svc.c revision 261053
1/*	$NetBSD: svc.c,v 1.21 2000/07/06 03:10:35 christos 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)
32static char *sccsid2 = "@(#)svc.c 1.44 88/02/08 Copyr 1984 Sun Micro";
33static char *sccsid = "@(#)svc.c	2.4 88/08/11 4.0 RPCSRC";
34#endif
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: stable/10/sys/rpc/svc.c 261053 2014-01-22 23:51:12Z mav $");
37
38/*
39 * svc.c, Server-side remote procedure call interface.
40 *
41 * There are two sets of procedures here.  The xprt routines are
42 * for handling transport handles.  The svc routines handle the
43 * list of service routines.
44 *
45 * Copyright (C) 1984, Sun Microsystems, Inc.
46 */
47
48#include <sys/param.h>
49#include <sys/lock.h>
50#include <sys/kernel.h>
51#include <sys/kthread.h>
52#include <sys/malloc.h>
53#include <sys/mbuf.h>
54#include <sys/mutex.h>
55#include <sys/proc.h>
56#include <sys/queue.h>
57#include <sys/socketvar.h>
58#include <sys/systm.h>
59#include <sys/ucred.h>
60
61#include <rpc/rpc.h>
62#include <rpc/rpcb_clnt.h>
63#include <rpc/replay.h>
64
65#include <rpc/rpc_com.h>
66
67#define SVC_VERSQUIET 0x0001		/* keep quiet about vers mismatch */
68#define version_keepquiet(xp) (SVC_EXT(xp)->xp_flags & SVC_VERSQUIET)
69
70static struct svc_callout *svc_find(SVCPOOL *pool, rpcprog_t, rpcvers_t,
71    char *);
72static void svc_new_thread(SVCPOOL *pool);
73static void xprt_unregister_locked(SVCXPRT *xprt);
74
75/* ***************  SVCXPRT related stuff **************** */
76
77static int svcpool_minthread_sysctl(SYSCTL_HANDLER_ARGS);
78static int svcpool_maxthread_sysctl(SYSCTL_HANDLER_ARGS);
79
80SVCPOOL*
81svcpool_create(const char *name, struct sysctl_oid_list *sysctl_base)
82{
83	SVCPOOL *pool;
84
85	pool = malloc(sizeof(SVCPOOL), M_RPC, M_WAITOK|M_ZERO);
86
87	mtx_init(&pool->sp_lock, "sp_lock", NULL, MTX_DEF);
88	pool->sp_name = name;
89	pool->sp_state = SVCPOOL_INIT;
90	pool->sp_proc = NULL;
91	TAILQ_INIT(&pool->sp_xlist);
92	TAILQ_INIT(&pool->sp_active);
93	TAILQ_INIT(&pool->sp_callouts);
94	LIST_INIT(&pool->sp_threads);
95	LIST_INIT(&pool->sp_idlethreads);
96	pool->sp_minthreads = 1;
97	pool->sp_maxthreads = 1;
98	pool->sp_threadcount = 0;
99
100	/*
101	 * Don't use more than a quarter of mbuf clusters or more than
102	 * 45Mb buffering requests.
103	 */
104	pool->sp_space_high = nmbclusters * MCLBYTES / 4;
105	if (pool->sp_space_high > 45 << 20)
106		pool->sp_space_high = 45 << 20;
107	pool->sp_space_low = 2 * pool->sp_space_high / 3;
108
109	sysctl_ctx_init(&pool->sp_sysctl);
110	if (sysctl_base) {
111		SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
112		    "minthreads", CTLTYPE_INT | CTLFLAG_RW,
113		    pool, 0, svcpool_minthread_sysctl, "I", "");
114		SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
115		    "maxthreads", CTLTYPE_INT | CTLFLAG_RW,
116		    pool, 0, svcpool_maxthread_sysctl, "I", "");
117		SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
118		    "threads", CTLFLAG_RD, &pool->sp_threadcount, 0, "");
119
120		SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
121		    "request_space_used", CTLFLAG_RD,
122		    &pool->sp_space_used, 0,
123		    "Space in parsed but not handled requests.");
124
125		SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
126		    "request_space_used_highest", CTLFLAG_RD,
127		    &pool->sp_space_used_highest, 0,
128		    "Highest space used since reboot.");
129
130		SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
131		    "request_space_high", CTLFLAG_RW,
132		    &pool->sp_space_high, 0,
133		    "Maximum space in parsed but not handled requests.");
134
135		SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
136		    "request_space_low", CTLFLAG_RW,
137		    &pool->sp_space_low, 0,
138		    "Low water mark for request space.");
139
140		SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
141		    "request_space_throttled", CTLFLAG_RD,
142		    &pool->sp_space_throttled, 0,
143		    "Whether nfs requests are currently throttled");
144
145		SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
146		    "request_space_throttle_count", CTLFLAG_RD,
147		    &pool->sp_space_throttle_count, 0,
148		    "Count of times throttling based on request space has occurred");
149	}
150
151	return pool;
152}
153
154void
155svcpool_destroy(SVCPOOL *pool)
156{
157	SVCXPRT *xprt, *nxprt;
158	struct svc_callout *s;
159	struct svcxprt_list cleanup;
160
161	TAILQ_INIT(&cleanup);
162	mtx_lock(&pool->sp_lock);
163
164	while (TAILQ_FIRST(&pool->sp_xlist)) {
165		xprt = TAILQ_FIRST(&pool->sp_xlist);
166		xprt_unregister_locked(xprt);
167		TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
168	}
169
170	while (TAILQ_FIRST(&pool->sp_callouts)) {
171		s = TAILQ_FIRST(&pool->sp_callouts);
172		mtx_unlock(&pool->sp_lock);
173		svc_unreg(pool, s->sc_prog, s->sc_vers);
174		mtx_lock(&pool->sp_lock);
175	}
176	mtx_unlock(&pool->sp_lock);
177
178	TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
179		SVC_RELEASE(xprt);
180	}
181
182	mtx_destroy(&pool->sp_lock);
183
184	if (pool->sp_rcache)
185		replay_freecache(pool->sp_rcache);
186
187	sysctl_ctx_free(&pool->sp_sysctl);
188	free(pool, M_RPC);
189}
190
191static bool_t
192svcpool_active(SVCPOOL *pool)
193{
194	enum svcpool_state state = pool->sp_state;
195
196	if (state == SVCPOOL_INIT || state == SVCPOOL_CLOSING)
197		return (FALSE);
198	return (TRUE);
199}
200
201/*
202 * Sysctl handler to set the minimum thread count on a pool
203 */
204static int
205svcpool_minthread_sysctl(SYSCTL_HANDLER_ARGS)
206{
207	SVCPOOL *pool;
208	int newminthreads, error, n;
209
210	pool = oidp->oid_arg1;
211	newminthreads = pool->sp_minthreads;
212	error = sysctl_handle_int(oidp, &newminthreads, 0, req);
213	if (error == 0 && newminthreads != pool->sp_minthreads) {
214		if (newminthreads > pool->sp_maxthreads)
215			return (EINVAL);
216		mtx_lock(&pool->sp_lock);
217		if (newminthreads > pool->sp_minthreads
218		    && svcpool_active(pool)) {
219			/*
220			 * If the pool is running and we are
221			 * increasing, create some more threads now.
