xref: /freebsd/sys/rpc/clnt_vc.c (revision 5ffd83dbcc34f10e07f6d3e968ae6365869615f4)
1 /*	$NetBSD: clnt_vc.c,v 1.4 2000/07/14 08:40:42 fvdl Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-3-Clause
5  *
6  * Copyright (c) 2009, Sun Microsystems, Inc.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions are met:
11  * - Redistributions of source code must retain the above copyright notice,
12  *   this list of conditions and the following disclaimer.
13  * - Redistributions in binary form must reproduce the above copyright notice,
14  *   this list of conditions and the following disclaimer in the documentation
15  *   and/or other materials provided with the distribution.
16  * - Neither the name of Sun Microsystems, Inc. nor the names of its
17  *   contributors may be used to endorse or promote products derived
18  *   from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
21  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
24  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char *sccsid2 = "@(#)clnt_tcp.c 1.37 87/10/05 Copyr 1984 Sun Micro";
35 static char *sccsid = "@(#)clnt_tcp.c	2.2 88/08/01 4.0 RPCSRC";
36 static char sccsid3[] = "@(#)clnt_vc.c 1.19 89/03/16 Copyr 1988 Sun Micro";
37 #endif
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 /*
42  * clnt_tcp.c, Implements a TCP/IP based, client side RPC.
43  *
44  * Copyright (C) 1984, Sun Microsystems, Inc.
45  *
46  * TCP based RPC supports 'batched calls'.
47  * A sequence of calls may be batched-up in a send buffer.  The rpc call
48  * return immediately to the client even though the call was not necessarily
49  * sent.  The batching occurs if the results' xdr routine is NULL (0) AND
50  * the rpc timeout value is zero (see clnt.h, rpc).
51  *
52  * Clients should NOT casually batch calls that in fact return results; that is,
53  * the server side should be aware that a call is batched and not produce any
54  * return message.  Batched calls that produce many result messages can
55  * deadlock (netlock) the client and the server....
56  *
57  * Now go hang yourself.
58  */
59 
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/lock.h>
64 #include <sys/malloc.h>
65 #include <sys/mbuf.h>
66 #include <sys/mutex.h>
67 #include <sys/pcpu.h>
68 #include <sys/proc.h>
69 #include <sys/protosw.h>
70 #include <sys/socket.h>
71 #include <sys/socketvar.h>
72 #include <sys/sx.h>
73 #include <sys/syslog.h>
74 #include <sys/time.h>
75 #include <sys/uio.h>
76 
77 #include <net/vnet.h>
78 
79 #include <netinet/tcp.h>
80 
81 #include <rpc/rpc.h>
82 #include <rpc/rpc_com.h>
83 #include <rpc/krpc.h>
84 
85 struct cmessage {
86         struct cmsghdr cmsg;
87         struct cmsgcred cmcred;
88 };
89 
90 static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
91     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
92 static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
93 static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
94 static void clnt_vc_abort(CLIENT *);
95 static bool_t clnt_vc_control(CLIENT *, u_int, void *);
96 static void clnt_vc_close(CLIENT *);
97 static void clnt_vc_destroy(CLIENT *);
98 static bool_t time_not_ok(struct timeval *);
99 static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
100 
101 static struct clnt_ops clnt_vc_ops = {
102 	.cl_call =	clnt_vc_call,
103 	.cl_abort =	clnt_vc_abort,
104 	.cl_geterr =	clnt_vc_geterr,
105 	.cl_freeres =	clnt_vc_freeres,
106 	.cl_close =	clnt_vc_close,
107 	.cl_destroy =	clnt_vc_destroy,
108 	.cl_control =	clnt_vc_control
109 };
110 
111 static void clnt_vc_upcallsdone(struct ct_data *);
112 
113 static int	fake_wchan;
114 
115 /*
116  * Create a client handle for a connection.
117  * Default options are set, which the user can change using clnt_control()'s.
118  * The rpc/vc package does buffering similar to stdio, so the client
119  * must pick send and receive buffer sizes, 0 => use the default.
120  * NB: fd is copied into a private area.
121  * NB: The rpch->cl_auth is set null authentication. Caller may wish to
122  * set this something more useful.
