRV)) 477 return -EACCES; 478 479 if (!m->msg_iovlen || (m->msg_iov[0].iov_len < sizeof(hdr))) 480 return -EMSGSIZE; 481 if (copy_from_user(&hdr, m->msg_iov[0].iov_base, sizeof(hdr))) 482 return -EFAULT; 483 if ((ntohs(hdr.tcm_type) & 0xC000) && (!capable(CAP_NET_ADMIN))) 484 return -EACCES; 485 486 return 0; 487 } 488 489 /** 490 * send_msg - send message in connectionless manner 491 * @iocb: if NULL, indicates that socket lock is already held 492 * @sock: socket structure 493 * @m: message to send 494 * @total_len: length of message 495 * 496 * Message must have an destination specified explicitly. 497 * Used for SOCK_RDM and SOCK_DGRAM messages, 498 * and for 'SYN' messages on SOCK_SEQPACKET and SOCK_STREAM connections. 499 * (Note: 'SYN+' is prohibited on SOCK_STREAM.) 500 * 501 * Returns the number of bytes sent on success, or errno otherwise 502 */ 503 static int send_msg(struct kiocb *iocb, struct socket *sock, 504 struct msghdr *m, size_t total_len) 505 { 506 struct sock *sk = sock->sk; 507 struct tipc_port *tport = tipc_sk_port(sk); 508 struct sockaddr_tipc *dest = (struct sockaddr_tipc *)m->msg_name; 509 int needs_conn; 510 long timeout_val; 511 int res = -EINVAL; 512 513 if (unlikely(!dest)) 514 return -EDESTADDRREQ; 515 if (unlikely((m->msg_namelen < sizeof(*dest)) || 516 (dest->family != AF_TIPC))) 517 return -EINVAL; 518 if ((total_len > TIPC_MAX_USER_MSG_SIZE) || 519 (m->msg_iovlen > (unsigned int)INT_MAX)) 520 return -EMSGSIZE; 521 522 if (iocb) 523 lock_sock(sk); 524 525 needs_conn = (sock->state != SS_READY); 526 if (unlikely(needs_conn)) { 527 if (sock->state == SS_LISTENING) { 528 res = -EPIPE; 529 goto exit; 530 } 531 if (sock->state != SS_UNCONNECTED) { 532 res = -EISCONN; 533 goto exit; 534 } 535 if ((tport->published) || 536 ((sock->type == SOCK_STREAM) && (total_len != 0))) { 537 res = -EOPNOTSUPP; 538 goto exit; 539 } 540 if (dest->addrtype == TIPC_ADDR_NAME) { 541 tport->conn_type = dest->addr.name.name.type; 542 tport->conn_instance = dest->addr.name.name.instance; 543 } 544 545 /* Abort any pending connection attempts (very unlikely) */ 546 reject_rx_queue(sk); 547 } 548 549 timeout_val = sock_sndtimeo(sk, m->msg_flags & MSG_DONTWAIT); 550 551 do { 552 if (dest->addrtype == TIPC_ADDR_NAME) { 553 res = dest_name_check(dest, m); 554 if (res) 555 break; 556 res = tipc_send2name(tport->ref, 557 &dest->addr.name.name, 558 dest->addr.name.domain, 559 m->msg_iovlen, 560 m->msg_iov, 561 total_len); 562 } else if (dest->addrtype == TIPC_ADDR_ID) { 563 res = tipc_send2port(tport->ref, 564 &dest->addr.id, 565 m->msg_iovlen, 566 m->msg_iov, 567 total_len); 568 } else if (dest->addrtype == TIPC_ADDR_MCAST) { 569 if (needs_conn) { 570 res = -EOPNOTSUPP; 571 break; 572 } 573 res = dest_name_check(dest, m); 574 if (res) 575 break; 576 res = tipc_multicast(tport->ref, 577 &dest->addr.nameseq, 578 m->msg_iovlen, 579 m->msg_iov, 580 total_len); 581 } 582 if (likely(res != -ELINKCONG)) { 583 if (needs_conn && (res >= 0)) 584 sock->state = SS_CONNECTING; 585 break; 586 } 587 if (timeout_val <= 0L) { 588 res = timeout_val ? timeout_val : -EWOULDBLOCK; 589 break; 590 } 591 release_sock(sk); 592 timeout_val = wait_event_interruptible_timeout(*sk_sleep(sk), 593 !tport->congested, timeout_val); 594 lock_sock(sk); 595 } while (1); 596 597 exit: 598 if (iocb) 599 release_sock(sk); 600 return res; 601 } 602 603 /** 604 * send_packet - send a connection-oriented message 605 * @iocb: if NULL, indicates that socket lock is already held 606 * @sock: socket structure 607 * @m: message to send 608 * @total_len: length of message 609 * 610 * Used for SOCK_SEQPACKET messages and SOCK_STREAM data. 611 * 612 * Returns the number of bytes sent on success, or errno otherwise 613 */ 614 static int send_packet(struct kiocb *iocb, struct socket *sock, 615 struct msghdr *m, size_t total_len) 616 { 617 struct sock *sk = sock->sk; 618 struct tipc_port *tport = tipc_sk_port(sk); 619 struct sockaddr_tipc *dest = (struct sockaddr_tipc *)m->msg_name; 620 long timeout_val; 621 int res; 622 623 /* Handle implied connection establishment */ 624 if (unlikely(dest)) 625 return send_msg(iocb, sock, m, total_len); 626 627 if ((total_len > TIPC_MAX_USER_MSG_SIZE) || 628 (m->msg_iovlen > (unsigned int)INT_MAX)) 629 return -EMSGSIZE; 630 631 if (iocb) 632 lock_sock(sk); 633 634 timeout_val = sock_sndtimeo(sk, m->msg_flags & MSG_DONTWAIT); 635 636 do { 637 if (unlikely(sock->state != SS_CONNECTED)) { 638 if (sock->state == SS_DISCONNECTING) 639 res = -EPIPE; 640 else 641 res = -ENOTCONN; 642 break; 643 } 644 645 res = tipc_send(tport->ref, m->msg_iovlen, m->msg_iov, 646 total_len); 647 if (likely(res != -ELINKCONG)) 648 break; 649 if (timeout_val <= 0L) { 650 res = timeout_val ? timeout_val : -EWOULDBLOCK; 651 break; 652 } 653 release_sock(sk); 654 timeout_val = wait_event_interruptible_timeout(*sk_sleep(sk), 655 (!tport->congested || !tport->connected), timeout_val); 656 lock_sock(sk); 657 } while (1); 658 659 if (iocb) 660 release_sock(sk); 661 return res; 662 } 663 664 /** 665 * send_stream - send stream-oriented data 666 * @iocb: (unused) 667 * @sock: socket structure 668 * @m: data to send 669 * @total_len: total length of data to be sent 670 * 671 * Used for SOCK_STREAM data. 672 * 673 * Returns the number of bytes sent on success (or partial success), 674 * or errno if no data sent 675 */ 676 static int send_stream(struct kiocb *iocb, struct socket *sock, 677 struct msghdr *m, size_t total_len) 678 { 679 struct sock *sk = sock->sk; 680 struct tipc_port *tport = tipc_sk_port(sk); 681 struct msghdr my_msg; 682 struct iovec my_iov; 683 struct iovec *curr_iov; 684 int curr_iovlen; 685 char __user *curr_start; 686 u32 hdr_size; 687 int curr_left; 688 int bytes_to_send; 689 int bytes_sent; 690 int res; 691 692 lock_sock(sk); 693 694 /* Handle special cases where there is no connection */ 695 if (unlikely(sock->state != SS_CONNECTED)) { 696 if (sock->state == SS_UNCONNECTED) { 697 res = send_packet(NULL, sock, m, total_len); 698 goto exit; 699 } else if (sock->state == SS_DISCONNECTING) { 700 res = -EPIPE; 701 goto exit; 702 } else { 703 res = -ENOTCONN; 704 goto exit; 705 } 706 } 707 708 if (unlikely(m->msg_name)) { 709 res = -EISCONN; 710 goto exit; 711 } 712 713 if ((total_len > (unsigned int)INT_MAX) || 714 (m->msg_iovlen > (unsigned int)INT_MAX)) { 715 res = -EMSGSIZE; 716 goto exit; 717 } 718 719 /* 720 * Send each iovec entry using one or more messages 721 * 722 * Note: This algorithm is good for the most likely case 723 * (i.e. one large iovec entry), but could be improved to pass sets 724 * of small iovec entries into send_packet(). 725 */ 726 curr_iov = m->msg_iov; 727 curr_iovlen = m->msg_iovlen; 728 my_msg.msg_iov = &my_iov; 729 my_msg.msg_iovlen = 1; 730 my_msg.msg_flags = m->msg_flags; 731 my_msg.msg_name = NULL; 732 bytes_sent = 0; 733 734 hdr_size = msg_hdr_sz(&tport->phdr); 735 736 while (curr_iovlen--) { 737 curr_start = curr_iov->iov_base; 738 curr_left = curr_iov->iov_len; 739 740 while (curr_left) { 741 bytes_to_send = tport->max_pkt - hdr_size; 742 if (bytes_to_send > TIPC_MAX_USER_MSG_SIZE) 743 bytes_to_send = TIPC_MAX_USER_MSG_SIZE; 744 if (curr_left < bytes_to_send) 745 bytes_to_send = curr_left; 746 my_iov.iov_base = curr_start; 747 my_iov.iov_len = bytes_to_send; 748 res = send_packet(NULL, sock, &my_msg, bytes_to_send); 749 if (res < 0) { 750 if (bytes_sent) 751 res = bytes_sent; 752 goto exit; 753 } 754 curr_left -= bytes_to_send; 755 curr_start += bytes_to_send; 756 bytes_sent += bytes_to_send; 757 } 758 759 curr_iov++; 760 } 761 res = bytes_sent; 762 exit: 763 release_sock(sk); 764 return res; 765 } 766 767 /** 768 * auto_connect - complete connection setup to a remote port 769 * @sock: socket structure 770 * @msg: peer's response message 771 * 772 * Returns 0 on success, errno otherwise 773 */ 774 static int auto_connect(struct socket *sock, struct tipc_msg *msg) 775 { 776 struct tipc_sock *tsock = tipc_sk(sock->sk); 777 778 if (msg_errcode(msg)) { 779 sock->state = SS_DISCONNECTING; 780 return -ECONNREFUSED; 781 } 782 783 tsock->peer_name.ref = msg_origport(msg); 784 tsock->peer_name.node = msg_orignode(msg); 785 tipc_connect2port(tsock->p->ref, &tsock->peer_name); 786 tipc_set_portimportance(tsock->p->ref, msg_importance(msg)); 787 sock->state = SS_CONNECTED; 788 return 0; 789 } 790 791 /** 792 * set_orig_addr - capture sender's address for received message 793 * @m: descriptor for message info 794 * @msg: received message header 795 * 796 * Note: Address is not captured if not requested by receiver. 797 */ 798 static void set_orig_addr(struct msghdr *m, struct tipc_msg *msg) 799 { 800 struct sockaddr_tipc *addr = (struct sockaddr_tipc *)m->msg_name; 801 802 if (addr) { 803 addr->family = AF_TIPC; 804 addr->addrtype = TIPC_ADDR_ID; 805 addr->addr.id.ref = msg_origport(msg); 806 addr->addr.id.node = msg_orignode(msg); 807 addr->addr.name.domain = 0; /* could leave uninitialized */ 808 addr->scope = 0; /* could leave uninitialized */ 809 m->msg_namelen = sizeof(struct sockaddr_tipc); 810 } 811 } 812 813 /** 814 * anc_data_recv - optionally capture ancillary data for received message 815 * @m: descriptor for message info 816 * @msg: received message header 817 * @tport: TIPC port associated with message 818 * 819 * Note: Ancillary data is not captured if not requested by receiver. 