clnt_vc.c revision 193437
1/*	$NetBSD: clnt_vc.c,v 1.4 2000/07/14 08:40:42 fvdl Exp $	*/
2
3/*
4 * Sun RPC is a product of Sun Microsystems, Inc. and is provided for
5 * unrestricted use provided that this legend is included on all tape
6 * media and as a part of the software program in whole or part.  Users
7 * may copy or modify Sun RPC without charge, but are not authorized
8 * to license or distribute it to anyone else except as part of a product or
9 * program developed by the user.
10 *
11 * SUN RPC IS PROVIDED AS IS WITH NO WARRANTIES OF ANY KIND INCLUDING THE
12 * WARRANTIES OF DESIGN, MERCHANTIBILITY AND FITNESS FOR A PARTICULAR
13 * PURPOSE, OR ARISING FROM A COURSE OF DEALING, USAGE OR TRADE PRACTICE.
14 *
15 * Sun RPC is provided with no support and without any obligation on the
16 * part of Sun Microsystems, Inc. to assist in its use, correction,
17 * modification or enhancement.
18 *
19 * SUN MICROSYSTEMS, INC. SHALL HAVE NO LIABILITY WITH RESPECT TO THE
20 * INFRINGEMENT OF COPYRIGHTS, TRADE SECRETS OR ANY PATENTS BY SUN RPC
21 * OR ANY PART THEREOF.
22 *
23 * In no event will Sun Microsystems, Inc. be liable for any lost revenue
24 * or profits or other special, indirect and consequential damages, even if
25 * Sun has been advised of the possibility of such damages.
26 *
27 * Sun Microsystems, Inc.
28 * 2550 Garcia Avenue
29 * Mountain View, California  94043
30 */
31
32#if defined(LIBC_SCCS) && !defined(lint)
33static char *sccsid2 = "@(#)clnt_tcp.c 1.37 87/10/05 Copyr 1984 Sun Micro";
34static char *sccsid = "@(#)clnt_tcp.c	2.2 88/08/01 4.0 RPCSRC";
35static char sccsid3[] = "@(#)clnt_vc.c 1.19 89/03/16 Copyr 1988 Sun Micro";
36#endif
37#include <sys/cdefs.h>
38__FBSDID("$FreeBSD: head/sys/rpc/clnt_vc.c 193437 2009-06-04 14:49:27Z rmacklem $");
39
40/*
41 * clnt_tcp.c, Implements a TCP/IP based, client side RPC.
42 *
43 * Copyright (C) 1984, Sun Microsystems, Inc.
44 *
45 * TCP based RPC supports 'batched calls'.
46 * A sequence of calls may be batched-up in a send buffer.  The rpc call
47 * return immediately to the client even though the call was not necessarily
48 * sent.  The batching occurs if the results' xdr routine is NULL (0) AND
49 * the rpc timeout value is zero (see clnt.h, rpc).
50 *
51 * Clients should NOT casually batch calls that in fact return results; that is,
52 * the server side should be aware that a call is batched and not produce any
53 * return message.  Batched calls that produce many result messages can
54 * deadlock (netlock) the client and the server....
55 *
56 * Now go hang yourself.
57 */
58
59#include <sys/param.h>
60#include <sys/systm.h>
61#include <sys/lock.h>
62#include <sys/malloc.h>
63#include <sys/mbuf.h>
64#include <sys/mutex.h>
65#include <sys/pcpu.h>
66#include <sys/proc.h>
67#include <sys/protosw.h>
68#include <sys/socket.h>
69#include <sys/socketvar.h>
70#include <sys/syslog.h>
71#include <sys/time.h>
72#include <sys/uio.h>
73#include <netinet/tcp.h>
74
75#include <rpc/rpc.h>
76#include <rpc/rpc_com.h>
77
78#define MCALL_MSG_SIZE 24
79
80struct cmessage {
81        struct cmsghdr cmsg;
82        struct cmsgcred cmcred;
83};
84
85static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
86    rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
87static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
88static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
89static void clnt_vc_abort(CLIENT *);
90static bool_t clnt_vc_control(CLIENT *, u_int, void *);
91static void clnt_vc_close(CLIENT *);
92static void clnt_vc_destroy(CLIENT *);
93static bool_t time_not_ok(struct timeval *);
94static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
95
96static struct clnt_ops clnt_vc_ops = {
97	.cl_call =	clnt_vc_call,
98	.cl_abort =	clnt_vc_abort,
99	.cl_geterr =	clnt_vc_geterr,
100	.cl_freeres =	clnt_vc_freeres,
101	.cl_close =	clnt_vc_close,
102	.cl_destroy =	clnt_vc_destroy,
103	.cl_control =	clnt_vc_control
104};
105
106/*
107 * A pending RPC request which awaits a reply. Requests which have
108 * received their reply will have cr_xid set to zero and cr_mrep to
109 * the mbuf chain of the reply.
