1 /* 2 * net/tipc/socket.c: TIPC socket API 3 * 4 * Copyright (c) 2001-2007, 2012-2014, Ericsson AB 5 * Copyright (c) 2004-2008, 2010-2013, 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 42 #define SS_LISTENING -1 /* socket is listening */ 43 #define SS_READY -2 /* socket is connectionless */ 44 45 #define CONN_TIMEOUT_DEFAULT 8000 /* default connect timeout = 8s */ 46 47 static int backlog_rcv(struct sock *sk, struct sk_buff *skb); 48 static void tipc_data_ready(struct sock *sk); 49 static void tipc_write_space(struct sock *sk); 50 static int tipc_release(struct socket *sock); 51 static int tipc_accept(struct socket *sock, struct socket *new_sock, int flags); 52 53 static const struct proto_ops packet_ops; 54 static const struct proto_ops stream_ops; 55 static const struct proto_ops msg_ops; 56 57 static struct proto tipc_proto; 58 static struct proto tipc_proto_kern; 59 60 /* 61 * Revised TIPC socket locking policy: 62 * 63 * Most socket operations take the standard socket lock when they start 64 * and hold it until they finish (or until they need to sleep). Acquiring 65 * this lock grants the owner exclusive access to the fields of the socket 66 * data structures, with the exception of the backlog queue. A few socket 67 * operations can be done without taking the socket lock because they only 68 * read socket information that never changes during the life of the socket. 69 * 70 * Socket operations may acquire the lock for the associated TIPC port if they 71 * need to perform an operation on the port. If any routine needs to acquire 72 * both the socket lock and the port lock it must take the socket lock first 73 * to avoid the risk of deadlock. 74 * 75 * The dispatcher handling incoming messages cannot grab the socket lock in 76 * the standard fashion, since invoked it runs at the BH level and cannot block. 77 * Instead, it checks to see if the socket lock is currently owned by someone, 78 * and either handles the message itself or adds it to the socket's backlog 79 * queue; in the latter case the queued message is processed once the process 80 * owning the socket lock releases it. 81 * 82 * NOTE: Releasing the socket lock while an operation is sleeping overcomes 83 * the problem of a blocked socket operation preventing any other operations 84 * from occurring. However, applications must be careful if they have 85 * multiple threads trying to send (or receive) on the same socket, as these 86 * operations might interfere with each other. For example, doing a connect 87 * and a receive at the same time might allow the receive to consume the 88 * ACK message meant for the connect. While additional work could be done 89 * to try and overcome this, it doesn't seem to be worthwhile at the present. 90 * 91 * NOTE: Releasing the socket lock while an operation is sleeping also ensures 92 * that another operation that must be performed in a non-blocking manner is 93 * not delayed for very long because the lock has already been taken. 94 * 95 * NOTE: This code assumes that certain fields of a port/socket pair are 96 * constant over its lifetime; such fields can be examined without taking 97 * the socket lock and/or port lock, and do not need to be re-read even 98 * after resuming processing after waiting. These fields include: 99 * - socket type 100 * - pointer to socket sk structure (aka tipc_sock structure) 101 * - pointer to port structure 102 * - port reference 103 */ 104 105 #include "socket.h" 106 107 /** 108 * advance_rx_queue - discard first buffer in socket receive queue 109 * 110 * Caller must hold socket lock 111 */ 112 static void advance_rx_queue(struct sock *sk) 113 { 114 kfree_skb(__skb_dequeue(&sk->sk_receive_queue)); 115 } 116 117 /** 118 * reject_rx_queue - reject all buffers in socket receive queue 119 * 120 * Caller must hold socket lock 121 */ 122 static void reject_rx_queue(struct sock *sk) 123 { 124 struct sk_buff *buf; 125 126 while ((buf = __skb_dequeue(&sk->sk_receive_queue))) 127 tipc_reject_msg(buf, TIPC_ERR_NO_PORT); 128 } 129 130 /** 131 * tipc_sk_create - create a TIPC socket 132 * @net: network namespace (must be default network) 133 * @sock: pre-allocated socket structure 134 * @protocol: protocol indicator (must be 0) 135 * @kern: caused by kernel or by userspace? 136 * 137 * This routine creates additional data structures used by the TIPC socket, 138 * initializes them, and links them together. 139 * 140 * Returns 0 on success, errno otherwise 141 */ 142 static int tipc_sk_create(struct net *net, struct socket *sock, 143 int protocol, int kern) 144 { 145 const struct proto_ops *ops; 146 socket_state state; 147 struct sock *sk; 148 struct tipc_sock *tsk; 149 struct tipc_port *port; 150 u32 ref; 151 152 /* Validate arguments */ 153 if (unlikely(protocol != 0)) 154 return -EPROTONOSUPPORT; 155 156 switch (sock->type) { 157 case SOCK_STREAM: 158 ops = &stream_ops; 159 state = SS_UNCONNECTED; 160 break; 161 case SOCK_SEQPACKET: 162 ops = &packet_ops; 163 state = SS_UNCONNECTED; 164 break; 165 case SOCK_DGRAM: 166 case SOCK_RDM: 167 ops = &msg_ops; 168 state = SS_READY; 169 break; 170 default: 171 return -EPROTOTYPE; 172 } 173 174 /* Allocate socket's protocol area */ 175 if (!kern) 176 sk = sk_alloc(net, AF_TIPC, GFP_KERNEL, &tipc_proto); 177 else 178 sk = sk_alloc(net, AF_TIPC, GFP_KERNEL, &tipc_proto_kern); 179 180 if (sk == NULL) 181 return -ENOMEM; 182 183 tsk = tipc_sk(sk); 184 port = &tsk->port; 185 186 ref = tipc_port_init(port, TIPC_LOW_IMPORTANCE); 187 if (!ref) { 188 pr_warn("Socket registration failed, ref. table exhausted\n"); 189 sk_free(sk); 190 return -ENOMEM; 191 } 192 193 /* Finish initializing socket data structures */ 194 sock->ops = ops; 195 sock->state = state; 196 197 sock_init_data(sock, sk); 198 sk->sk_backlog_rcv = backlog_rcv; 199 sk->sk_rcvbuf = sysctl_tipc_rmem[1]; 200 sk->sk_data_ready = tipc_data_ready; 201 sk->sk_write_space = tipc_write_space; 202 tipc_sk(sk)->conn_timeout = CONN_TIMEOUT_DEFAULT; 203 tipc_port_unlock(port); 204 205 if (sock->state == SS_READY) { 206 tipc_port_set_unreturnable(port, true); 207 if (sock->type == SOCK_DGRAM) 208 tipc_port_set_unreliable(port, true); 209 } 210 return 0; 211 } 212 213 /** 214 * tipc_sock_create_local - create TIPC socket from inside TIPC module 215 * @type: socket type - SOCK_RDM or SOCK_SEQPACKET 216 * 217 * We cannot use sock_creat_kern here because it bumps module user count. 218 * Since socket owner and creator is the same module we must make sure 219 * that module count remains zero for module local sockets, otherwise 220 * we cannot do rmmod. 221 * 222 * Returns 0 on success, errno otherwise 223 */ 224 int tipc_sock_create_local(int type, struct socket **res) 225 { 226 int rc; 227 228 rc = sock_create_lite(AF_TIPC, type, 0, res); 229 if (rc < 0) { 230 pr_err("Failed to create kernel socket\n"); 231 return rc; 232 } 233 tipc_sk_create(&init_net, *res, 0, 1); 234 235 return 0; 236 } 237 238 /** 239 * tipc_sock_release_local - release socket created by tipc_sock_create_local 240 * @sock: the socket to be released. 