1/*	$NetBSD: clnt_vc.c,v 1.4 2000/07/14 08:40:42 fvdl Exp $	*/
2
3/*-
4 * Copyright (c) 2009, Sun Microsystems, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are met:
9 * - Redistributions of source code must retain the above copyright notice,
10 *   this list of conditions and the following disclaimer.
11 * - Redistributions in binary form must reproduce the above copyright notice,
12 *   this list of conditions and the following disclaimer in the documentation
13 *   and/or other materials provided with the distribution.
14 * - Neither the name of Sun Microsystems, Inc. nor the names of its
15 *   contributors may be used to endorse or promote products derived
16 *   from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
19 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
29 */
30
31#if defined(LIBC_SCCS) && !defined(lint)
32static char *sccsid2 = "@(#)clnt_tcp.c 1.37 87/10/05 Copyr 1984 Sun Micro";
33static char *sccsid = "@(#)clnt_tcp.c	2.2 88/08/01 4.0 RPCSRC";
34static char sccsid3[] = "@(#)clnt_vc.c 1.19 89/03/16 Copyr 1988 Sun Micro";
35#endif
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD$");
38
39/*
40 * clnt_tcp.c, Implements a TCP/IP based, client side RPC.
41 *
42 * Copyright (C) 1984, Sun Microsystems, Inc.
43 *
44 * TCP based RPC supports 'batched calls'.
45 * A sequence of calls may be batched-up in a send buffer.  The rpc call
46 * return immediately to the client even though the call was not necessarily
47 * sent.  The batching occurs if the results' xdr routine is NULL (0) AND
48 * the rpc timeout value is zero (see clnt.h, rpc).
49 *
50 * Clients should NOT casually batch calls that in fact return results; that is,
51 * the server side should be aware that a call is batched and not produce any
52 * return message.  Batched calls that produce many result messages can
53 * deadlock (netlock) the client and the server....
54 *
55 * Now go hang yourself.
56 */
57
58#include <sys/param.h>
59#include <sys/systm.h>
60#include <sys/lock.h>
61#include <sys/malloc.h>
62#include <sys/mbuf.h>
63#include <sys/mutex.h>
64#include <sys/pcpu.h>
65#include <sys/proc.h>
66#include <sys/protosw.h>
67#include <sys/socket.h>
68#include <sys/socketvar.h>
69#include <sys/sx.h>
70#include <sys/syslog.h>
71#include <sys/time.h>
72#include <sys/uio.h>
73
74#include <net/vnet.h>
75
76#include <netinet/tcp.h>
77
78#include <rpc/rpc.h>
79#include <rpc/rpc_com.h>
80#include <rpc/krpc.h>
81
82struct cmessage {
83        struct cmsghdr cmsg;
84        struct cmsgcred cmcred;
85};
86
87static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
88    rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
89static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
90static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
91static void clnt_vc_abort(CLIENT *);
92static bool_t clnt_vc_control(CLIENT *, u_int, void *);
93static void clnt_vc_close(CLIENT *);
94static void clnt_vc_destroy(CLIENT *);
95static bool_t time_not_ok(struct timeval *);
96static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
97
98static struct clnt_ops clnt_vc_ops = {
99	.cl_call =	clnt_vc_call,
100	.cl_abort =	clnt_vc_abort,
101	.cl_geterr =	clnt_vc_geterr,
102	.cl_freeres =	clnt_vc_freeres,
103	.cl_close =	clnt_vc_close,
104	.cl_destroy =	clnt_vc_destroy,
105	.cl_control =	clnt_vc_control
106};
107
108static void clnt_vc_upcallsdone(struct ct_data *);
109
110/*
111 * Create a client handle for a connection.
112 * Default options are set, which the user can change using clnt_control()'s.
113 * The rpc/vc package does buffering similar to stdio, so the client
114 * must pick send and receive buffer sizes, 0 => use the default.
115 * NB: fd is copied into a private area.
116 * NB: The rpch->cl_auth is set null authentication. Caller may wish to
117 * set this something more useful.