110 */
111struct ct_request {
112	TAILQ_ENTRY(ct_request) cr_link;
113	uint32_t		cr_xid;		/* XID of request */
114	struct mbuf		*cr_mrep;	/* reply received by upcall */
115	int			cr_error;	/* any error from upcall */
116	char			cr_verf[MAX_AUTH_BYTES]; /* reply verf */
117};
118
119TAILQ_HEAD(ct_request_list, ct_request);
120
121struct ct_data {
122	struct mtx	ct_lock;
123	int		ct_threads;	/* number of threads in clnt_vc_call */
124	bool_t		ct_closing;	/* TRUE if we are closing */
125	bool_t		ct_closed;	/* TRUE if we are closed */
126	struct socket	*ct_socket;	/* connection socket */
127	bool_t		ct_closeit;	/* close it on destroy */
128	struct timeval	ct_wait;	/* wait interval in milliseconds */
129	struct sockaddr_storage	ct_addr; /* remote addr */
130	struct rpc_err	ct_error;
131	uint32_t	ct_xid;
132	char		ct_mcallc[MCALL_MSG_SIZE]; /* marshalled callmsg */
133	size_t		ct_mpos;	/* pos after marshal */
134	const char	*ct_waitchan;
135	int		ct_waitflag;
136	struct mbuf	*ct_record;	/* current reply record */
137	size_t		ct_record_resid; /* how much left of reply to read */
138	bool_t		ct_record_eor;	 /* true if reading last fragment */
139	struct ct_request_list ct_pending;
140	int		ct_upcallrefs;	/* Ref cnt of upcalls in prog. */
141};
142
143static void clnt_vc_upcallsdone(struct ct_data *);
144
145static const char clnt_vc_errstr[] = "%s : %s";
146static const char clnt_vc_str[] = "clnt_vc_create";
147static const char clnt_read_vc_str[] = "read_vc";
148static const char __no_mem_str[] = "out of memory";
149
150/*
151 * Create a client handle for a connection.
152 * Default options are set, which the user can change using clnt_control()'s.
153 * The rpc/vc package does buffering similar to stdio, so the client
154 * must pick send and receive buffer sizes, 0 => use the default.
155 * NB: fd is copied into a private area.
156 * NB: The rpch->cl_auth is set null authentication. Caller may wish to
157 * set this something more useful.
