xref: /freebsd/sys/rpc/clnt_dg.c (revision 39beb93c3f8bdbf72a61fda42300b5ebed7390c8)
1 /*	$NetBSD: clnt_dg.c,v 1.4 2000/07/14 08:40:41 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  * Copyright (c) 1986-1991 by Sun Microsystems Inc.
33  */
34 
35 #if defined(LIBC_SCCS) && !defined(lint)
36 #ident	"@(#)clnt_dg.c	1.23	94/04/22 SMI"
37 static char sccsid[] = "@(#)clnt_dg.c 1.19 89/03/16 Copyr 1988 Sun Micro";
38 #endif
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 /*
43  * Implements a connectionless client side RPC.
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mutex.h>
53 #include <sys/pcpu.h>
54 #include <sys/proc.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/time.h>
58 #include <sys/uio.h>
59 
60 #include <rpc/rpc.h>
61 #include <rpc/rpc_com.h>
62 
63 
64 #ifdef _FREEFALL_CONFIG
65 /*
66  * Disable RPC exponential back-off for FreeBSD.org systems.
67  */
68 #define	RPC_MAX_BACKOFF		1 /* second */
69 #else
70 #define	RPC_MAX_BACKOFF		30 /* seconds */
71 #endif
72 
73 static bool_t time_not_ok(struct timeval *);
74 static enum clnt_stat clnt_dg_call(CLIENT *, struct rpc_callextra *,
75     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
76 static void clnt_dg_geterr(CLIENT *, struct rpc_err *);
77 static bool_t clnt_dg_freeres(CLIENT *, xdrproc_t, void *);
78 static void clnt_dg_abort(CLIENT *);
79 static bool_t clnt_dg_control(CLIENT *, u_int, void *);
80 static void clnt_dg_close(CLIENT *);
81 static void clnt_dg_destroy(CLIENT *);
82 static void clnt_dg_soupcall(struct socket *so, void *arg, int waitflag);
83 
84 static struct clnt_ops clnt_dg_ops = {
85 	.cl_call =	clnt_dg_call,
86 	.cl_abort =	clnt_dg_abort,
87 	.cl_geterr =	clnt_dg_geterr,
88 	.cl_freeres =	clnt_dg_freeres,
89 	.cl_close =	clnt_dg_close,
90 	.cl_destroy =	clnt_dg_destroy,
91 	.cl_control =	clnt_dg_control
92 };
93 
94 static const char mem_err_clnt_dg[] = "clnt_dg_create: out of memory";
95 
96 /*
97  * A pending RPC request which awaits a reply. Requests which have
98  * received their reply will have cr_xid set to zero and cr_mrep to
99  * the mbuf chain of the reply.
100  */
101 struct cu_request {
102 	TAILQ_ENTRY(cu_request) cr_link;
103 	CLIENT			*cr_client;	/* owner */
104 	uint32_t		cr_xid;		/* XID of request */
105 	struct mbuf		*cr_mrep;	/* reply received by upcall */
106 	int			cr_error;	/* any error from upcall */
107 	char			cr_verf[MAX_AUTH_BYTES]; /* reply verf */
108 };
109 
110 TAILQ_HEAD(cu_request_list, cu_request);
111 
112 #define MCALL_MSG_SIZE 24
113 
114 /*
115  * This structure is pointed to by the socket's so_upcallarg
116  * member. It is separate from the client private data to facilitate
117  * multiple clients sharing the same socket. The cs_lock mutex is used
118  * to protect all fields of this structure, the socket's receive
119  * buffer SOCKBUF_LOCK is used to ensure that exactly one of these
120  * structures is installed on the socket.
121  */
122 struct cu_socket {
123 	struct mtx		cs_lock;
124 	int			cs_refs;	/* Count of clients */
125 	struct cu_request_list	cs_pending;	/* Requests awaiting replies */
126 };
127 
128 /*
129  * Private data kept per client handle
130  */
131 struct cu_data {
132 	int			cu_threads;	/* # threads in clnt_vc_call */
133 	bool_t			cu_closing;	/* TRUE if we are closing */
134 	bool_t			cu_closed;	/* TRUE if we are closed */
135 	struct socket		*cu_socket;	/* connection socket */
136 	bool_t			cu_closeit;	/* opened by library */
137 	struct sockaddr_storage	cu_raddr;	/* remote address */
138 	int			cu_rlen;
139 	struct timeval		cu_wait;	/* retransmit interval */
140 	struct timeval		cu_total;	/* total time for the call */
141 	struct rpc_err		cu_error;
142 	uint32_t		cu_xid;
143 	char			cu_mcallc[MCALL_MSG_SIZE]; /* marshalled callmsg */
144 	size_t			cu_mcalllen;
145 	size_t			cu_sendsz;	/* send size */
146 	size_t			cu_recvsz;	/* recv size */
147 	int			cu_async;
148 	int			cu_connect;	/* Use connect(). */
149 	int			cu_connected;	/* Have done connect(). */
150 	const char		*cu_waitchan;
151 	int			cu_waitflag;
152 	int			cu_cwnd;	/* congestion window */
153 	int			cu_sent;	/* number of in-flight RPCs */
154 	bool_t			cu_cwnd_wait;
155 };
156 
157 #define CWNDSCALE	256
158 #define MAXCWND		(32 * CWNDSCALE)
159 
160 /*
161  * Connection less client creation returns with client handle parameters.