123  *
124  * fd should be an open socket
125  */
126 CLIENT *
127 clnt_vc_create(
128 	struct socket *so,		/* open file descriptor */
129 	struct sockaddr *raddr,		/* servers address */
130 	const rpcprog_t prog,		/* program number */
131 	const rpcvers_t vers,		/* version number */
132 	size_t sendsz,			/* buffer recv size */
133 	size_t recvsz,			/* buffer send size */
134 	int intrflag)			/* interruptible */
135 {
136 	CLIENT *cl;			/* client handle */
137 	struct ct_data *ct = NULL;	/* client handle */
138 	struct timeval now;
139 	struct rpc_msg call_msg;
140 	static uint32_t disrupt;
141 	struct __rpc_sockinfo si;
142 	XDR xdrs;
143 	int error, interrupted, one = 1, sleep_flag;
144 	struct sockopt sopt;
145 
146 	if (disrupt == 0)
147 		disrupt = (uint32_t)(long)raddr;
148 
149 	cl = (CLIENT *)mem_alloc(sizeof (*cl));
150 	ct = (struct ct_data *)mem_alloc(sizeof (*ct));
151 
152 	mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
153 	ct->ct_threads = 0;
154 	ct->ct_closing = FALSE;
155 	ct->ct_closed = FALSE;
156 	ct->ct_upcallrefs = 0;
157 
158 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
159 		error = soconnect(so, raddr, curthread);
160 		SOCK_LOCK(so);
161 		interrupted = 0;
162 		sleep_flag = PSOCK;
163 		if (intrflag != 0)
164 			sleep_flag |= PCATCH;
165 		while ((so->so_state & SS_ISCONNECTING)
166 		    && so->so_error == 0) {
167 			error = msleep(&so->so_timeo, SOCK_MTX(so),
168 			    sleep_flag, "connec", 0);
169 			if (error) {
170 				if (error == EINTR || error == ERESTART)
171 					interrupted = 1;
172 				break;
173 			}
174 		}
175 		if (error == 0) {
176 			error = so->so_error;
177 			so->so_error = 0;
178 		}
179 		SOCK_UNLOCK(so);
180 		if (error) {
181 			if (!interrupted)
182 				so->so_state &= ~SS_ISCONNECTING;
183 			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
184 			rpc_createerr.cf_error.re_errno = error;
185 			goto err;
186 		}
187 	}
188 
189 	if (!__rpc_socket2sockinfo(so, &si)) {
190 		goto err;
191 	}
192 
193 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
194 		bzero(&sopt, sizeof(sopt));
195 		sopt.sopt_dir = SOPT_SET;
196 		sopt.sopt_level = SOL_SOCKET;
197 		sopt.sopt_name = SO_KEEPALIVE;
198 		sopt.sopt_val = &one;
199 		sopt.sopt_valsize = sizeof(one);
200 		sosetopt(so, &sopt);
201 	}
202 
203 	if (so->so_proto->pr_protocol == IPPROTO_TCP) {
204 		bzero(&sopt, sizeof(sopt));
205 		sopt.sopt_dir = SOPT_SET;
206 		sopt.sopt_level = IPPROTO_TCP;
207 		sopt.sopt_name = TCP_NODELAY;
208 		sopt.sopt_val = &one;
209 		sopt.sopt_valsize = sizeof(one);
210 		sosetopt(so, &sopt);
211 	}
212 
213 	ct->ct_closeit = FALSE;
214 
215 	/*
216 	 * Set up private data struct
217 	 */
218 	ct->ct_socket = so;
219 	ct->ct_wait.tv_sec = -1;
220 	ct->ct_wait.tv_usec = -1;
221 	memcpy(&ct->ct_addr, raddr, raddr->sa_len);
222 
223 	/*
224 	 * Initialize call message
225 	 */
226 	getmicrotime(&now);
227 	ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
228 	call_msg.rm_xid = ct->ct_xid;
229 	call_msg.rm_direction = CALL;
230 	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
231 	call_msg.rm_call.cb_prog = (uint32_t)prog;
232 	call_msg.rm_call.cb_vers = (uint32_t)vers;
233 
234 	/*
235 	 * pre-serialize the static part of the call msg and stash it away
236 	 */
237 	xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
238 	    XDR_ENCODE);
239 	if (! xdr_callhdr(&xdrs, &call_msg)) {
240 		if (ct->ct_closeit) {
241 			soclose(ct->ct_socket);
242 		}
243 		goto err;
244 	}
245 	ct->ct_mpos = XDR_GETPOS(&xdrs);
246 	XDR_DESTROY(&xdrs);
247 	ct->ct_waitchan = "rpcrecv";
248 	ct->ct_waitflag = 0;
249 
250 	/*
251 	 * Create a client handle which uses xdrrec for serialization
252 	 * and authnone for authentication.