158 *
159 * fd should be an open socket
160 */
161CLIENT *
162clnt_vc_create(
163	struct socket *so,		/* open file descriptor */
164	struct sockaddr *raddr,		/* servers address */
165	const rpcprog_t prog,		/* program number */
166	const rpcvers_t vers,		/* version number */
167	size_t sendsz,			/* buffer recv size */
168	size_t recvsz)			/* buffer send size */
169{
170	CLIENT *cl;			/* client handle */
171	struct ct_data *ct = NULL;	/* client handle */
172	struct timeval now;
173	struct rpc_msg call_msg;
174	static uint32_t disrupt;
175	struct __rpc_sockinfo si;
176	XDR xdrs;
177	int error, interrupted, one = 1;
178	struct sockopt sopt;
179
180	if (disrupt == 0)
181		disrupt = (uint32_t)(long)raddr;
182
183	cl = (CLIENT *)mem_alloc(sizeof (*cl));
184	ct = (struct ct_data *)mem_alloc(sizeof (*ct));
185
186	mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
187	ct->ct_threads = 0;
188	ct->ct_closing = FALSE;
189	ct->ct_closed = FALSE;
190	ct->ct_upcallrefs = 0;
191
192	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
193		error = soconnect(so, raddr, curthread);
194		SOCK_LOCK(so);
195		interrupted = 0;
196		while ((so->so_state & SS_ISCONNECTING)
197		    && so->so_error == 0) {
198			error = msleep(&so->so_timeo, SOCK_MTX(so),
199			    PSOCK | PCATCH, "connec", 0);
200			if (error) {
201				if (error == EINTR || error == ERESTART)
202					interrupted = 1;
203				break;
204			}
205		}
206		if (error == 0) {
207			error = so->so_error;
208			so->so_error = 0;
209		}
210		SOCK_UNLOCK(so);
211		if (error) {
212			if (!interrupted)
213				so->so_state &= ~SS_ISCONNECTING;
214			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
215			rpc_createerr.cf_error.re_errno = error;
216			goto err;
217		}
218	}
219
220	if (!__rpc_socket2sockinfo(so, &si))
221		goto err;
222
223	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
224		bzero(&sopt, sizeof(sopt));
225		sopt.sopt_dir = SOPT_SET;
226		sopt.sopt_level = SOL_SOCKET;
227		sopt.sopt_name = SO_KEEPALIVE;
228		sopt.sopt_val = &one;
229		sopt.sopt_valsize = sizeof(one);
230		sosetopt(so, &sopt);
231	}
232
233	if (so->so_proto->pr_protocol == IPPROTO_TCP) {
234		bzero(&sopt, sizeof(sopt));
235		sopt.sopt_dir = SOPT_SET;
236		sopt.sopt_level = IPPROTO_TCP;
237		sopt.sopt_name = TCP_NODELAY;
238		sopt.sopt_val = &one;
239		sopt.sopt_valsize = sizeof(one);
240		sosetopt(so, &sopt);
241	}
242
243	ct->ct_closeit = FALSE;
244
245	/*
246	 * Set up private data struct
247	 */
248	ct->ct_socket = so;
249	ct->ct_wait.tv_sec = -1;
250	ct->ct_wait.tv_usec = -1;
251	memcpy(&ct->ct_addr, raddr, raddr->sa_len);
252
253	/*
254	 * Initialize call message
255	 */
256	getmicrotime(&now);
257	ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
258	call_msg.rm_xid = ct->ct_xid;
259	call_msg.rm_direction = CALL;
260	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
261	call_msg.rm_call.cb_prog = (uint32_t)prog;
262	call_msg.rm_call.cb_vers = (uint32_t)vers;
263
264	/*
265	 * pre-serialize the static part of the call msg and stash it away
266	 */
267	xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
268	    XDR_ENCODE);
269	if (! xdr_callhdr(&xdrs, &call_msg)) {
270		if (ct->ct_closeit) {
271			soclose(ct->ct_socket);
272		}
273		goto err;
274	}
275	ct->ct_mpos = XDR_GETPOS(&xdrs);
276	XDR_DESTROY(&xdrs);
277	ct->ct_waitchan = "rpcrecv";
278	ct->ct_waitflag = 0;
279
280	/*
281	 * Create a client handle which uses xdrrec for serialization
282	 * and authnone for authentication.