820 * 821 * Returns 0 if successful, otherwise errno 822 */ 823 static int anc_data_recv(struct msghdr *m, struct tipc_msg *msg, 824 struct tipc_port *tport) 825 { 826 u32 anc_data[3]; 827 u32 err; 828 u32 dest_type; 829 int has_name; 830 int res; 831 832 if (likely(m->msg_controllen == 0)) 833 return 0; 834 835 /* Optionally capture errored message object(s) */ 836 err = msg ? msg_errcode(msg) : 0; 837 if (unlikely(err)) { 838 anc_data[0] = err; 839 anc_data[1] = msg_data_sz(msg); 840 res = put_cmsg(m, SOL_TIPC, TIPC_ERRINFO, 8, anc_data); 841 if (res) 842 return res; 843 if (anc_data[1]) { 844 res = put_cmsg(m, SOL_TIPC, TIPC_RETDATA, anc_data[1], 845 msg_data(msg)); 846 if (res) 847 return res; 848 } 849 } 850 851 /* Optionally capture message destination object */ 852 dest_type = msg ? msg_type(msg) : TIPC_DIRECT_MSG; 853 switch (dest_type) { 854 case TIPC_NAMED_MSG: 855 has_name = 1; 856 anc_data[0] = msg_nametype(msg); 857 anc_data[1] = msg_namelower(msg); 858 anc_data[2] = msg_namelower(msg); 859 break; 860 case TIPC_MCAST_MSG: 861 has_name = 1; 862 anc_data[0] = msg_nametype(msg); 863 anc_data[1] = msg_namelower(msg); 864 anc_data[2] = msg_nameupper(msg); 865 break; 866 case TIPC_CONN_MSG: 867 has_name = (tport->conn_type != 0); 868 anc_data[0] = tport->conn_type; 869 anc_data[1] = tport->conn_instance; 870 anc_data[2] = tport->conn_instance; 871 break; 872 default: 873 has_name = 0; 874 } 875 if (has_name) { 876 res = put_cmsg(m, SOL_TIPC, TIPC_DESTNAME, 12, anc_data); 877 if (res) 878 return res; 879 } 880 881 return 0; 882 } 883 884 /** 885 * recv_msg - receive packet-oriented message 886 * @iocb: (unused) 887 * @m: descriptor for message info 888 * @buf_len: total size of user buffer area 889 * @flags: receive flags 890 * 891 * Used for SOCK_DGRAM, SOCK_RDM, and SOCK_SEQPACKET messages. 892 * If the complete message doesn't fit in user area, truncate it. 893 * 894 * Returns size of returned message data, errno otherwise 895 */ 896 static int recv_msg(struct kiocb *iocb, struct socket *sock, 897 struct msghdr *m, size_t buf_len, int flags) 898 { 899 struct sock *sk = sock->sk; 900 struct tipc_port *tport = tipc_sk_port(sk); 901 struct sk_buff *buf; 902 struct tipc_msg *msg; 903 long timeout; 904 unsigned int sz; 905 u32 err; 906 int res; 907 908 /* Catch invalid receive requests */ 909 if (unlikely(!buf_len)) 910 return -EINVAL; 911 912 lock_sock(sk); 913 914 if (unlikely(sock->state == SS_UNCONNECTED)) { 915 res = -ENOTCONN; 916 goto exit; 917 } 918 919 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 920 restart: 921 922 /* Look for a message in receive queue; wait if necessary */ 923 while (skb_queue_empty(&sk->sk_receive_queue)) { 924 if (sock->state == SS_DISCONNECTING) { 925 res = -ENOTCONN; 926 goto exit; 927 } 928 if (timeout <= 0L) { 929 res = timeout ? timeout : -EWOULDBLOCK; 930 goto exit; 931 } 932 release_sock(sk); 933 timeout = wait_event_interruptible_timeout(*sk_sleep(sk), 934 tipc_rx_ready(sock), 935 timeout); 936 lock_sock(sk); 937 } 938 939 /* Look at first message in receive queue */ 940 buf = skb_peek(&sk->sk_receive_queue); 941 msg = buf_msg(buf); 942 sz = msg_data_sz(msg); 943 err = msg_errcode(msg); 944 945 /* Complete connection setup for an implied connect */ 946 if (unlikely(sock->state == SS_CONNECTING)) { 947 res = auto_connect(sock, msg); 948 if (res) 949 goto exit; 950 } 951 952 /* Discard an empty non-errored message & try again */ 953 if ((!sz) && (!err)) { 954 advance_rx_queue(sk); 955 goto restart; 956 } 957 958 /* Capture sender's address (optional) */ 959 set_orig_addr(m, msg); 960 961 /* Capture ancillary data (optional) */ 962 res = anc_data_recv(m, msg, tport); 963 if (res) 964 goto exit; 965 966 /* Capture message data (if valid) & compute return value (always) */ 967 if (!err) { 968 if (unlikely(buf_len < sz)) { 969 sz = buf_len; 970 m->msg_flags |= MSG_TRUNC; 971 } 972 res = skb_copy_datagram_iovec(buf, msg_hdr_sz(msg), 973 m->msg_iov, sz); 974 if (res) 975 goto exit; 976 res = sz; 977 } else { 978 if ((sock->state == SS_READY) || 979 ((err == TIPC_CONN_SHUTDOWN) || m->msg_control)) 980 res = 0; 981 else 982 res = -ECONNRESET; 983 } 984 985 /* Consume received message (optional) */ 986 if (likely(!