222			 */
223			n = newminthreads - pool->sp_threadcount;
224			if (n > 0) {
225				mtx_unlock(&pool->sp_lock);
226				while (n--)
227					svc_new_thread(pool);
228				mtx_lock(&pool->sp_lock);
229			}
230		}
231		pool->sp_minthreads = newminthreads;
232		mtx_unlock(&pool->sp_lock);
233	}
234	return (error);
235}
236
237/*
238 * Sysctl handler to set the maximum thread count on a pool
239 */
240static int
241svcpool_maxthread_sysctl(SYSCTL_HANDLER_ARGS)
242{
243	SVCPOOL *pool;
244	SVCTHREAD *st;
245	int newmaxthreads, error;
246
247	pool = oidp->oid_arg1;
248	newmaxthreads = pool->sp_maxthreads;
249	error = sysctl_handle_int(oidp, &newmaxthreads, 0, req);
250	if (error == 0 && newmaxthreads != pool->sp_maxthreads) {
251		if (newmaxthreads < pool->sp_minthreads)
252			return (EINVAL);
253		mtx_lock(&pool->sp_lock);
254		if (newmaxthreads < pool->sp_maxthreads
255		    && svcpool_active(pool)) {
256			/*
257			 * If the pool is running and we are
258			 * decreasing, wake up some idle threads to
259			 * encourage them to exit.
260			 */
261			LIST_FOREACH(st, &pool->sp_idlethreads, st_ilink)
262				cv_signal(&st->st_cond);
263		}
264		pool->sp_maxthreads = newmaxthreads;
265		mtx_unlock(&pool->sp_lock);
266	}
267	return (error);
268}
269
270/*
271 * Activate a transport handle.
272 */
273void
274xprt_register(SVCXPRT *xprt)
275{
276	SVCPOOL *pool = xprt->xp_pool;
277
278	SVC_ACQUIRE(xprt);
279	mtx_lock(&pool->sp_lock);
280	xprt->xp_registered = TRUE;
281	xprt->xp_active = FALSE;
282	TAILQ_INSERT_TAIL(&pool->sp_xlist, xprt, xp_link);
283	mtx_unlock(&pool->sp_lock);
284}
285
286/*
287 * De-activate a transport handle. Note: the locked version doesn't
288 * release the transport - caller must do that after dropping the pool
289 * lock.
290 */
291static void
292xprt_unregister_locked(SVCXPRT *xprt)
293{
294	SVCPOOL *pool = xprt->xp_pool;
295
296	mtx_assert(&pool->sp_lock, MA_OWNED);
297	KASSERT(xprt->xp_registered == TRUE,
298	    ("xprt_unregister_locked: not registered"));
299	xprt_inactive_locked(xprt);
300	TAILQ_REMOVE(&pool->sp_xlist, xprt, xp_link);
301	xprt->xp_registered = FALSE;
302}
303
304void
305xprt_unregister(SVCXPRT *xprt)
306{
307	SVCPOOL *pool = xprt->xp_pool;
308
309	mtx_lock(&pool->sp_lock);
310	if (xprt->xp_registered == FALSE) {
311		/* Already unregistered by another thread */
312		mtx_unlock(&pool->sp_lock);
313		return;
314	}
315	xprt_unregister_locked(xprt);
316	mtx_unlock(&pool->sp_lock);
317
318	SVC_RELEASE(xprt);
319}
320
321/*
322 * Attempt to assign a service thread to this transport.
323 */
324static int
325xprt_assignthread(SVCXPRT *xprt)
326{
327	SVCPOOL *pool = xprt->xp_pool;
328	SVCTHREAD *st;
329
330	mtx_assert(&pool->sp_lock, MA_OWNED);
331	st = LIST_FIRST(&pool->sp_idlethreads);
332	if (st) {
333		LIST_REMOVE(st, st_ilink);
334		st->st_idle = FALSE;
335		SVC_ACQUIRE(xprt);
336		xprt->xp_thread = st;
337		st->st_xprt = xprt;
338		cv_signal(&st->st_cond);
339		return (TRUE);
340	} else {
341		/*
342		 * See if we can create a new thread. The
343		 * actual thread creation happens in
344		 * svc_run_internal because our locking state
345		 * is poorly defined (we are typically called
346		 * from a socket upcall). Don't create more
347		 * than one thread per second.