253 	 */
254 	sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
255 	recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
256 	error = soreserve(ct->ct_socket, sendsz, recvsz);
257 	if (error != 0) {
258 		if (ct->ct_closeit) {
259 			soclose(ct->ct_socket);
260 		}
261 		goto err;
262 	}
263 	cl->cl_refs = 1;
264 	cl->cl_ops = &clnt_vc_ops;
265 	cl->cl_private = ct;
266 	cl->cl_auth = authnone_create();
267 
268 	SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
269 	soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
270 	SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
271 
272 	ct->ct_raw = NULL;
273 	ct->ct_record = NULL;
274 	ct->ct_record_resid = 0;
275 	TAILQ_INIT(&ct->ct_pending);
276 	return (cl);
277 
278 err:
279 	mtx_destroy(&ct->ct_lock);
280 	mem_free(ct, sizeof (struct ct_data));
281 	mem_free(cl, sizeof (CLIENT));
282 
283 	return ((CLIENT *)NULL);
284 }
285 
286 static enum clnt_stat
287 clnt_vc_call(
288 	CLIENT		*cl,		/* client handle */
289 	struct rpc_callextra *ext,	/* call metadata */
290 	rpcproc_t	proc,		/* procedure number */
291 	struct mbuf	*args,		/* pointer to args */
292 	struct mbuf	**resultsp,	/* pointer to results */
293 	struct timeval	utimeout)
294 {
295 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
296 	AUTH *auth;
297 	struct rpc_err *errp;
298 	enum clnt_stat stat;
299 	XDR xdrs;
300 	struct rpc_msg reply_msg;
301 	bool_t ok;
302 	int nrefreshes = 2;		/* number of times to refresh cred */
303 	struct timeval timeout;
304 	uint32_t xid;
305 	struct mbuf *mreq = NULL, *results;
306 	struct ct_request *cr;
307 	int error, trycnt;
308 
309 	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
310 
311 	mtx_lock(&ct->ct_lock);
312 
313 	if (ct->ct_closing || ct->ct_closed) {
314 		mtx_unlock(&ct->ct_lock);
315 		free(cr, M_RPC);
316 		return (RPC_CANTSEND);
317 	}
318 	ct->ct_threads++;
319 
320 	if (ext) {
321 		auth = ext->rc_auth;
322 		errp = &ext->rc_err;
323 	} else {
324 		auth = cl->cl_auth;
325 		errp = &ct->ct_error;
326 	}
327 
328 	cr->cr_mrep = NULL;
329 	cr->cr_error = 0;
330 
331 	if (ct->ct_wait.tv_usec == -1) {
332 		timeout = utimeout;	/* use supplied timeout */
333 	} else {
334 		timeout = ct->ct_wait;	/* use default timeout */
335 	}
336 
337 	/*
338 	 * After 15sec of looping, allow it to return RPC_CANTSEND, which will
339 	 * cause the clnt_reconnect layer to create a new TCP connection.
340 	 */
341 	trycnt = 15 * hz;
342 call_again:
343 	mtx_assert(&ct->ct_lock, MA_OWNED);
344 	if (ct->ct_closing || ct->ct_closed) {
345 		ct->ct_threads--;
346 		wakeup(ct);
347 		mtx_unlock(&ct->ct_lock);
348 		free(cr, M_RPC);
349 		return (RPC_CANTSEND);
350 	}
351 
352 	ct->ct_xid++;
353 	xid = ct->ct_xid;
354 
355 	mtx_unlock(&ct->ct_lock);
356 
357 	/*
358 	 * Leave space to pre-pend the record mark.
359 	 */
360 	mreq = m_gethdr(M_WAITOK, MT_DATA);
361 	mreq->m_data += sizeof(uint32_t);
362 	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
363 	    ("RPC header too big"));
364 	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
365 	mreq->m_len = ct->ct_mpos;
366 
367 	/*
368 	 * The XID is the first thing in the request.
369 	 */
370 	*mtod(mreq, uint32_t *) = htonl(xid);
371 
372 	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
373 
374 	errp->re_status = stat = RPC_SUCCESS;
375 
376 	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
377 	    (! AUTH_MARSHALL(auth, xid, &xdrs,
378 		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
379 		errp->re_status = stat = RPC_CANTENCODEARGS;
380 		mtx_lock(&ct->ct_lock);
381 		goto out;
382 	}
383 	mreq->m_pkthdr.len = m_length(mreq, NULL);
384 
385 	/*
386 	 * Prepend a record marker containing the packet length.
387 	 */
388 	M_PREPEND(mreq, sizeof(uint32_t), M_WAITOK);
389 	*mtod(mreq, uint32_t *) =
390 		htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
391 
392 	cr->cr_xid = xid;
393 	mtx_lock(&ct->ct_lock);
394 	/*
395 	 * Check to see if the other end has already started to close down
396 	 * the connection. The upcall will have set ct_error.re_status
397 	 * to RPC_CANTRECV if this is the case.
398 	 * If the other end starts to close down the connection after this
399 	 * point, it will be detected later when cr_error is checked,
400 	 * since the request is in the ct_pending queue.
