xref: /freebsd/sys/rpc/clnt_vc.c (revision b0b1dbdd)
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)
32 static char *sccsid2 = "@(#)clnt_tcp.c 1.37 87/10/05 Copyr 1984 Sun Micro";
33 static char *sccsid = "@(#)clnt_tcp.c	2.2 88/08/01 4.0 RPCSRC";
34 static 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 
82 struct cmessage {
83         struct cmsghdr cmsg;
84         struct cmsgcred cmcred;
85 };
86 
87 static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
88     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
89 static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
90 static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
91 static void clnt_vc_abort(CLIENT *);
92 static bool_t clnt_vc_control(CLIENT *, u_int, void *);
93 static void clnt_vc_close(CLIENT *);
94 static void clnt_vc_destroy(CLIENT *);
95 static bool_t time_not_ok(struct timeval *);
96 static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
97 
98 static 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 
108 static 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  */
121 CLIENT *
122 clnt_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;
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 
272 err:
273 	mtx_destroy(&ct->ct_lock);
274 	mem_free(ct, sizeof (struct ct_data));
275 	mem_free(cl, sizeof (CLIENT));
276 
277 	return ((CLIENT *)NULL);
278 }
279 
280 static enum clnt_stat
281 clnt_vc_call(
282 	CLIENT		*cl,		/* client handle */
283 	struct rpc_callextra *ext,	/* call metadata */
284 	rpcproc_t	proc,		/* procedure number */
285 	struct mbuf	*args,		/* pointer to args */
286 	struct mbuf	**resultsp,	/* pointer to results */
287 	struct timeval	utimeout)
288 {
289 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
290 	AUTH *auth;
291 	struct rpc_err *errp;
292 	enum clnt_stat stat;
293 	XDR xdrs;
294 	struct rpc_msg reply_msg;
295 	bool_t ok;
296 	int nrefreshes = 2;		/* number of times to refresh cred */
297 	struct timeval timeout;
298 	uint32_t xid;
299 	struct mbuf *mreq = NULL, *results;
300 	struct ct_request *cr;
301 	int error;
302 
303 	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
304 
305 	mtx_lock(&ct->ct_lock);
306 
307 	if (ct->ct_closing || ct->ct_closed) {
308 		mtx_unlock(&ct->ct_lock);
309 		free(cr, M_RPC);
310 		return (RPC_CANTSEND);
311 	}
312 	ct->ct_threads++;
313 
314 	if (ext) {
315 		auth = ext->rc_auth;
316 		errp = &ext->rc_err;
317 	} else {
318 		auth = cl->cl_auth;
319 		errp = &ct->ct_error;
320 	}
321 
322 	cr->cr_mrep = NULL;
323 	cr->cr_error = 0;
324 
325 	if (ct->ct_wait.tv_usec == -1) {
326 		timeout = utimeout;	/* use supplied timeout */
327 	} else {
328 		timeout = ct->ct_wait;	/* use default timeout */
329 	}
330 
331 call_again:
332 	mtx_assert(&ct->ct_lock, MA_OWNED);
333 
334 	ct->ct_xid++;
335 	xid = ct->ct_xid;
336 
337 	mtx_unlock(&ct->ct_lock);
338 
339 	/*
340 	 * Leave space to pre-pend the record mark.
341 	 */
342 	mreq = m_gethdr(M_WAITOK, MT_DATA);
343 	mreq->m_data += sizeof(uint32_t);
344 	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
345 	    ("RPC header too big"));
346 	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
347 	mreq->m_len = ct->ct_mpos;
348 
349 	/*
350 	 * The XID is the first thing in the request.
