clnt_vc.c revision 218757
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 218757 2011-02-16 21:29:13Z bz $");
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	if (!__rpc_socket2sockinfo(so, &si)) {
224		goto err;
225	}
226
227	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
228		bzero(&sopt, sizeof(sopt));
229		sopt.sopt_dir = SOPT_SET;
230		sopt.sopt_level = SOL_SOCKET;
231		sopt.sopt_name = SO_KEEPALIVE;
232		sopt.sopt_val = &one;
233		sopt.sopt_valsize = sizeof(one);
234		sosetopt(so, &sopt);
235	}
236
237	if (so->so_proto->pr_protocol == IPPROTO_TCP) {
238		bzero(&sopt, sizeof(sopt));
239		sopt.sopt_dir = SOPT_SET;
240		sopt.sopt_level = IPPROTO_TCP;
241		sopt.sopt_name = TCP_NODELAY;
242		sopt.sopt_val = &one;
243		sopt.sopt_valsize = sizeof(one);
244		sosetopt(so, &sopt);
245	}
246
247	ct->ct_closeit = FALSE;
248
249	/*
250	 * Set up private data struct
251	 */
252	ct->ct_socket = so;
253	ct->ct_wait.tv_sec = -1;
254	ct->ct_wait.tv_usec = -1;
255	memcpy(&ct->ct_addr, raddr, raddr->sa_len);
256
257	/*
258	 * Initialize call message
259	 */
260	getmicrotime(&now);
261	ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
262	call_msg.rm_xid = ct->ct_xid;
263	call_msg.rm_direction = CALL;
264	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
265	call_msg.rm_call.cb_prog = (uint32_t)prog;
266	call_msg.rm_call.cb_vers = (uint32_t)vers;
267
268	/*
269	 * pre-serialize the static part of the call msg and stash it away
270	 */
271	xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
272	    XDR_ENCODE);
273	if (! xdr_callhdr(&xdrs, &call_msg)) {
274		if (ct->ct_closeit) {
275			soclose(ct->ct_socket);
276		}
277		goto err;
278	}
279	ct->ct_mpos = XDR_GETPOS(&xdrs);
280	XDR_DESTROY(&xdrs);
281	ct->ct_waitchan = "rpcrecv";
282	ct->ct_waitflag = 0;
283
284	/*
285	 * Create a client handle which uses xdrrec for serialization
286	 * and authnone for authentication.
287	 */
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	error = soreserve(ct->ct_socket, sendsz, recvsz);
291	if (error != 0) {
292		if (ct->ct_closeit) {
293			soclose(ct->ct_socket);
294		}
295		goto err;
296	}
297	cl->cl_refs = 1;
298	cl->cl_ops = &clnt_vc_ops;
299	cl->cl_private = ct;
300	cl->cl_auth = authnone_create();
301
302	SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
303	soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
304	SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
305
306	ct->ct_record = NULL;
307	ct->ct_record_resid = 0;
308	TAILQ_INIT(&ct->ct_pending);
309	return (cl);
310
311err:
312	if (cl) {
313		if (ct) {
314			mtx_destroy(&ct->ct_lock);
315			mem_free(ct, sizeof (struct ct_data));
316		}
317		if (cl)
318			mem_free(cl, sizeof (CLIENT));
319	}
320	return ((CLIENT *)NULL);
321}
322
323static enum clnt_stat
324clnt_vc_call(
325	CLIENT		*cl,		/* client handle */
326	struct rpc_callextra *ext,	/* call metadata */
327	rpcproc_t	proc,		/* procedure number */
328	struct mbuf	*args,		/* pointer to args */
329	struct mbuf	**resultsp,	/* pointer to results */
330	struct timeval	utimeout)
331{
332	struct ct_data *ct = (struct ct_data *) cl->cl_private;
333	AUTH *auth;
334	struct rpc_err *errp;
335	enum clnt_stat stat;
336	XDR xdrs;
337	struct rpc_msg reply_msg;
338	bool_t ok;
339	int nrefreshes = 2;		/* number of times to refresh cred */
340	struct timeval timeout;
341	uint32_t xid;
342	struct mbuf *mreq = NULL, *results;
343	struct ct_request *cr;
344	int error;
345
346	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
347
348	mtx_lock(&ct->ct_lock);
349
350	if (ct->ct_closing || ct->ct_closed) {
351		mtx_unlock(&ct->ct_lock);
352		free(cr, M_RPC);
353		return (RPC_CANTSEND);
354	}
355	ct->ct_threads++;
356
357	if (ext) {
358		auth = ext->rc_auth;
359		errp = &ext->rc_err;
360	} else {
361		auth = cl->cl_auth;
362		errp = &ct->ct_error;
363	}
364
365	cr->cr_mrep = NULL;
366	cr->cr_error = 0;
367
368	if (ct->ct_wait.tv_usec == -1) {
369		timeout = utimeout;	/* use supplied timeout */
370	} else {
371		timeout = ct->ct_wait;	/* use default timeout */
372	}
373
374call_again:
375	mtx_assert(&ct->ct_lock, MA_OWNED);
376
377	ct->ct_xid++;
378	xid = ct->ct_xid;
379
380	mtx_unlock(&ct->ct_lock);
381
382	/*
383	 * Leave space to pre-pend the record mark.