283	 */
284	cl->cl_refs = 1;
285	cl->cl_ops = &clnt_vc_ops;
286	cl->cl_private = ct;
287	cl->cl_auth = authnone_create();
288	sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
289	recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
290	soreserve(ct->ct_socket, sendsz, recvsz);
291
292	SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
293	soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
294	SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
295
296	ct->ct_record = NULL;
297	ct->ct_record_resid = 0;
298	TAILQ_INIT(&ct->ct_pending);
299	return (cl);
300
301err:
302	if (cl) {
303		if (ct) {
304			mtx_destroy(&ct->ct_lock);
305			mem_free(ct, sizeof (struct ct_data));
306		}
307		if (cl)
308			mem_free(cl, sizeof (CLIENT));
309	}
310	return ((CLIENT *)NULL);
311}
312
313static enum clnt_stat
314clnt_vc_call(
315	CLIENT		*cl,		/* client handle */
316	struct rpc_callextra *ext,	/* call metadata */
317	rpcproc_t	proc,		/* procedure number */
318	struct mbuf	*args,		/* pointer to args */
319	struct mbuf	**resultsp,	/* pointer to results */
320	struct timeval	utimeout)
321{
322	struct ct_data *ct = (struct ct_data *) cl->cl_private;
323	AUTH *auth;
324	struct rpc_err *errp;
325	enum clnt_stat stat;
326	XDR xdrs;
327	struct rpc_msg reply_msg;
328	bool_t ok;
329	int nrefreshes = 2;		/* number of times to refresh cred */
330	struct timeval timeout;
331	uint32_t xid;
332	struct mbuf *mreq = NULL, *results;
333	struct ct_request *cr;
334	int error;
335
336	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
337
338	mtx_lock(&ct->ct_lock);
339
340	if (ct->ct_closing || ct->ct_closed) {
341		mtx_unlock(&ct->ct_lock);
342		free(cr, M_RPC);
343		return (RPC_CANTSEND);
344	}
345	ct->ct_threads++;
346
347	if (ext) {
348		auth = ext->rc_auth;
349		errp = &ext->rc_err;
350	} else {
351		auth = cl->cl_auth;
352		errp = &ct->ct_error;
353	}
354
355	cr->cr_mrep = NULL;
356	cr->cr_error = 0;
357
358	if (ct->ct_wait.tv_usec == -1) {
359		timeout = utimeout;	/* use supplied timeout */
360	} else {
361		timeout = ct->ct_wait;	/* use default timeout */
362	}
363
364call_again:
365	mtx_assert(&ct->ct_lock, MA_OWNED);
366
367	ct->ct_xid++;
368	xid = ct->ct_xid;
369
370	mtx_unlock(&ct->ct_lock);
371
372	/*
373	 * Leave space to pre-pend the record mark.
374	 */
375	MGETHDR(mreq, M_WAIT, MT_DATA);
376	mreq->m_data += sizeof(uint32_t);
377	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
378	    ("RPC header too big"));
379	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
380	mreq->m_len = ct->ct_mpos;
381
382	/*
383	 * The XID is the first thing in the request.
384	 */
385	*mtod(mreq, uint32_t *) = htonl(xid);
386
387	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
388
389	errp->re_status = stat = RPC_SUCCESS;
390
391	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
392	    (! AUTH_MARSHALL(auth, xid, &xdrs,
393		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
394		errp->re_status = stat = RPC_CANTENCODEARGS;
395		mtx_lock(&ct->ct_lock);
396		goto out;
397	}
398	mreq->m_pkthdr.len = m_length(mreq, NULL);
399
400	/*
401	 * Prepend a record marker containing the packet length.
402	 */
403	M_PREPEND(mreq, sizeof(uint32_t), M_WAIT);
404	*mtod(mreq, uint32_t *) =
405		htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
406
407	cr->cr_xid = xid;
408	mtx_lock(&ct->ct_lock);
409	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
410	mtx_unlock(&ct->ct_lock);
411
412	/*
413	 * sosend consumes mreq.
414	 */
415	error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
416	mreq = NULL;
417	if (error == EMSGSIZE) {
418		SOCKBUF_LOCK(&ct->ct_socket->so_snd);
419		sbwait(&ct->ct_socket->so_snd);
420		SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
421		AUTH_VALIDATE(auth, xid, NULL, NULL);
422		mtx_lock(&ct->ct_lock);
423		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
424		goto call_again;
425	}
426
427	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
428	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
429	reply_msg.acpted_rply.ar_verf.oa_length = 0;
430	reply_msg.acpted_rply.ar_results.where = NULL;
431	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
432
433	mtx_lock(&ct->ct_lock);
434	if (error) {
435		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
436		errp->re_errno = error;
437		errp->re_status = stat = RPC_CANTSEND;
438		goto out;
439	}
440
441	/*
442	 * Check to see if we got an upcall while waiting for the
443	 * lock. In both these cases, the request has been removed
444	 * from ct->ct_pending.