241 * 242 * Module reference count is not incremented when such sockets are created, 243 * so we must keep it from being decremented when they are released. 244 */ 245 void tipc_sock_release_local(struct socket *sock) 246 { 247 tipc_release(sock); 248 sock->ops = NULL; 249 sock_release(sock); 250 } 251 252 /** 253 * tipc_sock_accept_local - accept a connection on a socket created 254 * with tipc_sock_create_local. Use this function to avoid that 255 * module reference count is inadvertently incremented. 256 * 257 * @sock: the accepting socket 258 * @newsock: reference to the new socket to be created 259 * @flags: socket flags 260 */ 261 262 int tipc_sock_accept_local(struct socket *sock, struct socket **newsock, 263 int flags) 264 { 265 struct sock *sk = sock->sk; 266 int ret; 267 268 ret = sock_create_lite(sk->sk_family, sk->sk_type, 269 sk->sk_protocol, newsock); 270 if (ret < 0) 271 return ret; 272 273 ret = tipc_accept(sock, *newsock, flags); 274 if (ret < 0) { 275 sock_release(*newsock); 276 return ret; 277 } 278 (*newsock)->ops = sock->ops; 279 return ret; 280 } 281 282 /** 283 * tipc_release - destroy a TIPC socket 284 * @sock: socket to destroy 285 * 286 * This routine cleans up any messages that are still queued on the socket. 287 * For DGRAM and RDM socket types, all queued messages are rejected. 288 * For SEQPACKET and STREAM socket types, the first message is rejected 289 * and any others are discarded. (If the first message on a STREAM socket 290 * is partially-read, it is discarded and the next one is rejected instead.) 291 * 292 * NOTE: Rejected messages are not necessarily returned to the sender! They 293 * are returned or discarded according to the "destination droppable" setting 294 * specified for the message by the sender. 295 * 296 * Returns 0 on success, errno otherwise 297 */ 298 static int tipc_release(struct socket *sock) 299 { 300 struct sock *sk = sock->sk; 301 struct tipc_sock *tsk; 302 struct tipc_port *port; 303 struct sk_buff *buf; 304 305 /* 306 * Exit if socket isn't fully initialized (occurs when a failed accept() 307 * releases a pre-allocated child socket that was never used) 308 */ 309 if (sk == NULL) 310 return 0; 311 312 tsk = tipc_sk(sk); 313 port = &tsk->port; 314 lock_sock(sk); 315 316 /* 317 * Reject all unreceived messages, except on an active connection 318 * (which disconnects locally & sends a 'FIN+' to peer) 319 */ 320 while (sock->state != SS_DISCONNECTING) { 321 buf = __skb_dequeue(&sk->sk_receive_queue); 322 if (buf == NULL) 323 break; 324 if (TIPC_SKB_CB(buf)->handle != NULL) 325 kfree_skb(buf); 326 else { 327 if ((sock->state == SS_CONNECTING) || 328 (sock->state == SS_CONNECTED)) { 329 sock->state = SS_DISCONNECTING; 330 tipc_port_disconnect(port->ref); 331 } 332 tipc_reject_msg(buf, TIPC_ERR_NO_PORT); 333 } 334 } 335 336 /* Destroy TIPC port; also disconnects an active connection and 337 * sends a 'FIN-' to peer. 338 */ 339 tipc_port_destroy(port); 340 341 /* Discard any remaining (connection-based) messages in receive queue */ 342 __skb_queue_purge(&sk->sk_receive_queue); 343 344 /* Reject any messages that accumulated in backlog queue */ 345 sock->state = SS_DISCONNECTING; 346 release_sock(sk); 347 348 sock_put(sk); 349 sock->sk = NULL; 350 351 return 0; 352 } 353 354 /** 355 * tipc_bind - associate or disassocate TIPC name(s) with a socket 356 * @sock: socket structure 357 * @uaddr: socket address describing name(s) and desired operation 358 * @uaddr_len: size of socket address data structure 359 * 360 * Name and name sequence binding is indicated using a positive scope value; 361 * a negative scope value unbinds the specified name. Specifying no name 362 * (i.e. a socket address length of 0) unbinds all names from the socket. 363 * 364 * Returns 0 on success, errno otherwise 365 * 366 * NOTE: This routine doesn't need to take the socket lock since it doesn't 367 * access any non-constant socket information. 368 */ 369 static int tipc_bind(struct socket *sock, struct sockaddr *uaddr, 370 int uaddr_len) 371 { 372 struct sock *sk = sock->sk; 373 struct sockaddr_tipc *addr = (struct sockaddr_tipc *)uaddr; 374 struct tipc_sock *tsk = tipc_sk(sk); 375 int res = -EINVAL; 376 377 lock_sock(sk); 378 if (unlikely(!uaddr_len)) { 379 res = tipc_withdraw(&tsk->port, 0, NULL); 380 goto exit; 381 } 382 383 if (uaddr_len < sizeof(struct sockaddr_tipc)) { 384 res = -EINVAL; 385 goto exit; 386 } 387 if (addr->family != AF_TIPC) { 388 res = -EAFNOSUPPORT; 389 goto exit; 390 } 391 392 if (addr->addrtype == TIPC_ADDR_NAME) 393 addr->addr.nameseq.upper = addr->addr.nameseq.lower; 394 else if (addr->addrtype != TIPC_ADDR_NAMESEQ) { 395 res = -EAFNOSUPPORT; 396 goto exit; 397 } 398 399 if ((addr->addr.nameseq.type < TIPC_RESERVED_TYPES) && 400 (addr->addr.nameseq.type != TIPC_TOP_SRV) && 401 (addr->addr.nameseq.type != TIPC_CFG_SRV)) { 402 res = -EACCES; 403 goto exit; 404 } 405 406 res = (addr->scope > 0) ? 407 tipc_publish(&tsk->port, addr->scope, &addr->addr.nameseq) : 408 tipc_withdraw(&tsk->port, -addr->scope, &addr->addr.nameseq); 409 exit: 410 release_sock(sk); 411 return res; 412 } 413 414 /** 415 * tipc_getname - get port ID of socket or peer socket 416 * @sock: socket structure 417 * @uaddr: area for returned socket address 418 * @uaddr_len: area for returned length of socket address 419 * @peer: 0 = own ID, 1 = current peer ID, 2 = current/former peer ID 420 * 421 * Returns 0 on success, errno otherwise 422 * 423 * NOTE: This routine doesn't need to take the socket lock since it only 424 * accesses socket information that is unchanging (or which changes in 425 * a completely predictable manner). 426 */ 427 static int tipc_getname(struct socket *sock, struct sockaddr *uaddr, 428 int *uaddr_len, int peer) 429 { 430 struct sockaddr_tipc *addr = (struct sockaddr_tipc *)uaddr; 431 struct tipc_sock *tsk = tipc_sk(sock->sk); 432 433 memset(addr, 0, sizeof(*addr)); 434 if (peer) { 435 if ((sock->state != SS_CONNECTED) && 436 ((peer != 2) || (sock->state != SS_DISCONNECTING))) 437 return -ENOTCONN; 438 addr->addr.id.ref = tipc_port_peerport(&tsk->port); 439 addr->addr.id.node = tipc_port_peernode(&tsk->port); 440 } else { 441 addr->addr.id.ref = tsk->port.ref; 442 addr->addr.id.node = tipc_own_addr; 443 } 444 445 *uaddr_len = sizeof(*addr); 446 addr->addrtype = TIPC_ADDR_ID; 447 addr->family = AF_TIPC; 448 addr->scope = 0; 449 addr->addr.name.domain = 0; 450 451 return 0; 452 } 453 454 /** 455 * tipc_poll - read and possibly block on pollmask 456 * @file: file structure associated with the socket 457 * @sock: socket for which to calculate the poll bits 458 * @wait: ??? 459 * 460 * Returns pollmask value 461 * 462 * COMMENTARY: 463 * It appears that the usual socket locking mechanisms are not useful here 464 * since the pollmask info is potentially out-of-date the moment this routine 465 * exits. TCP and other protocols seem to rely on higher level poll routines 466 * to handle any preventable race conditions, so TIPC will do the same ... 