348		 */
349		if (pool->sp_state == SVCPOOL_ACTIVE
350		    && pool->sp_lastcreatetime < time_uptime
351		    && pool->sp_threadcount < pool->sp_maxthreads) {
352			pool->sp_state = SVCPOOL_THREADWANTED;
353		}
354	}
355	return (FALSE);
356}
357
358void
359xprt_active(SVCXPRT *xprt)
360{
361	SVCPOOL *pool = xprt->xp_pool;
362
363	mtx_lock(&pool->sp_lock);
364
365	if (!xprt->xp_registered) {
366		/*
367		 * Race with xprt_unregister - we lose.
368		 */
369		mtx_unlock(&pool->sp_lock);
370		return;
371	}
372
373	if (!xprt->xp_active) {
374		xprt->xp_active = TRUE;
375		if (xprt->xp_thread == NULL) {
376			if (!xprt_assignthread(xprt))
377				TAILQ_INSERT_TAIL(&pool->sp_active, xprt,
378				    xp_alink);
379		}
380	}
381
382	mtx_unlock(&pool->sp_lock);
383}
384
385void
386xprt_inactive_locked(SVCXPRT *xprt)
387{
388	SVCPOOL *pool = xprt->xp_pool;
389
390	mtx_assert(&pool->sp_lock, MA_OWNED);
391	if (xprt->xp_active) {
392		if (xprt->xp_thread == NULL)
393			TAILQ_REMOVE(&pool->sp_active, xprt, xp_alink);
394		xprt->xp_active = FALSE;
395	}
396}
397
398void
399xprt_inactive(SVCXPRT *xprt)
400{
401	SVCPOOL *pool = xprt->xp_pool;
402
403	mtx_lock(&pool->sp_lock);
404	xprt_inactive_locked(xprt);
405	mtx_unlock(&pool->sp_lock);
406}
407
408/*
409 * Variant of xprt_inactive() for use only when sure that port is
410 * assigned to thread. For example, withing receive handlers.
411 */
412void
413xprt_inactive_self(SVCXPRT *xprt)
414{
415
416	KASSERT(xprt->xp_thread != NULL,
417	    ("xprt_inactive_self(%p) with NULL xp_thread", xprt));
418	xprt->xp_active = FALSE;
419}
420
421/*
422 * Add a service program to the callout list.
423 * The dispatch routine will be called when a rpc request for this
424 * program number comes in.
425 */
426bool_t
427svc_reg(SVCXPRT *xprt, const rpcprog_t prog, const rpcvers_t vers,
428    void (*dispatch)(struct svc_req *, SVCXPRT *),
429    const struct netconfig *nconf)
430{
431	SVCPOOL *pool = xprt->xp_pool;
432	struct svc_callout *s;
433	char *netid = NULL;
434	int flag = 0;
435
436/* VARIABLES PROTECTED BY svc_lock: s, svc_head */
437
438	if (xprt->xp_netid) {
439		netid = strdup(xprt->xp_netid, M_RPC);
440		flag = 1;
441	} else if (nconf && nconf->nc_netid) {
442		netid = strdup(nconf->nc_netid, M_RPC);
443		flag = 1;
444	} /* must have been created with svc_raw_create */
445	if ((netid == NULL) && (flag == 1)) {
446		return (FALSE);
447	}
448
449	mtx_lock(&pool->sp_lock);
450	if ((s = svc_find(pool, prog, vers, netid)) != NULL) {
451		if (netid)
452			free(netid, M_RPC);
453		if (s->sc_dispatch == dispatch)
454			goto rpcb_it; /* he is registering another xptr */
455		mtx_unlock(&pool->sp_lock);
456		return (FALSE);
457	}
458	s = malloc(sizeof (struct svc_callout), M_RPC, M_NOWAIT);
459	if (s == NULL) {
460		if (netid)
461			free(netid, M_RPC);
462		mtx_unlock(&pool->sp_lock);
463		return (FALSE);
464	}
465
466	s->sc_prog = prog;
467	s->sc_vers = vers;
468	s->sc_dispatch = dispatch;
469	s->sc_netid = netid;
470	TAILQ_INSERT_TAIL(&pool->sp_callouts, s, sc_link);
471
472	if ((xprt->xp_netid == NULL) && (flag == 1) && netid)
473		((SVCXPRT *) xprt)->xp_netid = strdup(netid, M_RPC);
474
475rpcb_it:
476	mtx_unlock(&pool->sp_lock);
477	/* now register the information with the local binder service */
478	if (nconf) {
479		bool_t dummy;
480		struct netconfig tnc;
481		struct netbuf nb;
482		tnc = *nconf;
483		nb.buf = &xprt->xp_ltaddr;
484		nb.len = xprt->xp_ltaddr.ss_len;
485		dummy = rpcb_set(prog, vers, &tnc, &nb);
486		return (dummy);
487	}
488	return (TRUE);
489}
490
491/*
492 * Remove a service program from the callout list.
493 */
494void
495svc_unreg(SVCPOOL *pool, const rpcprog_t prog, const rpcvers_t vers)
496{
497	struct svc_callout *s;
498
499	/* unregister the information anyway */
500	(void) rpcb_unset(prog, vers, NULL);
501	mtx_lock(&pool->sp_lock);
502	while ((s = svc_find(pool, prog, vers, NULL)) != NULL) {
503		TAILQ_REMOVE(&pool->sp_callouts, s, sc_link);
504		if (s->sc_netid)
505			mem_free(s->sc_netid, sizeof (s->sc_netid) + 1);
506		mem_free(s, sizeof (struct svc_callout));
507	}
508	mtx_unlock(&pool->sp_lock);
509}
510
511/* ********************** CALLOUT list related stuff ************* */
512
513/*
514 * Search the callout list for a program number, return the callout
515 * struct.
