clnt_vc.c revision 196503
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 196503 2009-08-24 10:09:30Z zec $");
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
74#include <net/vnet.h>
75
76#include <netinet/tcp.h>
77
78#include <rpc/rpc.h>
79#include <rpc/rpc_com.h>
80
81#define MCALL_MSG_SIZE 24
82
83struct cmessage {
84        struct cmsghdr cmsg;
85        struct cmsgcred cmcred;
86};
87
88static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
89    rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
90static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
91static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
92static void clnt_vc_abort(CLIENT *);
93static bool_t clnt_vc_control(CLIENT *, u_int, void *);
94static void clnt_vc_close(CLIENT *);
95static void clnt_vc_destroy(CLIENT *);
96static bool_t time_not_ok(struct timeval *);
97static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
98
99static struct clnt_ops clnt_vc_ops = {
100	.cl_call =	clnt_vc_call,
101	.cl_abort =	clnt_vc_abort,
102	.cl_geterr =	clnt_vc_geterr,
103	.cl_freeres =	clnt_vc_freeres,
104	.cl_close =	clnt_vc_close,
105	.cl_destroy =	clnt_vc_destroy,
106	.cl_control =	clnt_vc_control
107};
108
109/*
110 * A pending RPC request which awaits a reply. Requests which have
111 * received their reply will have cr_xid set to zero and cr_mrep to
112 * the mbuf chain of the reply.
113 */
114struct ct_request {
115	TAILQ_ENTRY(ct_request) cr_link;
116	uint32_t		cr_xid;		/* XID of request */
117	struct mbuf		*cr_mrep;	/* reply received by upcall */
118	int			cr_error;	/* any error from upcall */
119	char			cr_verf[MAX_AUTH_BYTES]; /* reply verf */
120};
121
122TAILQ_HEAD(ct_request_list, ct_request);
123
124struct ct_data {
125	struct mtx	ct_lock;
126	int		ct_threads;	/* number of threads in clnt_vc_call */
127	bool_t		ct_closing;	/* TRUE if we are closing */
128	bool_t		ct_closed;	/* TRUE if we are closed */
129	struct socket	*ct_socket;	/* connection socket */
130	bool_t		ct_closeit;	/* close it on destroy */
131	struct timeval	ct_wait;	/* wait interval in milliseconds */
132	struct sockaddr_storage	ct_addr; /* remote addr */
133	struct rpc_err	ct_error;
134	uint32_t	ct_xid;
135	char		ct_mcallc[MCALL_MSG_SIZE]; /* marshalled callmsg */
136	size_t		ct_mpos;	/* pos after marshal */
137	const char	*ct_waitchan;
138	int		ct_waitflag;
139	struct mbuf	*ct_record;	/* current reply record */
140	size_t		ct_record_resid; /* how much left of reply to read */
141	bool_t		ct_record_eor;	 /* true if reading last fragment */
142	struct ct_request_list ct_pending;
143	int		ct_upcallrefs;	/* Ref cnt of upcalls in prog. */
144};
145
146static void clnt_vc_upcallsdone(struct ct_data *);
147
148static const char clnt_vc_errstr[] = "%s : %s";
149static const char clnt_vc_str[] = "clnt_vc_create";
150static const char clnt_read_vc_str[] = "read_vc";
151static const char __no_mem_str[] = "out of memory";
152
153/*
154 * Create a client handle for a connection.
155 * Default options are set, which the user can change using clnt_control()'s.
156 * The rpc/vc package does buffering similar to stdio, so the client
157 * must pick send and receive buffer sizes, 0 => use the default.
158 * NB: fd is copied into a private area.
159 * NB: The rpch->cl_auth is set null authentication. Caller may wish to
160 * set this something more useful.