162  * Default options are set, which the user can change using clnt_control().
163  * fd should be open and bound.
164  * NB: The rpch->cl_auth is initialized to null authentication.
165  * 	Caller may wish to set this something more useful.
166  *
167  * sendsz and recvsz are the maximum allowable packet sizes that can be
168  * sent and received. Normally they are the same, but they can be
169  * changed to improve the program efficiency and buffer allocation.
170  * If they are 0, use the transport default.
171  *
172  * If svcaddr is NULL, returns NULL.
173  */
174 CLIENT *
175 clnt_dg_create(
176 	struct socket *so,
177 	struct sockaddr *svcaddr,	/* servers address */
178 	rpcprog_t program,		/* program number */
179 	rpcvers_t version,		/* version number */
180 	size_t sendsz,			/* buffer recv size */
181 	size_t recvsz)			/* buffer send size */
182 {
183 	CLIENT *cl = NULL;		/* client handle */
184 	struct cu_data *cu = NULL;	/* private data */
185 	struct cu_socket *cs = NULL;
186 	struct timeval now;
187 	struct rpc_msg call_msg;
188 	struct __rpc_sockinfo si;
189 	XDR xdrs;
190 
191 	if (svcaddr == NULL) {
192 		rpc_createerr.cf_stat = RPC_UNKNOWNADDR;
193 		return (NULL);
194 	}
195 
196 	if (!__rpc_socket2sockinfo(so, &si)) {
197 		rpc_createerr.cf_stat = RPC_TLIERROR;
198 		rpc_createerr.cf_error.re_errno = 0;
199 		return (NULL);
200 	}
201 
202 	/*
203 	 * Find the receive and the send size
204 	 */
205 	sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
206 	recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
207 	if ((sendsz == 0) || (recvsz == 0)) {
208 		rpc_createerr.cf_stat = RPC_TLIERROR; /* XXX */
209 		rpc_createerr.cf_error.re_errno = 0;
210 		return (NULL);
211 	}
212 
213 	cl = mem_alloc(sizeof (CLIENT));
214 
215 	/*
216 	 * Should be multiple of 4 for XDR.
217 	 */
218 	sendsz = ((sendsz + 3) / 4) * 4;
219 	recvsz = ((recvsz + 3) / 4) * 4;
220 	cu = mem_alloc(sizeof (*cu));
221 	cu->cu_threads = 0;
222 	cu->cu_closing = FALSE;
223 	cu->cu_closed = FALSE;
224 	(void) memcpy(&cu->cu_raddr, svcaddr, (size_t)svcaddr->sa_len);
225 	cu->cu_rlen = svcaddr->sa_len;
226 	/* Other values can also be set through clnt_control() */
227 	cu->cu_wait.tv_sec = 3;	/* heuristically chosen */
228 	cu->cu_wait.tv_usec = 0;
229 	cu->cu_total.tv_sec = -1;
230 	cu->cu_total.tv_usec = -1;
231 	cu->cu_sendsz = sendsz;
232 	cu->cu_recvsz = recvsz;
233 	cu->cu_async = FALSE;
234 	cu->cu_connect = FALSE;
235 	cu->cu_connected = FALSE;
236 	cu->cu_waitchan = "rpcrecv";
237 	cu->cu_waitflag = 0;
238 	cu->cu_cwnd = MAXCWND / 2;
239 	cu->cu_sent = 0;
240 	cu->cu_cwnd_wait = FALSE;
241 	(void) getmicrotime(&now);
242 	cu->cu_xid = __RPC_GETXID(&now);
243 	call_msg.rm_xid = cu->cu_xid;
244 	call_msg.rm_call.cb_prog = program;
245 	call_msg.rm_call.cb_vers = version;
246 	xdrmem_create(&xdrs, cu->cu_mcallc, MCALL_MSG_SIZE, XDR_ENCODE);
247 	if (! xdr_callhdr(&xdrs, &call_msg)) {
248 		rpc_createerr.cf_stat = RPC_CANTENCODEARGS;  /* XXX */
249 		rpc_createerr.cf_error.re_errno = 0;
250 		goto err2;
251 	}
252 	cu->cu_mcalllen = XDR_GETPOS(&xdrs);;
253 
254 	/*
255 	 * By default, closeit is always FALSE. It is users responsibility
256 	 * to do a close on it, else the user may use clnt_control
257 	 * to let clnt_destroy do it for him/her.
258 	 */
259 	cu->cu_closeit = FALSE;
260 	cu->cu_socket = so;
261 	soreserve(so, 256*1024, 256*1024);
262 
263 	SOCKBUF_LOCK(&so->so_rcv);
264 recheck_socket:
265 	if (so->so_upcall) {
266 		if (so->so_upcall != clnt_dg_soupcall) {
267 			SOCKBUF_UNLOCK(&so->so_rcv);
268 			printf("clnt_dg_create(): socket already has an incompatible upcall\n");
269 			goto err2;
270 		}
271 		cs = (struct cu_socket *) so->so_upcallarg;
272 		mtx_lock(&cs->cs_lock);
273 		cs->cs_refs++;
274 		mtx_unlock(&cs->cs_lock);
275 	} else {
276 		/*
277 		 * We are the first on this socket - allocate the
278 		 * structure and install it in the socket.