445	 */
446	if (cr->cr_error) {
447		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
448		errp->re_errno = cr->cr_error;
449		errp->re_status = stat = RPC_CANTRECV;
450		goto out;
451	}
452	if (cr->cr_mrep) {
453		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
454		goto got_reply;
455	}
456
457	/*
458	 * Hack to provide rpc-based message passing
459	 */
460	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
461		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
462		errp->re_status = stat = RPC_TIMEDOUT;
463		goto out;
464	}
465
466	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
467	    tvtohz(&timeout));
468
469	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
470
471	if (error) {
472		/*
473		 * The sleep returned an error so our request is still
474		 * on the list. Turn the error code into an
475		 * appropriate client status.
476		 */
477		errp->re_errno = error;
478		switch (error) {
479		case EINTR:
480			stat = RPC_INTR;
481			break;
482		case EWOULDBLOCK:
483			stat = RPC_TIMEDOUT;
484			break;
485		default:
486			stat = RPC_CANTRECV;
487		}
488		errp->re_status = stat;
489		goto out;
490	} else {
491		/*
492		 * We were woken up by the upcall.  If the
493		 * upcall had a receive error, report that,
494		 * otherwise we have a reply.
495		 */
496		if (cr->cr_error) {
497			errp->re_errno = cr->cr_error;
498			errp->re_status = stat = RPC_CANTRECV;
499			goto out;
500		}
501	}
502
503got_reply:
504	/*
505	 * Now decode and validate the response. We need to drop the
506	 * lock since xdr_replymsg may end up sleeping in malloc.
507	 */
508	mtx_unlock(&ct->ct_lock);
509
510	if (ext && ext->rc_feedback)
511		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
512
513	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
514	ok = xdr_replymsg(&xdrs, &reply_msg);
515	cr->cr_mrep = NULL;
516
517	if (ok) {
518		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
519		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
520			errp->re_status = stat = RPC_SUCCESS;
521		else
522			stat = _seterr_reply(&reply_msg, errp);
523
524		if (stat == RPC_SUCCESS) {
525			results = xdrmbuf_getall(&xdrs);
526			if (!AUTH_VALIDATE(auth, xid,
527				&reply_msg.acpted_rply.ar_verf,
528				&results)) {
529				errp->re_status = stat = RPC_AUTHERROR;
530				errp->re_why = AUTH_INVALIDRESP;
531			} else {
532				KASSERT(results,
533				    ("auth validated but no result"));
534				*resultsp = results;
535			}
536		}		/* end successful completion */
537		/*
538		 * If unsuccesful AND error is an authentication error
539		 * then refresh credentials and try again, else break
540		 */
541		else if (stat == RPC_AUTHERROR)
542			/* maybe our credentials need to be refreshed ... */
543			if (nrefreshes > 0 &&
544			    AUTH_REFRESH(auth, &reply_msg)) {
545				nrefreshes--;
546				XDR_DESTROY(&xdrs);
547				mtx_lock(&ct->ct_lock);
548				goto call_again;
549			}
550		/* end of unsuccessful completion */
551	}	/* end of valid reply message */
552	else {
553		errp->re_status = stat = RPC_CANTDECODERES;
554	}
555	XDR_DESTROY(&xdrs);
556	mtx_lock(&ct->ct_lock);
557out:
558	mtx_assert(&ct->ct_lock, MA_OWNED);
559
560	KASSERT(stat != RPC_SUCCESS || *resultsp,
561	    ("RPC_SUCCESS without reply"));
562
563	if (mreq)
564		m_freem(mreq);
565	if (cr->cr_mrep)
566		m_freem(cr->cr_mrep);
567
568	ct->ct_threads--;
569	if (ct->ct_closing)
570		wakeup(ct);
571
572	mtx_unlock(&ct->ct_lock);
573
574	if (auth && stat != RPC_SUCCESS)
575		AUTH_VALIDATE(auth, xid, NULL, NULL);