467 * 468 * TIPC sets the returned events as follows: 469 * 470 * socket state flags set 471 * ------------ --------- 472 * unconnected no read flags 473 * POLLOUT if port is not congested 474 * 475 * connecting POLLIN/POLLRDNORM if ACK/NACK in rx queue 476 * no write flags 477 * 478 * connected POLLIN/POLLRDNORM if data in rx queue 479 * POLLOUT if port is not congested 480 * 481 * disconnecting POLLIN/POLLRDNORM/POLLHUP 482 * no write flags 483 * 484 * listening POLLIN if SYN in rx queue 485 * no write flags 486 * 487 * ready POLLIN/POLLRDNORM if data in rx queue 488 * [connectionless] POLLOUT (since port cannot be congested) 489 * 490 * IMPORTANT: The fact that a read or write operation is indicated does NOT 491 * imply that the operation will succeed, merely that it should be performed 492 * and will not block. 493 */ 494 static unsigned int tipc_poll(struct file *file, struct socket *sock, 495 poll_table *wait) 496 { 497 struct sock *sk = sock->sk; 498 struct tipc_sock *tsk = tipc_sk(sk); 499 u32 mask = 0; 500 501 sock_poll_wait(file, sk_sleep(sk), wait); 502 503 switch ((int)sock->state) { 504 case SS_UNCONNECTED: 505 if (!tsk->port.congested) 506 mask |= POLLOUT; 507 break; 508 case SS_READY: 509 case SS_CONNECTED: 510 if (!tsk->port.congested) 511 mask |= POLLOUT; 512 /* fall thru' */ 513 case SS_CONNECTING: 514 case SS_LISTENING: 515 if (!skb_queue_empty(&sk->sk_receive_queue)) 516 mask |= (POLLIN | POLLRDNORM); 517 break; 518 case SS_DISCONNECTING: 519 mask = (POLLIN | POLLRDNORM | POLLHUP); 520 break; 521 } 522 523 return mask; 524 } 525 526 /** 527 * dest_name_check - verify user is permitted to send to specified port name 528 * @dest: destination address 529 * @m: descriptor for message to be sent 530 * 531 * Prevents restricted configuration commands from being issued by 532 * unauthorized users. 533 * 534 * Returns 0 if permission is granted, otherwise errno 535 */ 536 static int dest_name_check(struct sockaddr_tipc *dest, struct msghdr *m) 537 { 538 struct tipc_cfg_msg_hdr hdr; 539 540 if (likely(dest->addr.name.name.type >= TIPC_RESERVED_TYPES)) 541 return 0; 542 if (likely(dest->addr.name.name.type == TIPC_TOP_SRV)) 543 return 0; 544 if (likely(dest->addr.name.name.type != TIPC_CFG_SRV)) 545 return -EACCES; 546 547 if (!m->msg_iovlen || (m->msg_iov[0].iov_len < sizeof(hdr))) 548 return -EMSGSIZE; 549 if (copy_from_user(&hdr, m->msg_iov[0].iov_base, sizeof(hdr))) 550 return -EFAULT; 551 if ((ntohs(hdr.tcm_type) & 0xC000) && (!capable(CAP_NET_ADMIN))) 552 return -EACCES; 553 554 return 0; 555 } 556 557 static int tipc_wait_for_sndmsg(struct socket *sock, long *timeo_p) 558 { 559 struct sock *sk = sock->sk; 560 struct tipc_sock *tsk = tipc_sk(sk); 561 DEFINE_WAIT(wait); 562 int done; 563 564 do { 565 int err = sock_error(sk); 566 if (err) 567 return err; 568 if (sock->state == SS_DISCONNECTING) 569 return -EPIPE; 570 if (!*timeo_p) 571 return -EAGAIN; 572 if (signal_pending(current)) 573 return sock_intr_errno(*timeo_p); 574 575 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 576 done = sk_wait_event(sk, timeo_p, !tsk->port.congested); 577 finish_wait(sk_sleep(sk), &wait); 578 } while (!done); 579 return 0; 580 } 581 582 583 /** 584 * tipc_sendmsg - send message in connectionless manner 585 * @iocb: if NULL, indicates that socket lock is already held 586 * @sock: socket structure 587 * @m: message to send 588 * @total_len: length of message 589 * 590 * Message must have an destination specified explicitly. 591 * Used for SOCK_RDM and SOCK_DGRAM messages, 592 * and for 'SYN' messages on SOCK_SEQPACKET and SOCK_STREAM connections. 593 * (Note: 'SYN+' is prohibited on SOCK_STREAM.) 594 * 595 * Returns the number of bytes sent on success, or errno otherwise 596 */ 597 static int tipc_sendmsg(struct kiocb *iocb, struct socket *sock, 598 struct msghdr *m, size_t total_len) 599 { 600 struct sock *sk = sock->sk; 601 struct tipc_sock *tsk = tipc_sk(sk); 602 struct tipc_port *port = &tsk->port; 603 DECLARE_SOCKADDR(struct sockaddr_tipc *, dest, m->msg_name); 604 int needs_conn; 605 long timeo; 606 int res = -EINVAL; 607 608 if (unlikely(!dest)) 609 return -EDESTADDRREQ; 610 if (unlikely((m->msg_namelen < sizeof(*dest)) || 611 (dest->family != AF_TIPC))) 612 return -EINVAL; 613 if (total_len > TIPC_MAX_USER_MSG_SIZE) 614 return -EMSGSIZE; 615 616 if (iocb) 617 lock_sock(sk); 618 619 needs_conn = (sock->state != SS_READY); 620 if (unlikely(needs_conn)) { 621 if (sock->state == SS_LISTENING) { 622 res = -EPIPE; 623 goto exit; 624 } 625 if (sock->state != SS_UNCONNECTED) { 626 res = -EISCONN; 627 goto exit; 628 } 629 if (tsk->port.published) { 630 res = -EOPNOTSUPP; 631 goto exit; 632 } 633 if (dest->addrtype == TIPC_ADDR_NAME) { 634 tsk->port.conn_type = dest->addr.name.name.type; 635 tsk->port.conn_instance = dest->addr.name.name.instance; 636 } 637 638 /* Abort any pending connection attempts (very unlikely) */ 639 reject_rx_queue(sk); 640 } 641 642 timeo = sock_sndtimeo(sk, m->msg_flags & MSG_DONTWAIT); 643 do { 644 if (dest->addrtype == TIPC_ADDR_NAME) { 645 res = dest_name_check(dest, m); 646 if (res) 647 break; 648 res = tipc_send2name(port, 649 &dest->addr.name.name, 650 dest->addr.name.domain, 651 m->msg_iov, 652 total_len); 653 } else if (dest->addrtype == TIPC_ADDR_ID) { 654 res = tipc_send2port(port, 655 &dest->addr.id, 656 m->msg_iov, 657 total_len); 658 } else if (dest->addrtype == TIPC_ADDR_MCAST) { 659 if (needs_conn) { 660 res = -EOPNOTSUPP; 661 break; 662 } 663 res = dest_name_check(dest, m); 664 if (res) 665 break; 666 res = tipc_port_mcast_xmit(port, 667 &dest->addr.nameseq, 668 m->msg_iov, 669 total_len); 670 } 671 if (likely(res != -ELINKCONG)) { 672 if (needs_conn && (res >= 0)) 673 sock->state = SS_CONNECTING; 674 break; 675 } 676 res = tipc_wait_for_sndmsg(sock, &timeo); 677 if (res) 678 break; 679 } while (1); 680 681 exit: 682 if (iocb) 683 release_sock(sk); 684 return res; 685 } 686 687 static int tipc_wait_for_sndpkt(struct socket *sock, long *timeo_p) 688 { 689 struct sock *sk = sock->sk; 690 struct tipc_sock *tsk = tipc_sk(sk); 691 struct tipc_port *port = &tsk->port; 692 DEFINE_WAIT(wait); 693 int done; 694 695 do { 696 int err = sock_error(sk); 697 if (err) 698 return err; 699 if (sock->state == SS_DISCONNECTING) 700 return -EPIPE; 701 else if (sock->state != SS_CONNECTED) 702 return -ENOTCONN; 703 if (!