401 	 */
402 	if (ct->ct_error.re_status == RPC_CANTRECV) {
403 		if (errp != &ct->ct_error) {
404 			errp->re_errno = ct->ct_error.re_errno;
405 			errp->re_status = RPC_CANTRECV;
406 		}
407 		stat = RPC_CANTRECV;
408 		goto out;
409 	}
410 	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
411 	mtx_unlock(&ct->ct_lock);
412 
413 	/*
414 	 * sosend consumes mreq.
415 	 */
416 	error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
417 	mreq = NULL;
418 	if (error == EMSGSIZE || (error == ERESTART &&
419 	    (ct->ct_waitflag & PCATCH) == 0 && trycnt-- > 0)) {
420 		SOCKBUF_LOCK(&ct->ct_socket->so_snd);
421 		sbwait(&ct->ct_socket->so_snd);
422 		SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
423 		AUTH_VALIDATE(auth, xid, NULL, NULL);
424 		mtx_lock(&ct->ct_lock);
425 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
426 		/* Sleep for 1 clock tick before trying the sosend() again. */
427 		msleep(&fake_wchan, &ct->ct_lock, 0, "rpclpsnd", 1);
428 		goto call_again;
429 	}
430 
431 	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
432 	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
433 	reply_msg.acpted_rply.ar_verf.oa_length = 0;
434 	reply_msg.acpted_rply.ar_results.where = NULL;
435 	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
436 
437 	mtx_lock(&ct->ct_lock);
438 	if (error) {
439 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
440 		errp->re_errno = error;
441 		errp->re_status = stat = RPC_CANTSEND;
442 		goto out;
443 	}
444 
445 	/*
446 	 * Check to see if we got an upcall while waiting for the
447 	 * lock. In both these cases, the request has been removed
448 	 * from ct->ct_pending.
449 	 */
450 	if (cr->cr_error) {
451 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
452 		errp->re_errno = cr->cr_error;
453 		errp->re_status = stat = RPC_CANTRECV;
454 		goto out;
455 	}
456 	if (cr->cr_mrep) {
457 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
458 		goto got_reply;
459 	}
460 
461 	/*
462 	 * Hack to provide rpc-based message passing
463 	 */
464 	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
465 		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
466 		errp->re_status = stat = RPC_TIMEDOUT;
467 		goto out;
468 	}
469 
470 	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
471 	    tvtohz(&timeout));
472 
473 	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
474 
475 	if (error) {
476 		/*
477 		 * The sleep returned an error so our request is still
478 		 * on the list. Turn the error code into an
479 		 * appropriate client status.
480 		 */
481 		errp->re_errno = error;
482 		switch (error) {
483 		case EINTR:
484 			stat = RPC_INTR;
485 			break;
486 		case EWOULDBLOCK:
487 			stat = RPC_TIMEDOUT;
488 			break;
489 		default:
490 			stat = RPC_CANTRECV;
491 		}
492 		errp->re_status = stat;
493 		goto out;
494 	} else {
495 		/*
496 		 * We were woken up by the upcall.  If the
497 		 * upcall had a receive error, report that,
498 		 * otherwise we have a reply.
499 		 */
500 		if (cr->cr_error) {
501 			errp->re_errno = cr->cr_error;
502 			errp->re_status = stat = RPC_CANTRECV;
503 			goto out;
504 		}
505 	}
506 
507 got_reply:
508 	/*
509 	 * Now decode and validate the response. We need to drop the
510 	 * lock since xdr_replymsg may end up sleeping in malloc.