118 *
119 * fd should be an open socket
120 */
121CLIENT *
122clnt_vc_create(
123	struct socket *so,		/* open file descriptor */
124	struct sockaddr *raddr,		/* servers address */
125	const rpcprog_t prog,		/* program number */
126	const rpcvers_t vers,		/* version number */
127	size_t sendsz,			/* buffer recv size */
128	size_t recvsz,			/* buffer send size */
129	int intrflag)			/* interruptible */
130{
131	CLIENT *cl;			/* client handle */
132	struct ct_data *ct = NULL;	/* client handle */
133	struct timeval now;
134	struct rpc_msg call_msg;
135	static uint32_t disrupt;
136	struct __rpc_sockinfo si;
137	XDR xdrs;
138	int error, interrupted, one = 1, sleep_flag;
139	struct sockopt sopt;
140
141	if (disrupt == 0)
142		disrupt = (uint32_t)(long)raddr;
143
144	cl = (CLIENT *)mem_alloc(sizeof (*cl));
145	ct = (struct ct_data *)mem_alloc(sizeof (*ct));
146
147	mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
148	ct->ct_threads = 0;
149	ct->ct_closing = FALSE;
150	ct->ct_closed = FALSE;
151	ct->ct_upcallrefs = 0;
152
153	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
154		error = soconnect(so, raddr, curthread);
155		SOCK_LOCK(so);
156		interrupted = 0;
157		sleep_flag = PSOCK;
158		if (intrflag != 0)
159			sleep_flag |= (PCATCH | PBDRY);
160		while ((so->so_state & SS_ISCONNECTING)
161		    && so->so_error == 0) {
162			error = msleep(&so->so_timeo, SOCK_MTX(so),
163			    sleep_flag, "connec", 0);
164			if (error) {
165				if (error == EINTR || error == ERESTART)
166					interrupted = 1;
167				break;
168			}
169		}
170		if (error == 0) {
171			error = so->so_error;
172			so->so_error = 0;
173		}
174		SOCK_UNLOCK(so);
175		if (error) {
176			if (!interrupted)
177				so->so_state &= ~SS_ISCONNECTING;
178			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
179			rpc_createerr.cf_error.re_errno = error;
180			goto err;
181		}
182	}
183
184	if (!__rpc_socket2sockinfo(so, &si)) {
185		goto err;
186	}
187
188	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
189		bzero(&sopt, sizeof(sopt));
190		sopt.sopt_dir = SOPT_SET;
191		sopt.sopt_level = SOL_SOCKET;
192		sopt.sopt_name = SO_KEEPALIVE;
193		sopt.sopt_val = &one;
194		sopt.sopt_valsize = sizeof(one);
195		sosetopt(so, &sopt);
196	}
197
198	if (so->so_proto->pr_protocol == IPPROTO_TCP) {
199		bzero(&sopt, sizeof(sopt));
200		sopt.sopt_dir = SOPT_SET;
201		sopt.sopt_level = IPPROTO_TCP;
202		sopt.sopt_name = TCP_NODELAY;
203		sopt.sopt_val = &one;
204		sopt.sopt_valsize = sizeof(one);
205		sosetopt(so, &sopt);
206	}
207
208	ct->ct_closeit = FALSE;
209
210	/*
211	 * Set up private data struct
212	 */
213	ct->ct_socket = so;
214	ct->ct_wait.tv_sec = -1;
215	ct->ct_wait.tv_usec = -1;
216	memcpy(&ct->ct_addr, raddr, raddr->sa_len);
217
218	/*
219	 * Initialize call message
220	 */
221	getmicrotime(&now);
222	ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
223	call_msg.rm_xid = ct->ct_xid;
224	call_msg.rm_direction = CALL;
225	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
226	call_msg.rm_call.cb_prog = (uint32_t)prog;
227	call_msg.rm_call.cb_vers = (uint32_t)vers;
228
229	/*
230	 * pre-serialize the static part of the call msg and stash it away
231	 */
232	xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
233	    XDR_ENCODE);
234	if (! xdr_callhdr(&xdrs, &call_msg)) {
235		if (ct->ct_closeit) {
236			soclose(ct->ct_socket);
237		}
238		goto err;
239	}
240	ct->ct_mpos = XDR_GETPOS(&xdrs);
241	XDR_DESTROY(&xdrs);
242	ct->ct_waitchan = "rpcrecv";
243	ct->ct_waitflag = 0;
244
245	/*
246	 * Create a client handle which uses xdrrec for serialization
247	 * and authnone for authentication.