*timeo_p) 704 return -EAGAIN; 705 if (signal_pending(current)) 706 return sock_intr_errno(*timeo_p); 707 708 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 709 done = sk_wait_event(sk, timeo_p, 710 (!port->congested || !port->connected)); 711 finish_wait(sk_sleep(sk), &wait); 712 } while (!done); 713 return 0; 714 } 715 716 /** 717 * tipc_send_packet - send a connection-oriented message 718 * @iocb: if NULL, indicates that socket lock is already held 719 * @sock: socket structure 720 * @m: message to send 721 * @total_len: length of message 722 * 723 * Used for SOCK_SEQPACKET messages and SOCK_STREAM data. 724 * 725 * Returns the number of bytes sent on success, or errno otherwise 726 */ 727 static int tipc_send_packet(struct kiocb *iocb, struct socket *sock, 728 struct msghdr *m, size_t total_len) 729 { 730 struct sock *sk = sock->sk; 731 struct tipc_sock *tsk = tipc_sk(sk); 732 DECLARE_SOCKADDR(struct sockaddr_tipc *, dest, m->msg_name); 733 int res = -EINVAL; 734 long timeo; 735 736 /* Handle implied connection establishment */ 737 if (unlikely(dest)) 738 return tipc_sendmsg(iocb, sock, m, total_len); 739 740 if (total_len > TIPC_MAX_USER_MSG_SIZE) 741 return -EMSGSIZE; 742 743 if (iocb) 744 lock_sock(sk); 745 746 if (unlikely(sock->state != SS_CONNECTED)) { 747 if (sock->state == SS_DISCONNECTING) 748 res = -EPIPE; 749 else 750 res = -ENOTCONN; 751 goto exit; 752 } 753 754 timeo = sock_sndtimeo(sk, m->msg_flags & MSG_DONTWAIT); 755 do { 756 res = tipc_send(&tsk->port, m->msg_iov, total_len); 757 if (likely(res != -ELINKCONG)) 758 break; 759 res = tipc_wait_for_sndpkt(sock, &timeo); 760 if (res) 761 break; 762 } while (1); 763 exit: 764 if (iocb) 765 release_sock(sk); 766 return res; 767 } 768 769 /** 770 * tipc_send_stream - send stream-oriented data 771 * @iocb: (unused) 772 * @sock: socket structure 773 * @m: data to send 774 * @total_len: total length of data to be sent 775 * 776 * Used for SOCK_STREAM data. 777 * 778 * Returns the number of bytes sent on success (or partial success), 779 * or errno if no data sent 780 */ 781 static int tipc_send_stream(struct kiocb *iocb, struct socket *sock, 782 struct msghdr *m, size_t total_len) 783 { 784 struct sock *sk = sock->sk; 785 struct tipc_sock *tsk = tipc_sk(sk); 786 struct msghdr my_msg; 787 struct iovec my_iov; 788 struct iovec *curr_iov; 789 int curr_iovlen; 790 char __user *curr_start; 791 u32 hdr_size; 792 int curr_left; 793 int bytes_to_send; 794 int bytes_sent; 795 int res; 796 797 lock_sock(sk); 798 799 /* Handle special cases where there is no connection */ 800 if (unlikely(sock->state != SS_CONNECTED)) { 801 if (sock->state == SS_UNCONNECTED) 802 res = tipc_send_packet(NULL, sock, m, total_len); 803 else 804 res = sock->state == SS_DISCONNECTING ? -EPIPE : -ENOTCONN; 805 goto exit; 806 } 807 808 if (unlikely(m->msg_name)) { 809 res = -EISCONN; 810 goto exit; 811 } 812 813 if (total_len > (unsigned int)INT_MAX) { 814 res = -EMSGSIZE; 815 goto exit; 816 } 817 818 /* 819 * Send each iovec entry using one or more messages 820 * 821 * Note: This algorithm is good for the most likely case 822 * (i.e. one large iovec entry), but could be improved to pass sets 823 * of small iovec entries into send_packet(). 824 */ 825 curr_iov = m->msg_iov; 826 curr_iovlen = m->msg_iovlen; 827 my_msg.msg_iov = &my_iov; 828 my_msg.msg_iovlen = 1; 829 my_msg.msg_flags = m->msg_flags; 830 my_msg.msg_name = NULL; 831 bytes_sent = 0; 832 833 hdr_size = msg_hdr_sz(&tsk->port.phdr); 834 835 while (curr_iovlen--) { 836 curr_start = curr_iov->iov_base; 837 curr_left = curr_iov->iov_len; 838 839 while (curr_left) { 840 bytes_to_send = tsk->port.max_pkt - hdr_size; 841 if (bytes_to_send > TIPC_MAX_USER_MSG_SIZE) 842 bytes_to_send = TIPC_MAX_USER_MSG_SIZE; 843 if (curr_left < bytes_to_send) 844 bytes_to_send = curr_left; 845 my_iov.iov_base = curr_start; 846 my_iov.iov_len = bytes_to_send; 847 res = tipc_send_packet(NULL, sock, &my_msg, 848 bytes_to_send); 849 if (res < 0) { 850 if (bytes_sent) 851 res = bytes_sent; 852 goto exit; 853 } 854 curr_left -= bytes_to_send; 855 curr_start += bytes_to_send; 856 bytes_sent += bytes_to_send; 857 } 858 859 curr_iov++; 860 } 861 res = bytes_sent; 862 exit: 863 release_sock(sk); 864 return res; 865 } 866 867 /** 868 * auto_connect - complete connection setup to a remote port 869 * @tsk: tipc socket structure 870 * @msg: peer's response message 871 * 872 * Returns 0 on success, errno otherwise 873 */ 874 static int auto_connect(struct tipc_sock *tsk, struct tipc_msg *msg) 875 { 876 struct tipc_port *port = &tsk->port; 877 struct socket *sock = tsk->sk.sk_socket; 878 struct tipc_portid peer; 879 880 peer.ref = msg_origport(msg); 881 peer.node = msg_orignode(msg); 882 883 __tipc_port_connect(port->ref, port, &peer); 884 885 if (msg_importance(msg) > TIPC_CRITICAL_IMPORTANCE) 886 return -EINVAL; 887 msg_set_importance(&port->phdr, (u32)msg_importance(msg)); 888 sock->state = SS_CONNECTED; 889 return 0; 890 } 891 892 /** 893 * set_orig_addr - capture sender's address for received message 894 * @m: descriptor for message info 895 * @msg: received message header 896 * 897 * Note: Address is not captured if not requested by receiver. 898 */ 899 static void set_orig_addr(struct msghdr *m, struct tipc_msg *msg) 900 { 901 DECLARE_SOCKADDR(struct sockaddr_tipc *, addr, m->msg_name); 902 903 if (addr) { 904 addr->family = AF_TIPC; 905 addr->addrtype = TIPC_ADDR_ID; 906 memset(&addr->addr, 0, sizeof(addr->addr)); 907 addr->addr.id.ref = msg_origport(msg); 908 addr->addr.id.node = msg_orignode(msg); 909 addr->addr.name.domain = 0; /* could leave uninitialized */ 910 addr->scope = 0; /* could leave uninitialized */ 911 m->msg_namelen = sizeof(struct sockaddr_tipc); 912 } 913 } 914 915 /** 916 * anc_data_recv - optionally capture ancillary data for received message 917 * @m: descriptor for message info 918 * @msg: received message header 919 * @tport: TIPC port associated with message 920 * 921 * Note: Ancillary data is not captured if not requested by receiver. 922 * 923 * Returns 0 if successful, otherwise errno 924 */ 925 static int anc_data_recv(struct msghdr *m, struct tipc_msg *msg, 926 struct tipc_port *tport) 927 { 928 u32 anc_data[3]; 929 u32 err; 930 u32 dest_type; 931 int has_name; 932 int res; 933 934 if (likely(m->msg_controllen == 0)) 935 return 0; 936 937 /* Optionally capture errored message object(s) */ 938 err = msg ? msg_errcode(msg) : 0; 939 if (unlikely(err)) { 940 anc_data[0] = err; 941 anc_data[1] = msg_data_sz(msg); 942 res = put_cmsg(m, SOL_TIPC, TIPC_ERRINFO, 8, anc_data); 943 if (res) 944 return res; 945 if (anc_data[1]) { 946 res = put_cmsg(m, SOL_TIPC, TIPC_RETDATA, anc_data[1], 947 msg_data(msg)); 948 if (res) 949 return res; 950 } 951 } 952 953 /* Optionally capture message destination object */ 954 dest_type = msg ? msg_type(msg) : TIPC_DIRECT_MSG; 955 switch (dest_type) { 956 case TIPC_NAMED_MSG: 957 has_name = 1; 958 anc_data[0] = msg_nametype(msg); 959 anc_data[1] = msg_namelower(msg); 960 anc_data[2] = msg_namelower(msg); 961 break; 962 case TIPC_MCAST_MSG: 963 has_name = 1; 964 anc_data[0] = msg_nametype(msg); 965 anc_data[1] = msg_namelower(msg); 966 anc_data[2] = msg_nameupper(msg); 967 break; 968 case TIPC_CONN_MSG: 969 has_name = (tport->conn_type != 0); 970 anc_data[0] = tport->conn_type; 971 anc_data[1] = tport->conn_instance; 972 anc_data[2] = tport->conn_instance; 973 break; 974 default: 975 has_name = 0; 976 } 977 if (has_name) { 978 res = put_cmsg(m, SOL_TIPC, TIPC_DESTNAME, 12, anc_data); 979 if (res) 980 return res; 981 } 982 983 return 0; 984 } 985 986 static int tipc_wait_for_rcvmsg(struct socket *sock, long timeo) 987 { 988 struct sock *sk = sock->sk; 989 DEFINE_WAIT(wait); 990 int err; 991 992 for (;;) { 993 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 994 if (timeo && skb_queue_empty(&sk->sk_receive_queue)) { 995 if (sock->state == SS_DISCONNECTING) { 996 err = -ENOTCONN; 997 break; 998 } 999 release_sock(sk); 1000 timeo = schedule_timeout(timeo); 1001 lock_sock(sk); 1002 } 1003 err = 0; 1004 if (!skb_queue_empty(&sk->sk_receive_queue)) 1005 break; 1006 err = sock_intr_errno(timeo); 1007 if (signal_pending(current)) 1008 break; 1009 err = -EAGAIN; 1010 if (!timeo) 1011 break; 1012 } 1013 finish_wait(sk_sleep(sk), &wait); 1014 return err; 1015 } 1016 1017 /** 1018 * tipc_recvmsg - receive packet-oriented message 1019 * @iocb: (unused) 1020 * @m: descriptor for message info 1021 * @buf_len: total size of user buffer area 1022 * @flags: receive flags 1023 * 1024 * Used for SOCK_DGRAM, SOCK_RDM, and SOCK_SEQPACKET messages. 