516 */
517static struct svc_callout *
518svc_find(SVCPOOL *pool, rpcprog_t prog, rpcvers_t vers, char *netid)
519{
520	struct svc_callout *s;
521
522	mtx_assert(&pool->sp_lock, MA_OWNED);
523	TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
524		if (s->sc_prog == prog && s->sc_vers == vers
525		    && (netid == NULL || s->sc_netid == NULL ||
526			strcmp(netid, s->sc_netid) == 0))
527			break;
528	}
529
530	return (s);
531}
532
533/* ******************* REPLY GENERATION ROUTINES  ************ */
534
535static bool_t
536svc_sendreply_common(struct svc_req *rqstp, struct rpc_msg *rply,
537    struct mbuf *body)
538{
539	SVCXPRT *xprt = rqstp->rq_xprt;
540	bool_t ok;
541
542	if (rqstp->rq_args) {
543		m_freem(rqstp->rq_args);
544		rqstp->rq_args = NULL;
545	}
546
547	if (xprt->xp_pool->sp_rcache)
548		replay_setreply(xprt->xp_pool->sp_rcache,
549		    rply, svc_getrpccaller(rqstp), body);
550
551	if (!SVCAUTH_WRAP(&rqstp->rq_auth, &body))
552		return (FALSE);
553
554	ok = SVC_REPLY(xprt, rply, rqstp->rq_addr, body);
555	if (rqstp->rq_addr) {
556		free(rqstp->rq_addr, M_SONAME);
557		rqstp->rq_addr = NULL;
558	}
559
560	return (ok);
561}
562
563/*
564 * Send a reply to an rpc request
565 */
566bool_t
567svc_sendreply(struct svc_req *rqstp, xdrproc_t xdr_results, void * xdr_location)
568{
569	struct rpc_msg rply;
570	struct mbuf *m;
571	XDR xdrs;
572	bool_t ok;
573
574	rply.rm_xid = rqstp->rq_xid;
575	rply.rm_direction = REPLY;
576	rply.rm_reply.rp_stat = MSG_ACCEPTED;
577	rply.acpted_rply.ar_verf = rqstp->rq_verf;
578	rply.acpted_rply.ar_stat = SUCCESS;
579	rply.acpted_rply.ar_results.where = NULL;
580	rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
581
582	m = m_getcl(M_WAITOK, MT_DATA, 0);
583	xdrmbuf_create(&xdrs, m, XDR_ENCODE);
584	ok = xdr_results(&xdrs, xdr_location);
585	XDR_DESTROY(&xdrs);
586
587	if (ok) {
588		return (svc_sendreply_common(rqstp, &rply, m));
589	} else {
590		m_freem(m);
591		return (FALSE);
592	}
593}
594
595bool_t
596svc_sendreply_mbuf(struct svc_req *rqstp, struct mbuf *m)
597{
598	struct rpc_msg rply;
599
600	rply.rm_xid = rqstp->rq_xid;
601	rply.rm_direction = REPLY;
602	rply.rm_reply.rp_stat = MSG_ACCEPTED;
603	rply.acpted_rply.ar_verf = rqstp->rq_verf;
604	rply.acpted_rply.ar_stat = SUCCESS;
605	rply.acpted_rply.ar_results.where = NULL;
606	rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
607
608	return (svc_sendreply_common(rqstp, &rply, m));
609}
610
611/*
612 * No procedure error reply
613 */
614void
615svcerr_noproc(struct svc_req *rqstp)
616{
617	SVCXPRT *xprt = rqstp->rq_xprt;
618	struct rpc_msg rply;
619
620	rply.rm_xid = rqstp->rq_xid;
621	rply.rm_direction = REPLY;
622	rply.rm_reply.rp_stat = MSG_ACCEPTED;
623	rply.acpted_rply.ar_verf = rqstp->rq_verf;
624	rply.acpted_rply.ar_stat = PROC_UNAVAIL;
625
626	if (xprt->xp_pool->sp_rcache)
627		replay_setreply(xprt->xp_pool->sp_rcache,
628		    &rply, svc_getrpccaller(rqstp), NULL);
629
630	svc_sendreply_common(rqstp, &rply, NULL);
631}
632
633/*
634 * Can't decode args error reply
635 */
636void
637svcerr_decode(struct svc_req *rqstp)
638{
639	SVCXPRT *xprt = rqstp->rq_xprt;
640	struct rpc_msg rply;
641
642	rply.rm_xid = rqstp->rq_xid;
643	rply.rm_direction = REPLY;
644	rply.rm_reply.rp_stat = MSG_ACCEPTED;
645	rply.acpted_rply.ar_verf = rqstp->rq_verf;
646	rply.acpted_rply.ar_stat = GARBAGE_ARGS;
647
648	if (xprt->xp_pool->sp_rcache)
649		replay_setreply(xprt->xp_pool->sp_rcache,
650		    &rply, (struct sockaddr *) &xprt->xp_rtaddr, NULL);
651
652	svc_sendreply_common(rqstp, &rply, NULL);
653}
654
655/*
656 * Some system error
657 */
658void
659svcerr_systemerr(struct svc_req *rqstp)
660{
661	SVCXPRT *xprt = rqstp->rq_xprt;
662	struct rpc_msg rply;
663
664	rply.rm_xid = rqstp->rq_xid;
665	rply.rm_direction = REPLY;
666	rply.rm_reply.rp_stat = MSG_ACCEPTED;
667	rply.acpted_rply.ar_verf = rqstp->rq_verf;
668	rply.acpted_rply.ar_stat = SYSTEM_ERR;
669