576
577	free(cr, M_RPC);
578
579	return (stat);
580}
581
582static void
583clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
584{
585	struct ct_data *ct = (struct ct_data *) cl->cl_private;
586
587	*errp = ct->ct_error;
588}
589
590static bool_t
591clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
592{
593	XDR xdrs;
594	bool_t dummy;
595
596	xdrs.x_op = XDR_FREE;
597	dummy = (*xdr_res)(&xdrs, res_ptr);
598
599	return (dummy);
600}
601
602/*ARGSUSED*/
603static void
604clnt_vc_abort(CLIENT *cl)
605{
606}
607
608static bool_t
609clnt_vc_control(CLIENT *cl, u_int request, void *info)
610{
611	struct ct_data *ct = (struct ct_data *)cl->cl_private;
612	void *infop = info;
613
614	mtx_lock(&ct->ct_lock);
615
616	switch (request) {
617	case CLSET_FD_CLOSE:
618		ct->ct_closeit = TRUE;
619		mtx_unlock(&ct->ct_lock);
620		return (TRUE);
621	case CLSET_FD_NCLOSE:
622		ct->ct_closeit = FALSE;
623		mtx_unlock(&ct->ct_lock);
624		return (TRUE);
625	default:
626		break;
627	}
628
629	/* for other requests which use info */
630	if (info == NULL) {
631		mtx_unlock(&ct->ct_lock);
632		return (FALSE);
633	}
634	switch (request) {
635	case CLSET_TIMEOUT:
636		if (time_not_ok((struct timeval *)info)) {
637			mtx_unlock(&ct->ct_lock);
638			return (FALSE);
639		}
640		ct->ct_wait = *(struct timeval *)infop;
641		break;
642	case CLGET_TIMEOUT:
643		*(struct timeval *)infop = ct->ct_wait;
644		break;
645	case CLGET_SERVER_ADDR:
646		(void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
647		break;
648	case CLGET_SVC_ADDR:
649		/*
650		 * Slightly different semantics to userland - we use
651		 * sockaddr instead of netbuf.
652		 */
653		memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
654		break;
655	case CLSET_SVC_ADDR:		/* set to new address */
656		mtx_unlock(&ct->ct_lock);
657		return (FALSE);
658	case CLGET_XID:
659		*(uint32_t *)info = ct->ct_xid;
660		break;
661	case CLSET_XID:
662		/* This will set the xid of the NEXT call */
663		/* decrement by 1 as clnt_vc_call() increments once */
664		ct->ct_xid = *(uint32_t *)info - 1;
665		break;
666	case CLGET_VERS:
667		/*
668		 * This RELIES on the information that, in the call body,
669		 * the version number field is the fifth field from the
670		 * begining of the RPC header. MUST be changed if the
671		 * call_struct is changed
672		 */
673		*(uint32_t *)info =
674		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
675		    4 * BYTES_PER_XDR_UNIT));
676		break;
677
678	case CLSET_VERS:
679		*(uint32_t *)(void *)(ct->ct_mcallc +
680		    4 * BYTES_PER_XDR_UNIT) =
681		    htonl(*(uint32_t *)info);
682		break;
683
684	case CLGET_PROG:
685		/*
686		 * This RELIES on the information that, in the call body,
687		 * the program number field is the fourth field from the
688		 * begining of the RPC header. MUST be changed if the
689		 * call_struct is changed
690		 */
691		*(uint32_t *)info =
692		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
693		    3 * BYTES_PER_XDR_UNIT));
694		break;
695
696	case CLSET_PROG:
697		*(uint32_t *)(void *)(ct->ct_mcallc +
698		    3 * BYTES_PER_XDR_UNIT) =
699		    htonl(*(uint32_t *)info);
700		break;
701
702	case CLSET_WAITCHAN:
703		ct->ct_waitchan = (const char *)info;
704		break;
705
706	case CLGET_WAITCHAN:
707		*(const char **) info = ct->ct_waitchan;
708		break;
709
710	case CLSET_INTERRUPTIBLE:
711		if (*(int *) info)
712			ct->ct_waitflag = PCATCH;
713		else
714			ct->ct_waitflag = 0;
715		break;
716
717	case CLGET_INTERRUPTIBLE:
718		if (ct->ct_waitflag)