1025 * If the complete message doesn't fit in user area, truncate it. 1026 * 1027 * Returns size of returned message data, errno otherwise 1028 */ 1029 static int tipc_recvmsg(struct kiocb *iocb, struct socket *sock, 1030 struct msghdr *m, size_t buf_len, int flags) 1031 { 1032 struct sock *sk = sock->sk; 1033 struct tipc_sock *tsk = tipc_sk(sk); 1034 struct tipc_port *port = &tsk->port; 1035 struct sk_buff *buf; 1036 struct tipc_msg *msg; 1037 long timeo; 1038 unsigned int sz; 1039 u32 err; 1040 int res; 1041 1042 /* Catch invalid receive requests */ 1043 if (unlikely(!buf_len)) 1044 return -EINVAL; 1045 1046 lock_sock(sk); 1047 1048 if (unlikely(sock->state == SS_UNCONNECTED)) { 1049 res = -ENOTCONN; 1050 goto exit; 1051 } 1052 1053 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1054 restart: 1055 1056 /* Look for a message in receive queue; wait if necessary */ 1057 res = tipc_wait_for_rcvmsg(sock, timeo); 1058 if (res) 1059 goto exit; 1060 1061 /* Look at first message in receive queue */ 1062 buf = skb_peek(&sk->sk_receive_queue); 1063 msg = buf_msg(buf); 1064 sz = msg_data_sz(msg); 1065 err = msg_errcode(msg); 1066 1067 /* Discard an empty non-errored message & try again */ 1068 if ((!sz) && (!err)) { 1069 advance_rx_queue(sk); 1070 goto restart; 1071 } 1072 1073 /* Capture sender's address (optional) */ 1074 set_orig_addr(m, msg); 1075 1076 /* Capture ancillary data (optional) */ 1077 res = anc_data_recv(m, msg, port); 1078 if (res) 1079 goto exit; 1080 1081 /* Capture message data (if valid) & compute return value (always) */ 1082 if (!err) { 1083 if (unlikely(buf_len < sz)) { 1084 sz = buf_len; 1085 m->msg_flags |= MSG_TRUNC; 1086 } 1087 res = skb_copy_datagram_iovec(buf, msg_hdr_sz(msg), 1088 m->msg_iov, sz); 1089 if (res) 1090 goto exit; 1091 res = sz; 1092 } else { 1093 if ((sock->state == SS_READY) || 1094 ((err == TIPC_CONN_SHUTDOWN) || m->msg_control)) 1095 res = 0; 1096 else 1097 res = -ECONNRESET; 1098 } 1099 1100 /* Consume received message (optional) */ 1101 if (likely(!(flags & MSG_PEEK))) { 1102 if ((sock->state != SS_READY) && 1103 (++port->conn_unacked >= TIPC_FLOW_CONTROL_WIN)) 1104 tipc_acknowledge(port->ref, port->conn_unacked); 1105 advance_rx_queue(sk); 1106 } 1107 exit: 1108 release_sock(sk); 1109 return res; 1110 } 1111 1112 /** 1113 * tipc_recv_stream - receive stream-oriented data 1114 * @iocb: (unused) 1115 * @m: descriptor for message info 1116 * @buf_len: total size of user buffer area 1117 * @flags: receive flags 1118 * 1119 * Used for SOCK_STREAM messages only. If not enough data is available 1120 * will optionally wait for more; never truncates data. 1121 * 1122 * Returns size of returned message data, errno otherwise 1123 */ 1124 static int tipc_recv_stream(struct kiocb *iocb, struct socket *sock, 1125 struct msghdr *m, size_t buf_len, int flags) 1126 { 1127 struct sock *sk = sock->sk; 1128 struct tipc_sock *tsk = tipc_sk(sk); 1129 struct tipc_port *port = &tsk->port; 1130 struct sk_buff *buf; 1131 struct tipc_msg *msg; 1132 long timeo; 1133 unsigned int sz; 1134 int sz_to_copy, target, needed; 1135 int sz_copied = 0; 1136 u32 err; 1137 int res = 0; 1138 1139 /* Catch invalid receive attempts */ 1140 if (unlikely(!buf_len)) 1141 return -EINVAL; 1142 1143 lock_sock(sk); 1144 1145 if (unlikely(sock->state == SS_UNCONNECTED)) { 1146 res = -ENOTCONN; 1147 goto exit; 1148 } 1149 1150 target = sock_rcvlowat(sk, flags & MSG_WAITALL, buf_len); 1151 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1152 1153 restart: 1154 /* Look for a message in receive queue; wait if necessary */ 1155 res = tipc_wait_for_rcvmsg(sock, timeo); 1156 if (res) 1157 goto exit; 1158 1159 /* Look at first message in receive queue */ 1160 buf = skb_peek(&sk->sk_receive_queue); 1161 msg = buf_msg(buf); 1162 sz = msg_data_sz(msg); 1163 err = msg_errcode(msg); 1164 1165 /* Discard an empty non-errored message & try again */ 1166 if ((!sz) && (!err)) { 1167 advance_rx_queue(sk); 1168 goto restart; 1169 } 1170 1171 /* Optionally capture sender's address & ancillary data of first msg */ 1172 if (sz_copied == 0) { 1173 set_orig_addr(m, msg); 1174 res = anc_data_recv(m, msg, port); 1175 if (res) 1176 goto exit; 1177 } 1178 1179 /* Capture message data (if valid) & compute return value (always) */ 1180 if (!err) { 1181 u32 offset = (u32)(unsigned long)(TIPC_SKB_CB(buf)->handle); 1182 1183 sz -= offset; 1184 needed = (buf_len - sz_copied); 1185 sz_to_copy = (sz <= needed) ? sz : needed; 1186 1187 res = skb_copy_datagram_iovec(buf, msg_hdr_sz(msg) + offset, 1188 m->msg_iov, sz_to_copy); 1189 if (res) 1190 goto exit; 1191 1192 sz_copied += sz_to_copy; 1193 1194 if (sz_to_copy < sz) { 1195 if (!(flags & MSG_PEEK)) 1196 TIPC_SKB_CB(buf)->handle = 1197 (void *)(unsigned long)(offset + sz_to_copy); 1198 goto exit; 1199 } 1200 } else { 1201 if (sz_copied != 0) 1202 goto exit; /* can't add error msg to valid data */ 1203 1204 if ((err == TIPC_CONN_SHUTDOWN) || m->msg_control) 1205 res = 0; 1206 else 1207 res = -ECONNRESET; 1208 } 1209 1210 /* Consume received message (optional) */ 1211 if (likely(!(flags & MSG_PEEK))) { 1212 if (unlikely(++port->conn_unacked >= TIPC_FLOW_CONTROL_WIN)) 1213 tipc_acknowledge(port->ref, port->conn_unacked); 1214 advance_rx_queue(sk); 1215 } 1216 1217 /* Loop around if more data is required */ 1218 if ((sz_copied < buf_len) && /* didn't get all requested data */ 1219 (!skb_queue_empty(&sk->sk_receive_queue) || 1220 (sz_copied < target)) && /* and more is ready or required */ 1221 (!