xref: /linux/net/tipc/socket.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
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