511 	 */
512 	mtx_unlock(&ct->ct_lock);
513 
514 	if (ext && ext->rc_feedback)
515 		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
516 
517 	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
518 	ok = xdr_replymsg(&xdrs, &reply_msg);
519 	cr->cr_mrep = NULL;
520 
521 	if (ok) {
522 		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
523 		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
524 			errp->re_status = stat = RPC_SUCCESS;
525 		else
526 			stat = _seterr_reply(&reply_msg, errp);
527 
528 		if (stat == RPC_SUCCESS) {
529 			results = xdrmbuf_getall(&xdrs);
530 			if (!AUTH_VALIDATE(auth, xid,
531 				&reply_msg.acpted_rply.ar_verf,
532 				&results)) {
533 				errp->re_status = stat = RPC_AUTHERROR;
534 				errp->re_why = AUTH_INVALIDRESP;
535 			} else {
536 				KASSERT(results,
537 				    ("auth validated but no result"));
538 				*resultsp = results;
539 			}
540 		}		/* end successful completion */
541 		/*
542 		 * If unsuccessful AND error is an authentication error
543 		 * then refresh credentials and try again, else break
544 		 */
545 		else if (stat == RPC_AUTHERROR)
546 			/* maybe our credentials need to be refreshed ... */
547 			if (nrefreshes > 0 &&
548 			    AUTH_REFRESH(auth, &reply_msg)) {
549 				nrefreshes--;
550 				XDR_DESTROY(&xdrs);
551 				mtx_lock(&ct->ct_lock);
552 				goto call_again;
553 			}
554 		/* end of unsuccessful completion */
555 	}	/* end of valid reply message */
556 	else {
557 		errp->re_status = stat = RPC_CANTDECODERES;
558 	}
559 	XDR_DESTROY(&xdrs);
560 	mtx_lock(&ct->ct_lock);
561 out:
562 	mtx_assert(&ct->ct_lock, MA_OWNED);
563 
564 	KASSERT(stat != RPC_SUCCESS || *resultsp,
565 	    ("RPC_SUCCESS without reply"));
566 
567 	if (mreq)
568 		m_freem(mreq);
569 	if (cr->cr_mrep)
570 		m_freem(cr->cr_mrep);
571 
572 	ct->ct_threads--;
573 	if (ct->ct_closing)
574 		wakeup(ct);
575 
576 	mtx_unlock(&ct->ct_lock);
577 
578 	if (auth && stat != RPC_SUCCESS)
579 		AUTH_VALIDATE(auth, xid, NULL, NULL);
580 
581 	free(cr, M_RPC);
582 
583 	return (stat);
584 }
585 
586 static void
587 clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
588 {
589 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
590 
591 	*errp = ct->ct_error;
592 }
593 
594 static bool_t
595 clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
596 {
597 	XDR xdrs;
598 	bool_t dummy;
599 
600 	xdrs.x_op = XDR_FREE;
601 	dummy = (*xdr_res)(&xdrs, res_ptr);
602 
603 	return (dummy);
604 }
605 
606 /*ARGSUSED*/
607 static void
608 clnt_vc_abort(CLIENT *cl)
609 {
610 }
611 
612 static bool_t
613 clnt_vc_control(CLIENT *cl, u_int request, void *info)
614 {
615 	struct ct_data *ct = (struct ct_data *)cl->cl_private;
616 	void *infop = info;
617 	SVCXPRT *xprt;
618 
619 	mtx_lock(&ct->ct_lock);
620 
621 	switch (request) {
622 	case CLSET_FD_CLOSE:
623 		ct->ct_closeit = TRUE;
624 		mtx_unlock(&ct->ct_lock);
625 		return (TRUE);
626 	case CLSET_FD_NCLOSE:
627 		ct->ct_closeit = FALSE;
628 		mtx_unlock(&ct->ct_lock);
629 		return (TRUE);
630 	default:
631 		break;
632 	}
633 
634 	/* for other requests which use info */
635 	if (info == NULL) {
636 		mtx_unlock(&ct->ct_lock);
637 		return (FALSE);
638 	}
639 	switch (request) {
640 	case CLSET_TIMEOUT:
641 		if (time_not_ok((struct timeval *)info)) {
642 			mtx_unlock(&ct->ct_lock);
643 			return (FALSE);
644 		}
645 		ct->ct_wait = *(struct timeval *)infop;
646 		break;
647 	case CLGET_TIMEOUT:
648 		*(struct timeval *)infop = ct->ct_wait;
649 		break;
650 	case CLGET_SERVER_ADDR:
651 		(void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
652 		break;
653 	case CLGET_SVC_ADDR:
654 		/*
655 		 * Slightly different semantics to userland - we use
656 		 * sockaddr instead of netbuf.