(flags & MSG_PEEK)) && /* and aren't just peeking at data */ 1222 (!err)) /* and haven't reached a FIN */ 1223 goto restart; 1224 1225 exit: 1226 release_sock(sk); 1227 return sz_copied ? sz_copied : res; 1228 } 1229 1230 /** 1231 * tipc_write_space - wake up thread if port congestion is released 1232 * @sk: socket 1233 */ 1234 static void tipc_write_space(struct sock *sk) 1235 { 1236 struct socket_wq *wq; 1237 1238 rcu_read_lock(); 1239 wq = rcu_dereference(sk->sk_wq); 1240 if (wq_has_sleeper(wq)) 1241 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 1242 POLLWRNORM | POLLWRBAND); 1243 rcu_read_unlock(); 1244 } 1245 1246 /** 1247 * tipc_data_ready - wake up threads to indicate messages have been received 1248 * @sk: socket 1249 * @len: the length of messages 1250 */ 1251 static void tipc_data_ready(struct sock *sk) 1252 { 1253 struct socket_wq *wq; 1254 1255 rcu_read_lock(); 1256 wq = rcu_dereference(sk->sk_wq); 1257 if (wq_has_sleeper(wq)) 1258 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | 1259 POLLRDNORM | POLLRDBAND); 1260 rcu_read_unlock(); 1261 } 1262 1263 /** 1264 * filter_connect - Handle all incoming messages for a connection-based socket 1265 * @tsk: TIPC socket 1266 * @msg: message 1267 * 1268 * Returns TIPC error status code and socket error status code 1269 * once it encounters some errors 1270 */ 1271 static u32 filter_connect(struct tipc_sock *tsk, struct sk_buff **buf) 1272 { 1273 struct sock *sk = &tsk->sk; 1274 struct tipc_port *port = &tsk->port; 1275 struct socket *sock = sk->sk_socket; 1276 struct tipc_msg *msg = buf_msg(*buf); 1277 1278 u32 retval = TIPC_ERR_NO_PORT; 1279 int res; 1280 1281 if (msg_mcast(msg)) 1282 return retval; 1283 1284 switch ((int)sock->state) { 1285 case SS_CONNECTED: 1286 /* Accept only connection-based messages sent by peer */ 1287 if (msg_connected(msg) && tipc_port_peer_msg(port, msg)) { 1288 if (unlikely(msg_errcode(msg))) { 1289 sock->state = SS_DISCONNECTING; 1290 __tipc_port_disconnect(port); 1291 } 1292 retval = TIPC_OK; 1293 } 1294 break; 1295 case SS_CONNECTING: 1296 /* Accept only ACK or NACK message */ 1297 if (unlikely(msg_errcode(msg))) { 1298 sock->state = SS_DISCONNECTING; 1299 sk->sk_err = ECONNREFUSED; 1300 retval = TIPC_OK; 1301 break; 1302 } 1303 1304 if (unlikely(!msg_connected(msg))) 1305 break; 1306 1307 res = auto_connect(tsk, msg); 1308 if (res) { 1309 sock->state = SS_DISCONNECTING; 1310 sk->sk_err = -res; 1311 retval = TIPC_OK; 1312 break; 1313 } 1314 1315 /* If an incoming message is an 'ACK-', it should be 1316 * discarded here because it doesn't contain useful 1317 * data. In addition, we should try to wake up 1318 * connect() routine if sleeping. 1319 */ 1320 if (msg_data_sz(msg) == 0) { 1321 kfree_skb(*buf); 1322 *buf = NULL; 1323 if (waitqueue_active(sk_sleep(sk))) 1324 wake_up_interruptible(sk_sleep(sk)); 1325 } 1326 retval = TIPC_OK; 1327 break; 1328 case SS_LISTENING: 1329 case SS_UNCONNECTED: 1330 /* Accept only SYN message */ 1331 if (!msg_connected(msg) && !(msg_errcode(msg))) 1332 retval = TIPC_OK; 1333 break; 1334 case SS_DISCONNECTING: 1335 break; 1336 default: 1337 pr_err("Unknown socket state %u\n", sock->state); 1338 } 1339 return retval; 1340 } 1341 1342 /** 1343 * rcvbuf_limit - get proper overload limit of socket receive queue 1344 * @sk: socket 1345 * @buf: message 1346 * 1347 * For all connection oriented messages, irrespective of importance, 1348 * the default overload value (i.e. 67MB) is set as limit. 1349 * 1350 * For all connectionless messages, by default new queue limits are 1351 * as belows: 1352 * 1353 * TIPC_LOW_IMPORTANCE (4 MB) 1354 * TIPC_MEDIUM_IMPORTANCE (8 MB) 1355 * TIPC_HIGH_IMPORTANCE (16 MB) 1356 * TIPC_CRITICAL_IMPORTANCE (32 MB) 1357 * 1358 * Returns overload limit according to corresponding message importance 1359 */ 1360 static unsigned int rcvbuf_limit(struct sock *sk, struct sk_buff *buf) 1361 { 1362 struct tipc_msg *msg = buf_msg(buf); 1363 1364 if (msg_connected(msg)) 1365 return sysctl_tipc_rmem[2]; 1366 1367 return sk->sk_rcvbuf >> TIPC_CRITICAL_IMPORTANCE << 1368 msg_importance(msg); 1369 } 1370 1371 /** 1372 * filter_rcv - validate incoming message 1373 * @sk: socket 1374 * @buf: message 1375 * 1376 * Enqueues message on receive queue if acceptable; optionally handles 1377 * disconnect indication for a connected socket. 1378 * 1379 * Called with socket lock already taken; port lock may also be taken. 1380 * 1381 * Returns TIPC error status code (TIPC_OK if message is not to be rejected) 1382 */ 1383 static u32 filter_rcv(struct sock *sk, struct sk_buff *buf) 1384 { 1385 struct socket *sock = sk->sk_socket; 1386 struct tipc_sock *tsk = tipc_sk(sk); 1387 struct tipc_msg *msg = buf_msg(buf); 1388 unsigned int limit = rcvbuf_limit(sk, buf); 1389 u32 res = TIPC_OK; 1390 1391 /* Reject message if it is wrong sort of message for socket */ 1392 if (msg_type(msg) > TIPC_DIRECT_MSG) 1393 return TIPC_ERR_NO_PORT; 1394 1395 if (sock->state == SS_READY) { 1396 if (msg_connected(msg)) 1397 return TIPC_ERR_NO_PORT; 1398 } else { 1399 res = filter_connect(tsk, &buf); 1400 if (res != TIPC_OK || buf == NULL) 1401 return res; 1402 } 1403 1404 /* Reject message if there isn't room to queue it */ 1405 if (sk_rmem_alloc_get(sk) + buf->truesize >= limit) 1406 return TIPC_ERR_OVERLOAD; 1407 1408 /* Enqueue message */ 1409 TIPC_SKB_CB(buf)->handle = NULL; 1410 __skb_queue_tail(&sk->sk_receive_queue, buf); 1411 skb_set_owner_r(buf, sk); 1412 1413 sk->sk_data_ready(sk); 1414 return TIPC_OK; 1415 } 1416 1417 /** 1418 * backlog_rcv - handle incoming message from backlog queue 1419 * @sk: socket 1420 * @buf: message 1421 * 1422 * Caller must hold socket lock, but not port lock. 1423 * 1424 * Returns 0 1425 */ 1426 static int backlog_rcv(struct sock *sk, struct sk_buff *buf) 1427 { 1428 u32 res; 1429 1430 res = filter_rcv(sk, buf); 1431 if (res) 1432 tipc_reject_msg(buf, res); 1433 return 0; 1434 } 1435 1436 /** 1437 * tipc_sk_rcv - handle incoming message 1438 * @sk: socket receiving message 1439 * @buf: message 1440 * 1441 * Called with port lock already taken. 1442 * 1443 * Returns TIPC error status code (TIPC_OK if message is not to be rejected) 1444 */ 1445 u32 tipc_sk_rcv(struct sock *sk, struct sk_buff *buf) 1446 { 1447 u32 res; 1448 1449 /* 1450 * Process message if socket is unlocked; otherwise add to backlog queue 1451 * 1452 * This code is based on sk_receive_skb(), but must be distinct from it 1453 * since a TIPC-specific filter/reject mechanism is utilized 1454 */ 1455 bh_lock_sock(sk); 1456 if (!sock_owned_by_user(sk)) { 1457 res = filter_rcv(sk, buf); 1458 } else { 1459 if (sk_add_backlog(sk, buf, rcvbuf_limit(sk, buf))) 1460 res = TIPC_ERR_OVERLOAD; 1461 else 1462 res = TIPC_OK; 1463 } 1464 bh_unlock_sock(sk); 1465 1466 return res; 1467 } 1468 1469 static int tipc_wait_for_connect(struct socket *sock, long *timeo_p) 1470 { 1471 struct sock *sk = sock->sk; 1472 DEFINE_WAIT(wait); 1473 int done; 1474 1475 do { 1476 int err = sock_error(sk); 1477 if (err) 1478 return err; 1479 if (!