384	 */
385	MGETHDR(mreq, M_WAIT, MT_DATA);
386	mreq->m_data += sizeof(uint32_t);
387	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
388	    ("RPC header too big"));
389	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
390	mreq->m_len = ct->ct_mpos;
391
392	/*
393	 * The XID is the first thing in the request.
394	 */
395	*mtod(mreq, uint32_t *) = htonl(xid);
396
397	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
398
399	errp->re_status = stat = RPC_SUCCESS;
400
401	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
402	    (! AUTH_MARSHALL(auth, xid, &xdrs,
403		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
404		errp->re_status = stat = RPC_CANTENCODEARGS;
405		mtx_lock(&ct->ct_lock);
406		goto out;
407	}
408	mreq->m_pkthdr.len = m_length(mreq, NULL);
409
410	/*
411	 * Prepend a record marker containing the packet length.
412	 */
413	M_PREPEND(mreq, sizeof(uint32_t), M_WAIT);
414	*mtod(mreq, uint32_t *) =
415		htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
416
417	cr->cr_xid = xid;
418	mtx_lock(&ct->ct_lock);
419	/*
420	 * Check to see if the other end has already started to close down
421	 * the connection. The upcall will have set ct_error.re_status
422	 * to RPC_CANTRECV if this is the case.
423	 * If the other end starts to close down the connection after this
424	 * point, it will be detected later when cr_error is checked,
425	 * since the request is in the ct_pending queue.
426	 */
427	if (ct->ct_error.re_status == RPC_CANTRECV) {
428		if (errp != &ct->ct_error) {
429			errp->re_errno = ct->ct_error.re_errno;
430			errp->re_status = RPC_CANTRECV;
431		}
432		stat = RPC_CANTRECV;
433		goto out;
434	}
435	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
436	mtx_unlock(&ct->ct_lock);
437
438	/*
439	 * sosend consumes mreq.
440	 */
441	error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
442	mreq = NULL;
443	if (error == EMSGSIZE) {
444		SOCKBUF_LOCK(&ct->ct_socket->so_snd);
445		sbwait(&ct->ct_socket->so_snd);
446		SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
447		AUTH_VALIDATE(auth, xid, NULL, NULL);
448		mtx_lock(&ct->ct_lock);
449		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
450		goto call_again;
451	}
452
453	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
454	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
455	reply_msg.acpted_rply.ar_verf.oa_length = 0;
456	reply_msg.acpted_rply.ar_results.where = NULL;
457	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
458
459	mtx_lock(&ct->ct_lock);
460	if (error) {
461		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
462		errp->re_errno = error;
463		errp->re_status = stat = RPC_CANTSEND;
464		goto out;
465	}
466
467	/*
468	 * Check to see if we got an upcall while waiting for the
469	 * lock. In both these cases, the request has been removed
470	 * from ct->ct_pending.