(flags & MSG_PEEK))) { 987 if ((sock->state != SS_READY) && 988 (++tport->conn_unacked >= TIPC_FLOW_CONTROL_WIN)) 989 tipc_acknowledge(tport->ref, tport->conn_unacked); 990 advance_rx_queue(sk); 991 } 992 exit: 993 release_sock(sk); 994 return res; 995 } 996 997 /** 998 * recv_stream - receive stream-oriented data 999 * @iocb: (unused) 1000 * @m: descriptor for message info 1001 * @buf_len: total size of user buffer area 1002 * @flags: receive flags 1003 * 1004 * Used for SOCK_STREAM messages only. If not enough data is available 1005 * will optionally wait for more; never truncates data. 1006 * 1007 * Returns size of returned message data, errno otherwise 1008 */ 1009 static int recv_stream(struct kiocb *iocb, struct socket *sock, 1010 struct msghdr *m, size_t buf_len, int flags) 1011 { 1012 struct sock *sk = sock->sk; 1013 struct tipc_port *tport = tipc_sk_port(sk); 1014 struct sk_buff *buf; 1015 struct tipc_msg *msg; 1016 long timeout; 1017 unsigned int sz; 1018 int sz_to_copy, target, needed; 1019 int sz_copied = 0; 1020 u32 err; 1021 int res = 0; 1022 1023 /* Catch invalid receive attempts */ 1024 if (unlikely(!buf_len)) 1025 return -EINVAL; 1026 1027 lock_sock(sk); 1028 1029 if (unlikely((sock->state == SS_UNCONNECTED) || 1030 (sock->state == SS_CONNECTING))) { 1031 res = -ENOTCONN; 1032 goto exit; 1033 } 1034 1035 target = sock_rcvlowat(sk, flags & MSG_WAITALL, buf_len); 1036 timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1037 1038 restart: 1039 /* Look for a message in receive queue; wait if necessary */ 1040 while (skb_queue_empty(&sk->sk_receive_queue)) { 1041 if (sock->state == SS_DISCONNECTING) { 1042 res = -ENOTCONN; 1043 goto exit; 1044 } 1045 if (timeout <= 0L) { 1046 res = timeout ? timeout : -EWOULDBLOCK; 1047 goto exit; 1048 } 1049 release_sock(sk); 1050 timeout = wait_event_interruptible_timeout(*sk_sleep(sk), 1051 tipc_rx_ready(sock), 1052 timeout); 1053 lock_sock(sk); 1054 } 1055 1056 /* Look at first message in receive queue */ 1057 buf = skb_peek(&sk->sk_receive_queue); 1058 msg = buf_msg(buf); 1059 sz = msg_data_sz(msg); 1060 err = msg_errcode(msg); 1061 1062 /* Discard an empty non-errored message & try again */ 1063 if ((!sz) && (!err)) { 1064 advance_rx_queue(sk); 1065 goto restart; 1066 } 1067 1068 /* Optionally capture sender's address & ancillary data of first msg */ 1069 if (sz_copied == 0) { 1070 set_orig_addr(m, msg); 1071 res = anc_data_recv(m, msg, tport); 1072 if (res) 1073 goto exit; 1074 } 1075 1076 /* Capture message data (if valid) & compute return value (always) */ 1077 if (!err) { 1078 u32 offset = (u32)(unsigned long)(TIPC_SKB_CB(buf)->handle); 1079 1080 sz -= offset; 1081 needed = (buf_len - sz_copied); 1082 sz_to_copy = (sz <= needed) ? sz : needed; 1083 1084 res = skb_copy_datagram_iovec(buf, msg_hdr_sz(msg) + offset, 1085 m->msg_iov, sz_to_copy); 1086 if (res) 1087 goto exit; 1088 1089 sz_copied += sz_to_copy; 1090 1091 if (sz_to_copy < sz) { 1092 if (!(flags & MSG_PEEK)) 1093 TIPC_SKB_CB(buf)->handle = 1094 (void *)(unsigned long)(offset + sz_to_copy); 1095 goto exit; 1096 } 1097 } else { 1098 if (sz_copied != 0) 1099 goto exit; /* can't add error msg to valid data */ 1100 1101 if ((err == TIPC_CONN_SHUTDOWN) || m->msg_control) 1102 res = 0; 1103 else 1104 res = -ECONNRESET; 1105 } 1106 1107 /* Consume received message (optional) */ 1108 if (likely(!(flags & MSG_PEEK))) { 1109 if (unlikely(++tport->conn_unacked >= TIPC_FLOW_CONTROL_WIN)) 1110 tipc_acknowledge(tport->ref, tport->conn_unacked); 1111 advance_rx_queue(sk); 1112 } 1113 1114 /* Loop around if more data is required */ 1115 if ((sz_copied < buf_len) && /* didn't get all requested data */ 1116 (!skb_queue_empty(&sk->sk_receive_queue) || 1117 (sz_copied < target)) && /* and more is ready or required */ 1118 (!(flags & MSG_PEEK)) && /* and aren't just peeking at data */ 1119 (!err)) /* and haven't reached a FIN */ 1120 goto restart; 1121 1122 exit: 1123 release_sock(sk); 1124 return sz_copied ? sz_copied : res; 1125 } 1126 1127 /** 1128 * rx_queue_full - determine if receive queue can accept another message 1129 * @msg: message to be added to queue 1130 * @queue_size: current size of queue 1131 * @base: nominal maximum size of queue 1132 * 1133 * Returns 1 if queue is unable to accept message, 0 otherwise 1134 */ 1135 static int rx_queue_full(struct tipc_msg *msg, u32 queue_size, u32 base) 1136 { 1137 u32 threshold; 1138 u32 imp = msg_importance(msg); 1139 1140 if (imp == TIPC_LOW_IMPORTANCE) 1141 threshold = base; 1142 else if (imp == TIPC_MEDIUM_IMPORTANCE) 1143 threshold = base * 2; 1144 else if (imp == TIPC_HIGH_IMPORTANCE) 1145 threshold = base * 100; 1146 else 1147 return 0; 1148 1149 if (msg_connected(msg)) 1150 threshold *= 4; 1151 1152 return queue_size >= threshold; 1153 } 1154 1155 /** 1156 * filter_rcv - validate incoming message 1157 * @sk: socket 1158 * @buf: message 1159 * 1160 * Enqueues message on receive queue if acceptable; optionally handles 1161 * disconnect indication for a connected socket. 