279 		 */
280 		SOCKBUF_UNLOCK(&cu->cu_socket->so_rcv);
281 		cs = mem_alloc(sizeof(*cs));
282 		SOCKBUF_LOCK(&cu->cu_socket->so_rcv);
283 		if (so->so_upcall) {
284 			/*
285 			 * We have lost a race with some other client.
286 			 */
287 			mem_free(cs, sizeof(*cs));
288 			goto recheck_socket;
289 		}
290 		mtx_init(&cs->cs_lock, "cs->cs_lock", NULL, MTX_DEF);
291 		cs->cs_refs = 1;
292 		TAILQ_INIT(&cs->cs_pending);
293 		so->so_upcallarg = cs;
294 		so->so_upcall = clnt_dg_soupcall;
295 		so->so_rcv.sb_flags |= SB_UPCALL;
296 	}
297 	SOCKBUF_UNLOCK(&so->so_rcv);
298 
299 	cl->cl_refs = 1;
300 	cl->cl_ops = &clnt_dg_ops;
301 	cl->cl_private = (caddr_t)(void *)cu;
302 	cl->cl_auth = authnone_create();
303 	cl->cl_tp = NULL;
304 	cl->cl_netid = NULL;
305 	return (cl);
306 err2:
307 	if (cl) {
308 		mem_free(cl, sizeof (CLIENT));
309 		if (cu)
310 			mem_free(cu, sizeof (*cu));
311 	}
312 	return (NULL);
313 }
314 
315 static enum clnt_stat
316 clnt_dg_call(
317 	CLIENT		*cl,		/* client handle */
318 	struct rpc_callextra *ext,	/* call metadata */
319 	rpcproc_t	proc,		/* procedure number */
320 	struct mbuf	*args,		/* pointer to args */
321 	struct mbuf	**resultsp,	/* pointer to results */
322 	struct timeval	utimeout)	/* seconds to wait before giving up */
323 {
324 	struct cu_data *cu = (struct cu_data *)cl->cl_private;
325 	struct cu_socket *cs = (struct cu_socket *) cu->cu_socket->so_upcallarg;
326 	struct rpc_timers *rt;
327 	AUTH *auth;
328 	struct rpc_err *errp;
329 	enum clnt_stat stat;
330 	XDR xdrs;
331 	struct rpc_msg reply_msg;
332 	bool_t ok;
333 	int retrans;			/* number of re-transmits so far */
334 	int nrefreshes = 2;		/* number of times to refresh cred */
335 	struct timeval *tvp;
336 	int timeout;
337 	int retransmit_time;
338 	int next_sendtime, starttime, rtt, time_waited, tv = 0;
339 	struct sockaddr *sa;
340 	socklen_t salen;
341 	uint32_t xid = 0;
342 	struct mbuf *mreq = NULL, *results;
343 	struct cu_request *cr;
344 	int error;
345 
346 	cr = malloc(sizeof(struct cu_request), M_RPC, M_WAITOK);
347 
348 	mtx_lock(&cs->cs_lock);
349 
350 	if (cu->cu_closing || cu->cu_closed) {
351 		mtx_unlock(&cs->cs_lock);
352 		free(cr, M_RPC);
353 		return (RPC_CANTSEND);
354 	}
355 	cu->cu_threads++;
356 
357 	if (ext) {
358 		auth = ext->rc_auth;
359 		errp = &ext->rc_err;
360 	} else {
361 		auth = cl->cl_auth;
362 		errp = &cu->cu_error;
363 	}
364 
365 	cr->cr_client = cl;
366 	cr->cr_mrep = NULL;
367 	cr->cr_error = 0;
368 
369 	if (cu->cu_total.tv_usec == -1) {
370 		tvp = &utimeout; /* use supplied timeout */
371 	} else {
372 		tvp = &cu->cu_total; /* use default timeout */
373 	}
374 	if (tvp->tv_sec || tvp->tv_usec)
375 		timeout = tvtohz(tvp);
376 	else
377 		timeout = 0;
378 
379 	if (cu->cu_connect && !cu->cu_connected) {
380 		mtx_unlock(&cs->cs_lock);
381 		error = soconnect(cu->cu_socket,
382 		    (struct sockaddr *)&cu->cu_raddr, curthread);
383 		mtx_lock(&cs->cs_lock);
384 		if (error) {
385 			errp->re_errno = error;
386 			errp->re_status = stat = RPC_CANTSEND;
387 			goto out;
388 		}
389 		cu->cu_connected = 1;
390 	}
391 	if (cu->cu_connected) {
392 		sa = NULL;
393 		salen = 0;
394 	} else {
395 		sa = (struct sockaddr *)&cu->cu_raddr;
396 		salen = cu->cu_rlen;
397 	}
398 	time_waited = 0;
399 	retrans = 0;
400 	if (ext && ext->rc_timers) {
401 		rt = ext->rc_timers;
402 		if (!rt->rt_rtxcur)
403 			rt->rt_rtxcur = tvtohz(&cu->cu_wait);
404 		retransmit_time = next_sendtime = rt->rt_rtxcur;
405 	} else {
406 		rt = NULL;
407 		retransmit_time = next_sendtime = tvtohz(&cu->cu_wait);
408 	}
409 
410 	starttime = ticks;
411 
412 call_again:
413 	mtx_assert(&cs->cs_lock, MA_OWNED);
414 
415 	cu->cu_xid++;
416 	xid = cu->cu_xid;
417 
418 send_again:
419 	mtx_unlock(&cs->cs_lock);
420 
421 	MGETHDR(mreq, M_WAIT, MT_DATA);
422 	KASSERT(cu->cu_mcalllen <= MHLEN, ("RPC header too big"));
423 	bcopy(cu->cu_mcallc, mreq->m_data, cu->cu_mcalllen);
424 	mreq->m_len = cu->cu_mcalllen;
425 
426 	/*
427 	 * The XID is the first thing in the request.