471	 */
472	if (cr->cr_error) {
473		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
474		errp->re_errno = cr->cr_error;
475		errp->re_status = stat = RPC_CANTRECV;
476		goto out;
477	}
478	if (cr->cr_mrep) {
479		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
480		goto got_reply;
481	}
482
483	/*
484	 * Hack to provide rpc-based message passing
485	 */
486	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
487		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
488		errp->re_status = stat = RPC_TIMEDOUT;
489		goto out;
490	}
491
492	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
493	    tvtohz(&timeout));
494
495	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
496
497	if (error) {
498		/*
499		 * The sleep returned an error so our request is still
500		 * on the list. Turn the error code into an
501		 * appropriate client status.
502		 */
503		errp->re_errno = error;
504		switch (error) {
505		case EINTR:
506		case ERESTART:
507			stat = RPC_INTR;
508			break;
509		case EWOULDBLOCK:
510			stat = RPC_TIMEDOUT;
511			break;
512		default:
513			stat = RPC_CANTRECV;
514		}
515		errp->re_status = stat;
516		goto out;
517	} else {
518		/*
519		 * We were woken up by the upcall.  If the
520		 * upcall had a receive error, report that,
521		 * otherwise we have a reply.
522		 */
523		if (cr->cr_error) {
524			errp->re_errno = cr->cr_error;
525			errp->re_status = stat = RPC_CANTRECV;
526			goto out;
527		}
528	}
529
530got_reply:
531	/*
532	 * Now decode and validate the response. We need to drop the
533	 * lock since xdr_replymsg may end up sleeping in malloc.
534	 */
535	mtx_unlock(&ct->ct_lock);
536
537	if (ext && ext->rc_feedback)
538		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
539
540	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
541	ok = xdr_replymsg(&xdrs, &reply_msg);
542	cr->cr_mrep = NULL;
543
544	if (ok) {
545		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
546		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
547			errp->re_status = stat = RPC_SUCCESS;
548		else
549			stat = _seterr_reply(&reply_msg, errp);
550
551		if (stat == RPC_SUCCESS) {
552			results = xdrmbuf_getall(&xdrs);
553			if (!AUTH_VALIDATE(auth, xid,
554				&reply_msg.acpted_rply.ar_verf,
555				&results)) {
556				errp->re_status = stat = RPC_AUTHERROR;
557				errp->re_why = AUTH_INVALIDRESP;
558			} else {
559				KASSERT(results,
560				    ("auth validated but no result"));
561				*resultsp = results;
562			}
563		}		/* end successful completion */
564		/*
565		 * If unsuccesful AND error is an authentication error
566		 * then refresh credentials and try again, else break
567		 */
568		else if (stat == RPC_AUTHERROR)
569			/* maybe our credentials need to be refreshed ... */
570			if (nrefreshes > 0 &&
571			    AUTH_REFRESH(auth, &reply_msg)) {
572				nrefreshes--;
573				XDR_DESTROY(&xdrs);
574				mtx_lock(&ct->ct_lock);
575				goto call_again;
576			}
577		/* end of unsuccessful completion */
578	}	/* end of valid reply message */
579	else {
580		errp->re_status = stat = RPC_CANTDECODERES;
581	}
582	XDR_DESTROY(&xdrs);
583	mtx_lock(&ct->ct_lock);
584out:
585	mtx_assert(&ct->ct_lock, MA_OWNED);
586
587	KASSERT(stat != RPC_SUCCESS || *resultsp,
588	    ("RPC_SUCCESS without reply"));
589
590	if (mreq)
591		m_freem(mreq);
592	if (cr->cr_mrep)
593		m_freem(cr->cr_mrep);
594
595	ct->ct_threads--;
596	if (ct->ct_closing)
597		wakeup(ct);
598
599	mtx_unlock(&ct->ct_lock);
600
601	if (auth && stat != RPC_SUCCESS)
602		AUTH_VALIDATE(auth, xid, NULL, NULL);