248	 */
249	sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
250	recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
251	error = soreserve(ct->ct_socket, sendsz, recvsz);
252	if (error != 0) {
253		if (ct->ct_closeit) {
254			soclose(ct->ct_socket);
255		}
256		goto err;
257	}
258	cl->cl_refs = 1;
259	cl->cl_ops = &clnt_vc_ops;
260	cl->cl_private = ct;
261	cl->cl_auth = authnone_create();
262
263	SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
264	soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
265	SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
266
267	ct->ct_record = NULL;
268	ct->ct_record_resid = 0;
269	TAILQ_INIT(&ct->ct_pending);
270	return (cl);
271
272err:
273	if (ct) {
274		mtx_destroy(&ct->ct_lock);
275		mem_free(ct, sizeof (struct ct_data));
276	}
277	if (cl)
278		mem_free(cl, sizeof (CLIENT));
279	return ((CLIENT *)NULL);
280}
281
282static enum clnt_stat
283clnt_vc_call(
284	CLIENT		*cl,		/* client handle */
285	struct rpc_callextra *ext,	/* call metadata */
286	rpcproc_t	proc,		/* procedure number */
287	struct mbuf	*args,		/* pointer to args */
288	struct mbuf	**resultsp,	/* pointer to results */
289	struct timeval	utimeout)
290{
291	struct ct_data *ct = (struct ct_data *) cl->cl_private;
292	AUTH *auth;
293	struct rpc_err *errp;
294	enum clnt_stat stat;
295	XDR xdrs;
296	struct rpc_msg reply_msg;
297	bool_t ok;
298	int nrefreshes = 2;		/* number of times to refresh cred */
299	struct timeval timeout;
300	uint32_t xid;
301	struct mbuf *mreq = NULL, *results;
302	struct ct_request *cr;
303	int error;
304
305	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
306
307	mtx_lock(&ct->ct_lock);
308
309	if (ct->ct_closing || ct->ct_closed) {
310		mtx_unlock(&ct->ct_lock);
311		free(cr, M_RPC);
312		return (RPC_CANTSEND);
313	}
314	ct->ct_threads++;
315
316	if (ext) {
317		auth = ext->rc_auth;
318		errp = &ext->rc_err;
319	} else {
320		auth = cl->cl_auth;
321		errp = &ct->ct_error;
322	}
323
324	cr->cr_mrep = NULL;
325	cr->cr_error = 0;
326
327	if (ct->ct_wait.tv_usec == -1) {
328		timeout = utimeout;	/* use supplied timeout */
329	} else {
330		timeout = ct->ct_wait;	/* use default timeout */
331	}
332
333call_again:
334	mtx_assert(&ct->ct_lock, MA_OWNED);
335
336	ct->ct_xid++;
337	xid = ct->ct_xid;
338
339	mtx_unlock(&ct->ct_lock);
340
341	/*
342	 * Leave space to pre-pend the record mark.
343	 */
344	MGETHDR(mreq, M_WAIT, MT_DATA);
345	mreq->m_data += sizeof(uint32_t);
346	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
347	    ("RPC header too big"));
348	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
349	mreq->m_len = ct->ct_mpos;
350
351	/*
352	 * The XID is the first thing in the request.
353	 */
354	*mtod(mreq, uint32_t *) = htonl(xid);
355
356	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
357
358	errp->re_status = stat = RPC_SUCCESS;
359
360	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
361	    (! AUTH_MARSHALL(auth, xid, &xdrs,
362		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
363		errp->re_status = stat = RPC_CANTENCODEARGS;
364		mtx_lock(&ct->ct_lock);
365		goto out;
366	}
367	mreq->m_pkthdr.len = m_length(mreq, NULL);
368
369	/*
370	 * Prepend a record marker containing the packet length.
371	 */
372	M_PREPEND(mreq, sizeof(uint32_t), M_WAIT);
373	*mtod(mreq, uint32_t *) =
374		htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
375
376	cr->cr_xid = xid;
377	mtx_lock(&ct->ct_lock);
378	/*
379	 * Check to see if the other end has already started to close down
380	 * the connection. The upcall will have set ct_error.re_status
381	 * to RPC_CANTRECV if this is the case.