161 *
162 * fd should be an open socket
163 */
164CLIENT *
165clnt_vc_create(
166	struct socket *so,		/* open file descriptor */
167	struct sockaddr *raddr,		/* servers address */
168	const rpcprog_t prog,		/* program number */
169	const rpcvers_t vers,		/* version number */
170	size_t sendsz,			/* buffer recv size */
171	size_t recvsz)			/* buffer send size */
172{
173	CLIENT *cl;			/* client handle */
174	struct ct_data *ct = NULL;	/* client handle */
175	struct timeval now;
176	struct rpc_msg call_msg;
177	static uint32_t disrupt;
178	struct __rpc_sockinfo si;
179	XDR xdrs;
180	int error, interrupted, one = 1;
181	struct sockopt sopt;
182
183	if (disrupt == 0)
184		disrupt = (uint32_t)(long)raddr;
185
186	cl = (CLIENT *)mem_alloc(sizeof (*cl));
187	ct = (struct ct_data *)mem_alloc(sizeof (*ct));
188
189	mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
190	ct->ct_threads = 0;
191	ct->ct_closing = FALSE;
192	ct->ct_closed = FALSE;
193	ct->ct_upcallrefs = 0;
194
195	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
196		error = soconnect(so, raddr, curthread);
197		SOCK_LOCK(so);
198		interrupted = 0;
199		while ((so->so_state & SS_ISCONNECTING)
200		    && so->so_error == 0) {
201			error = msleep(&so->so_timeo, SOCK_MTX(so),
202			    PSOCK | PCATCH | PBDRY, "connec", 0);
203			if (error) {
204				if (error == EINTR || error == ERESTART)
205					interrupted = 1;
206				break;
207			}
208		}
209		if (error == 0) {
210			error = so->so_error;
211			so->so_error = 0;
212		}
213		SOCK_UNLOCK(so);
214		if (error) {
215			if (!interrupted)
216				so->so_state &= ~SS_ISCONNECTING;
217			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
218			rpc_createerr.cf_error.re_errno = error;
219			goto err;
220		}
221	}
222
223	CURVNET_SET(so->so_vnet);
224	if (!__rpc_socket2sockinfo(so, &si)) {
225		CURVNET_RESTORE();
226		goto err;
227	}
228
229	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
230		bzero(&sopt, sizeof(sopt));
231		sopt.sopt_dir = SOPT_SET;
232		sopt.sopt_level = SOL_SOCKET;
233		sopt.sopt_name = SO_KEEPALIVE;
234		sopt.sopt_val = &one;
235		sopt.sopt_valsize = sizeof(one);
236		sosetopt(so, &sopt);
237	}
238
239	if (so->so_proto->pr_protocol == IPPROTO_TCP) {
240		bzero(&sopt, sizeof(sopt));
241		sopt.sopt_dir = SOPT_SET;
242		sopt.sopt_level = IPPROTO_TCP;
243		sopt.sopt_name = TCP_NODELAY;
244		sopt.sopt_val = &one;
245		sopt.sopt_valsize = sizeof(one);
246		sosetopt(so, &sopt);
247	}
248	CURVNET_RESTORE();
249
250	ct->ct_closeit = FALSE;
251
252	/*
253	 * Set up private data struct
254	 */
255	ct->ct_socket = so;
256	ct->ct_wait.tv_sec = -1;
257	ct->ct_wait.tv_usec = -1;
258	memcpy(&ct->ct_addr, raddr, raddr->sa_len);
259
260	/*
261	 * Initialize call message
262	 */
263	getmicrotime(&now);
264	ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
265	call_msg.rm_xid = ct->ct_xid;
266	call_msg.rm_direction = CALL;
267	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
268	call_msg.rm_call.cb_prog = (uint32_t)prog;
269	call_msg.rm_call.cb_vers = (uint32_t)vers;
270
271	/*
272	 * pre-serialize the static part of the call msg and stash it away
273	 */
274	xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
275	    XDR_ENCODE);
276	if (! xdr_callhdr(&xdrs, &call_msg)) {
277		if (ct->ct_closeit) {
278			soclose(ct->ct_socket);
279		}
280		goto err;
281	}
282	ct->ct_mpos = XDR_GETPOS(&xdrs);
283	XDR_DESTROY(&xdrs);
284	ct->ct_waitchan = "rpcrecv";
285	ct->ct_waitflag = 0;
286
287	/*
288	 * Create a client handle which uses xdrrec for serialization
289	 * and authnone for authentication.