*timeo_p) 1480 return -ETIMEDOUT; 1481 if (signal_pending(current)) 1482 return sock_intr_errno(*timeo_p); 1483 1484 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1485 done = sk_wait_event(sk, timeo_p, sock->state != SS_CONNECTING); 1486 finish_wait(sk_sleep(sk), &wait); 1487 } while (!done); 1488 return 0; 1489 } 1490 1491 /** 1492 * tipc_connect - establish a connection to another TIPC port 1493 * @sock: socket structure 1494 * @dest: socket address for destination port 1495 * @destlen: size of socket address data structure 1496 * @flags: file-related flags associated with socket 1497 * 1498 * Returns 0 on success, errno otherwise 1499 */ 1500 static int tipc_connect(struct socket *sock, struct sockaddr *dest, 1501 int destlen, int flags) 1502 { 1503 struct sock *sk = sock->sk; 1504 struct sockaddr_tipc *dst = (struct sockaddr_tipc *)dest; 1505 struct msghdr m = {NULL,}; 1506 long timeout = (flags & O_NONBLOCK) ? 0 : tipc_sk(sk)->conn_timeout; 1507 socket_state previous; 1508 int res; 1509 1510 lock_sock(sk); 1511 1512 /* For now, TIPC does not allow use of connect() with DGRAM/RDM types */ 1513 if (sock->state == SS_READY) { 1514 res = -EOPNOTSUPP; 1515 goto exit; 1516 } 1517 1518 /* 1519 * Reject connection attempt using multicast address 1520 * 1521 * Note: send_msg() validates the rest of the address fields, 1522 * so there's no need to do it here 1523 */ 1524 if (dst->addrtype == TIPC_ADDR_MCAST) { 1525 res = -EINVAL; 1526 goto exit; 1527 } 1528 1529 previous = sock->state; 1530 switch (sock->state) { 1531 case SS_UNCONNECTED: 1532 /* Send a 'SYN-' to destination */ 1533 m.msg_name = dest; 1534 m.msg_namelen = destlen; 1535 1536 /* If connect is in non-blocking case, set MSG_DONTWAIT to 1537 * indicate send_msg() is never blocked. 1538 */ 1539 if (!timeout) 1540 m.msg_flags = MSG_DONTWAIT; 1541 1542 res = tipc_sendmsg(NULL, sock, &m, 0); 1543 if ((res < 0) && (res != -EWOULDBLOCK)) 1544 goto exit; 1545 1546 /* Just entered SS_CONNECTING state; the only 1547 * difference is that return value in non-blocking 1548 * case is EINPROGRESS, rather than EALREADY. 1549 */ 1550 res = -EINPROGRESS; 1551 case SS_CONNECTING: 1552 if (previous == SS_CONNECTING) 1553 res = -EALREADY; 1554 if (!timeout) 1555 goto exit; 1556 timeout = msecs_to_jiffies(timeout); 1557 /* Wait until an 'ACK' or 'RST' arrives, or a timeout occurs */ 1558 res = tipc_wait_for_connect(sock, &timeout); 1559 break; 1560 case SS_CONNECTED: 1561 res = -EISCONN; 1562 break; 1563 default: 1564 res = -EINVAL; 1565 break; 1566 } 1567 exit: 1568 release_sock(sk); 1569 return res; 1570 } 1571 1572 /** 1573 * tipc_listen - allow socket to listen for incoming connections 1574 * @sock: socket structure 1575 * @len: (unused) 1576 * 1577 * Returns 0 on success, errno otherwise 1578 */ 1579 static int tipc_listen(struct socket *sock, int len) 1580 { 1581 struct sock *sk = sock->sk; 1582 int res; 1583 1584 lock_sock(sk); 1585 1586 if (sock->state != SS_UNCONNECTED) 1587 res = -EINVAL; 1588 else { 1589 sock->state = SS_LISTENING; 1590 res = 0; 1591 } 1592 1593 release_sock(sk); 1594 return res; 1595 } 1596 1597 static int tipc_wait_for_accept(struct socket *sock, long timeo) 1598 { 1599 struct sock *sk = sock->sk; 1600 DEFINE_WAIT(wait); 1601 int err; 1602 1603 /* True wake-one mechanism for incoming connections: only 1604 * one process gets woken up, not the 'whole herd'. 1605 * Since we do not 'race & poll' for established sockets 1606 * anymore, the common case will execute the loop only once. 1607 */ 1608 for (;;) { 1609 prepare_to_wait_exclusive(sk_sleep(sk), &wait, 1610 TASK_INTERRUPTIBLE); 1611 if (timeo && skb_queue_empty(&sk->sk_receive_queue)) { 1612 release_sock(sk); 1613 timeo = schedule_timeout(timeo); 1614 lock_sock(sk); 1615 } 1616 err = 0; 1617 if (!skb_queue_empty(&sk->sk_receive_queue)) 1618 break; 1619 err = -EINVAL; 1620 if (sock->state != SS_LISTENING) 1621 break; 1622 err = sock_intr_errno(timeo); 1623 if (signal_pending(current)) 1624 break; 1625 err = -EAGAIN; 1626 if (!timeo) 1627 break; 1628 } 1629 finish_wait(sk_sleep(sk), &wait); 1630 return err; 1631 } 1632 1633 /** 1634 * tipc_accept - wait for connection request 1635 * @sock: listening socket 1636 * @newsock: new socket that is to be connected 1637 * @flags: file-related flags associated with socket 1638 * 1639 * Returns 0 on success, errno otherwise 1640 */ 1641 static int tipc_accept(struct socket *sock, struct socket *new_sock, int flags) 1642 { 1643 struct sock *new_sk, *sk = sock->sk; 1644 struct sk_buff *buf; 1645 struct tipc_port *new_port; 1646 struct tipc_msg *msg; 1647 struct tipc_portid peer; 1648 u32 new_ref; 1649 long timeo; 1650 int res; 1651 1652 lock_sock(sk); 1653 1654 if (sock->state != SS_LISTENING) { 1655 res = -EINVAL; 1656 goto exit; 1657 } 1658 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1659 res = tipc_wait_for_accept(sock, timeo); 1660 if (res) 1661 goto exit; 1662 1663 buf = skb_peek(&sk->sk_receive_queue); 1664 1665 res = tipc_sk_create(sock_net(sock->sk), new_sock, 0, 1); 1666 if (res) 1667 goto exit; 1668 1669 new_sk = new_sock->sk; 1670 new_port = &tipc_sk(new_sk)->port; 1671 new_ref = new_port->ref; 1672 msg = buf_msg(buf); 1673 1674 /* we lock on new_sk; but lockdep sees the lock on sk */ 1675 lock_sock_nested(new_sk, SINGLE_DEPTH_NESTING); 1676 1677 /* 1678 * Reject any stray messages received by new socket 1679 * before the socket lock was taken (very, very unlikely) 1680 */ 1681 reject_rx_queue(new_sk); 1682 1683 /* Connect new socket to it's peer */ 1684 peer.ref = msg_origport(msg); 1685 peer.node = msg_orignode(msg); 1686 tipc_port_connect(new_ref, &peer); 1687 new_sock->state = SS_CONNECTED; 1688 1689 tipc_port_set_importance(new_port, msg_importance(msg)); 1690 if (msg_named(msg)) { 1691 new_port->conn_type = msg_nametype(msg); 1692 new_port->conn_instance = msg_nameinst(msg); 1693 } 1694 1695 /* 1696 * Respond to 'SYN-' by discarding it & returning 'ACK'-. 1697 * Respond to 'SYN+' by queuing it on new socket. 1698 */ 1699 if (!msg_data_sz(msg)) { 1700 struct msghdr m = {NULL,}; 1701 1702 advance_rx_queue(sk); 1703 tipc_send_packet(NULL, new_sock, &m, 0); 1704 } else { 1705 __skb_dequeue(&sk->sk_receive_queue); 1706 __skb_queue_head(&new_sk->sk_receive_queue, buf); 1707 skb_set_owner_r(buf, new_sk); 1708 } 1709 release_sock(new_sk); 1710 exit: 1711 release_sock(sk); 1712 return res; 1713 } 1714 1715 /** 1716 * tipc_shutdown - shutdown socket connection 1717 * @sock: socket structure 1718 * @how: direction to close (must be SHUT_RDWR) 1719 * 1720 * Terminates connection (if necessary), then purges socket's receive queue. 