382	 * If the other end starts to close down the connection after this
383	 * point, it will be detected later when cr_error is checked,
384	 * since the request is in the ct_pending queue.
385	 */
386	if (ct->ct_error.re_status == RPC_CANTRECV) {
387		if (errp != &ct->ct_error) {
388			errp->re_errno = ct->ct_error.re_errno;
389			errp->re_status = RPC_CANTRECV;
390		}
391		stat = RPC_CANTRECV;
392		goto out;
393	}
394	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
395	mtx_unlock(&ct->ct_lock);
396
397	/*
398	 * sosend consumes mreq.
399	 */
400	error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
401	mreq = NULL;
402	if (error == EMSGSIZE) {
403		SOCKBUF_LOCK(&ct->ct_socket->so_snd);
404		sbwait(&ct->ct_socket->so_snd);
405		SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
406		AUTH_VALIDATE(auth, xid, NULL, NULL);
407		mtx_lock(&ct->ct_lock);
408		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
409		goto call_again;
410	}
411
412	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
413	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
414	reply_msg.acpted_rply.ar_verf.oa_length = 0;
415	reply_msg.acpted_rply.ar_results.where = NULL;
416	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
417
418	mtx_lock(&ct->ct_lock);
419	if (error) {
420		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
421		errp->re_errno = error;
422		errp->re_status = stat = RPC_CANTSEND;
423		goto out;
424	}
425
426	/*
427	 * Check to see if we got an upcall while waiting for the
428	 * lock. In both these cases, the request has been removed
429	 * from ct->ct_pending.
430	 */
431	if (cr->cr_error) {
432		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
433		errp->re_errno = cr->cr_error;
434		errp->re_status = stat = RPC_CANTRECV;
435		goto out;
436	}
437	if (cr->cr_mrep) {
438		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
439		goto got_reply;
440	}
441
442	/*
443	 * Hack to provide rpc-based message passing
444	 */
445	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
446		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
447		errp->re_status = stat = RPC_TIMEDOUT;
448		goto out;
449	}
450
451	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
452	    tvtohz(&timeout));
453
454	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
455
456	if (error) {
457		/*
458		 * The sleep returned an error so our request is still
459		 * on the list. Turn the error code into an
460		 * appropriate client status.
461		 */
462		errp->re_errno = error;
463		switch (error) {
464		case EINTR:
465		case ERESTART:
466			stat = RPC_INTR;
467			break;
468		case EWOULDBLOCK:
469			stat = RPC_TIMEDOUT;
470			break;
471		default:
472			stat = RPC_CANTRECV;
473		}
474		errp->re_status = stat;
475		goto out;
476	} else {
477		/*
478		 * We were woken up by the upcall.  If the
479		 * upcall had a receive error, report that,
480		 * otherwise we have a reply.
481		 */
482		if (cr->cr_error) {
483			errp->re_errno = cr->cr_error;
484			errp->re_status = stat = RPC_CANTRECV;
485			goto out;
486		}
487	}
488
489got_reply:
490	/*
491	 * Now decode and validate the response. We need to drop the
492	 * lock since xdr_replymsg may end up sleeping in malloc.