657 		 */
658 		memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
659 		break;
660 	case CLSET_SVC_ADDR:		/* set to new address */
661 		mtx_unlock(&ct->ct_lock);
662 		return (FALSE);
663 	case CLGET_XID:
664 		*(uint32_t *)info = ct->ct_xid;
665 		break;
666 	case CLSET_XID:
667 		/* This will set the xid of the NEXT call */
668 		/* decrement by 1 as clnt_vc_call() increments once */
669 		ct->ct_xid = *(uint32_t *)info - 1;
670 		break;
671 	case CLGET_VERS:
672 		/*
673 		 * This RELIES on the information that, in the call body,
674 		 * the version number field is the fifth field from the
675 		 * beginning of the RPC header. MUST be changed if the
676 		 * call_struct is changed
677 		 */
678 		*(uint32_t *)info =
679 		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
680 		    4 * BYTES_PER_XDR_UNIT));
681 		break;
682 
683 	case CLSET_VERS:
684 		*(uint32_t *)(void *)(ct->ct_mcallc +
685 		    4 * BYTES_PER_XDR_UNIT) =
686 		    htonl(*(uint32_t *)info);
687 		break;
688 
689 	case CLGET_PROG:
690 		/*
691 		 * This RELIES on the information that, in the call body,
692 		 * the program number field is the fourth field from the
693 		 * beginning of the RPC header. MUST be changed if the
694 		 * call_struct is changed
695 		 */
696 		*(uint32_t *)info =
697 		    ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
698 		    3 * BYTES_PER_XDR_UNIT));
699 		break;
700 
701 	case CLSET_PROG:
702 		*(uint32_t *)(void *)(ct->ct_mcallc +
703 		    3 * BYTES_PER_XDR_UNIT) =
704 		    htonl(*(uint32_t *)info);
705 		break;
706 
707 	case CLSET_WAITCHAN:
708 		ct->ct_waitchan = (const char *)info;
709 		break;
710 
711 	case CLGET_WAITCHAN:
712 		*(const char **) info = ct->ct_waitchan;
713 		break;
714 
715 	case CLSET_INTERRUPTIBLE:
716 		if (*(int *) info)
717 			ct->ct_waitflag = PCATCH;
718 		else
719 			ct->ct_waitflag = 0;
720 		break;
721 
722 	case CLGET_INTERRUPTIBLE:
723 		if (ct->ct_waitflag)
724 			*(int *) info = TRUE;
725 		else
726 			*(int *) info = FALSE;
727 		break;
728 
729 	case CLSET_BACKCHANNEL:
730 		xprt = (SVCXPRT *)info;
731 		if (ct->ct_backchannelxprt == NULL) {
732 			xprt->xp_p2 = ct;
733 			ct->ct_backchannelxprt = xprt;
734 		}
735 		break;
736 
737 	default:
738 		mtx_unlock(&ct->ct_lock);
739 		return (FALSE);
740 	}
741 
742 	mtx_unlock(&ct->ct_lock);
743 	return (TRUE);
744 }
745 
746 static void
747 clnt_vc_close(CLIENT *cl)
748 {
749 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
750 	struct ct_request *cr;
751 
752 	mtx_lock(&ct->ct_lock);
753 
754 	if (ct->ct_closed) {
755 		mtx_unlock(&ct->ct_lock);
756 		return;
757 	}
758 
759 	if (ct->ct_closing) {
760 		while (ct->ct_closing)
761 			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
762 		KASSERT(ct->ct_closed, ("client should be closed"));
763 		mtx_unlock(&ct->ct_lock);
764 		return;
765 	}
766 
767 	if (ct->ct_socket) {
768 		ct->ct_closing = TRUE;
769 		mtx_unlock(&ct->ct_lock);
770 
771 		SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
772 		soupcall_clear(ct->ct_socket, SO_RCV);
773 		clnt_vc_upcallsdone(ct);
774 		SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
775 
776 		/*
777 		 * Abort any pending requests and wait until everyone
778 		 * has finished with clnt_vc_call.
779 		 */
780 		mtx_lock(&ct->ct_lock);
781 		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
782 			cr->cr_xid = 0;
783 			cr->cr_error = ESHUTDOWN;
784 			wakeup(cr);
785 		}
786 
787 		while (ct->ct_threads)
788 			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
789 	}
790 
791 	ct->ct_closing = FALSE;
792 	ct->ct_closed = TRUE;
793 	mtx_unlock(&ct->ct_lock);
794 	wakeup(ct);
795 }
796 
797 static void
798 clnt_vc_destroy(CLIENT *cl)
799 {
800 	struct ct_data *ct = (struct ct_data *) cl->cl_private;
801 	struct socket *so = NULL;
802 	SVCXPRT *xprt;
803 
804 	clnt_vc_close(cl);
805 
806 	mtx_lock(&ct->ct_lock);
807 	xprt = ct->ct_backchannelxprt;
808 	ct->ct_backchannelxprt = NULL;
809 	if (xprt != NULL) {
810 		mtx_unlock(&ct->ct_lock);	/* To avoid a LOR. */
811 		sx_xlock(&xprt->xp_lock);
812 		mtx_lock(&ct->ct_lock);
813 		xprt->xp_p2 = NULL;
814 		sx_xunlock(&xprt->xp_lock);
815 	}
816 
817 	if (ct->ct_socket) {
818 		if (ct->ct_closeit) {
819 			so = ct->ct_socket;
820 		}
821 	}
822 
823 	mtx_unlock(&ct->ct_lock);
824 
825 	mtx_destroy(&ct->ct_lock);
826 	if (so) {
827 		soshutdown(so, SHUT_WR);
828 		soclose(so);
829 	}
830 	m_freem(ct->ct_record);
831 	m_freem(ct->ct_raw);
832 	mem_free(ct, sizeof(struct ct_data));
833 	if (cl->cl_netid && cl->cl_netid[0])
834 		mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
835 	if (cl->cl_tp && cl->cl_tp[0])
836 		mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
837 	mem_free(cl, sizeof(CLIENT));
838 }
839 
840 /*
841  * Make sure that the time is not garbage.   -1 value is disallowed.