603
604	free(cr, M_RPC);
605
606	return (stat);
607}
608
609static void
610clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
611{
612	struct ct_data *ct = (struct ct_data *) cl->cl_private;
613
614	*errp = ct->ct_error;
615}
616
617static bool_t
618clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
619{
620	XDR xdrs;
621	bool_t dummy;
622
623	xdrs.x_op = XDR_FREE;
624	dummy = (*xdr_res)(&xdrs, res_ptr);
625
626	return (dummy);
627}
628
629/*ARGSUSED*/
630static void
631clnt_vc_abort(CLIENT *cl)
632{
633}
634
635static bool_t
636clnt_vc_control(CLIENT *cl, u_int request, void *info)
637{
638	struct ct_data *ct = (struct ct_data *)cl->cl_private;
639	void *infop = info;
640
641	mtx_lock(&ct->ct_lock);
642
643	switch (request) {
644	case CLSET_FD_CLOSE:
645		ct->ct_closeit = TRUE;
646		mtx_unlock(&ct->ct_lock);
647		return (TRUE);
648	case CLSET_FD_NCLOSE:
649		ct->ct_closeit = FALSE;
650		mtx_unlock(&ct->ct_lock);
651		return (TRUE);
652	default:
653		break;
654	}
655
656	/* for other requests which use info */
657	if (info == NULL) {
658		mtx_unlock(&ct->ct_lock);
659		return (FALSE);
660	}
661	switch (request) {
662	case CLSET_TIMEOUT:
663		if (time_not_ok((struct timeval *)info)) {
664			mtx_unlock(&ct->ct_lock);
665			return (FALSE);
666		}
667		ct->ct_wait = *(struct timeval *)infop;
668		break;
669	case CLGET_TIMEOUT:
670		*(struct timeval *)infop = ct->ct_wait;
671		break;
672	case CLGET_SERVER_ADDR:
673		(void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
674		break;
675	case CLGET_SVC_ADDR:
676		/*
677		 * Slightly different semantics to userland - we use
678		 * sockaddr instead of netbuf.
679		 */
680		memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
681		break;
682	case CLSET_SVC_ADDR:		/* set to new address */
683		mtx_unlock(&ct->ct_lock);
684		return (FALSE);
685	case CLGET_XID:
686		*(uint32_t *)info = ct->ct_xid;
687		break;
688	case CLSET_XID:
689		/* This will set the xid of the NEXT call */
690		/* decrement by 1 as clnt_vc_call() increments once */
691		ct->ct_xid = *(uint32_t *)info - 1;
692		break;
693	case CLGET_VERS:
694		/*
695		 * This RELIES on the information that, in the call body,
696		 * the version number field is the fifth field from the
697		 * begining of the RPC header. MUST be changed if the
698		 * call_struct is changed
699		 */
700		*(uint32_t *)info =
701		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
702		    4 * BYTES_PER_XDR_UNIT));
703		break;
704
705	case CLSET_VERS:
706		*(uint32_t *)(void *)(ct->ct_mcallc +
707		    4 * BYTES_PER_XDR_UNIT) =
708		    htonl(*(uint32_t *)info);
709		break;
710
711	case CLGET_PROG:
712		/*
713		 * This RELIES on the information that, in the call body,
714		 * the program number field is the fourth field from the
715		 * begining of the RPC header. MUST be changed if the
716		 * call_struct is changed
717		 */
718		*(uint32_t *)info =
719		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
720		    3 * BYTES_PER_XDR_UNIT));
721		break;
722
723	case CLSET_PROG:
724		*(uint32_t *)(void *)(ct->ct_mcallc +
725		    3 * BYTES_PER_XDR_UNIT) =
726		    htonl(*(uint32_t *)info);
727		break;
728
729	case CLSET_WAITCHAN:
730		ct->ct_waitchan = (const char *)info;
731		break;
732
733	case CLGET_WAITCHAN:
734		*(const char **) info = ct->ct_waitchan;
735		break;
736
737	case CLSET_INTERRUPTIBLE:
738		if (*(int *) info)
739			ct->ct_waitflag = PCATCH | PBDRY;
740		else
741			ct->ct_waitflag = 0;
742		break;
743
744	case CLGET_INTERRUPTIBLE:
745		if (ct->ct_waitflag)