351 	 */
352 	*mtod(mreq, uint32_t *) = htonl(xid);
353 
354 	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
355 
356 	errp->re_status = stat = RPC_SUCCESS;
357 
358 	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
359 	    (! AUTH_MARSHALL(auth, xid, &xdrs,
360 		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
361 		errp->re_status = stat = RPC_CANTENCODEARGS;
362 		mtx_lock(&ct->ct_lock);
363 		goto out;
364 	}
365 	mreq->m_pkthdr.len = m_length(mreq, NULL);
366 
367 	/*
368 	 * Prepend a record marker containing the packet length.
369 	 */
370 	M_PREPEND(mreq, sizeof(uint32_t), M_WAITOK);
371 	*mtod(mreq, uint32_t *) =
372 		htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
373 
374 	cr->cr_xid = xid;
375 	mtx_lock(&ct->ct_lock);
376 	/*
377 	 * Check to see if the other end has already started to close down
378 	 * the connection. The upcall will have set ct_error.re_status
379 	 * to RPC_CANTRECV if this is the case.
380 	 * If the other end starts to close down the connection after this
381 	 * point, it will be detected later when cr_error is checked,
382 	 * since the request is in the ct_pending queue.
383 	 */
384 	if (ct->ct_error.re_status == RPC_CANTRECV) {
385 		if (errp != &ct->ct_error) {
386 			errp->re_errno = ct->ct_error.re_errno;
387 			errp->re_status = RPC_CANTRECV;
388 		}
389 		stat = RPC_CANTRECV;
390 		goto out;
391 	}
392 	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
393 	mtx_unlock(&ct->ct_lock);
394 
395 	/*
396 	 * sosend consumes mreq.
397 	 */
398 	error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
399 	mreq = NULL;
400 	if (error == EMSGSIZE) {
401 		SOCKBUF_LOCK(&ct->ct_socket->so_snd);
402 		sbwait(&ct->ct_socket->so_snd);
403 		SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
404 		AUTH_VALIDATE(auth, xid, NULL, NULL);
405 		mtx_lock(&ct->ct_lock);
406 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
407 		goto call_again;
408 	}
409 
410 	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
411 	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
412 	reply_msg.acpted_rply.ar_verf.oa_length = 0;
413 	reply_msg.acpted_rply.ar_results.where = NULL;
414 	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
415 
416 	mtx_lock(&ct->ct_lock);
417 	if (error) {
418 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
419 		errp->re_errno = error;
420 		errp->re_status = stat = RPC_CANTSEND;
421 		goto out;
422 	}
423 
424 	/*
425 	 * Check to see if we got an upcall while waiting for the
426 	 * lock. In both these cases, the request has been removed
427 	 * from ct->ct_pending.
428 	 */
429 	if (cr->cr_error) {
430 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
431 		errp->re_errno = cr->cr_error;
432 		errp->re_status = stat = RPC_CANTRECV;
433 		goto out;
434 	}
435 	if (cr->cr_mrep) {
436 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
437 		goto got_reply;
438 	}
439 
440 	/*
441 	 * Hack to provide rpc-based message passing
442 	 */
443 	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
444 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
445 		errp->re_status = stat = RPC_TIMEDOUT;
446 		goto out;
447 	}
448 
449 	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
450 	    tvtohz(&timeout));
451 
452 	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
453 
454 	if (error) {
455 		/*
456 		 * The sleep returned an error so our request is still
457 		 * on the list. Turn the error code into an
458 		 * appropriate client status.
459 		 */
460 		errp->re_errno = error;
461 		switch (error) {
462 		case EINTR:
463 			stat = RPC_INTR;
464 			break;
465 		case EWOULDBLOCK:
466 			stat = RPC_TIMEDOUT;
467 			break;
468 		default:
469 			stat = RPC_CANTRECV;
470 		}
471 		errp->re_status = stat;
472 		goto out;
473 	} else {
474 		/*
475 		 * We were woken up by the upcall.  If the
476 		 * upcall had a receive error, report that,
477 		 * otherwise we have a reply.