428 	 */
429 	*mtod(mreq, uint32_t *) = htonl(xid);
430 
431 	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
432 
433 	if (cu->cu_async == TRUE && args == NULL)
434 		goto get_reply;
435 
436 	if ((! XDR_PUTINT32(&xdrs, &proc)) ||
437 	    (! AUTH_MARSHALL(auth, xid, &xdrs,
438 		m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
439 		errp->re_status = stat = RPC_CANTENCODEARGS;
440 		mtx_lock(&cs->cs_lock);
441 		goto out;
442 	}
443 	mreq->m_pkthdr.len = m_length(mreq, NULL);
444 
445 	cr->cr_xid = xid;
446 	mtx_lock(&cs->cs_lock);
447 
448 	/*
449 	 * Try to get a place in the congestion window.
450 	 */
451 	while (cu->cu_sent >= cu->cu_cwnd) {
452 		cu->cu_cwnd_wait = TRUE;
453 		error = msleep(&cu->cu_cwnd_wait, &cs->cs_lock,
454 		    cu->cu_waitflag, "rpccwnd", 0);
455 		if (error) {
456 			errp->re_errno = error;
457 			errp->re_status = stat = RPC_CANTSEND;
458 			goto out;
459 		}
460 	}
461 	cu->cu_sent += CWNDSCALE;
462 
463 	TAILQ_INSERT_TAIL(&cs->cs_pending, cr, cr_link);
464 	mtx_unlock(&cs->cs_lock);
465 
466 	/*
467 	 * sosend consumes mreq.
468 	 */
469 	error = sosend(cu->cu_socket, sa, NULL, mreq, NULL, 0, curthread);
470 	mreq = NULL;
471 
472 	/*
473 	 * sub-optimal code appears here because we have
474 	 * some clock time to spare while the packets are in flight.
475 	 * (We assume that this is actually only executed once.)
476 	 */
477 	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
478 	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
479 	reply_msg.acpted_rply.ar_verf.oa_length = 0;
480 	reply_msg.acpted_rply.ar_results.where = NULL;
481 	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
482 
483 	mtx_lock(&cs->cs_lock);
484 	if (error) {
485 		TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
486 		errp->re_errno = error;
487 		errp->re_status = stat = RPC_CANTSEND;
488 		cu->cu_sent -= CWNDSCALE;
489 		if (cu->cu_cwnd_wait) {
490 			cu->cu_cwnd_wait = FALSE;
491 			wakeup(&cu->cu_cwnd_wait);
492 		}
493 		goto out;
494 	}
495 
496 	/*
497 	 * Check to see if we got an upcall while waiting for the
498 	 * lock.
499 	 */
500 	if (cr->cr_error) {
501 		TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
502 		errp->re_errno = cr->cr_error;
503 		errp->re_status = stat = RPC_CANTRECV;
504 		cu->cu_sent -= CWNDSCALE;
505 		if (cu->cu_cwnd_wait) {
506 			cu->cu_cwnd_wait = FALSE;
507 			wakeup(&cu->cu_cwnd_wait);
508 		}
509 		goto out;
510 	}
511 	if (cr->cr_mrep) {
512 		TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
513 		cu->cu_sent -= CWNDSCALE;
514 		if (cu->cu_cwnd_wait) {
515 			cu->cu_cwnd_wait = FALSE;
516 			wakeup(&cu->cu_cwnd_wait);
517 		}
518 		goto got_reply;
519 	}
520 
521 	/*
522 	 * Hack to provide rpc-based message passing
523 	 */
524 	if (timeout == 0) {
525 		TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
526 		errp->re_status = stat = RPC_TIMEDOUT;
527 		cu->cu_sent -= CWNDSCALE;
528 		if (cu->cu_cwnd_wait) {
529 			cu->cu_cwnd_wait = FALSE;
530 			wakeup(&cu->cu_cwnd_wait);
531 		}
532 		goto out;
533 	}
534 
535 get_reply:
536 	for (;;) {
537 		/* Decide how long to wait. */
538 		if (next_sendtime < timeout)
539 			tv = next_sendtime;
540 		else
541 			tv = timeout;
542 		tv -= time_waited;
543 
544 		if (tv > 0) {
545 			if (cu->cu_closing || cu->cu_closed)
546 				error = 0;
547 			else
548 				error = msleep(cr, &cs->cs_lock,
549 				    cu->cu_waitflag, cu->cu_waitchan, tv);
550 		} else {
551 			error = EWOULDBLOCK;
552 		}
553 
554 		TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
555 		cu->cu_sent -= CWNDSCALE;
556 		if (cu->cu_cwnd_wait) {
557 			cu->cu_cwnd_wait = FALSE;
558 			wakeup(&cu->cu_cwnd_wait);
559 		}
560 
561 		if (!error) {
562 			/*
563 			 * We were woken up by the upcall.  If the
564 			 * upcall had a receive error, report that,
565 			 * otherwise we have a reply.