746			*(int *) info = TRUE;
747		else
748			*(int *) info = FALSE;
749		break;
750
751	default:
752		mtx_unlock(&ct->ct_lock);
753		return (FALSE);
754	}
755
756	mtx_unlock(&ct->ct_lock);
757	return (TRUE);
758}
759
760static void
761clnt_vc_close(CLIENT *cl)
762{
763	struct ct_data *ct = (struct ct_data *) cl->cl_private;
764	struct ct_request *cr;
765
766	mtx_lock(&ct->ct_lock);
767
768	if (ct->ct_closed) {
769		mtx_unlock(&ct->ct_lock);
770		return;
771	}
772
773	if (ct->ct_closing) {
774		while (ct->ct_closing)
775			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
776		KASSERT(ct->ct_closed, ("client should be closed"));
777		mtx_unlock(&ct->ct_lock);
778		return;
779	}
780
781	if (ct->ct_socket) {
782		ct->ct_closing = TRUE;
783		mtx_unlock(&ct->ct_lock);
784
785		SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
786		soupcall_clear(ct->ct_socket, SO_RCV);
787		clnt_vc_upcallsdone(ct);
788		SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
789
790		/*
791		 * Abort any pending requests and wait until everyone
792		 * has finished with clnt_vc_call.
793		 */
794		mtx_lock(&ct->ct_lock);
795		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
796			cr->cr_xid = 0;
797			cr->cr_error = ESHUTDOWN;
798			wakeup(cr);
799		}
800
801		while (ct->ct_threads)
802			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
803	}
804
805	ct->ct_closing = FALSE;
806	ct->ct_closed = TRUE;
807	mtx_unlock(&ct->ct_lock);
808	wakeup(ct);
809}
810
811static void
812clnt_vc_destroy(CLIENT *cl)
813{
814	struct ct_data *ct = (struct ct_data *) cl->cl_private;
815	struct socket *so = NULL;
816
817	clnt_vc_close(cl);
818
819	mtx_lock(&ct->ct_lock);
820
821	if (ct->ct_socket) {
822		if (ct->ct_closeit) {
823			so = ct->ct_socket;
824		}
825	}
826
827	mtx_unlock(&ct->ct_lock);
828
829	mtx_destroy(&ct->ct_lock);
830	if (so) {
831		soshutdown(so, SHUT_WR);
832		soclose(so);
833	}
834	mem_free(ct, sizeof(struct ct_data));
835	mem_free(cl, sizeof(CLIENT));
836}
837
838/*
839 * Make sure that the time is not garbage.   -1 value is disallowed.
840 * Note this is different from time_not_ok in clnt_dg.c
841 */
842static bool_t
843time_not_ok(struct timeval *t)
844{
845	return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
846		t->tv_usec <= -1 || t->tv_usec > 1000000);
847}
848
849int
850clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
851{
852	struct ct_data *ct = (struct ct_data *) arg;
853	struct uio uio;
854	struct mbuf *m;
855	struct ct_request *cr;
856	int error, rcvflag, foundreq;
857	uint32_t xid, header;
858	bool_t do_read;
859
860	ct->ct_upcallrefs++;
861	uio.uio_td = curthread;
862	do {
863		/*
864		 * If ct_record_resid is zero, we are waiting for a
865		 * record mark.
866		 */
867		if (ct->ct_record_resid == 0) {
868
869			/*
870			 * Make sure there is either a whole record
871			 * mark in the buffer or there is some other
872			 * error condition
873			 */
874			do_read = FALSE;
875			if (so->so_rcv.sb_cc >= sizeof(uint32_t)
876			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
877			    || so->so_error)
878				do_read = TRUE;
879
880			if (!do_read)
881				break;
882
883			SOCKBUF_UNLOCK(&so->so_rcv);
884			uio.uio_resid = sizeof(uint32_t);
885			m = NULL;
886			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
887			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
888			SOCKBUF_LOCK(&so->so_rcv);
889
890			if (error == EWOULDBLOCK)
891				break;
892
893			/*
894			 * If there was an error, wake up all pending
895			 * requests.