670	if (xprt->xp_pool->sp_rcache)
671		replay_setreply(xprt->xp_pool->sp_rcache,
672		    &rply, svc_getrpccaller(rqstp), NULL);
673
674	svc_sendreply_common(rqstp, &rply, NULL);
675}
676
677/*
678 * Authentication error reply
679 */
680void
681svcerr_auth(struct svc_req *rqstp, enum auth_stat why)
682{
683	SVCXPRT *xprt = rqstp->rq_xprt;
684	struct rpc_msg rply;
685
686	rply.rm_xid = rqstp->rq_xid;
687	rply.rm_direction = REPLY;
688	rply.rm_reply.rp_stat = MSG_DENIED;
689	rply.rjcted_rply.rj_stat = AUTH_ERROR;
690	rply.rjcted_rply.rj_why = why;
691
692	if (xprt->xp_pool->sp_rcache)
693		replay_setreply(xprt->xp_pool->sp_rcache,
694		    &rply, svc_getrpccaller(rqstp), NULL);
695
696	svc_sendreply_common(rqstp, &rply, NULL);
697}
698
699/*
700 * Auth too weak error reply
701 */
702void
703svcerr_weakauth(struct svc_req *rqstp)
704{
705
706	svcerr_auth(rqstp, AUTH_TOOWEAK);
707}
708
709/*
710 * Program unavailable error reply
711 */
712void
713svcerr_noprog(struct svc_req *rqstp)
714{
715	SVCXPRT *xprt = rqstp->rq_xprt;
716	struct rpc_msg rply;
717
718	rply.rm_xid = rqstp->rq_xid;
719	rply.rm_direction = REPLY;
720	rply.rm_reply.rp_stat = MSG_ACCEPTED;
721	rply.acpted_rply.ar_verf = rqstp->rq_verf;
722	rply.acpted_rply.ar_stat = PROG_UNAVAIL;
723
724	if (xprt->xp_pool->sp_rcache)
725		replay_setreply(xprt->xp_pool->sp_rcache,
726		    &rply, svc_getrpccaller(rqstp), NULL);
727
728	svc_sendreply_common(rqstp, &rply, NULL);
729}
730
731/*
732 * Program version mismatch error reply
733 */
734void
735svcerr_progvers(struct svc_req *rqstp, rpcvers_t low_vers, rpcvers_t high_vers)
736{
737	SVCXPRT *xprt = rqstp->rq_xprt;
738	struct rpc_msg rply;
739
740	rply.rm_xid = rqstp->rq_xid;
741	rply.rm_direction = REPLY;
742	rply.rm_reply.rp_stat = MSG_ACCEPTED;
743	rply.acpted_rply.ar_verf = rqstp->rq_verf;
744	rply.acpted_rply.ar_stat = PROG_MISMATCH;
745	rply.acpted_rply.ar_vers.low = (uint32_t)low_vers;
746	rply.acpted_rply.ar_vers.high = (uint32_t)high_vers;
747
748	if (xprt->xp_pool->sp_rcache)
749		replay_setreply(xprt->xp_pool->sp_rcache,
750		    &rply, svc_getrpccaller(rqstp), NULL);
751
752	svc_sendreply_common(rqstp, &rply, NULL);
753}
754
755/*
756 * Allocate a new server transport structure. All fields are
757 * initialized to zero and xp_p3 is initialized to point at an
758 * extension structure to hold various flags and authentication
759 * parameters.
760 */
761SVCXPRT *
762svc_xprt_alloc()
763{
764	SVCXPRT *xprt;
765	SVCXPRT_EXT *ext;
766
767	xprt = mem_alloc(sizeof(SVCXPRT));
768	memset(xprt, 0, sizeof(SVCXPRT));
769	ext = mem_alloc(sizeof(SVCXPRT_EXT));
770	memset(ext, 0, sizeof(SVCXPRT_EXT));
771	xprt->xp_p3 = ext;
772	refcount_init(&xprt->xp_refs, 1);
773
774	return (xprt);
775}
776
777/*
778 * Free a server transport structure.
779 */
780void
781svc_xprt_free(xprt)
782	SVCXPRT *xprt;
783{
784
785	mem_free(xprt->xp_p3, sizeof(SVCXPRT_EXT));
786	mem_free(xprt, sizeof(SVCXPRT));
787}
788
789/* ******************* SERVER INPUT STUFF ******************* */
790
791/*
792 * Read RPC requests from a transport and queue them to be
793 * executed. We handle authentication and replay cache replies here.
794 * Actually dispatching the RPC is deferred till svc_executereq.
795 */
796static enum xprt_stat
797svc_getreq(SVCXPRT *xprt, struct svc_req **rqstp_ret)
798{
799	SVCPOOL *pool = xprt->xp_pool;
800	struct svc_req *r;
801	struct rpc_msg msg;
802	struct mbuf *args;
803	enum xprt_stat stat;
804
805	/* now receive msgs from xprtprt (support batch calls) */
806	r = malloc(sizeof(*r), M_RPC, M_WAITOK|M_ZERO);
807
808	msg.rm_call.cb_cred.oa_base = r->rq_credarea;
809	msg.rm_call.cb_verf.oa_base = &r->rq_credarea[MAX_AUTH_BYTES];
810	r->rq_clntcred = &r->rq_credarea[2*MAX_AUTH_BYTES];
811	if (SVC_RECV(xprt, &msg, &r->rq_addr, &args)) {
812		enum auth_stat why;
813
814		/*
815		 * Handle replays and authenticate before queuing the
816		 * request to be executed.