566 			 */
567 			if (cr->cr_error) {
568 				errp->re_errno = cr->cr_error;
569 				errp->re_status = stat = RPC_CANTRECV;
570 				goto out;
571 			}
572 
573 			cu->cu_cwnd += (CWNDSCALE * CWNDSCALE
574 			    + cu->cu_cwnd / 2) / cu->cu_cwnd;
575 			if (cu->cu_cwnd > MAXCWND)
576 				cu->cu_cwnd = MAXCWND;
577 
578 			if (rt) {
579 				/*
580 				 * Add one to the time since a tick
581 				 * count of N means that the actual
582 				 * time taken was somewhere between N
583 				 * and N+1.
584 				 */
585 				rtt = ticks - starttime + 1;
586 
587 				/*
588 				 * Update our estimate of the round
589 				 * trip time using roughly the
590 				 * algorithm described in RFC
591 				 * 2988. Given an RTT sample R:
592 				 *
593 				 * RTTVAR = (1-beta) * RTTVAR + beta * |SRTT-R|
594 				 * SRTT = (1-alpha) * SRTT + alpha * R
595 				 *
596 				 * where alpha = 0.125 and beta = 0.25.
597 				 *
598 				 * The initial retransmit timeout is
599 				 * SRTT + 4*RTTVAR and doubles on each
600 				 * retransmision.
601 				 */
602 				if (rt->rt_srtt == 0) {
603 					rt->rt_srtt = rtt;
604 					rt->rt_deviate = rtt / 2;
605 				} else {
606 					int32_t error = rtt - rt->rt_srtt;
607 					rt->rt_srtt += error / 8;
608 					error = abs(error) - rt->rt_deviate;
609 					rt->rt_deviate += error / 4;
610 				}
611 				rt->rt_rtxcur = rt->rt_srtt + 4*rt->rt_deviate;
612 			}
613 
614 			break;
615 		}
616 
617 		/*
618 		 * The sleep returned an error so our request is still
619 		 * on the list. If we got EWOULDBLOCK, we may want to
620 		 * re-send the request.
621 		 */
622 		if (error != EWOULDBLOCK) {
623 			errp->re_errno = error;
624 			if (error == EINTR)
625 				errp->re_status = stat = RPC_INTR;
626 			else
627 				errp->re_status = stat = RPC_CANTRECV;
628 			goto out;
629 		}
630 
631 		time_waited = ticks - starttime;
632 
633 		/* Check for timeout. */
634 		if (time_waited > timeout) {
635 			errp->re_errno = EWOULDBLOCK;
636 			errp->re_status = stat = RPC_TIMEDOUT;
637 			goto out;
638 		}
639 
640 		/* Retransmit if necessary. */
641 		if (time_waited >= next_sendtime) {
642 			cu->cu_cwnd /= 2;
643 			if (cu->cu_cwnd < CWNDSCALE)
644 				cu->cu_cwnd = CWNDSCALE;
645 			if (ext && ext->rc_feedback) {
646 				mtx_unlock(&cs->cs_lock);
647 				if (retrans == 0)
648 					ext->rc_feedback(FEEDBACK_REXMIT1,
649 					    proc, ext->rc_feedback_arg);
650 				else
651 					ext->rc_feedback(FEEDBACK_REXMIT2,
652 					    proc, ext->rc_feedback_arg);
653 				mtx_lock(&cs->cs_lock);
654 			}
655 			if (cu->cu_closing || cu->cu_closed) {
656 				errp->re_errno = ESHUTDOWN;
657 				errp->re_status = stat = RPC_CANTRECV;
658 				goto out;
659 			}
660 			retrans++;
661 			/* update retransmit_time */
662 			if (retransmit_time < RPC_MAX_BACKOFF * hz)
663 				retransmit_time = 2 * retransmit_time;
664 			next_sendtime += retransmit_time;
665 			goto send_again;
666 		}
667 		TAILQ_INSERT_TAIL(&cs->cs_pending, cr, cr_link);
668 	}
669 
670 got_reply:
671 	/*
672 	 * Now decode and validate the response. We need to drop the
673 	 * lock since xdr_replymsg may end up sleeping in malloc.