1162 * 1163 * Called with socket lock already taken; port lock may also be taken. 1164 * 1165 * Returns TIPC error status code (TIPC_OK if message is not to be rejected) 1166 */ 1167 static u32 filter_rcv(struct sock *sk, struct sk_buff *buf) 1168 { 1169 struct socket *sock = sk->sk_socket; 1170 struct tipc_msg *msg = buf_msg(buf); 1171 u32 recv_q_len; 1172 1173 /* Reject message if it is wrong sort of message for socket */ 1174 if (msg_type(msg) > TIPC_DIRECT_MSG) 1175 return TIPC_ERR_NO_PORT; 1176 1177 if (sock->state == SS_READY) { 1178 if (msg_connected(msg)) 1179 return TIPC_ERR_NO_PORT; 1180 } else { 1181 if (msg_mcast(msg)) 1182 return TIPC_ERR_NO_PORT; 1183 if (sock->state == SS_CONNECTED) { 1184 if (!msg_connected(msg) || 1185 !tipc_port_peer_msg(tipc_sk_port(sk), msg)) 1186 return TIPC_ERR_NO_PORT; 1187 } else if (sock->state == SS_CONNECTING) { 1188 if (!msg_connected(msg) && (msg_errcode(msg) == 0)) 1189 return TIPC_ERR_NO_PORT; 1190 } else if (sock->state == SS_LISTENING) { 1191 if (msg_connected(msg) || msg_errcode(msg)) 1192 return TIPC_ERR_NO_PORT; 1193 } else if (sock->state == SS_DISCONNECTING) { 1194 return TIPC_ERR_NO_PORT; 1195 } else /* (sock->state == SS_UNCONNECTED) */ { 1196 if (msg_connected(msg) || msg_errcode(msg)) 1197 return TIPC_ERR_NO_PORT; 1198 } 1199 } 1200 1201 /* Reject message if there isn't room to queue it */ 1202 recv_q_len = (u32)atomic_read(&tipc_queue_size); 1203 if (unlikely(recv_q_len >= OVERLOAD_LIMIT_BASE)) { 1204 if (rx_queue_full(msg, recv_q_len, OVERLOAD_LIMIT_BASE)) 1205 return TIPC_ERR_OVERLOAD; 1206 } 1207 recv_q_len = skb_queue_len(&sk->sk_receive_queue); 1208 if (unlikely(recv_q_len >= (OVERLOAD_LIMIT_BASE / 2))) { 1209 if (rx_queue_full(msg, recv_q_len, OVERLOAD_LIMIT_BASE / 2)) 1210 return TIPC_ERR_OVERLOAD; 1211 } 1212 1213 /* Enqueue message (finally!) */ 1214 TIPC_SKB_CB(buf)->handle = 0; 1215 atomic_inc(&tipc_queue_size); 1216 __skb_queue_tail(&sk->sk_receive_queue, buf); 1217 1218 /* Initiate connection termination for an incoming 'FIN' */ 1219 if (unlikely(msg_errcode(msg) && (sock->state == SS_CONNECTED))) { 1220 sock->state = SS_DISCONNECTING; 1221 tipc_disconnect_port(tipc_sk_port(sk)); 1222 } 1223 1224 if (waitqueue_active(sk_sleep(sk))) 1225 wake_up_interruptible(sk_sleep(sk)); 1226 return TIPC_OK; 1227 } 1228 1229 /** 1230 * backlog_rcv - handle incoming message from backlog queue 1231 * @sk: socket 1232 * @buf: message 1233 * 1234 * Caller must hold socket lock, but not port lock. 1235 * 1236 * Returns 0 1237 */ 1238 static int backlog_rcv(struct sock *sk, struct sk_buff *buf) 1239 { 1240 u32 res; 1241 1242 res = filter_rcv(sk, buf); 1243 if (res) 1244 tipc_reject_msg(buf, res); 1245 return 0; 1246 } 1247 1248 /** 1249 * dispatch - handle incoming message 1250 * @tport: TIPC port that received message 1251 * @buf: message 1252 * 1253 * Called with port lock already taken. 1254 * 1255 * Returns TIPC error status code (TIPC_OK if message is not to be rejected) 1256 */ 1257 static u32 dispatch(struct tipc_port *tport, struct sk_buff *buf) 1258 { 1259 struct sock *sk = (struct sock *)tport->usr_handle; 1260 u32 res; 1261 1262 /* 1263 * Process message if socket is unlocked; otherwise add to backlog queue 1264 * 1265 * This code is based on sk_receive_skb(), but must be distinct from it 1266 * since a TIPC-specific filter/reject mechanism is utilized 1267 */ 1268 bh_lock_sock(sk); 1269 if (!sock_owned_by_user(sk)) { 1270 res = filter_rcv(sk, buf); 1271 } else { 1272 if (sk_add_backlog(sk, buf, sk->sk_rcvbuf)) 1273 res = TIPC_ERR_OVERLOAD; 1274 else 1275 res = TIPC_OK; 1276 } 1277 bh_unlock_sock(sk); 1278 1279 return res; 1280 } 1281 1282 /** 1283 * wakeupdispatch - wake up port after congestion 1284 * @tport: port to wakeup 1285 * 1286 * Called with port lock already taken. 