719			*(int *) info = TRUE;
720		else
721			*(int *) info = FALSE;
722		break;
723
724	default:
725		mtx_unlock(&ct->ct_lock);
726		return (FALSE);
727	}
728
729	mtx_unlock(&ct->ct_lock);
730	return (TRUE);
731}
732
733static void
734clnt_vc_close(CLIENT *cl)
735{
736	struct ct_data *ct = (struct ct_data *) cl->cl_private;
737	struct ct_request *cr;
738
739	mtx_lock(&ct->ct_lock);
740
741	if (ct->ct_closed) {
742		mtx_unlock(&ct->ct_lock);
743		return;
744	}
745
746	if (ct->ct_closing) {
747		while (ct->ct_closing)
748			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
749		KASSERT(ct->ct_closed, ("client should be closed"));
750		mtx_unlock(&ct->ct_lock);
751		return;
752	}
753
754	if (ct->ct_socket) {
755		ct->ct_closing = TRUE;
756		mtx_unlock(&ct->ct_lock);
757
758		SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
759		soupcall_clear(ct->ct_socket, SO_RCV);
760		clnt_vc_upcallsdone(ct);
761		SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
762
763		/*
764		 * Abort any pending requests and wait until everyone
765		 * has finished with clnt_vc_call.
766		 */
767		mtx_lock(&ct->ct_lock);
768		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
769			cr->cr_xid = 0;
770			cr->cr_error = ESHUTDOWN;
771			wakeup(cr);
772		}
773
774		while (ct->ct_threads)
775			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
776	}
777
778	ct->ct_closing = FALSE;
779	ct->ct_closed = TRUE;
780	mtx_unlock(&ct->ct_lock);
781	wakeup(ct);
782}
783
784static void
785clnt_vc_destroy(CLIENT *cl)
786{
787	struct ct_data *ct = (struct ct_data *) cl->cl_private;
788	struct socket *so = NULL;
789
790	clnt_vc_close(cl);
791
792	mtx_lock(&ct->ct_lock);
793
794	if (ct->ct_socket) {
795		if (ct->ct_closeit) {
796			so = ct->ct_socket;
797		}
798	}
799
800	mtx_unlock(&ct->ct_lock);
801
802	mtx_destroy(&ct->ct_lock);
803	if (so) {
804		soshutdown(so, SHUT_WR);
805		soclose(so);
806	}
807	mem_free(ct, sizeof(struct ct_data));
808	mem_free(cl, sizeof(CLIENT));
809}
810
811/*
812 * Make sure that the time is not garbage.   -1 value is disallowed.
813 * Note this is different from time_not_ok in clnt_dg.c
814 */
815static bool_t
816time_not_ok(struct timeval *t)
817{
818	return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
819		t->tv_usec <= -1 || t->tv_usec > 1000000);
820}
821
822int
823clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
824{
825	struct ct_data *ct = (struct ct_data *) arg;
826	struct uio uio;
827	struct mbuf *m;
828	struct ct_request *cr;
829	int error, rcvflag, foundreq;
830	uint32_t xid, header;
831	bool_t do_read;
832
833	ct->ct_upcallrefs++;
834	uio.uio_td = curthread;
835	do {
836		/*
837		 * If ct_record_resid is zero, we are waiting for a
838		 * record mark.
839		 */
840		if (ct->ct_record_resid == 0) {
841
842			/*
843			 * Make sure there is either a whole record
844			 * mark in the buffer or there is some other
845			 * error condition
846			 */
847			do_read = FALSE;
848			if (so->so_rcv.sb_cc >= sizeof(uint32_t)
849			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
850			    || so->so_error)
851				do_read = TRUE;
852
853			if (!do_read)
854				break;
855
856			SOCKBUF_UNLOCK(&so->so_rcv);
857			uio.uio_resid = sizeof(uint32_t);
858			m = NULL;
859			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
860			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
861			SOCKBUF_LOCK(&so->so_rcv);
862
863			if (error == EWOULDBLOCK)
864				break;
865
866			/*
867			 * If there was an error, wake up all pending
868			 * requests.