478 		 */
479 		if (cr->cr_error) {
480 			errp->re_errno = cr->cr_error;
481 			errp->re_status = stat = RPC_CANTRECV;
482 			goto out;
483 		}
484 	}
485 
486 got_reply:
487 	/*
488 	 * Now decode and validate the response. We need to drop the
489 	 * lock since xdr_replymsg may end up sleeping in malloc.
490 	 */
491 	mtx_unlock(&ct->ct_lock);
492 
493 	if (ext && ext->rc_feedback)
494 		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
495 
496 	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
497 	ok = xdr_replymsg(&xdrs, &reply_msg);
498 	cr->cr_mrep = NULL;
499 
500 	if (ok) {
501 		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
502 		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
503 			errp->re_status = stat = RPC_SUCCESS;
504 		else
505 			stat = _seterr_reply(&reply_msg, errp);
506 
507 		if (stat == RPC_SUCCESS) {
508 			results = xdrmbuf_getall(&xdrs);
509 			if (!AUTH_VALIDATE(auth, xid,
510 				&reply_msg.acpted_rply.ar_verf,
511 				&results)) {
512 				errp->re_status = stat = RPC_AUTHERROR;
513 				errp->re_why = AUTH_INVALIDRESP;
514 			} else {
515 				KASSERT(results,
516 				    ("auth validated but no result"));
517 				*resultsp = results;
518 			}
519 		}		/* end successful completion */
520 		/*
521 		 * If unsuccessful AND error is an authentication error
522 		 * then refresh credentials and try again, else break
523 		 */
524 		else if (stat == RPC_AUTHERROR)
525 			/* maybe our credentials need to be refreshed ... */
526 			if (nrefreshes > 0 &&
527 			    AUTH_REFRESH(auth, &reply_msg)) {
528 				nrefreshes--;
529 				XDR_DESTROY(&xdrs);
530 				mtx_lock(&ct->ct_lock);
531 				goto call_again;
532 			}
533 		/* end of unsuccessful completion */
534 	}	/* end of valid reply message */
535 	else {
536 		errp->re_status = stat = RPC_CANTDECODERES;
537 	}
538 	XDR_DESTROY(&xdrs);
539 	mtx_lock(&ct->ct_lock);
540 out:
541 	mtx_assert(&ct->ct_lock, MA_OWNED);
542 
543 	KASSERT(stat != RPC_SUCCESS || *resultsp,
544 	    ("RPC_SUCCESS without reply"));
545 
546 	if (mreq)
547 		m_freem(mreq);
548 	if (cr->cr_mrep)
549 		m_freem(cr->cr_mrep);
550 
551 	ct->ct_threads--;
552 	if (ct->ct_closing)
553 		wakeup(ct);
554 
555 	mtx_unlock(&ct->ct_lock);
556 
557 	if (auth && stat != RPC_SUCCESS)
558 		AUTH_VALIDATE(auth, xid, NULL, NULL);
559 
560 	free(cr, M_RPC);
561 
562 	return (stat);
563 }
564 
565 static void
566 clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
567 {
568 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
569 
570 	*errp = ct->ct_error;
571 }
572 
573 static bool_t
574 clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
575 {
576 	XDR xdrs;
577 	bool_t dummy;
578 
579 	xdrs.x_op = XDR_FREE;
580 	dummy = (*xdr_res)(&xdrs, res_ptr);
581 
582 	return (dummy);
583 }
584 
585 /*ARGSUSED*/
586 static void
587 clnt_vc_abort(CLIENT *cl)
588 {
589 }
590 
591 static bool_t
592 clnt_vc_control(CLIENT *cl, u_int request, void *info)
593 {