493	 */
494	mtx_unlock(&ct->ct_lock);
495
496	if (ext && ext->rc_feedback)
497		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
498
499	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
500	ok = xdr_replymsg(&xdrs, &reply_msg);
501	cr->cr_mrep = NULL;
502
503	if (ok) {
504		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
505		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
506			errp->re_status = stat = RPC_SUCCESS;
507		else
508			stat = _seterr_reply(&reply_msg, errp);
509
510		if (stat == RPC_SUCCESS) {
511			results = xdrmbuf_getall(&xdrs);
512			if (!AUTH_VALIDATE(auth, xid,
513				&reply_msg.acpted_rply.ar_verf,
514				&results)) {
515				errp->re_status = stat = RPC_AUTHERROR;
516				errp->re_why = AUTH_INVALIDRESP;
517			} else {
518				KASSERT(results,
519				    ("auth validated but no result"));
520				*resultsp = results;
521			}
522		}		/* end successful completion */
523		/*
524		 * If unsuccesful AND error is an authentication error
525		 * then refresh credentials and try again, else break
526		 */
527		else if (stat == RPC_AUTHERROR)
528			/* maybe our credentials need to be refreshed ... */
529			if (nrefreshes > 0 &&
530			    AUTH_REFRESH(auth, &reply_msg)) {
531				nrefreshes--;
532				XDR_DESTROY(&xdrs);
533				mtx_lock(&ct->ct_lock);
534				goto call_again;
535			}
536		/* end of unsuccessful completion */
537	}	/* end of valid reply message */
538	else {
539		errp->re_status = stat = RPC_CANTDECODERES;
540	}
541	XDR_DESTROY(&xdrs);
542	mtx_lock(&ct->ct_lock);
543out:
544	mtx_assert(&ct->ct_lock, MA_OWNED);
545
546	KASSERT(stat != RPC_SUCCESS || *resultsp,
547	    ("RPC_SUCCESS without reply"));
548
549	if (mreq)
550		m_freem(mreq);
551	if (cr->cr_mrep)
552		m_freem(cr->cr_mrep);
553
554	ct->ct_threads--;
555	if (ct->ct_closing)
556		wakeup(ct);
557
558	mtx_unlock(&ct->ct_lock);
559
560	if (auth && stat != RPC_SUCCESS)
561		AUTH_VALIDATE(auth, xid, NULL, NULL);
562
563	free(cr, M_RPC);
564
565	return (stat);
566}
567
568static void
569clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
570{
571	struct ct_data *ct = (struct ct_data *) cl->cl_private;
572
573	*errp = ct->ct_error;
574}
575
576static bool_t
577clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
578{
579	XDR xdrs;
580	bool_t dummy;
581
582	xdrs.x_op = XDR_FREE;
583	dummy = (*xdr_res)(&xdrs, res_ptr);
584
585	return (dummy);
586}
587
588/*ARGSUSED*/
589static void
590clnt_vc_abort(CLIENT *cl)
591{
592}
593
594static bool_t
595clnt_vc_control(CLIENT *cl, u_int request, void *info)
596{
597	struct ct_data *ct = (struct ct_data *)cl->cl_private;
598	void *infop = info;
599	SVCXPRT *xprt;
600
601	mtx_lock(&ct->ct_lock);
602
603	switch (request) {
604	case CLSET_FD_CLOSE:
605		ct->ct_closeit = TRUE;
606		mtx_unlock(&ct->ct_lock);
607		return (TRUE);
608	case CLSET_FD_NCLOSE:
609		ct->ct_closeit = FALSE;
610		mtx_unlock(&ct->ct_lock);
611		return (TRUE);
612	default:
613		break;
614	}
615
616	/* for other requests which use info */
617	if (info == NULL) {
618		mtx_unlock(&ct->ct_lock);
619		return (FALSE);
620	}
621	switch (request) {
622	case CLSET_TIMEOUT:
623		if (time_not_ok((struct timeval *)info)) {
624			mtx_unlock(&ct->ct_lock);
625			return (FALSE);
626		}
627		ct->ct_wait = *(struct timeval *)infop;
628		break;
629	case CLGET_TIMEOUT:
630		*(struct timeval *)infop = ct->ct_wait;
631		break;
632	case CLGET_SERVER_ADDR:
633		(void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
634		break;
635	case CLGET_SVC_ADDR:
636		/*
637		 * Slightly different semantics to userland - we use
638		 * sockaddr instead of netbuf.