674 	 */
675 	mtx_unlock(&cs->cs_lock);
676 
677 	if (ext && ext->rc_feedback)
678 		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
679 
680 	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
681 	ok = xdr_replymsg(&xdrs, &reply_msg);
682 	cr->cr_mrep = NULL;
683 
684 	if (ok) {
685 		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
686 		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
687 			errp->re_status = stat = RPC_SUCCESS;
688 		else
689 			stat = _seterr_reply(&reply_msg, &(cu->cu_error));
690 
691 		if (errp->re_status == RPC_SUCCESS) {
692 			results = xdrmbuf_getall(&xdrs);
693 			if (! AUTH_VALIDATE(auth, xid,
694 				&reply_msg.acpted_rply.ar_verf,
695 				&results)) {
696 				errp->re_status = stat = RPC_AUTHERROR;
697 				errp->re_why = AUTH_INVALIDRESP;
698 				if (retrans &&
699 				    auth->ah_cred.oa_flavor == RPCSEC_GSS) {
700 					/*
701 					 * If we retransmitted, its
702 					 * possible that we will
703 					 * receive a reply for one of
704 					 * the earlier transmissions
705 					 * (which will use an older
706 					 * RPCSEC_GSS sequence
707 					 * number). In this case, just
708 					 * go back and listen for a
709 					 * new reply. We could keep a
710 					 * record of all the seq
711 					 * numbers we have transmitted
712 					 * so far so that we could
713 					 * accept a reply for any of
714 					 * them here.
715 					 */
716 					XDR_DESTROY(&xdrs);
717 					mtx_lock(&cs->cs_lock);
718 					TAILQ_INSERT_TAIL(&cs->cs_pending,
719 					    cr, cr_link);
720 					cr->cr_mrep = NULL;
721 					goto get_reply;
722 				}
723 			} else {
724 				*resultsp = results;
725 			}
726 		}		/* end successful completion */
727 		/*
728 		 * If unsuccesful AND error is an authentication error
729 		 * then refresh credentials and try again, else break
730 		 */
731 		else if (stat == RPC_AUTHERROR)
732 			/* maybe our credentials need to be refreshed ... */
733 			if (nrefreshes > 0 &&
734 			    AUTH_REFRESH(auth, &reply_msg)) {
735 				nrefreshes--;
736 				XDR_DESTROY(&xdrs);
737 				mtx_lock(&cs->cs_lock);
738 				goto call_again;
739 			}
740 		/* end of unsuccessful completion */
741 	}	/* end of valid reply message */
742 	else {
743 		errp->re_status = stat = RPC_CANTDECODERES;
744 
745 	}
746 	XDR_DESTROY(&xdrs);
747 	mtx_lock(&cs->cs_lock);
748 out:
749 	mtx_assert(&cs->cs_lock, MA_OWNED);
750 
751 	if (mreq)
752 		m_freem(mreq);
753 	if (cr->cr_mrep)
754 		m_freem(cr->cr_mrep);
755 
756 	cu->cu_threads--;
757 	if (cu->cu_closing)
758 		wakeup(cu);
759 
760 	mtx_unlock(&cs->cs_lock);
761 
762 	if (auth && stat != RPC_SUCCESS)
763 		AUTH_VALIDATE(auth, xid, NULL, NULL);
764 
765 	free(cr, M_RPC);
766 
767 	return (stat);
768 }
769 
770 static void
771 clnt_dg_geterr(CLIENT *cl, struct rpc_err *errp)
772 {
773 	struct cu_data *cu = (struct cu_data *)cl->cl_private;
774 
775 	*errp = cu->cu_error;
776 }
777 
778 static bool_t
779 clnt_dg_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
780 {
781 	XDR xdrs;
782 	bool_t dummy;
783 
784 	xdrs.x_op = XDR_FREE;
785 	dummy = (*xdr_res)(&xdrs, res_ptr);
786 
787 	return (dummy);
788 }
789 
790 /*ARGSUSED*/
791 static void
792 clnt_dg_abort(CLIENT *h)
793 {
794 }
795 
796 static bool_t
797 clnt_dg_control(CLIENT *cl, u_int request, void *info)
798 {
799 	struct cu_data *cu = (struct cu_data *)cl->cl_private;
800 	struct cu_socket *cs = (struct cu_socket *) cu->cu_socket->so_upcallarg;
801 	struct sockaddr *addr;
802 
803 	mtx_lock(&cs->cs_lock);
804 
805 	switch (request) {
806 	case CLSET_FD_CLOSE:
807 		cu->cu_closeit = TRUE;
808 		mtx_unlock(&cs->cs_lock);
809 		return (TRUE);
810 	case CLSET_FD_NCLOSE:
811 		cu->cu_closeit = FALSE;
812 		mtx_unlock(&cs->cs_lock);
813 		return (TRUE);
814 	}
815 
816 	/* for other requests which use info */
817 	if (info == NULL) {
818 		mtx_unlock(&cs->cs_lock);
819 		return (FALSE);
820 	}
821 	switch (request) {
822 	case CLSET_TIMEOUT:
823 		if (time_not_ok((struct timeval *)info)) {
824 			mtx_unlock(&cs->cs_lock);
825 			return (FALSE);
826 		}
827 		cu->cu_total = *(struct timeval *)info;
828 		break;
829 	case CLGET_TIMEOUT:
830 		*(struct timeval *)info = cu->cu_total;
831 		break;
832 	case CLSET_RETRY_TIMEOUT:
833 		if (time_not_ok((struct timeval *)info)) {
834 			mtx_unlock(&cs->cs_lock);
835 			return (FALSE);
836 		}
837 		cu->cu_wait = *(struct timeval *)info;
838 		break;
839 	case CLGET_RETRY_TIMEOUT:
840 		*(struct timeval *)info = cu->cu_wait;
841 		break;
842 	case CLGET_SVC_ADDR:
843 		/*
844 		 * Slightly different semantics to userland - we use
845 		 * sockaddr instead of netbuf.