594 	struct ct_data *ct = (struct ct_data *)cl->cl_private;
595 	void *infop = info;
596 	SVCXPRT *xprt;
597 
598 	mtx_lock(&ct->ct_lock);
599 
600 	switch (request) {
601 	case CLSET_FD_CLOSE:
602 		ct->ct_closeit = TRUE;
603 		mtx_unlock(&ct->ct_lock);
604 		return (TRUE);
605 	case CLSET_FD_NCLOSE:
606 		ct->ct_closeit = FALSE;
607 		mtx_unlock(&ct->ct_lock);
608 		return (TRUE);
609 	default:
610 		break;
611 	}
612 
613 	/* for other requests which use info */
614 	if (info == NULL) {
615 		mtx_unlock(&ct->ct_lock);
616 		return (FALSE);
617 	}
618 	switch (request) {
619 	case CLSET_TIMEOUT:
620 		if (time_not_ok((struct timeval *)info)) {
621 			mtx_unlock(&ct->ct_lock);
622 			return (FALSE);
623 		}
624 		ct->ct_wait = *(struct timeval *)infop;
625 		break;
626 	case CLGET_TIMEOUT:
627 		*(struct timeval *)infop = ct->ct_wait;
628 		break;
629 	case CLGET_SERVER_ADDR:
630 		(void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
631 		break;
632 	case CLGET_SVC_ADDR:
633 		/*
634 		 * Slightly different semantics to userland - we use
635 		 * sockaddr instead of netbuf.
636 		 */
637 		memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
638 		break;
639 	case CLSET_SVC_ADDR:		/* set to new address */
640 		mtx_unlock(&ct->ct_lock);
641 		return (FALSE);
642 	case CLGET_XID:
643 		*(uint32_t *)info = ct->ct_xid;
644 		break;
645 	case CLSET_XID:
646 		/* This will set the xid of the NEXT call */
647 		/* decrement by 1 as clnt_vc_call() increments once */
648 		ct->ct_xid = *(uint32_t *)info - 1;
649 		break;
650 	case CLGET_VERS:
651 		/*
652 		 * This RELIES on the information that, in the call body,
653 		 * the version number field is the fifth field from the
654 		 * beginning of the RPC header. MUST be changed if the
655 		 * call_struct is changed
656 		 */
657 		*(uint32_t *)info =
658 		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
659 		    4 * BYTES_PER_XDR_UNIT));
660 		break;
661 
662 	case CLSET_VERS:
663 		*(uint32_t *)(void *)(ct->ct_mcallc +
664 		    4 * BYTES_PER_XDR_UNIT) =
665 		    htonl(*(uint32_t *)info);
666 		break;
667 
668 	case CLGET_PROG:
669 		/*
670 		 * This RELIES on the information that, in the call body,
671 		 * the program number field is the fourth field from the
672 		 * beginning of the RPC header. MUST be changed if the
673 		 * call_struct is changed
674 		 */
675 		*(uint32_t *)info =
676 		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
677 		    3 * BYTES_PER_XDR_UNIT));
678 		break;
679 
680 	case CLSET_PROG:
681 		*(uint32_t *)(void *)(ct->ct_mcallc +
682 		    3 * BYTES_PER_XDR_UNIT) =
683 		    htonl(*(uint32_t *)info);
684 		break;
685 
686 	case CLSET_WAITCHAN:
687 		ct->ct_waitchan = (const char *)info;
688 		break;
689 
690 	case CLGET_WAITCHAN:
691 		*(const char **) info = ct->ct_waitchan;
692 		break;
693 
694 	case CLSET_INTERRUPTIBLE:
695 		if (*(int *) info)
696 			ct->ct_waitflag = PCATCH;
697 		else
698 			ct->ct_waitflag = 0;
699 		break;
700 