290	 */
291	cl->cl_refs = 1;
292	cl->cl_ops = &clnt_vc_ops;
293	cl->cl_private = ct;
294	cl->cl_auth = authnone_create();
295	sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
296	recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
297	soreserve(ct->ct_socket, sendsz, recvsz);
298
299	SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
300	soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
301	SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
302
303	ct->ct_record = NULL;
304	ct->ct_record_resid = 0;
305	TAILQ_INIT(&ct->ct_pending);
306	return (cl);
307
308err:
309	if (cl) {
310		if (ct) {
311			mtx_destroy(&ct->ct_lock);
312			mem_free(ct, sizeof (struct ct_data));
313		}
314		if (cl)
315			mem_free(cl, sizeof (CLIENT));
316	}
317	return ((CLIENT *)NULL);
318}
319
320static enum clnt_stat
321clnt_vc_call(
322	CLIENT		*cl,		/* client handle */
323	struct rpc_callextra *ext,	/* call metadata */
324	rpcproc_t	proc,		/* procedure number */
325	struct mbuf	*args,		/* pointer to args */
326	struct mbuf	**resultsp,	/* pointer to results */
327	struct timeval	utimeout)
328{
329	struct ct_data *ct = (struct ct_data *) cl->cl_private;
330	AUTH *auth;
331	struct rpc_err *errp;
332	enum clnt_stat stat;
333	XDR xdrs;
334	struct rpc_msg reply_msg;
335	bool_t ok;
336	int nrefreshes = 2;		/* number of times to refresh cred */
337	struct timeval timeout;
338	uint32_t xid;
339	struct mbuf *mreq = NULL, *results;
340	struct ct_request *cr;
341	int error;
342
343	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
344
345	mtx_lock(&ct->ct_lock);
346
347	if (ct->ct_closing || ct->ct_closed) {
348		mtx_unlock(&ct->ct_lock);
349		free(cr, M_RPC);
350		return (RPC_CANTSEND);
351	}
352	ct->ct_threads++;
353
354	if (ext) {
355		auth = ext->rc_auth;
356		errp = &ext->rc_err;
357	} else {
358		auth = cl->cl_auth;
359		errp = &ct->ct_error;
360	}
361
362	cr->cr_mrep = NULL;
363	cr->cr_error = 0;
364
365	if (ct->ct_wait.tv_usec == -1) {
366		timeout = utimeout;	/* use supplied timeout */
367	} else {
368		timeout = ct->ct_wait;	/* use default timeout */
369	}
370
371call_again:
372	mtx_assert(&ct->ct_lock, MA_OWNED);
373
374	ct->ct_xid++;
375	xid = ct->ct_xid;
376
377	mtx_unlock(&ct->ct_lock);
378
379	/*
380	 * Leave space to pre-pend the record mark.
381	 */
382	MGETHDR(mreq, M_WAIT, MT_DATA);
383	mreq->m_data += sizeof(uint32_t);
384	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
385	    ("RPC header too big"));
386	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
387	mreq->m_len = ct->ct_mpos;
388
389	/*
390	 * The XID is the first thing in the request.
391	 */
392	*mtod(mreq, uint32_t *) = htonl(xid);
393
394	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
395
396	errp->re_status = stat = RPC_SUCCESS;
397
398	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
399	    (! AUTH_MARSHALL(auth, xid, &xdrs,
400		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
401		errp->re_status = stat = RPC_CANTENCODEARGS;
402		mtx_lock(&ct->ct_lock);
403		goto out;
404	}
405	mreq->m_pkthdr.len = m_length(mreq, NULL);
406
407	/*
408	 * Prepend a record marker containing the packet length.
409	 */
410	M_PREPEND(mreq, sizeof(uint32_t), M_WAIT);
411	*mtod(mreq, uint32_t *) =
412		htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
413
414	cr->cr_xid = xid;
415	mtx_lock(&ct->ct_lock);
416	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
417	mtx_unlock(&ct->ct_lock);
418
419	/*
420	 * sosend consumes mreq.