817		 */
818		SVC_ACQUIRE(xprt);
819		r->rq_xprt = xprt;
820		if (pool->sp_rcache) {
821			struct rpc_msg repmsg;
822			struct mbuf *repbody;
823			enum replay_state rs;
824			rs = replay_find(pool->sp_rcache, &msg,
825			    svc_getrpccaller(r), &repmsg, &repbody);
826			switch (rs) {
827			case RS_NEW:
828				break;
829			case RS_DONE:
830				SVC_REPLY(xprt, &repmsg, r->rq_addr,
831				    repbody);
832				if (r->rq_addr) {
833					free(r->rq_addr, M_SONAME);
834					r->rq_addr = NULL;
835				}
836				m_freem(args);
837				goto call_done;
838
839			default:
840				m_freem(args);
841				goto call_done;
842			}
843		}
844
845		r->rq_xid = msg.rm_xid;
846		r->rq_prog = msg.rm_call.cb_prog;
847		r->rq_vers = msg.rm_call.cb_vers;
848		r->rq_proc = msg.rm_call.cb_proc;
849		r->rq_size = sizeof(*r) + m_length(args, NULL);
850		r->rq_args = args;
851		if ((why = _authenticate(r, &msg)) != AUTH_OK) {
852			/*
853			 * RPCSEC_GSS uses this return code
854			 * for requests that form part of its
855			 * context establishment protocol and
856			 * should not be dispatched to the
857			 * application.
858			 */
859			if (why != RPCSEC_GSS_NODISPATCH)
860				svcerr_auth(r, why);
861			goto call_done;
862		}
863
864		if (!SVCAUTH_UNWRAP(&r->rq_auth, &r->rq_args)) {
865			svcerr_decode(r);
866			goto call_done;
867		}
868
869		/*
870		 * Everything checks out, return request to caller.
871		 */
872		*rqstp_ret = r;
873		r = NULL;
874	}
875call_done:
876	if (r) {
877		svc_freereq(r);
878		r = NULL;
879	}
880	if ((stat = SVC_STAT(xprt)) == XPRT_DIED) {
881		xprt_unregister(xprt);
882	}
883
884	return (stat);
885}
886
887static void
888svc_executereq(struct svc_req *rqstp)
889{
890	SVCXPRT *xprt = rqstp->rq_xprt;
891	SVCPOOL *pool = xprt->xp_pool;
892	int prog_found;
893	rpcvers_t low_vers;
894	rpcvers_t high_vers;
895	struct svc_callout *s;
896
897	/* now match message with a registered service*/
898	prog_found = FALSE;
899	low_vers = (rpcvers_t) -1L;
900	high_vers = (rpcvers_t) 0L;
901	TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
902		if (s->sc_prog == rqstp->rq_prog) {
903			if (s->sc_vers == rqstp->rq_vers) {
904				/*
905				 * We hand ownership of r to the
906				 * dispatch method - they must call
907				 * svc_freereq.
908				 */
909				(*s->sc_dispatch)(rqstp, xprt);
910				return;
911			}  /* found correct version */
912			prog_found = TRUE;
913			if (s->sc_vers < low_vers)
914				low_vers = s->sc_vers;
915			if (s->sc_vers > high_vers)
916				high_vers = s->sc_vers;
917		}   /* found correct program */
918	}
919
920	/*
921	 * if we got here, the program or version
922	 * is not served ...
923	 */
924	if (prog_found)
925		svcerr_progvers(rqstp, low_vers, high_vers);
926	else
927		svcerr_noprog(rqstp);
928
929	svc_freereq(rqstp);
930}
931
932static void
933svc_checkidle(SVCPOOL *pool)
934{
935	SVCXPRT *xprt, *nxprt;
936	time_t timo;
937	struct svcxprt_list cleanup;
938
939	TAILQ_INIT(&cleanup);
940	TAILQ_FOREACH_SAFE(xprt, &pool->sp_xlist, xp_link, nxprt) {
941		/*
942		 * Only some transports have idle timers. Don't time
943		 * something out which is just waking up.
944		 */
945		if (!xprt->xp_idletimeout || xprt->xp_thread)
946			continue;
947
948		timo = xprt->xp_lastactive + xprt->xp_idletimeout;
949		if (time_uptime > timo) {
950			xprt_unregister_locked(xprt);
951			TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
952		}
953	}
954
955	mtx_unlock(&pool->sp_lock);
956	TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
957		SVC_RELEASE(xprt);
958	}
959	mtx_lock(&pool->sp_lock);
960
961}
962
963static void
964svc_assign_waiting_sockets(SVCPOOL *pool)
965{
966	SVCXPRT *xprt;
967
968	while ((xprt = TAILQ_FIRST(&pool->sp_active)) != NULL) {
969		if (xprt_assignthread(xprt))
970			TAILQ_REMOVE(&pool->sp_active, xprt, xp_alink);
971		else
972			break;
973	}
974}
975
976static bool_t
977svc_request_space_available(SVCPOOL *pool)
978{
979
980	mtx_assert(&pool->sp_lock, MA_OWNED);
981
982	if (pool->sp_space_throttled) {
983		/*
984		 * Below the low-water yet? If so, assign any waiting sockets.
985		 */
986		if (pool->sp_space_used < pool->sp_space_low) {
987			pool->sp_space_throttled = FALSE;
988			svc_assign_waiting_sockets(pool);
989			return TRUE;
990		}
991
992		return FALSE;
993	} else {
994		if (pool->sp_space_used
995		    >= pool->sp_space_high) {
996			pool->sp_space_throttled = TRUE;
997			pool->sp_space_throttle_count++;
998			return FALSE;
999		}
1000
1001		return TRUE;
1002	}
1003}
1004
1005static void
1006svc_run_internal(SVCPOOL *pool, bool_t ismaster)
1007{
1008	SVCTHREAD *st, *stpref;
1009	SVCXPRT *xprt;
1010	enum xprt_stat stat;
1011	struct svc_req *rqstp;
1012	int error;
1013
1014	st = mem_alloc(sizeof(*st));
1015	st->st_xprt = NULL;
1016	STAILQ_INIT(&st->st_reqs);
1017	cv_init(&st->st_cond, "rpcsvc");
1018
1019	mtx_lock(&pool->sp_lock);
1020	LIST_INSERT_HEAD(&pool->sp_threads, st, st_link);
1021
1022	/*
1023	 * If we are a new thread which was spawned to cope with
1024	 * increased load, set the state back to SVCPOOL_ACTIVE.