701 	case CLGET_INTERRUPTIBLE:
702 		if (ct->ct_waitflag)
703 			*(int *) info = TRUE;
704 		else
705 			*(int *) info = FALSE;
706 		break;
707 
708 	case CLSET_BACKCHANNEL:
709 		xprt = (SVCXPRT *)info;
710 		if (ct->ct_backchannelxprt == NULL) {
711 			xprt->xp_p2 = ct;
712 			ct->ct_backchannelxprt = xprt;
713 		}
714 		break;
715 
716 	default:
717 		mtx_unlock(&ct->ct_lock);
718 		return (FALSE);
719 	}
720 
721 	mtx_unlock(&ct->ct_lock);
722 	return (TRUE);
723 }
724 
725 static void
726 clnt_vc_close(CLIENT *cl)
727 {
728 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
729 	struct ct_request *cr;
730 
731 	mtx_lock(&ct->ct_lock);
732 
733 	if (ct->ct_closed) {
734 		mtx_unlock(&ct->ct_lock);
735 		return;
736 	}
737 
738 	if (ct->ct_closing) {
739 		while (ct->ct_closing)
740 			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
741 		KASSERT(ct->ct_closed, ("client should be closed"));
742 		mtx_unlock(&ct->ct_lock);
743 		return;
744 	}
745 
746 	if (ct->ct_socket) {
747 		ct->ct_closing = TRUE;
748 		mtx_unlock(&ct->ct_lock);
749 
750 		SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
751 		soupcall_clear(ct->ct_socket, SO_RCV);
752 		clnt_vc_upcallsdone(ct);
753 		SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
754 
755 		/*
756 		 * Abort any pending requests and wait until everyone
757 		 * has finished with clnt_vc_call.
758 		 */
759 		mtx_lock(&ct->ct_lock);
760 		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
761 			cr->cr_xid = 0;
762 			cr->cr_error = ESHUTDOWN;
763 			wakeup(cr);
764 		}
765 
766 		while (ct->ct_threads)
767 			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
768 	}
769 
770 	ct->ct_closing = FALSE;
771 	ct->ct_closed = TRUE;
772 	mtx_unlock(&ct->ct_lock);
773 	wakeup(ct);
774 }
775 
776 static void
777 clnt_vc_destroy(CLIENT *cl)
778 {
779 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
780 	struct socket *so = NULL;
781 	SVCXPRT *xprt;
782 
783 	clnt_vc_close(cl);
784 
785 	mtx_lock(&ct->ct_lock);
786 	xprt = ct->ct_backchannelxprt;
787 	ct->ct_backchannelxprt = NULL;
788 	if (xprt != NULL) {
789 		mtx_unlock(&ct->ct_lock);	/* To avoid a LOR. */
790 		sx_xlock(&xprt->xp_lock);
791 		mtx_lock(&ct->ct_lock);
792 		xprt->xp_p2 = NULL;
793 		xprt_unregister(xprt);
794 	}
795 
796 	if (ct->ct_socket) {
797 		if (ct->ct_closeit) {
798 			so = ct->ct_socket;
799 		}
800 	}
801 
802 	mtx_unlock(&ct->ct_lock);
803 	if (xprt != NULL) {
804 		sx_xunlock(&xprt->xp_lock);
805 		SVC_RELEASE(xprt);
806 	}
807 
808 	mtx_destroy(&ct->ct_lock);
809 	if (so) {
810 		soshutdown(so, SHUT_WR);
811 		soclose(so);
812 	}
813 	mem_free(ct, sizeof(struct ct_data));
814 	if (cl->cl_netid && cl->cl_netid[0])
815 		mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
816 	if (cl->cl_tp && cl->cl_tp[0])
817 		mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
818 	mem_free(cl, sizeof(CLIENT));
819 }
820 
821 /*
822  * Make sure that the time is not garbage.   -1 value is disallowed.