869			 */
870			if (error || uio.uio_resid > 0) {
871			wakeup_all:
872				mtx_lock(&ct->ct_lock);
873				if (!error) {
874					/*
875					 * We must have got EOF trying
876					 * to read from the stream.
877					 */
878					error = ECONNRESET;
879				}
880				ct->ct_error.re_status = RPC_CANTRECV;
881				ct->ct_error.re_errno = error;
882				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
883					cr->cr_error = error;
884					wakeup(cr);
885				}
886				mtx_unlock(&ct->ct_lock);
887				break;
888			}
889			bcopy(mtod(m, uint32_t *), &header, sizeof(uint32_t));
890			header = ntohl(header);
891			ct->ct_record = NULL;
892			ct->ct_record_resid = header & 0x7fffffff;
893			ct->ct_record_eor = ((header & 0x80000000) != 0);
894			m_freem(m);
895		} else {
896			/*
897			 * Wait until the socket has the whole record
898			 * buffered.
899			 */
900			do_read = FALSE;
901			if (so->so_rcv.sb_cc >= ct->ct_record_resid
902			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
903			    || so->so_error)
904				do_read = TRUE;
905
906			if (!do_read)
907				break;
908
909			/*
910			 * We have the record mark. Read as much as
911			 * the socket has buffered up to the end of
912			 * this record.
913			 */
914			SOCKBUF_UNLOCK(&so->so_rcv);
915			uio.uio_resid = ct->ct_record_resid;
916			m = NULL;
917			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
918			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
919			SOCKBUF_LOCK(&so->so_rcv);
920
921			if (error == EWOULDBLOCK)
922				break;
923
924			if (error || uio.uio_resid == ct->ct_record_resid)
925				goto wakeup_all;
926
927			/*
928			 * If we have part of the record already,
929			 * chain this bit onto the end.
930			 */
931			if (ct->ct_record)
932				m_last(ct->ct_record)->m_next = m;
933			else
934				ct->ct_record = m;
935
936			ct->ct_record_resid = uio.uio_resid;
937
938			/*
939			 * If we have the entire record, see if we can
940			 * match it to a request.
941			 */
942			if (ct->ct_record_resid == 0
943			    && ct->ct_record_eor) {
944				/*
945				 * The XID is in the first uint32_t of
946				 * the reply.
947				 */
948				if (ct->ct_record->m_len < sizeof(xid))
949					ct->ct_record =
950						m_pullup(ct->ct_record,
951						    sizeof(xid));
952				if (!ct->ct_record)
953					break;
954				bcopy(mtod(ct->ct_record, uint32_t *),
955				    &xid, sizeof(uint32_t));
956				xid = ntohl(xid);
957
958				mtx_lock(&ct->ct_lock);
959				foundreq = 0;
960				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
961					if (cr->cr_xid == xid) {
962						/*
963						 * This one
964						 * matches. We leave
965						 * the reply mbuf in
966						 * cr->cr_mrep. Set
967						 * the XID to zero so
968						 * that we will ignore
969						 * any duplicaed
970						 * replies.
971						 */
972						cr->cr_xid = 0;
973						cr->cr_mrep = ct->ct_record;
974						cr->cr_error = 0;
975						foundreq = 1;
976						wakeup(cr);
977						break;
978					}
979				}
980				mtx_unlock(&ct->ct_lock);
981
982				if (!foundreq)
983					m_freem(ct->ct_record);
984				ct->ct_record = NULL;
985			}
986		}
987	} while (m);
988	ct->ct_upcallrefs--;
989	if (ct->ct_upcallrefs < 0)
990		panic("rpcvc upcall refcnt");
991	if (ct->ct_upcallrefs == 0)
992		wakeup(&ct->ct_upcallrefs);
993	return (SU_OK);
994}
995
996/*
997 * Wait for all upcalls in progress to complete.
998 */
999static void
1000clnt_vc_upcallsdone(struct ct_data *ct)
1001{
1002
1003	SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1004
1005	while (ct->ct_upcallrefs > 0)
1006		(void) msleep(&ct->ct_upcallrefs,
1007		    SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1008}
1009