639		 */
640		memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
641		break;
642	case CLSET_SVC_ADDR:		/* set to new address */
643		mtx_unlock(&ct->ct_lock);
644		return (FALSE);
645	case CLGET_XID:
646		*(uint32_t *)info = ct->ct_xid;
647		break;
648	case CLSET_XID:
649		/* This will set the xid of the NEXT call */
650		/* decrement by 1 as clnt_vc_call() increments once */
651		ct->ct_xid = *(uint32_t *)info - 1;
652		break;
653	case CLGET_VERS:
654		/*
655		 * This RELIES on the information that, in the call body,
656		 * the version number field is the fifth field from the
657		 * begining of the RPC header. MUST be changed if the
658		 * call_struct is changed
659		 */
660		*(uint32_t *)info =
661		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
662		    4 * BYTES_PER_XDR_UNIT));
663		break;
664
665	case CLSET_VERS:
666		*(uint32_t *)(void *)(ct->ct_mcallc +
667		    4 * BYTES_PER_XDR_UNIT) =
668		    htonl(*(uint32_t *)info);
669		break;
670
671	case CLGET_PROG:
672		/*
673		 * This RELIES on the information that, in the call body,
674		 * the program number field is the fourth field from the
675		 * begining of the RPC header. MUST be changed if the
676		 * call_struct is changed
677		 */
678		*(uint32_t *)info =
679		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
680		    3 * BYTES_PER_XDR_UNIT));
681		break;
682
683	case CLSET_PROG:
684		*(uint32_t *)(void *)(ct->ct_mcallc +
685		    3 * BYTES_PER_XDR_UNIT) =
686		    htonl(*(uint32_t *)info);
687		break;
688
689	case CLSET_WAITCHAN:
690		ct->ct_waitchan = (const char *)info;
691		break;
692
693	case CLGET_WAITCHAN:
694		*(const char **) info = ct->ct_waitchan;
695		break;
696
697	case CLSET_INTERRUPTIBLE:
698		if (*(int *) info)
699			ct->ct_waitflag = PCATCH | PBDRY;
700		else
701			ct->ct_waitflag = 0;
702		break;
703
704	case CLGET_INTERRUPTIBLE:
705		if (ct->ct_waitflag)
706			*(int *) info = TRUE;
707		else
708			*(int *) info = FALSE;
709		break;
710
711	case CLSET_BACKCHANNEL:
712		xprt = (SVCXPRT *)info;
713		if (ct->ct_backchannelxprt == NULL) {
714			xprt->xp_p2 = ct;
715			ct->ct_backchannelxprt = xprt;
716		}
717		break;
718
719	default:
720		mtx_unlock(&ct->ct_lock);
721		return (FALSE);
722	}
723
724	mtx_unlock(&ct->ct_lock);
725	return (TRUE);
726}
727
728static void
729clnt_vc_close(CLIENT *cl)
730{
731	struct ct_data *ct = (struct ct_data *) cl->cl_private;
732	struct ct_request *cr;
733
734	mtx_lock(&ct->ct_lock);
735
736	if (ct->ct_closed) {
737		mtx_unlock(&ct->ct_lock);
738		return;
739	}
740
741	if (ct->ct_closing) {
742		while (ct->ct_closing)
743			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
744		KASSERT(ct->ct_closed, ("client should be closed"));
745		mtx_unlock(&ct->ct_lock);
746		return;
747	}
748
749	if (ct->ct_socket) {
750		ct->ct_closing = TRUE;
751		mtx_unlock(&ct->ct_lock);
752
753		SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
754		soupcall_clear(ct->ct_socket, SO_RCV);
755		clnt_vc_upcallsdone(ct);
756		SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
757
758		/*
759		 * Abort any pending requests and wait until everyone
760		 * has finished with clnt_vc_call.
761		 */
762		mtx_lock(&ct->ct_lock);
763		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
764			cr->cr_xid = 0;
765			cr->cr_error = ESHUTDOWN;
766			wakeup(cr);
767		}
768
769		while (ct->ct_threads)
770			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
771	}
772
773	ct->ct_closing = FALSE;
774	ct->ct_closed = TRUE;
775	mtx_unlock(&ct->ct_lock);
776	wakeup(ct);
777}
778
779static void
780clnt_vc_destroy(CLIENT *cl)
781{
782	struct ct_data *ct = (struct ct_data *) cl->cl_private;
783	struct socket *so = NULL;
784	SVCXPRT *xprt;
785
786	clnt_vc_close(cl);
787
788	mtx_lock(&ct->ct_lock);
789	xprt = ct->ct_backchannelxprt;
790	ct->ct_backchannelxprt = NULL;
791	if (xprt != NULL) {
792		mtx_unlock(&ct->ct_lock);	/* To avoid a LOR. */
793		sx_xlock(&xprt->xp_lock);
794		mtx_lock(&ct->ct_lock);
795		xprt->xp_p2 = NULL;
796		xprt_unregister(xprt);
797	}
798
799	if (ct->ct_socket) {
800		if (ct->ct_closeit) {
801			so = ct->ct_socket;
802		}
803	}
804
805	mtx_unlock(&ct->ct_lock);
806	if (xprt != NULL) {
807		sx_xunlock(&xprt->xp_lock);
808		SVC_RELEASE(xprt);
809	}
810
811	mtx_destroy(&ct->ct_lock);
812	if (so) {
813		soshutdown(so, SHUT_WR);
814		soclose(so);
815	}
816	mem_free(ct, sizeof(struct ct_data));
817	if (cl->cl_netid && cl->cl_netid[0])
818		mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
819	if (cl->cl_tp && cl->cl_tp[0])
820		mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
821	mem_free(cl, sizeof(CLIENT));
822}
823
824/*
825 * Make sure that the time is not garbage.   -1 value is disallowed.