823  * Note this is different from time_not_ok in clnt_dg.c
824  */
825 static bool_t
826 time_not_ok(struct timeval *t)
827 {
828 	return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
829 		t->tv_usec <= -1 || t->tv_usec > 1000000);
830 }
831 
832 int
833 clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
834 {
835 	struct ct_data *ct = (struct ct_data *) arg;
836 	struct uio uio;
837 	struct mbuf *m, *m2;
838 	struct ct_request *cr;
839 	int error, rcvflag, foundreq;
840 	uint32_t xid_plus_direction[2], header;
841 	bool_t do_read;
842 	SVCXPRT *xprt;
843 	struct cf_conn *cd;
844 
845 	CTASSERT(sizeof(xid_plus_direction) == 2 * sizeof(uint32_t));
846 	ct->ct_upcallrefs++;
847 	uio.uio_td = curthread;
848 	do {
849 		/*
850 		 * If ct_record_resid is zero, we are waiting for a
851 		 * record mark.
852 		 */
853 		if (ct->ct_record_resid == 0) {
854 
855 			/*
856 			 * Make sure there is either a whole record
857 			 * mark in the buffer or there is some other
858 			 * error condition
859 			 */
860 			do_read = FALSE;
861 			if (sbavail(&so->so_rcv) >= sizeof(uint32_t)
862 			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
863 			    || so->so_error)
864 				do_read = TRUE;
865 
866 			if (!do_read)
867 				break;
868 
869 			SOCKBUF_UNLOCK(&so->so_rcv);
870 			uio.uio_resid = sizeof(uint32_t);
871 			m = NULL;
872 			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
873 			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
874 			SOCKBUF_LOCK(&so->so_rcv);
875 
876 			if (error == EWOULDBLOCK)
877 				break;
878 
879 			/*
880 			 * If there was an error, wake up all pending
881 			 * requests.
882 			 */
883 			if (error || uio.uio_resid > 0) {
884 			wakeup_all:
885 				mtx_lock(&ct->ct_lock);
886 				if (!error) {
887 					/*
888 					 * We must have got EOF trying
889 					 * to read from the stream.
890 					 */
891 					error = ECONNRESET;
892 				}
893 				ct->ct_error.re_status = RPC_CANTRECV;
894 				ct->ct_error.re_errno = error;
895 				TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
896 					cr->cr_error = error;
897 					wakeup(cr);
898 				}
899 				mtx_unlock(&ct->ct_lock);
900 				break;
901 			}
902 			m_copydata(m, 0, sizeof(uint32_t), (char *)&header);
903 			header = ntohl(header);
904 			ct->ct_record = NULL;
905 			ct->ct_record_resid = header & 0x7fffffff;
906 			ct->ct_record_eor = ((header & 0x80000000) != 0);
907 			m_freem(m);
908 		} else {
909 			/*
910 			 * Wait until the socket has the whole record
911 			 * buffered.
912 			 */
913 			do_read = FALSE;
914 			if (sbavail(&so->so_rcv) >= ct->ct_record_resid
915 			    || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
916 			    || so->so_error)
917 				do_read = TRUE;
918 
919 			if (!do_read)
920 				break;
921 
922 			/*
923 			 * We have the record mark. Read as much as
924 			 * the socket has buffered up to the end of
925 			 * this record.
926 			 */
927 			SOCKBUF_UNLOCK(&so->so_rcv);
928 			uio.uio_resid = ct->ct_record_resid;
929 			m = NULL;
930 			rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
931 			error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
932 			SOCKBUF_LOCK(&so->so_rcv);
933 
934 			if (error == EWOULDBLOCK)
935 				break;
936 
937 			if (error || uio.uio_resid == ct->ct_record_resid)
938 				goto wakeup_all;
939 
940 			/*
941 			 * If we have part of the record already,
942 			 * chain this bit onto the end.
943 			 */
944 			if (ct->ct_record)
945 				m_last(ct->ct_record)->m_next = m;
946 			else
947 				ct->ct_record = m;
948 
949 			ct->ct_record_resid = uio.uio_resid;
950 
951 			/*
952 			 * If we have the entire record, see if we can
953 			 * match it to a request.