421	 */
422	error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
423	mreq = NULL;
424	if (error == EMSGSIZE) {
425		SOCKBUF_LOCK(&ct->ct_socket->so_snd);
426		sbwait(&ct->ct_socket->so_snd);
427		SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
428		AUTH_VALIDATE(auth, xid, NULL, NULL);
429		mtx_lock(&ct->ct_lock);
430		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
431		goto call_again;
432	}
433
434	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
435	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
436	reply_msg.acpted_rply.ar_verf.oa_length = 0;
437	reply_msg.acpted_rply.ar_results.where = NULL;
438	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
439
440	mtx_lock(&ct->ct_lock);
441	if (error) {
442		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
443		errp->re_errno = error;
444		errp->re_status = stat = RPC_CANTSEND;
445		goto out;
446	}
447
448	/*
449	 * Check to see if we got an upcall while waiting for the
450	 * lock. In both these cases, the request has been removed
451	 * from ct->ct_pending.
452	 */
453	if (cr->cr_error) {
454		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
455		errp->re_errno = cr->cr_error;
456		errp->re_status = stat = RPC_CANTRECV;
457		goto out;
458	}
459	if (cr->cr_mrep) {
460		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
461		goto got_reply;
462	}
463
464	/*
465	 * Hack to provide rpc-based message passing
466	 */
467	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
468		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
469		errp->re_status = stat = RPC_TIMEDOUT;
470		goto out;
471	}
472
473	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
474	    tvtohz(&timeout));
475
476	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
477
478	if (error) {
479		/*
480		 * The sleep returned an error so our request is still
481		 * on the list. Turn the error code into an
482		 * appropriate client status.
483		 */
484		errp->re_errno = error;
485		switch (error) {
486		case EINTR:
487		case ERESTART:
488			stat = RPC_INTR;
489			break;
490		case EWOULDBLOCK:
491			stat = RPC_TIMEDOUT;
492			break;
493		default:
494			stat = RPC_CANTRECV;
495		}
496		errp->re_status = stat;
497		goto out;
498	} else {
499		/*
500		 * We were woken up by the upcall.  If the
501		 * upcall had a receive error, report that,
502		 * otherwise we have a reply.
503		 */
504		if (cr->cr_error) {
505			errp->re_errno = cr->cr_error;
506			errp->re_status = stat = RPC_CANTRECV;
507			goto out;
508		}
509	}
510
511got_reply:
512	/*
513	 * Now decode and validate the response. We need to drop the
514	 * lock since xdr_replymsg may end up sleeping in malloc.
515	 */
516	mtx_unlock(&ct->ct_lock);
517
518	if (ext && ext->rc_feedback)
519		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
520
521	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
522	ok = xdr_replymsg(&xdrs, &reply_msg);
523	cr->cr_mrep = NULL;
524
525	if (ok) {
526		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
527		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
528			errp->re_status = stat = RPC_SUCCESS;
529		else
530			stat = _seterr_reply(&reply_msg, errp);
531
532		if (stat == RPC_SUCCESS) {
533			results = xdrmbuf_getall(&xdrs);
534			if (!AUTH_VALIDATE(auth, xid,
535				&reply_msg.acpted_rply.ar_verf,
536				&results)) {
537				errp->re_status = stat = RPC_AUTHERROR;
538				errp->re_why = AUTH_INVALIDRESP;
539			} else {
540				KASSERT(results,
541				    ("auth validated but no result"));
542				*resultsp = results;
543			}
544		}		/* end successful completion */
545		/*
546		 * If unsuccesful AND error is an authentication error
547		 * then refresh credentials and try again, else break
548		 */
549		else if (stat == RPC_AUTHERROR)
550			/* maybe our credentials need to be refreshed ... */
551			if (nrefreshes > 0 &&
552			    AUTH_REFRESH(auth, &reply_msg)) {