826 * Note this is different from time_not_ok in clnt_dg.c
827 */
828static bool_t
829time_not_ok(struct timeval *t)
830{
831	return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
832		t->tv_usec <= -1 || t->tv_usec > 1000000);
833}
834
835int
836clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
837{
838	struct ct_data *ct = (struct ct_data *) arg;
839	struct uio uio;
840	struct mbuf *m, *m2;
841	struct ct_request *cr;
842	int error, rcvflag, foundreq;
843	uint32_t xid_plus_direction[2], header;
844	bool_t do_read;
845	SVCXPRT *xprt;
846	struct cf_conn *cd;
847
848	CTASSERT(sizeof(xid_plus_direction) == 2 * sizeof(uint32_t));
849	ct->ct_upcallrefs++;
850	uio.uio_td = curthread;
851	do {
852		/*
853		 * If ct_record_resid is zero, we are waiting for a
854		 * record mark.
855		 */
856		if (ct->ct_record_resid == 0) {
857
858			/*
859			 * Make sure there is either a whole record
860			 * mark in the buffer or there is some other
861			 * error condition
862			 */
863			do_read = FALSE;
864			if (so->so_rcv.sb_cc >= sizeof(uint32_t)
865			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
866			    || so->so_error)
867				do_read = TRUE;
868
869			if (!do_read)
870				break;
871
872			SOCKBUF_UNLOCK(&so->so_rcv);
873			uio.uio_resid = sizeof(uint32_t);
874			m = NULL;
875			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
876			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
877			SOCKBUF_LOCK(&so->so_rcv);
878
879			if (error == EWOULDBLOCK)
880				break;
881
882			/*
883			 * If there was an error, wake up all pending
884			 * requests.
885			 */
886			if (error || uio.uio_resid > 0) {
887			wakeup_all:
888				mtx_lock(&ct->ct_lock);
889				if (!error) {
890					/*
891					 * We must have got EOF trying
892					 * to read from the stream.
893					 */
894					error = ECONNRESET;
895				}
896				ct->ct_error.re_status = RPC_CANTRECV;
897				ct->ct_error.re_errno = error;
898				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
899					cr->cr_error = error;
900					wakeup(cr);
901				}
902				mtx_unlock(&ct->ct_lock);
903				break;
904			}
905			m_copydata(m, 0, sizeof(uint32_t), (char *)&header);
906			header = ntohl(header);
907			ct->ct_record = NULL;
908			ct->ct_record_resid = header & 0x7fffffff;
909			ct->ct_record_eor = ((header & 0x80000000) != 0);
910			m_freem(m);
911		} else {
912			/*
913			 * Wait until the socket has the whole record
914			 * buffered.
915			 */
916			do_read = FALSE;
917			if (so->so_rcv.sb_cc >= ct->ct_record_resid
918			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
919			    || so->so_error)
920				do_read = TRUE;
921
922			if (!do_read)
923				break;
924
925			/*
926			 * We have the record mark. Read as much as
927			 * the socket has buffered up to the end of
928			 * this record.
929			 */
930			SOCKBUF_UNLOCK(&so->so_rcv);
931			uio.uio_resid = ct->ct_record_resid;
932			m = NULL;
933			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
934			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
935			SOCKBUF_LOCK(&so->so_rcv);
936
937			if (error == EWOULDBLOCK)
938				break;
939
940			if (error || uio.uio_resid == ct->ct_record_resid)
941				goto wakeup_all;
942
943			/*
944			 * If we have part of the record already,
945			 * chain this bit onto the end.