1287 */ 1288 static void wakeupdispatch(struct tipc_port *tport) 1289 { 1290 struct sock *sk = (struct sock *)tport->usr_handle; 1291 1292 if (waitqueue_active(sk_sleep(sk))) 1293 wake_up_interruptible(sk_sleep(sk)); 1294 } 1295 1296 /** 1297 * connect - establish a connection to another TIPC port 1298 * @sock: socket structure 1299 * @dest: socket address for destination port 1300 * @destlen: size of socket address data structure 1301 * @flags: file-related flags associated with socket 1302 * 1303 * Returns 0 on success, errno otherwise 1304 */ 1305 static int connect(struct socket *sock, struct sockaddr *dest, int destlen, 1306 int flags) 1307 { 1308 struct sock *sk = sock->sk; 1309 struct sockaddr_tipc *dst = (struct sockaddr_tipc *)dest; 1310 struct msghdr m = {NULL,}; 1311 struct sk_buff *buf; 1312 struct tipc_msg *msg; 1313 unsigned int timeout; 1314 int res; 1315 1316 lock_sock(sk); 1317 1318 /* For now, TIPC does not allow use of connect() with DGRAM/RDM types */ 1319 if (sock->state == SS_READY) { 1320 res = -EOPNOTSUPP; 1321 goto exit; 1322 } 1323 1324 /* For now, TIPC does not support the non-blocking form of connect() */ 1325 if (flags & O_NONBLOCK) { 1326 res = -EOPNOTSUPP; 1327 goto exit; 1328 } 1329 1330 /* Issue Posix-compliant error code if socket is in the wrong state */ 1331 if (sock->state == SS_LISTENING) { 1332 res = -EOPNOTSUPP; 1333 goto exit; 1334 } 1335 if (sock->state == SS_CONNECTING) { 1336 res = -EALREADY; 1337 goto exit; 1338 } 1339 if (sock->state != SS_UNCONNECTED) { 1340 res = -EISCONN; 1341 goto exit; 1342 } 1343 1344 /* 1345 * Reject connection attempt using multicast address 1346 * 1347 * Note: send_msg() validates the rest of the address fields, 1348 * so there's no need to do it here 1349 */ 1350 if (dst->addrtype == TIPC_ADDR_MCAST) { 1351 res = -EINVAL; 1352 goto exit; 1353 } 1354 1355 /* Reject any messages already in receive queue (very unlikely) */ 1356 reject_rx_queue(sk); 1357 1358 /* Send a 'SYN-' to destination */ 1359 m.msg_name = dest; 1360 m.msg_namelen = destlen; 1361 res = send_msg(NULL, sock, &m, 0); 1362 if (res < 0) 1363 goto exit; 1364 1365 /* Wait until an 'ACK' or 'RST' arrives, or a timeout occurs */ 1366 timeout = tipc_sk(sk)->conn_timeout; 1367 release_sock(sk); 1368 res = wait_event_interruptible_timeout(*sk_sleep(sk), 1369 (!skb_queue_empty(&sk->sk_receive_queue) || 1370 (sock->state != SS_CONNECTING)), 1371 timeout ? (long)msecs_to_jiffies(timeout) 1372 : MAX_SCHEDULE_TIMEOUT); 1373 lock_sock(sk); 1374 1375 if (res > 0) { 1376 buf = skb_peek(&sk->sk_receive_queue); 1377 if (buf != NULL) { 1378 msg = buf_msg(buf); 1379 res = auto_connect(sock, msg); 1380 if (!res) { 1381 if (!msg_data_sz(msg)) 1382 advance_rx_queue(sk); 1383 } 1384 } else { 1385 if (sock->state == SS_CONNECTED) 1386 res = -EISCONN; 1387 else 1388 res = -ECONNREFUSED; 1389 } 1390 } else { 1391 if (res == 0) 1392 res = -ETIMEDOUT; 1393 else 1394 ; /* leave "res" unchanged */ 1395 sock->state = SS_DISCONNECTING; 1396 } 1397 1398 exit: 1399 release_sock(sk); 1400 return res; 1401 } 1402 1403 /** 1404 * listen - allow socket to listen for incoming connections 1405 * @sock: socket structure 1406 * @len: (unused) 1407 * 1408 * Returns 0 on success, errno otherwise 1409 */ 1410 static int listen(struct socket *sock, int len) 1411 { 1412 struct sock *sk = sock->sk; 1413 int res; 1414 1415 lock_sock(sk); 1416 1417 if (sock->state != SS_UNCONNECTED) 1418 res = -EINVAL; 1419 else { 1420 sock->state = SS_LISTENING; 1421 res = 0; 1422 } 1423 1424 release_sock(sk); 1425 return res; 1426 } 1427 1428 /** 1429 * accept - wait for connection request 1430 * @sock: listening socket 1431 * @newsock: new socket that is to be connected 1432 * @flags: file-related flags associated with socket 1433 * 1434 * Returns 0 on success, errno otherwise 1435 */ 1436 static int accept(struct socket *sock, struct socket *new_sock, int flags) 1437 { 1438 struct sock *sk = sock->sk; 1439 struct sk_buff *buf; 1440 int res; 1441 1442 lock_sock(sk); 1443 1444 if (sock->state != SS_LISTENING) { 1445 res = -EINVAL; 1446 goto exit; 1447 } 1448 1449 while (skb_queue_empty(&sk->sk_receive_queue)) { 1450 if (flags & O_NONBLOCK) { 1451 res = -EWOULDBLOCK; 1452 goto exit; 1453 } 1454 release_sock(sk); 1455 res = wait_event_interruptible(*sk_sleep(sk), 1456 (!skb_queue_empty(&sk->sk_receive_queue))); 1457 lock_sock(sk); 1458 if (res) 1459 goto exit; 1460 } 1461 1462 buf = skb_peek(&sk->sk_receive_queue); 1463 1464 res = tipc_create(sock_net(sock->sk), new_sock, 0, 0); 1465 if (!res) { 1466 struct sock *new_sk = new_sock->sk; 1467 struct tipc_sock *new_tsock = tipc_sk(new_sk); 1468 struct tipc_port *new_tport = new_tsock->p; 1469 u32 new_ref = new_tport->ref; 1470 struct tipc_msg *msg = buf_msg(buf); 1471 1472 lock_sock(new_sk); 1473 1474 /* 1475 * Reject any stray messages received by new socket 1476 * before the socket lock was taken (very, very unlikely) 1477 */ 1478 reject_rx_queue(new_sk); 1479 1480 /* Connect new socket to it's peer */ 1481 new_tsock->peer_name.ref = msg_origport(msg); 1482 new_tsock->peer_name.node = msg_orignode(msg); 1483 tipc_connect2port(new_ref, &new_tsock->peer_name); 1484 new_sock->state = SS_CONNECTED; 1485 1486 tipc_set_portimportance(new_ref, msg_importance(msg)); 1487 if (msg_named(msg)) { 1488 new_tport->conn_type = msg_nametype(msg); 1489 new_tport->conn_instance = msg_nameinst(msg); 1490 } 1491 1492 /* 1493 * Respond to 'SYN-' by discarding it & returning 'ACK'-. 1494 * Respond to 'SYN+' by queuing it on new socket. 1495 */ 1496 if (!msg_data_sz(msg)) { 1497 struct msghdr m = {NULL,}; 1498 1499 advance_rx_queue(sk); 1500 send_packet(NULL, new_sock, &m, 0); 1501 } else { 1502 __skb_dequeue(&sk->sk_receive_queue); 1503 __skb_queue_head(&new_sk->sk_receive_queue, buf); 1504 } 1505 release_sock(new_sk); 1506 } 1507 exit: 1508 release_sock(sk); 1509 return res; 1510 } 1511 1512 /** 1513 * shutdown - shutdown socket connection 1514 * @sock: socket structure 1515 * @how: direction to close (must be SHUT_RDWR) 1516 * 1517 * Terminates connection (if necessary), then purges socket's receive queue. 1518 * 1519 * Returns 0 on success, errno otherwise 1520 */ 1521 static int shutdown(struct socket *sock, int how) 1522 { 1523 struct sock *sk = sock->sk; 1524 struct tipc_port *tport = tipc_sk_port(sk); 1525 struct sk_buff *buf; 1526 int res; 1527 1528 if (how != SHUT_RDWR) 1529 return -EINVAL; 1530 1531 lock_sock(sk); 1532 1533 switch (sock->state) { 1534 case SS_CONNECTING: 1535 case SS_CONNECTED: 1536 1537 restart: 1538 /* Disconnect and send a 'FIN+' or 'FIN-' message to peer */ 1539 buf = __skb_dequeue(&sk->sk_receive_queue); 1540 if (buf) { 1541 atomic_dec(&tipc_queue_size); 1542 if (TIPC_SKB_CB(buf)->handle != 0) { 1543 kfree_skb(buf); 1544 goto restart; 1545 } 1546 tipc_disconnect(tport->ref); 1547 tipc_reject_msg(buf, TIPC_CONN_SHUTDOWN); 1548 } else { 1549 tipc_shutdown(tport->ref); 1550 } 1551 1552 sock->state = SS_DISCONNECTING; 1553 1554 /* fall through */ 1555 1556 case SS_DISCONNECTING: 1557 1558 /* Discard any unreceived messages; wake up sleeping tasks */ 1559 discard_rx_queue(sk); 1560 if (waitqueue_active(sk_sleep(sk))) 1561 wake_up_interruptible(sk_sleep(sk)); 1562 res = 0; 1563 break; 1564 1565 default: 1566 res = -ENOTCONN; 1567 } 1568 1569 release_sock(sk); 1570 return res; 1571 } 1572 1573 /** 1574 * setsockopt - set socket option 1575 * @sock: socket structure 1576 * @lvl: option level 1577 * @opt: option identifier 1578 * @ov: pointer to new option value 1579 * @ol: length of option value 1580 * 1581 * For stream sockets only, accepts and ignores all IPPROTO_TCP options 1582 * (to ease compatibility). 1583 * 1584 * Returns 0 on success, errno otherwise 1585 */ 1586 static int setsockopt(struct socket *sock, 1587 int lvl, int opt, char __user *ov, unsigned int ol) 1588 { 1589 struct sock *sk = sock->sk; 1590 struct tipc_port *tport = tipc_sk_port(sk); 1591 u32 value; 1592 int res; 1593 1594 if ((lvl == IPPROTO_TCP) && (sock->type == SOCK_STREAM)) 1595 return 0; 1596 if (lvl != SOL_TIPC) 1597 return -ENOPROTOOPT; 1598 if (ol < sizeof(value)) 1599 return -EINVAL; 1600 res = get_user(value, (u32 __user *)ov); 1601 if (res) 1602 return res; 1603 1604 lock_sock(sk); 1605 1606 switch (opt) { 1607 case TIPC_IMPORTANCE: 1608 res = tipc_set_portimportance(tport->ref, value); 1609 break; 1610 case TIPC_SRC_DROPPABLE: 1611 if (sock->type != SOCK_STREAM) 1612 res = tipc_set_portunreliable(tport->ref, value); 1613 else 1614 res = -ENOPROTOOPT; 1615 break; 1616 case TIPC_DEST_DROPPABLE: 1617 res = tipc_set_portunreturnable(tport->ref, value); 1618 break; 1619 case TIPC_CONN_TIMEOUT: 1620 tipc_sk(sk)->conn_timeout = value; 1621 /* no need to set "res", since already 0 at this point */ 1622 break; 1623 default: 1624 res = -EINVAL; 1625 } 1626 1627 release_sock(sk); 1628 1629 return res; 1630 } 1631 1632 /** 1633 * getsockopt - get socket option 1634 * @sock: socket structure 1635 * @lvl: option level 1636 * @opt: option identifier 1637 * @ov: receptacle for option value 1638 * @ol: receptacle for length of option value 1639 * 1640 * For stream sockets only, returns 0 length result for all IPPROTO_TCP options 1641 * (to ease compatibility). 