954 			 */
955 			if (ct->ct_record_resid == 0
956 			    && ct->ct_record_eor) {
957 				/*
958 				 * The XID is in the first uint32_t of
959 				 * the reply and the message direction
960 				 * is the second one.
961 				 */
962 				if (ct->ct_record->m_len <
963 				    sizeof(xid_plus_direction) &&
964 				    m_length(ct->ct_record, NULL) <
965 				    sizeof(xid_plus_direction)) {
966 					m_freem(ct->ct_record);
967 					break;
968 				}
969 				m_copydata(ct->ct_record, 0,
970 				    sizeof(xid_plus_direction),
971 				    (char *)xid_plus_direction);
972 				xid_plus_direction[0] =
973 				    ntohl(xid_plus_direction[0]);
974 				xid_plus_direction[1] =
975 				    ntohl(xid_plus_direction[1]);
976 				/* Check message direction. */
977 				if (xid_plus_direction[1] == CALL) {
978 					/* This is a backchannel request. */
979 					mtx_lock(&ct->ct_lock);
980 					xprt = ct->ct_backchannelxprt;
981 					if (xprt == NULL) {
982 						mtx_unlock(&ct->ct_lock);
983 						/* Just throw it away. */
984 						m_freem(ct->ct_record);
985 						ct->ct_record = NULL;
986 					} else {
987 						cd = (struct cf_conn *)
988 						    xprt->xp_p1;
989 						m2 = cd->mreq;
990 						/*
991 						 * The requests are chained
992 						 * in the m_nextpkt list.
993 						 */
994 						while (m2 != NULL &&
995 						    m2->m_nextpkt != NULL)
996 							/* Find end of list. */
997 							m2 = m2->m_nextpkt;
998 						if (m2 != NULL)
999 							m2->m_nextpkt =
1000 							    ct->ct_record;
1001 						else
1002 							cd->mreq =
1003 							    ct->ct_record;
1004 						ct->ct_record->m_nextpkt =
1005 						    NULL;
1006 						ct->ct_record = NULL;
1007 						xprt_active(xprt);
1008 						mtx_unlock(&ct->ct_lock);
1009 					}
1010 				} else {
1011 					mtx_lock(&ct->ct_lock);
1012 					foundreq = 0;
1013 					TAILQ_FOREACH(cr, &ct->ct_pending,
1014 					    cr_link) {
1015 						if (cr->cr_xid ==
1016 						    xid_plus_direction[0]) {
1017 							/*
1018 							 * This one
1019 							 * matches. We leave
1020 							 * the reply mbuf in
1021 							 * cr->cr_mrep. Set
1022 							 * the XID to zero so
1023 							 * that we will ignore
1024 							 * any duplicated
1025 							 * replies.
1026 							 */
1027 							cr->cr_xid = 0;
1028 							cr->cr_mrep =
1029 							    ct->ct_record;
1030 							cr->cr_error = 0;
1031 							foundreq = 1;
1032 							wakeup(cr);
1033 							break;
1034 						}
1035 					}
1036 					mtx_unlock(&ct->ct_lock);
1037 
1038 					if (!foundreq)
1039 						m_freem(ct->ct_record);
1040 					ct->ct_record = NULL;
1041 				}
1042 			}
1043 		}
1044 	} while (m);
1045 	ct->ct_upcallrefs--;
1046 	if (ct->ct_upcallrefs < 0)
1047 		panic("rpcvc upcall refcnt");
1048 	if (ct->ct_upcallrefs == 0)
1049 		wakeup(&ct->ct_upcallrefs);
1050 	return (SU_OK);
1051 }
1052 
1053 /*
1054  * Wait for all upcalls in progress to complete.
1055  */
1056 static void
1057 clnt_vc_upcallsdone(struct ct_data *ct)
1058 {
1059 
1060 	SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1061 
1062 	while (ct->ct_upcallrefs > 0)
1063 		(void) msleep(&ct->ct_upcallrefs,
1064 		    SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1065 }
1066