553				nrefreshes--;
554				XDR_DESTROY(&xdrs);
555				mtx_lock(&ct->ct_lock);
556				goto call_again;
557			}
558		/* end of unsuccessful completion */
559	}	/* end of valid reply message */
560	else {
561		errp->re_status = stat = RPC_CANTDECODERES;
562	}
563	XDR_DESTROY(&xdrs);
564	mtx_lock(&ct->ct_lock);
565out:
566	mtx_assert(&ct->ct_lock, MA_OWNED);
567
568	KASSERT(stat != RPC_SUCCESS || *resultsp,
569	    ("RPC_SUCCESS without reply"));
570
571	if (mreq)
572		m_freem(mreq);
573	if (cr->cr_mrep)
574		m_freem(cr->cr_mrep);
575
576	ct->ct_threads--;
577	if (ct->ct_closing)
578		wakeup(ct);
579
580	mtx_unlock(&ct->ct_lock);
581
582	if (auth && stat != RPC_SUCCESS)
583		AUTH_VALIDATE(auth, xid, NULL, NULL);
584
585	free(cr, M_RPC);
586
587	return (stat);
588}
589
590static void
591clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
592{
593	struct ct_data *ct = (struct ct_data *) cl->cl_private;
594
595	*errp = ct->ct_error;
596}
597
598static bool_t
599clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
600{
601	XDR xdrs;
602	bool_t dummy;
603
604	xdrs.x_op = XDR_FREE;
605	dummy = (*xdr_res)(&xdrs, res_ptr);
606
607	return (dummy);
608}
609
610/*ARGSUSED*/
611static void
612clnt_vc_abort(CLIENT *cl)
613{
614}
615
616static bool_t
617clnt_vc_control(CLIENT *cl, u_int request, void *info)
618{
619	struct ct_data *ct = (struct ct_data *)cl->cl_private;
620	void *infop = info;
621
622	mtx_lock(&ct->ct_lock);
623
624	switch (request) {
625	case CLSET_FD_CLOSE:
626		ct->ct_closeit = TRUE;
627		mtx_unlock(&ct->ct_lock);
628		return (TRUE);
629	case CLSET_FD_NCLOSE:
630		ct->ct_closeit = FALSE;
631		mtx_unlock(&ct->ct_lock);
632		return (TRUE);
633	default:
634		break;
635	}
636
637	/* for other requests which use info */
638	if (info == NULL) {
639		mtx_unlock(&ct->ct_lock);
640		return (FALSE);
641	}
642	switch (request) {
643	case CLSET_TIMEOUT:
644		if (time_not_ok((struct timeval *)info)) {
645			mtx_unlock(&ct->ct_lock);
646			return (FALSE);
647		}
648		ct->ct_wait = *(struct timeval *)infop;
649		break;
650	case CLGET_TIMEOUT:
651		*(struct timeval *)infop = ct->ct_wait;
652		break;
653	case CLGET_SERVER_ADDR:
654		(void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
655		break;
656	case CLGET_SVC_ADDR:
657		/*
658		 * Slightly different semantics to userland - we use
659		 * sockaddr instead of netbuf.
660		 */
661		memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
662		break;
663	case CLSET_SVC_ADDR:		/* set to new address */
664		mtx_unlock(&ct->ct_lock);
665		return (FALSE);
666	case CLGET_XID:
667		*(uint32_t *)info = ct->ct_xid;
668		break;
669	case CLSET_XID:
670		/* This will set the xid of the NEXT call */
671		/* decrement by 1 as clnt_vc_call() increments once */
672		ct->ct_xid = *(uint32_t *)info - 1;
673		break;
674	case CLGET_VERS:
675		/*
676		 * This RELIES on the information that, in the call body,
677		 * the version number field is the fifth field from the
678		 * begining of the RPC header. MUST be changed if the
679		 * call_struct is changed
680		 */
681		*(uint32_t *)info =
682		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
683		    4 * BYTES_PER_XDR_UNIT));
684		break;
685
686	case CLSET_VERS:
687		*(uint32_t *)(void *)(ct->ct_mcallc +
688		    4 * BYTES_PER_XDR_UNIT) =
689		    htonl(*(uint32_t *)info);
690		break;
691
692	case CLGET_PROG:
693		/*
694		 * This RELIES on the information that, in the call body,
695		 * the program number field is the fourth field from the
696		 * begining of the RPC header. MUST be changed if the
697		 * call_struct is changed
698		 */
699		*(uint32_t *)info =
700		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
701		    3 * BYTES_PER_XDR_UNIT));
702		break;
703
704	case CLSET_PROG:
705		*(uint32_t *)(void *)(ct->ct_mcallc +
706		    3 * BYTES_PER_XDR_UNIT) =
707		    htonl(*(uint32_t *)info);
708		break;
709