946			 */
947			if (ct->ct_record)
948				m_last(ct->ct_record)->m_next = m;
949			else
950				ct->ct_record = m;
951
952			ct->ct_record_resid = uio.uio_resid;
953
954			/*
955			 * If we have the entire record, see if we can
956			 * match it to a request.
957			 */
958			if (ct->ct_record_resid == 0
959			    && ct->ct_record_eor) {
960				/*
961				 * The XID is in the first uint32_t of
962				 * the reply and the message direction
963				 * is the second one.
964				 */
965				if (ct->ct_record->m_len <
966				    sizeof(xid_plus_direction) &&
967				    m_length(ct->ct_record, NULL) <
968				    sizeof(xid_plus_direction)) {
969					m_freem(ct->ct_record);
970					break;
971				}
972				m_copydata(ct->ct_record, 0,
973				    sizeof(xid_plus_direction),
974				    (char *)xid_plus_direction);
975				xid_plus_direction[0] =
976				    ntohl(xid_plus_direction[0]);
977				xid_plus_direction[1] =
978				    ntohl(xid_plus_direction[1]);
979				/* Check message direction. */
980				if (xid_plus_direction[1] == CALL) {
981					/* This is a backchannel request. */
982					mtx_lock(&ct->ct_lock);
983					xprt = ct->ct_backchannelxprt;
984					if (xprt == NULL) {
985						mtx_unlock(&ct->ct_lock);
986						/* Just throw it away. */
987						m_freem(ct->ct_record);
988						ct->ct_record = NULL;
989					} else {
990						cd = (struct cf_conn *)
991						    xprt->xp_p1;
992						m2 = cd->mreq;
993						/*
994						 * The requests are chained
995						 * in the m_nextpkt list.
996						 */
997						while (m2 != NULL &&
998						    m2->m_nextpkt != NULL)
999							/* Find end of list. */
1000							m2 = m2->m_nextpkt;
1001						if (m2 != NULL)
1002							m2->m_nextpkt =
1003							    ct->ct_record;
1004						else
1005							cd->mreq =
1006							    ct->ct_record;
1007						ct->ct_record->m_nextpkt =
1008						    NULL;
1009						ct->ct_record = NULL;
1010						xprt_active(xprt);
1011						mtx_unlock(&ct->ct_lock);
1012					}
1013				} else {
1014					mtx_lock(&ct->ct_lock);
1015					foundreq = 0;
1016					TAILQ_FOREACH(cr, &ct->ct_pending,
1017					    cr_link) {
1018						if (cr->cr_xid ==
1019						    xid_plus_direction[0]) {
1020							/*
1021							 * This one
1022							 * matches. We leave
1023							 * the reply mbuf in
1024							 * cr->cr_mrep. Set
1025							 * the XID to zero so
1026							 * that we will ignore
1027							 * any duplicated
1028							 * replies.
1029							 */
1030							cr->cr_xid = 0;
1031							cr->cr_mrep =
1032							    ct->ct_record;
1033							cr->cr_error = 0;
1034							foundreq = 1;
1035							wakeup(cr);
1036							break;
1037						}
1038					}
1039					mtx_unlock(&ct->ct_lock);
1040
1041					if (!foundreq)
1042						m_freem(ct->ct_record);
1043					ct->ct_record = NULL;
1044				}
1045			}
1046		}
1047	} while (m);
1048	ct->ct_upcallrefs--;
1049	if (ct->ct_upcallrefs < 0)
1050		panic("rpcvc upcall refcnt");
1051	if (ct->ct_upcallrefs == 0)
1052		wakeup(&ct->ct_upcallrefs);
1053	return (SU_OK);
1054}
1055
1056/*
1057 * Wait for all upcalls in progress to complete.
1058 */
1059static void
1060clnt_vc_upcallsdone(struct ct_data *ct)
1061{
1062
1063	SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1064
1065	while (ct->ct_upcallrefs > 0)
1066		(void) msleep(&ct->ct_upcallrefs,
1067		    SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1068}
1069