1025	 */
1026	if (pool->sp_state == SVCPOOL_THREADSTARTING)
1027		pool->sp_state = SVCPOOL_ACTIVE;
1028
1029	while (pool->sp_state != SVCPOOL_CLOSING) {
1030		/*
1031		 * Create new thread if requested.
1032		 */
1033		if (pool->sp_state == SVCPOOL_THREADWANTED) {
1034			pool->sp_state = SVCPOOL_THREADSTARTING;
1035			pool->sp_lastcreatetime = time_uptime;
1036			mtx_unlock(&pool->sp_lock);
1037			svc_new_thread(pool);
1038			mtx_lock(&pool->sp_lock);
1039			continue;
1040		}
1041
1042		/*
1043		 * Check for idle transports once per second.
1044		 */
1045		if (time_uptime > pool->sp_lastidlecheck) {
1046			pool->sp_lastidlecheck = time_uptime;
1047			svc_checkidle(pool);
1048		}
1049
1050		xprt = st->st_xprt;
1051		if (!xprt && STAILQ_EMPTY(&st->st_reqs)) {
1052			/*
1053			 * Enforce maxthreads count.
1054			 */
1055			if (pool->sp_threadcount > pool->sp_maxthreads)
1056				break;
1057
1058			/*
1059			 * Before sleeping, see if we can find an
1060			 * active transport which isn't being serviced
1061			 * by a thread.
1062			 */
1063			if (svc_request_space_available(pool) &&
1064			    (xprt = TAILQ_FIRST(&pool->sp_active)) != NULL) {
1065				TAILQ_REMOVE(&pool->sp_active, xprt, xp_alink);
1066				SVC_ACQUIRE(xprt);
1067				xprt->xp_thread = st;
1068				st->st_xprt = xprt;
1069				continue;
1070			}
1071
1072			LIST_INSERT_HEAD(&pool->sp_idlethreads, st, st_ilink);
1073			st->st_idle = TRUE;
1074			if (ismaster || (!ismaster &&
1075			    pool->sp_threadcount > pool->sp_minthreads))
1076				error = cv_timedwait_sig(&st->st_cond,
1077				    &pool->sp_lock, 5 * hz);
1078			else
1079				error = cv_wait_sig(&st->st_cond,
1080				    &pool->sp_lock);
1081			if (st->st_idle) {
1082				LIST_REMOVE(st, st_ilink);
1083				st->st_idle = FALSE;
1084			}
1085
1086			/*
1087			 * Reduce worker thread count when idle.
1088			 */
1089			if (error == EWOULDBLOCK) {
1090				if (!ismaster
1091				    && (pool->sp_threadcount
1092					> pool->sp_minthreads)
1093					&& !st->st_xprt
1094					&& STAILQ_EMPTY(&st->st_reqs))
1095					break;
1096			} else if (error) {
1097				mtx_unlock(&pool->sp_lock);
1098				svc_exit(pool);
1099				mtx_lock(&pool->sp_lock);
1100				break;
1101			}
1102			continue;
1103		}
1104
1105		if (xprt) {
1106			/*
1107			 * Drain the transport socket and queue up any
1108			 * RPCs.
1109			 */
1110			xprt->xp_lastactive = time_uptime;
1111			stat = XPRT_IDLE;
1112			do {
1113				if (!svc_request_space_available(pool))
1114					break;
1115				rqstp = NULL;
1116				mtx_unlock(&pool->sp_lock);
1117				stat = svc_getreq(xprt, &rqstp);
1118				mtx_lock(&pool->sp_lock);
1119				if (rqstp) {
1120					/*
1121					 * See if the application has
1122					 * a preference for some other
1123					 * thread.
1124					 */
1125					stpref = st;
1126					if (pool->sp_assign)
1127						stpref = pool->sp_assign(st,
1128						    rqstp);
1129
1130					pool->sp_space_used +=
1131						rqstp->rq_size;
1132					if (pool->sp_space_used
1133					    > pool->sp_space_used_highest)
1134						pool->sp_space_used_highest =
1135							pool->sp_space_used;
1136					rqstp->rq_thread = stpref;
1137					STAILQ_INSERT_TAIL(&stpref->st_reqs,
1138					    rqstp, rq_link);
1139					stpref->st_reqcount++;
1140
1141					/*
1142					 * If we assigned the request
1143					 * to another thread, make
1144					 * sure its awake and continue
1145					 * reading from the
1146					 * socket. Otherwise, try to
1147					 * find some other thread to
1148					 * read from the socket and
1149					 * execute the request
1150					 * immediately.
1151					 */
1152					if (stpref == st)
1153						break;
1154					if (stpref->st_idle) {
1155						LIST_REMOVE(stpref, st_ilink);
1156						stpref->st_idle = FALSE;
1157						cv_signal(&stpref->st_cond);
1158					}
1159				}
1160			} while (stat == XPRT_MOREREQS
1161			    && pool->sp_state != SVCPOOL_CLOSING);
1162
1163			/*
1164			 * Move this transport to the end of the
1165			 * active list to ensure fairness when
1166			 * multiple transports are active. If this was
1167			 * the last queued request, svc_getreq will
1168			 * end up calling xprt_inactive to remove from
1169			 * the active list.