710	case CLSET_WAITCHAN:
711		ct->ct_waitchan = (const char *)info;
712		break;
713
714	case CLGET_WAITCHAN:
715		*(const char **) info = ct->ct_waitchan;
716		break;
717
718	case CLSET_INTERRUPTIBLE:
719		if (*(int *) info)
720			ct->ct_waitflag = PCATCH | PBDRY;
721		else
722			ct->ct_waitflag = 0;
723		break;
724
725	case CLGET_INTERRUPTIBLE:
726		if (ct->ct_waitflag)
727			*(int *) info = TRUE;
728		else
729			*(int *) info = FALSE;
730		break;
731
732	default:
733		mtx_unlock(&ct->ct_lock);
734		return (FALSE);
735	}
736
737	mtx_unlock(&ct->ct_lock);
738	return (TRUE);
739}
740
741static void
742clnt_vc_close(CLIENT *cl)
743{
744	struct ct_data *ct = (struct ct_data *) cl->cl_private;
745	struct ct_request *cr;
746
747	mtx_lock(&ct->ct_lock);
748
749	if (ct->ct_closed) {
750		mtx_unlock(&ct->ct_lock);
751		return;
752	}
753
754	if (ct->ct_closing) {
755		while (ct->ct_closing)
756			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
757		KASSERT(ct->ct_closed, ("client should be closed"));
758		mtx_unlock(&ct->ct_lock);
759		return;
760	}
761
762	if (ct->ct_socket) {
763		ct->ct_closing = TRUE;
764		mtx_unlock(&ct->ct_lock);
765
766		SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
767		soupcall_clear(ct->ct_socket, SO_RCV);
768		clnt_vc_upcallsdone(ct);
769		SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
770
771		/*
772		 * Abort any pending requests and wait until everyone
773		 * has finished with clnt_vc_call.
774		 */
775		mtx_lock(&ct->ct_lock);
776		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
777			cr->cr_xid = 0;
778			cr->cr_error = ESHUTDOWN;
779			wakeup(cr);
780		}
781
782		while (ct->ct_threads)
783			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
784	}
785
786	ct->ct_closing = FALSE;
787	ct->ct_closed = TRUE;
788	mtx_unlock(&ct->ct_lock);
789	wakeup(ct);
790}
791
792static void
793clnt_vc_destroy(CLIENT *cl)
794{
795	struct ct_data *ct = (struct ct_data *) cl->cl_private;
796	struct socket *so = NULL;
797
798	clnt_vc_close(cl);
799
800	mtx_lock(&ct->ct_lock);
801
802	if (ct->ct_socket) {
803		if (ct->ct_closeit) {
804			so = ct->ct_socket;
805		}
806	}
807
808	mtx_unlock(&ct->ct_lock);
809
810	mtx_destroy(&ct->ct_lock);
811	if (so) {
812		soshutdown(so, SHUT_WR);
813		soclose(so);
814	}
815	mem_free(ct, sizeof(struct ct_data));
816	mem_free(cl, sizeof(CLIENT));
817}
818
819/*
820 * Make sure that the time is not garbage.   -1 value is disallowed.
821 * Note this is different from time_not_ok in clnt_dg.c
822 */
823static bool_t
824time_not_ok(struct timeval *t)
825{
826	return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
827		t->tv_usec <= -1 || t->tv_usec > 1000000);
828}
829
830int
831clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
832{
833	struct ct_data *ct = (struct ct_data *) arg;
834	struct uio uio;
835	struct mbuf *m;
836	struct ct_request *cr;
837	int error, rcvflag, foundreq;
838	uint32_t xid, header;
839	bool_t do_read;
840
841	ct->ct_upcallrefs++;
842	uio.uio_td = curthread;
843	do {
844		/*
845		 * If ct_record_resid is zero, we are waiting for a
846		 * record mark.
847		 */
848		if (ct->ct_record_resid == 0) {
849
850			/*
851			 * Make sure there is either a whole record
852			 * mark in the buffer or there is some other
853			 * error condition
854			 */
855			do_read = FALSE;
856			if (so->so_rcv.sb_cc >= sizeof(uint32_t)
857			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
858			    || so->so_error)
859				do_read = TRUE;
860
861			if (!do_read)
862				break;
863
864			SOCKBUF_UNLOCK(&so->so_rcv);
865			uio.uio_resid = sizeof(uint32_t);
866			m = NULL;
867			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
868			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
869			SOCKBUF_LOCK(&so->so_rcv);
870
871			if (error == EWOULDBLOCK)
872				break;
873
874			/*
875			 * If there was an error, wake up all pending
876			 * requests.