846 		 */
847 		memcpy(info, &cu->cu_raddr, cu->cu_raddr.ss_len);
848 		break;
849 	case CLSET_SVC_ADDR:		/* set to new address */
850 		addr = (struct sockaddr *)info;
851 		(void) memcpy(&cu->cu_raddr, addr, addr->sa_len);
852 		break;
853 	case CLGET_XID:
854 		*(uint32_t *)info = cu->cu_xid;
855 		break;
856 
857 	case CLSET_XID:
858 		/* This will set the xid of the NEXT call */
859 		/* decrement by 1 as clnt_dg_call() increments once */
860 		cu->cu_xid = *(uint32_t *)info - 1;
861 		break;
862 
863 	case CLGET_VERS:
864 		/*
865 		 * This RELIES on the information that, in the call body,
866 		 * the version number field is the fifth field from the
867 		 * begining of the RPC header. MUST be changed if the
868 		 * call_struct is changed
869 		 */
870 		*(uint32_t *)info =
871 		    ntohl(*(uint32_t *)(void *)(cu->cu_mcallc +
872 		    4 * BYTES_PER_XDR_UNIT));
873 		break;
874 
875 	case CLSET_VERS:
876 		*(uint32_t *)(void *)(cu->cu_mcallc + 4 * BYTES_PER_XDR_UNIT)
877 			= htonl(*(uint32_t *)info);
878 		break;
879 
880 	case CLGET_PROG:
881 		/*
882 		 * This RELIES on the information that, in the call body,
883 		 * the program number field is the fourth field from the
884 		 * begining of the RPC header. MUST be changed if the
885 		 * call_struct is changed
886 		 */
887 		*(uint32_t *)info =
888 		    ntohl(*(uint32_t *)(void *)(cu->cu_mcallc +
889 		    3 * BYTES_PER_XDR_UNIT));
890 		break;
891 
892 	case CLSET_PROG:
893 		*(uint32_t *)(void *)(cu->cu_mcallc + 3 * BYTES_PER_XDR_UNIT)
894 			= htonl(*(uint32_t *)info);
895 		break;
896 	case CLSET_ASYNC:
897 		cu->cu_async = *(int *)info;
898 		break;
899 	case CLSET_CONNECT:
900 		cu->cu_connect = *(int *)info;
901 		break;
902 	case CLSET_WAITCHAN:
903 		cu->cu_waitchan = (const char *)info;
904 		break;
905 	case CLGET_WAITCHAN:
906 		*(const char **) info = cu->cu_waitchan;
907 		break;
908 	case CLSET_INTERRUPTIBLE:
909 		if (*(int *) info)
910 			cu->cu_waitflag = PCATCH;
911 		else
912 			cu->cu_waitflag = 0;
913 		break;
914 	case CLGET_INTERRUPTIBLE:
915 		if (cu->cu_waitflag)
916 			*(int *) info = TRUE;
917 		else
918 			*(int *) info = FALSE;
919 		break;
920 	default:
921 		mtx_unlock(&cs->cs_lock);
922 		return (FALSE);
923 	}
924 	mtx_unlock(&cs->cs_lock);
925 	return (TRUE);
926 }
927 
928 static void
929 clnt_dg_close(CLIENT *cl)
930 {
931 	struct cu_data *cu = (struct cu_data *)cl->cl_private;
932 	struct cu_socket *cs = (struct cu_socket *) cu->cu_socket->so_upcallarg;
933 	struct cu_request *cr;
934 
935 	mtx_lock(&cs->cs_lock);
936 
937 	if (cu->cu_closed) {
938 		mtx_unlock(&cs->cs_lock);
939 		return;
940 	}
941 
942 	if (cu->cu_closing) {
943 		while (cu->cu_closing)
944 			msleep(cu, &cs->cs_lock, 0, "rpcclose", 0);
945 		KASSERT(cu->cu_closed, ("client should be closed"));
946 		mtx_unlock(&cs->cs_lock);
947 		return;
948 	}
949 
950 	/*
951 	 * Abort any pending requests and wait until everyone
952 	 * has finished with clnt_vc_call.