896			 */
897			if (error || uio.uio_resid > 0) {
898			wakeup_all:
899				mtx_lock(&ct->ct_lock);
900				if (!error) {
901					/*
902					 * We must have got EOF trying
903					 * to read from the stream.
904					 */
905					error = ECONNRESET;
906				}
907				ct->ct_error.re_status = RPC_CANTRECV;
908				ct->ct_error.re_errno = error;
909				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
910					cr->cr_error = error;
911					wakeup(cr);
912				}
913				mtx_unlock(&ct->ct_lock);
914				break;
915			}
916			m_copydata(m, 0, sizeof(uint32_t), (char *)&header);
917			header = ntohl(header);
918			ct->ct_record = NULL;
919			ct->ct_record_resid = header & 0x7fffffff;
920			ct->ct_record_eor = ((header & 0x80000000) != 0);
921			m_freem(m);
922		} else {
923			/*
924			 * Wait until the socket has the whole record
925			 * buffered.
926			 */
927			do_read = FALSE;
928			if (so->so_rcv.sb_cc >= ct->ct_record_resid
929			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
930			    || so->so_error)
931				do_read = TRUE;
932
933			if (!do_read)
934				break;
935
936			/*
937			 * We have the record mark. Read as much as
938			 * the socket has buffered up to the end of
939			 * this record.
940			 */
941			SOCKBUF_UNLOCK(&so->so_rcv);
942			uio.uio_resid = ct->ct_record_resid;
943			m = NULL;
944			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
945			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
946			SOCKBUF_LOCK(&so->so_rcv);
947
948			if (error == EWOULDBLOCK)
949				break;
950
951			if (error || uio.uio_resid == ct->ct_record_resid)
952				goto wakeup_all;
953
954			/*
955			 * If we have part of the record already,
956			 * chain this bit onto the end.
957			 */
958			if (ct->ct_record)
959				m_last(ct->ct_record)->m_next = m;
960			else
961				ct->ct_record = m;
962
963			ct->ct_record_resid = uio.uio_resid;
964
965			/*
966			 * If we have the entire record, see if we can
967			 * match it to a request.
968			 */
969			if (ct->ct_record_resid == 0
970			    && ct->ct_record_eor) {
971				/*
972				 * The XID is in the first uint32_t of
973				 * the reply.
974				 */
975				if (ct->ct_record->m_len < sizeof(xid) &&
976				    m_length(ct->ct_record, NULL) < sizeof(xid))
977					break;
978				m_copydata(ct->ct_record, 0, sizeof(xid),
979				    (char *)&xid);
980				xid = ntohl(xid);
981
982				mtx_lock(&ct->ct_lock);
983				foundreq = 0;
984				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
985					if (cr->cr_xid == xid) {
986						/*
987						 * This one
988						 * matches. We leave
989						 * the reply mbuf in
990						 * cr->cr_mrep. Set
991						 * the XID to zero so
992						 * that we will ignore
993						 * any duplicaed
994						 * replies.
995						 */
996						cr->cr_xid = 0;
997						cr->cr_mrep = ct->ct_record;
998						cr->cr_error = 0;
999						foundreq = 1;
1000						wakeup(cr);
1001						break;
1002					}
1003				}
1004				mtx_unlock(&ct->ct_lock);
1005
1006				if (!foundreq)
1007					m_freem(ct->ct_record);
1008				ct->ct_record = NULL;
1009			}
1010		}
1011	} while (m);
1012	ct->ct_upcallrefs--;
1013	if (ct->ct_upcallrefs < 0)
1014		panic("rpcvc upcall refcnt");
1015	if (ct->ct_upcallrefs == 0)
1016		wakeup(&ct->ct_upcallrefs);
1017	return (SU_OK);
1018}
1019
1020/*
1021 * Wait for all upcalls in progress to complete.
1022 */
1023static void
1024clnt_vc_upcallsdone(struct ct_data *ct)
1025{
1026
1027	SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1028
1029	while (ct->ct_upcallrefs > 0)
1030		(void) msleep(&ct->ct_upcallrefs,
1031		    SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1032}
1033