877			 */
878			if (error || uio.uio_resid > 0) {
879			wakeup_all:
880				mtx_lock(&ct->ct_lock);
881				if (!error) {
882					/*
883					 * We must have got EOF trying
884					 * to read from the stream.
885					 */
886					error = ECONNRESET;
887				}
888				ct->ct_error.re_status = RPC_CANTRECV;
889				ct->ct_error.re_errno = error;
890				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
891					cr->cr_error = error;
892					wakeup(cr);
893				}
894				mtx_unlock(&ct->ct_lock);
895				break;
896			}
897			bcopy(mtod(m, uint32_t *), &header, sizeof(uint32_t));
898			header = ntohl(header);
899			ct->ct_record = NULL;
900			ct->ct_record_resid = header & 0x7fffffff;
901			ct->ct_record_eor = ((header & 0x80000000) != 0);
902			m_freem(m);
903		} else {
904			/*
905			 * Wait until the socket has the whole record
906			 * buffered.
907			 */
908			do_read = FALSE;
909			if (so->so_rcv.sb_cc >= ct->ct_record_resid
910			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
911			    || so->so_error)
912				do_read = TRUE;
913
914			if (!do_read)
915				break;
916
917			/*
918			 * We have the record mark. Read as much as
919			 * the socket has buffered up to the end of
920			 * this record.
921			 */
922			SOCKBUF_UNLOCK(&so->so_rcv);
923			uio.uio_resid = ct->ct_record_resid;
924			m = NULL;
925			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
926			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
927			SOCKBUF_LOCK(&so->so_rcv);
928
929			if (error == EWOULDBLOCK)
930				break;
931
932			if (error || uio.uio_resid == ct->ct_record_resid)
933				goto wakeup_all;
934
935			/*
936			 * If we have part of the record already,
937			 * chain this bit onto the end.
938			 */
939			if (ct->ct_record)
940				m_last(ct->ct_record)->m_next = m;
941			else
942				ct->ct_record = m;
943
944			ct->ct_record_resid = uio.uio_resid;
945
946			/*
947			 * If we have the entire record, see if we can
948			 * match it to a request.
949			 */
950			if (ct->ct_record_resid == 0
951			    && ct->ct_record_eor) {
952				/*
953				 * The XID is in the first uint32_t of
954				 * the reply.
955				 */
956				if (ct->ct_record->m_len < sizeof(xid))
957					ct->ct_record =
958						m_pullup(ct->ct_record,
959						    sizeof(xid));
960				if (!ct->ct_record)
961					break;
962				bcopy(mtod(ct->ct_record, uint32_t *),
963				    &xid, sizeof(uint32_t));
964				xid = ntohl(xid);
965
966				mtx_lock(&ct->ct_lock);
967				foundreq = 0;
968				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
969					if (cr->cr_xid == xid) {
970						/*
971						 * This one
972						 * matches. We leave
973						 * the reply mbuf in
974						 * cr->cr_mrep. Set
975						 * the XID to zero so
976						 * that we will ignore
977						 * any duplicaed
978						 * replies.
979						 */
980						cr->cr_xid = 0;
981						cr->cr_mrep = ct->ct_record;
982						cr->cr_error = 0;
983						foundreq = 1;
984						wakeup(cr);
985						break;
986					}
987				}
988				mtx_unlock(&ct->ct_lock);
989
990				if (!foundreq)
991					m_freem(ct->ct_record);
992				ct->ct_record = NULL;
993			}
994		}
995	} while (m);
996	ct->ct_upcallrefs--;
997	if (ct->ct_upcallrefs < 0)
998		panic("rpcvc upcall refcnt");
999	if (ct->ct_upcallrefs == 0)
1000		wakeup(&ct->ct_upcallrefs);
1001	return (SU_OK);
1002}
1003
1004/*
1005 * Wait for all upcalls in progress to complete.
1006 */
1007static void
1008clnt_vc_upcallsdone(struct ct_data *ct)
1009{
1010
1011	SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1012
1013	while (ct->ct_upcallrefs > 0)
1014		(void) msleep(&ct->ct_upcallrefs,
1015		    SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1016}
1017