842  * Note this is different from time_not_ok in clnt_dg.c
843  */
844 static bool_t
845 time_not_ok(struct timeval *t)
846 {
847 	return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
848 		t->tv_usec <= -1 || t->tv_usec > 1000000);
849 }
850 
851 int
852 clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
853 {
854 	struct ct_data *ct = (struct ct_data *) arg;
855 	struct uio uio;
856 	struct mbuf *m, *m2;
857 	struct ct_request *cr;
858 	int error, rcvflag, foundreq;
859 	uint32_t xid_plus_direction[2], header;
860 	SVCXPRT *xprt;
861 	struct cf_conn *cd;
862 	u_int rawlen;
863 
864 	/*
865 	 * If another thread is already here, it must be in
866 	 * soreceive(), so just return to avoid races with it.
867 	 * ct_upcallrefs is protected by the SOCKBUF_LOCK(),
868 	 * which is held in this function, except when
869 	 * soreceive() is called.
870 	 */
871 	if (ct->ct_upcallrefs > 0)
872 		return (SU_OK);
873 	ct->ct_upcallrefs++;
874 
875 	/*
876 	 * Read as much as possible off the socket and link it
877 	 * onto ct_raw.
878 	 */
879 	for (;;) {
880 		uio.uio_resid = 1000000000;
881 		uio.uio_td = curthread;
882 		m2 = m = NULL;
883 		rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
884 		SOCKBUF_UNLOCK(&so->so_rcv);
885 		error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
886 		SOCKBUF_LOCK(&so->so_rcv);
887 
888 		if (error == EWOULDBLOCK) {
889 			/*
890 			 * We must re-test for readability after
891 			 * taking the lock to protect us in the case
892 			 * where a new packet arrives on the socket
893 			 * after our call to soreceive fails with
894 			 * EWOULDBLOCK.
895 			 */
896 			error = 0;
897 			if (!soreadable(so))
898 				break;
899 			continue;
900 		}
901 		if (error == 0 && m == NULL) {
902 			/*
903 			 * We must have got EOF trying
904 			 * to read from the stream.
905 			 */
906 			error = ECONNRESET;
907 		}
908 		if (error != 0)
909 			break;
910 
911 		if (ct->ct_raw != NULL)
912 			m_last(ct->ct_raw)->m_next = m;
913 		else
914 			ct->ct_raw = m;
915 	}
916 	rawlen = m_length(ct->ct_raw, NULL);
917 
918 	/* Now, process as much of ct_raw as possible. */
919 	for (;;) {
920 		/*
921 		 * If ct_record_resid is zero, we are waiting for a
922 		 * record mark.
923 		 */
924 		if (ct->ct_record_resid == 0) {
925 			if (rawlen < sizeof(uint32_t))
926 				break;
927 			m_copydata(ct->ct_raw, 0, sizeof(uint32_t),
928 			    (char *)&header);
929 			header = ntohl(header);
930 			ct->ct_record_resid = header & 0x7fffffff;
931 			ct->ct_record_eor = ((header & 0x80000000) != 0);
932 			m_adj(ct->ct_raw, sizeof(uint32_t));
933 			rawlen -= sizeof(uint32_t);
934 		} else {
935 			/*
936 			 * Move as much of the record as possible to
937 			 * ct_record.
938 			 */
939 			if (rawlen == 0)
940 				break;
941 			if (rawlen <= ct->ct_record_resid) {
942 				if (ct->ct_record != NULL)
943 					m_last(ct->ct_record)->m_next =
944 					    ct->ct_raw;
945 				else
946 					ct->ct_record = ct->ct_raw;
947 				ct->ct_raw = NULL;
948 				ct->ct_record_resid -= rawlen;
949 				rawlen = 0;
950 			} else {
951 				m = m_split(ct->ct_raw, ct->ct_record_resid,
952 				    M_NOWAIT);
953 				if (m == NULL)
954 					break;
955 				if (ct->ct_record != NULL)
956 					m_last(ct->ct_record)->m_next =
957 					    ct->ct_raw;
958 				else
959 					ct->ct_record = ct->ct_raw;
960 				rawlen -= ct->ct_record_resid;
961 				ct->ct_record_resid = 0;
962 				ct->ct_raw = m;
963 			}
964 			if (ct->ct_record_resid > 0)
965 				break;
966 
967 			/*
968 			 * If we have the entire record, see if we can
969 			 * match it to a request.
970 			 */
971 			if (ct->ct_record_eor) {
972 				/*
973 				 * The XID is in the first uint32_t of
974 				 * the reply and the message direction
975 				 * is the second one.
976 				 */
977 				if (ct->ct_record->m_len <
978 				    sizeof(xid_plus_direction) &&
979 				    m_length(ct->ct_record, NULL) <
980 				    sizeof(xid_plus_direction)) {
981 					/*
982 					 * What to do now?