1170			 */
1171			xprt->xp_thread = NULL;
1172			st->st_xprt = NULL;
1173			if (xprt->xp_active) {
1174				if (!xprt_assignthread(xprt))
1175					TAILQ_INSERT_TAIL(&pool->sp_active,
1176					    xprt, xp_alink);
1177			}
1178			mtx_unlock(&pool->sp_lock);
1179			SVC_RELEASE(xprt);
1180			mtx_lock(&pool->sp_lock);
1181		}
1182
1183		/*
1184		 * Execute what we have queued.
1185		 */
1186		while ((rqstp = STAILQ_FIRST(&st->st_reqs)) != NULL) {
1187			size_t sz = rqstp->rq_size;
1188			mtx_unlock(&pool->sp_lock);
1189			svc_executereq(rqstp);
1190			mtx_lock(&pool->sp_lock);
1191			pool->sp_space_used -= sz;
1192		}
1193	}
1194
1195	if (st->st_xprt) {
1196		xprt = st->st_xprt;
1197		st->st_xprt = NULL;
1198		SVC_RELEASE(xprt);
1199	}
1200
1201	KASSERT(STAILQ_EMPTY(&st->st_reqs), ("stray reqs on exit"));
1202	LIST_REMOVE(st, st_link);
1203	pool->sp_threadcount--;
1204
1205	mtx_unlock(&pool->sp_lock);
1206
1207	cv_destroy(&st->st_cond);
1208	mem_free(st, sizeof(*st));
1209
1210	if (!ismaster)
1211		wakeup(pool);
1212}
1213
1214static void
1215svc_thread_start(void *arg)
1216{
1217
1218	svc_run_internal((SVCPOOL *) arg, FALSE);
1219	kthread_exit();
1220}
1221
1222static void
1223svc_new_thread(SVCPOOL *pool)
1224{
1225	struct thread *td;
1226
1227	pool->sp_threadcount++;
1228	kthread_add(svc_thread_start, pool,
1229	    pool->sp_proc, &td, 0, 0,
1230	    "%s: service", pool->sp_name);
1231}
1232
1233void
1234svc_run(SVCPOOL *pool)
1235{
1236	int i;
1237	struct proc *p;
1238	struct thread *td;
1239
1240	p = curproc;
1241	td = curthread;
1242	snprintf(td->td_name, sizeof(td->td_name),
1243	    "%s: master", pool->sp_name);
1244	pool->sp_state = SVCPOOL_ACTIVE;
1245	pool->sp_proc = p;
1246	pool->sp_lastcreatetime = time_uptime;
1247	pool->sp_threadcount = 1;
1248
1249	for (i = 1; i < pool->sp_minthreads; i++) {
1250		svc_new_thread(pool);
1251	}
1252
1253	svc_run_internal(pool, TRUE);
1254
1255	mtx_lock(&pool->sp_lock);
1256	while (pool->sp_threadcount > 0)
1257		msleep(pool, &pool->sp_lock, 0, "svcexit", 0);
1258	mtx_unlock(&pool->sp_lock);
1259}
1260
1261void
1262svc_exit(SVCPOOL *pool)
1263{
1264	SVCTHREAD *st;
1265
1266	mtx_lock(&pool->sp_lock);
1267
1268	if (pool->sp_state != SVCPOOL_CLOSING) {
1269		pool->sp_state = SVCPOOL_CLOSING;
1270		LIST_FOREACH(st, &pool->sp_idlethreads, st_ilink)
1271			cv_signal(&st->st_cond);
1272	}
1273
1274	mtx_unlock(&pool->sp_lock);
1275}
1276
1277bool_t
1278svc_getargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
1279{
1280	struct mbuf *m;
1281	XDR xdrs;
1282	bool_t stat;
1283
1284	m = rqstp->rq_args;
1285	rqstp->rq_args = NULL;
1286
1287	xdrmbuf_create(&xdrs, m, XDR_DECODE);
1288	stat = xargs(&xdrs, args);
1289	XDR_DESTROY(&xdrs);
1290
1291	return (stat);
1292}
1293
1294bool_t
1295svc_freeargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
1296{
1297	XDR xdrs;
1298
1299	if (rqstp->rq_addr) {
1300		free(rqstp->rq_addr, M_SONAME);
1301		rqstp->rq_addr = NULL;
1302	}
1303
1304	xdrs.x_op = XDR_FREE;
1305	return (xargs(&xdrs, args));
1306}
1307
1308void
1309svc_freereq(struct svc_req *rqstp)
1310{
1311	SVCTHREAD *st;
1312	SVCXPRT *xprt;
1313	SVCPOOL *pool;
1314
1315	st = rqstp->rq_thread;
1316	xprt = rqstp->rq_xprt;
1317	if (xprt)
1318		pool = xprt->xp_pool;
1319	else
1320		pool = NULL;
1321	if (st) {
1322		mtx_lock(&pool->sp_lock);
1323		KASSERT(rqstp == STAILQ_FIRST(&st->st_reqs),
1324		    ("Freeing request out of order"));
1325		STAILQ_REMOVE_HEAD(&st->st_reqs, rq_link);
1326		st->st_reqcount--;
1327		if (pool->sp_done)
1328			pool->sp_done(st, rqstp);
1329		mtx_unlock(&pool->sp_lock);
1330	}
1331
1332	if (rqstp->rq_auth.svc_ah_ops)
1333		SVCAUTH_RELEASE(&rqstp->rq_auth);
1334
1335	if (rqstp->rq_xprt) {
1336		SVC_RELEASE(rqstp->rq_xprt);
1337	}
1338
1339	if (rqstp->rq_addr)
1340		free(rqstp->rq_addr, M_SONAME);
1341
1342	if (rqstp->rq_args)
1343		m_freem(rqstp->rq_args);
1344
1345	free(rqstp, M_RPC);
1346}
1347