953 	 */
954 	cu->cu_closing = TRUE;
955 	TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
956 		if (cr->cr_client == cl) {
957 			cr->cr_xid = 0;
958 			cr->cr_error = ESHUTDOWN;
959 			wakeup(cr);
960 		}
961 	}
962 
963 	while (cu->cu_threads)
964 		msleep(cu, &cs->cs_lock, 0, "rpcclose", 0);
965 
966 	cu->cu_closing = FALSE;
967 	cu->cu_closed = TRUE;
968 
969 	mtx_unlock(&cs->cs_lock);
970 	wakeup(cu);
971 }
972 
973 static void
974 clnt_dg_destroy(CLIENT *cl)
975 {
976 	struct cu_data *cu = (struct cu_data *)cl->cl_private;
977 	struct cu_socket *cs = (struct cu_socket *) cu->cu_socket->so_upcallarg;
978 	struct socket *so = NULL;
979 	bool_t lastsocketref;
980 
981 	clnt_dg_close(cl);
982 
983 	mtx_lock(&cs->cs_lock);
984 
985 	cs->cs_refs--;
986 	if (cs->cs_refs == 0) {
987 		mtx_destroy(&cs->cs_lock);
988 		SOCKBUF_LOCK(&cu->cu_socket->so_rcv);
989 		cu->cu_socket->so_upcallarg = NULL;
990 		cu->cu_socket->so_upcall = NULL;
991 		cu->cu_socket->so_rcv.sb_flags &= ~SB_UPCALL;
992 		SOCKBUF_UNLOCK(&cu->cu_socket->so_rcv);
993 		mem_free(cs, sizeof(*cs));
994 		lastsocketref = TRUE;
995 	} else {
996 		mtx_unlock(&cs->cs_lock);
997 		lastsocketref = FALSE;
998 	}
999 
1000 	if (cu->cu_closeit && lastsocketref) {
1001 		so = cu->cu_socket;
1002 		cu->cu_socket = NULL;
1003 	}
1004 
1005 	if (so)
1006 		soclose(so);
1007 
1008 	if (cl->cl_netid && cl->cl_netid[0])
1009 		mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
1010 	if (cl->cl_tp && cl->cl_tp[0])
1011 		mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
1012 	mem_free(cu, sizeof (*cu));
1013 	mem_free(cl, sizeof (CLIENT));
1014 }
1015 
1016 /*
1017  * Make sure that the time is not garbage.  -1 value is allowed.
1018  */
1019 static bool_t
1020 time_not_ok(struct timeval *t)
1021 {
1022 	return (t->tv_sec < -1 || t->tv_sec > 100000000 ||
1023 		t->tv_usec < -1 || t->tv_usec > 1000000);
1024 }
1025 
1026 void
1027 clnt_dg_soupcall(struct socket *so, void *arg, int waitflag)
1028 {
1029 	struct cu_socket *cs = (struct cu_socket *) arg;
1030 	struct uio uio;
1031 	struct mbuf *m;
1032 	struct mbuf *control;
1033 	struct cu_request *cr;
1034 	int error, rcvflag, foundreq;
1035 	uint32_t xid;
1036 
1037 	uio.uio_resid = 1000000000;
1038 	uio.uio_td = curthread;
1039 	do {
1040 		m = NULL;
1041 		control = NULL;
1042 		rcvflag = MSG_DONTWAIT;
1043 		error = soreceive(so, NULL, &uio, &m, &control, &rcvflag);
1044 		if (control)
1045 			m_freem(control);
1046 
1047 		if (error == EWOULDBLOCK)
1048 			break;
1049 
1050 		/*
1051 		 * If there was an error, wake up all pending
1052 		 * requests.
1053 		 */
1054 		if (error) {
1055 			mtx_lock(&cs->cs_lock);
1056 			TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
1057 				cr->cr_xid = 0;
1058 				cr->cr_error = error;
1059 				wakeup(cr);
1060 			}
1061 			mtx_unlock(&cs->cs_lock);
1062 			break;
1063 		}
1064 
1065 		/*
1066 		 * The XID is in the first uint32_t of the reply.
1067 		 */
1068 		if (m->m_len < sizeof(xid))
1069 			m = m_pullup(m, sizeof(xid));
1070 		if (!m)
1071 			/*
1072 			 * Should never happen.
1073 			 */
1074 			continue;
1075 
1076 		xid = ntohl(*mtod(m, uint32_t *));
1077 
1078 		/*
1079 		 * Attempt to match this reply with a pending request.
1080 		 */
1081 		mtx_lock(&cs->cs_lock);
1082 		foundreq = 0;
1083 		TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
1084 			if (cr->cr_xid == xid) {
1085 				/*
1086 				 * This one matches. We leave the
1087 				 * reply mbuf in cr->cr_mrep. Set the
1088 				 * XID to zero so that we will ignore
1089 				 * any duplicated replies that arrive
1090 				 * before clnt_dg_call removes it from
1091 				 * the queue.
1092 				 */
1093 				cr->cr_xid = 0;
1094 				cr->cr_mrep = m;
1095 				cr->cr_error = 0;
1096 				foundreq = 1;
1097 				wakeup(cr);
1098 				break;
1099 			}
1100 		}
1101 		mtx_unlock(&cs->cs_lock);
1102 
1103 		/*
1104 		 * If we didn't find the matching request, just drop
1105 		 * it - its probably a repeated reply.
1106 		 */
1107 		if (!foundreq)
1108 			m_freem(m);
1109 	} while (m);
1110 }
1111 
1112