983 					 * The data in the TCP stream is
984 					 * corrupted such that there is no
985 					 * valid RPC message to parse.
986 					 * I think it best to close this
987 					 * connection and allow
988 					 * clnt_reconnect_call() to try
989 					 * and establish a new one.
990 					 */
991 					printf("clnt_vc_soupcall: "
992 					    "connection data corrupted\n");
993 					error = ECONNRESET;
994 					goto wakeup_all;
995 				}
996 				m_copydata(ct->ct_record, 0,
997 				    sizeof(xid_plus_direction),
998 				    (char *)xid_plus_direction);
999 				xid_plus_direction[0] =
1000 				    ntohl(xid_plus_direction[0]);
1001 				xid_plus_direction[1] =
1002 				    ntohl(xid_plus_direction[1]);
1003 				/* Check message direction. */
1004 				if (xid_plus_direction[1] == CALL) {
1005 					/* This is a backchannel request. */
1006 					mtx_lock(&ct->ct_lock);
1007 					xprt = ct->ct_backchannelxprt;
1008 					if (xprt == NULL) {
1009 						mtx_unlock(&ct->ct_lock);
1010 						/* Just throw it away. */
1011 						m_freem(ct->ct_record);
1012 						ct->ct_record = NULL;
1013 					} else {
1014 						cd = (struct cf_conn *)
1015 						    xprt->xp_p1;
1016 						m2 = cd->mreq;
1017 						/*
1018 						 * The requests are chained
1019 						 * in the m_nextpkt list.
1020 						 */
1021 						while (m2 != NULL &&
1022 						    m2->m_nextpkt != NULL)
1023 							/* Find end of list. */
1024 							m2 = m2->m_nextpkt;
1025 						if (m2 != NULL)
1026 							m2->m_nextpkt =
1027 							    ct->ct_record;
1028 						else
1029 							cd->mreq =
1030 							    ct->ct_record;
1031 						ct->ct_record->m_nextpkt =
1032 						    NULL;
1033 						ct->ct_record = NULL;
1034 						xprt_active(xprt);
1035 						mtx_unlock(&ct->ct_lock);
1036 					}
1037 				} else {
1038 					mtx_lock(&ct->ct_lock);
1039 					foundreq = 0;
1040 					TAILQ_FOREACH(cr, &ct->ct_pending,
1041 					    cr_link) {
1042 						if (cr->cr_xid ==
1043 						    xid_plus_direction[0]) {
1044 							/*
1045 							 * This one
1046 							 * matches. We leave
1047 							 * the reply mbuf in
1048 							 * cr->cr_mrep. Set
1049 							 * the XID to zero so
1050 							 * that we will ignore
1051 							 * any duplicated
1052 							 * replies.
1053 							 */
1054 							cr->cr_xid = 0;
1055 							cr->cr_mrep =
1056 							    ct->ct_record;
1057 							cr->cr_error = 0;
1058 							foundreq = 1;
1059 							wakeup(cr);
1060 							break;
1061 						}
1062 					}
1063 					mtx_unlock(&ct->ct_lock);
1064 
1065 					if (!foundreq)
1066 						m_freem(ct->ct_record);
1067 					ct->ct_record = NULL;
1068 				}
1069 			}
1070 		}
1071 	}
1072 
1073 	if (error != 0) {
1074 	wakeup_all:
1075 		/*
1076 		 * This socket is broken, so mark that it cannot
1077 		 * receive and fail all RPCs waiting for a reply
1078 		 * on it, so that they will be retried on a new
1079 		 * TCP connection created by clnt_reconnect_X().
1080 		 */
1081 		mtx_lock(&ct->ct_lock);
1082 		ct->ct_error.re_status = RPC_CANTRECV;
1083 		ct->ct_error.re_errno = error;
1084 		TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
1085 			cr->cr_error = error;
1086 			wakeup(cr);
1087 		}
1088 		mtx_unlock(&ct->ct_lock);
1089 	}
1090 
1091 	ct->ct_upcallrefs--;
1092 	if (ct->ct_upcallrefs < 0)
1093 		panic("rpcvc upcall refcnt");
1094 	if (ct->ct_upcallrefs == 0)
1095 		wakeup(&ct->ct_upcallrefs);
1096 	return (SU_OK);
1097 }
1098 
1099 /*
1100  * Wait for all upcalls in progress to complete.
1101  */
1102 static void
1103 clnt_vc_upcallsdone(struct ct_data *ct)
1104 {
1105 
1106 	SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1107 
1108 	while (ct->ct_upcallrefs > 0)
1109 		(void) msleep(&ct->ct_upcallrefs,
1110 		    SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1111 }
1112