1642 * 1643 * Returns 0 on success, errno otherwise 1644 */ 1645 static int getsockopt(struct socket *sock, 1646 int lvl, int opt, char __user *ov, int __user *ol) 1647 { 1648 struct sock *sk = sock->sk; 1649 struct tipc_port *tport = tipc_sk_port(sk); 1650 int len; 1651 u32 value; 1652 int res; 1653 1654 if ((lvl == IPPROTO_TCP) && (sock->type == SOCK_STREAM)) 1655 return put_user(0, ol); 1656 if (lvl != SOL_TIPC) 1657 return -ENOPROTOOPT; 1658 res = get_user(len, ol); 1659 if (res) 1660 return res; 1661 1662 lock_sock(sk); 1663 1664 switch (opt) { 1665 case TIPC_IMPORTANCE: 1666 res = tipc_portimportance(tport->ref, &value); 1667 break; 1668 case TIPC_SRC_DROPPABLE: 1669 res = tipc_portunreliable(tport->ref, &value); 1670 break; 1671 case TIPC_DEST_DROPPABLE: 1672 res = tipc_portunreturnable(tport->ref, &value); 1673 break; 1674 case TIPC_CONN_TIMEOUT: 1675 value = tipc_sk(sk)->conn_timeout; 1676 /* no need to set "res", since already 0 at this point */ 1677 break; 1678 case TIPC_NODE_RECVQ_DEPTH: 1679 value = (u32)atomic_read(&tipc_queue_size); 1680 break; 1681 case TIPC_SOCK_RECVQ_DEPTH: 1682 value = skb_queue_len(&sk->sk_receive_queue); 1683 break; 1684 default: 1685 res = -EINVAL; 1686 } 1687 1688 release_sock(sk); 1689 1690 if (res) 1691 return res; /* "get" failed */ 1692 1693 if (len < sizeof(value)) 1694 return -EINVAL; 1695 1696 if (copy_to_user(ov, &value, sizeof(value))) 1697 return -EFAULT; 1698 1699 return put_user(sizeof(value), ol); 1700 } 1701 1702 /* Protocol switches for the various types of TIPC sockets */ 1703 1704 static const struct proto_ops msg_ops = { 1705 .owner = THIS_MODULE, 1706 .family = AF_TIPC, 1707 .release = release, 1708 .bind = bind, 1709 .connect = connect, 1710 .socketpair = sock_no_socketpair, 1711 .accept = sock_no_accept, 1712 .getname = get_name, 1713 .poll = poll, 1714 .ioctl = sock_no_ioctl, 1715 .listen = sock_no_listen, 1716 .shutdown = shutdown, 1717 .setsockopt = setsockopt, 1718 .getsockopt = getsockopt, 1719 .sendmsg = send_msg, 1720 .recvmsg = recv_msg, 1721 .mmap = sock_no_mmap, 1722 .sendpage = sock_no_sendpage 1723 }; 1724 1725 static const struct proto_ops packet_ops = { 1726 .owner = THIS_MODULE, 1727 .family = AF_TIPC, 1728 .release = release, 1729 .bind = bind, 1730 .connect = connect, 1731 .socketpair = sock_no_socketpair, 1732 .accept = accept, 1733 .getname = get_name, 1734 .poll = poll, 1735 .ioctl = sock_no_ioctl, 1736 .listen = listen, 1737 .shutdown = shutdown, 1738 .setsockopt = setsockopt, 1739 .getsockopt = getsockopt, 1740 .sendmsg = send_packet, 1741 .recvmsg = recv_msg, 1742 .mmap = sock_no_mmap, 1743 .sendpage = sock_no_sendpage 1744 }; 1745 1746 static const struct proto_ops stream_ops = { 1747 .owner = THIS_MODULE, 1748 .family = AF_TIPC, 1749 .release = release, 1750 .bind = bind, 1751 .connect = connect, 1752 .socketpair = sock_no_socketpair, 1753 .accept = accept, 1754 .getname = get_name, 1755 .poll = poll, 1756 .ioctl = sock_no_ioctl, 1757 .listen = listen, 1758 .shutdown = shutdown, 1759 .setsockopt = setsockopt, 1760 .getsockopt = getsockopt, 1761 .sendmsg = send_stream, 1762 .recvmsg = recv_stream, 1763 .mmap = sock_no_mmap, 1764 .sendpage = sock_no_sendpage 1765 }; 1766 1767 static const struct net_proto_family tipc_family_ops = { 1768 .owner = THIS_MODULE, 1769 .family = AF_TIPC, 1770 .create = tipc_create 1771 }; 1772 1773 static struct proto tipc_proto = { 1774 .name = "TIPC", 1775 .owner = THIS_MODULE, 1776 .obj_size = sizeof(struct tipc_sock) 1777 }; 1778 1779 /** 1780 * tipc_socket_init - initialize TIPC socket interface 1781 * 1782 * Returns 0 on success, errno otherwise 1783 */ 1784 int tipc_socket_init(void) 1785 { 1786 int res; 1787 1788 res = proto_register(&tipc_proto, 1); 1789 if (res) { 1790 pr_err("Failed to register TIPC protocol type\n"); 1791 goto out; 1792 } 1793 1794 res = sock_register(&tipc_family_ops); 1795 if (res) { 1796 pr_err("Failed to register TIPC socket type\n"); 1797 proto_unregister(&tipc_proto); 1798 goto out; 1799 } 1800 1801 sockets_enabled = 1; 1802 out: 1803 return res; 1804 } 1805 1806 /** 1807 * tipc_socket_stop - stop TIPC socket interface 1808 */ 1809 void tipc_socket_stop(void) 1810 { 1811 if (!sockets_enabled) 1812 return; 1813 1814 sockets_enabled = 0; 1815 sock_unregister(tipc_family_ops.family); 1816 proto_unregister(&tipc_proto); 1817 } 1818