1721 * 1722 * Returns 0 on success, errno otherwise 1723 */ 1724 static int tipc_shutdown(struct socket *sock, int how) 1725 { 1726 struct sock *sk = sock->sk; 1727 struct tipc_sock *tsk = tipc_sk(sk); 1728 struct tipc_port *port = &tsk->port; 1729 struct sk_buff *buf; 1730 int res; 1731 1732 if (how != SHUT_RDWR) 1733 return -EINVAL; 1734 1735 lock_sock(sk); 1736 1737 switch (sock->state) { 1738 case SS_CONNECTING: 1739 case SS_CONNECTED: 1740 1741 restart: 1742 /* Disconnect and send a 'FIN+' or 'FIN-' message to peer */ 1743 buf = __skb_dequeue(&sk->sk_receive_queue); 1744 if (buf) { 1745 if (TIPC_SKB_CB(buf)->handle != NULL) { 1746 kfree_skb(buf); 1747 goto restart; 1748 } 1749 tipc_port_disconnect(port->ref); 1750 tipc_reject_msg(buf, TIPC_CONN_SHUTDOWN); 1751 } else { 1752 tipc_port_shutdown(port->ref); 1753 } 1754 1755 sock->state = SS_DISCONNECTING; 1756 1757 /* fall through */ 1758 1759 case SS_DISCONNECTING: 1760 1761 /* Discard any unreceived messages */ 1762 __skb_queue_purge(&sk->sk_receive_queue); 1763 1764 /* Wake up anyone sleeping in poll */ 1765 sk->sk_state_change(sk); 1766 res = 0; 1767 break; 1768 1769 default: 1770 res = -ENOTCONN; 1771 } 1772 1773 release_sock(sk); 1774 return res; 1775 } 1776 1777 /** 1778 * tipc_setsockopt - set socket option 1779 * @sock: socket structure 1780 * @lvl: option level 1781 * @opt: option identifier 1782 * @ov: pointer to new option value 1783 * @ol: length of option value 1784 * 1785 * For stream sockets only, accepts and ignores all IPPROTO_TCP options 1786 * (to ease compatibility). 1787 * 1788 * Returns 0 on success, errno otherwise 1789 */ 1790 static int tipc_setsockopt(struct socket *sock, int lvl, int opt, 1791 char __user *ov, unsigned int ol) 1792 { 1793 struct sock *sk = sock->sk; 1794 struct tipc_sock *tsk = tipc_sk(sk); 1795 struct tipc_port *port = &tsk->port; 1796 u32 value; 1797 int res; 1798 1799 if ((lvl == IPPROTO_TCP) && (sock->type == SOCK_STREAM)) 1800 return 0; 1801 if (lvl != SOL_TIPC) 1802 return -ENOPROTOOPT; 1803 if (ol < sizeof(value)) 1804 return -EINVAL; 1805 res = get_user(value, (u32 __user *)ov); 1806 if (res) 1807 return res; 1808 1809 lock_sock(sk); 1810 1811 switch (opt) { 1812 case TIPC_IMPORTANCE: 1813 tipc_port_set_importance(port, value); 1814 break; 1815 case TIPC_SRC_DROPPABLE: 1816 if (sock->type != SOCK_STREAM) 1817 tipc_port_set_unreliable(port, value); 1818 else 1819 res = -ENOPROTOOPT; 1820 break; 1821 case TIPC_DEST_DROPPABLE: 1822 tipc_port_set_unreturnable(port, value); 1823 break; 1824 case TIPC_CONN_TIMEOUT: 1825 tipc_sk(sk)->conn_timeout = value; 1826 /* no need to set "res", since already 0 at this point */ 1827 break; 1828 default: 1829 res = -EINVAL; 1830 } 1831 1832 release_sock(sk); 1833 1834 return res; 1835 } 1836 1837 /** 1838 * tipc_getsockopt - get socket option 1839 * @sock: socket structure 1840 * @lvl: option level 1841 * @opt: option identifier 1842 * @ov: receptacle for option value 1843 * @ol: receptacle for length of option value 1844 * 1845 * For stream sockets only, returns 0 length result for all IPPROTO_TCP options 1846 * (to ease compatibility). 1847 * 1848 * Returns 0 on success, errno otherwise 1849 */ 1850 static int tipc_getsockopt(struct socket *sock, int lvl, int opt, 1851 char __user *ov, int __user *ol) 1852 { 1853 struct sock *sk = sock->sk; 1854 struct tipc_sock *tsk = tipc_sk(sk); 1855 struct tipc_port *port = &tsk->port; 1856 int len; 1857 u32 value; 1858 int res; 1859 1860 if ((lvl == IPPROTO_TCP) && (sock->type == SOCK_STREAM)) 1861 return put_user(0, ol); 1862 if (lvl != SOL_TIPC) 1863 return -ENOPROTOOPT; 1864 res = get_user(len, ol); 1865 if (res) 1866 return res; 1867 1868 lock_sock(sk); 1869 1870 switch (opt) { 1871 case TIPC_IMPORTANCE: 1872 value = tipc_port_importance(port); 1873 break; 1874 case TIPC_SRC_DROPPABLE: 1875 value = tipc_port_unreliable(port); 1876 break; 1877 case TIPC_DEST_DROPPABLE: 1878 value = tipc_port_unreturnable(port); 1879 break; 1880 case TIPC_CONN_TIMEOUT: 1881 value = tipc_sk(sk)->conn_timeout; 1882 /* no need to set "res", since already 0 at this point */ 1883 break; 1884 case TIPC_NODE_RECVQ_DEPTH: 1885 value = 0; /* was tipc_queue_size, now obsolete */ 1886 break; 1887 case TIPC_SOCK_RECVQ_DEPTH: 1888 value = skb_queue_len(&sk->sk_receive_queue); 1889 break; 1890 default: 1891 res = -EINVAL; 1892 } 1893 1894 release_sock(sk); 1895 1896 if (res) 1897 return res; /* "get" failed */ 1898 1899 if (len < sizeof(value)) 1900 return -EINVAL; 1901 1902 if (copy_to_user(ov, &value, sizeof(value))) 1903 return -EFAULT; 1904 1905 return put_user(sizeof(value), ol); 1906 } 1907 1908 /* Protocol switches for the various types of TIPC sockets */ 1909 1910 static const struct proto_ops msg_ops = { 1911 .owner = THIS_MODULE, 1912 .family = AF_TIPC, 1913 .release = tipc_release, 1914 .bind = tipc_bind, 1915 .connect = tipc_connect, 1916 .socketpair = sock_no_socketpair, 1917 .accept = sock_no_accept, 1918 .getname = tipc_getname, 1919 .poll = tipc_poll, 1920 .ioctl = sock_no_ioctl, 1921 .listen = sock_no_listen, 1922 .shutdown = tipc_shutdown, 1923 .setsockopt = tipc_setsockopt, 1924 .getsockopt = tipc_getsockopt, 1925 .sendmsg = tipc_sendmsg, 1926 .recvmsg = tipc_recvmsg, 1927 .mmap = sock_no_mmap, 1928 .sendpage = sock_no_sendpage 1929 }; 1930 1931 static const struct proto_ops packet_ops = { 1932 .owner = THIS_MODULE, 1933 .family = AF_TIPC, 1934 .release = tipc_release, 1935 .bind = tipc_bind, 1936 .connect = tipc_connect, 1937 .socketpair = sock_no_socketpair, 1938 .accept = tipc_accept, 1939 .getname = tipc_getname, 1940 .poll = tipc_poll, 1941 .ioctl = sock_no_ioctl, 1942 .listen = tipc_listen, 1943 .shutdown = tipc_shutdown, 1944 .setsockopt = tipc_setsockopt, 1945 .getsockopt = tipc_getsockopt, 1946 .sendmsg = tipc_send_packet, 1947 .recvmsg = tipc_recvmsg, 1948 .mmap = sock_no_mmap, 1949 .sendpage = sock_no_sendpage 1950 }; 1951 1952 static const struct proto_ops stream_ops = { 1953 .owner = THIS_MODULE, 1954 .family = AF_TIPC, 1955 .release = tipc_release, 1956 .bind = tipc_bind, 1957 .connect = tipc_connect, 1958 .socketpair = sock_no_socketpair, 1959 .accept = tipc_accept, 1960 .getname = tipc_getname, 1961 .poll = tipc_poll, 1962 .ioctl = sock_no_ioctl, 1963 .listen = tipc_listen, 1964 .shutdown = tipc_shutdown, 1965 .setsockopt = tipc_setsockopt, 1966 .getsockopt = tipc_getsockopt, 1967 .sendmsg = tipc_send_stream, 1968 .recvmsg = tipc_recv_stream, 1969 .mmap = sock_no_mmap, 1970 .sendpage = sock_no_sendpage 1971 }; 1972 1973 static const struct net_proto_family tipc_family_ops = { 1974 .owner = THIS_MODULE, 1975 .family = AF_TIPC, 1976 .create = tipc_sk_create 1977 }; 1978 1979 static struct proto tipc_proto = { 1980 .name = "TIPC", 1981 .owner = THIS_MODULE, 1982 .obj_size = sizeof(struct tipc_sock), 1983 .sysctl_rmem = sysctl_tipc_rmem 1984 }; 1985 1986 static struct proto tipc_proto_kern = { 1987 .name = "TIPC", 1988 .obj_size = sizeof(struct tipc_sock), 1989 .sysctl_rmem = sysctl_tipc_rmem 1990 }; 1991 1992 /** 1993 * tipc_socket_init - initialize TIPC socket interface 1994 * 1995 * Returns 0 on success, errno otherwise 1996 */ 1997 int tipc_socket_init(void) 1998 { 1999 int res; 2000 2001 res = proto_register(&tipc_proto, 1); 2002 if (res) { 2003 pr_err("Failed to register TIPC protocol type\n"); 2004 goto out; 2005 } 2006 2007 res = sock_register(&tipc_family_ops); 2008 if (res) { 2009 pr_err("Failed to register TIPC socket type\n"); 2010 proto_unregister(&tipc_proto); 2011 goto out; 2012 } 2013 out: 2014 return res; 2015 } 2016 2017 /** 2018 * tipc_socket_stop - stop TIPC socket interface 2019 */ 2020 void tipc_socket_stop(void) 2021 { 2022 sock_unregister(tipc_family_ops.family); 2023 proto_unregister(&tipc_proto); 2024 } 2025