svc.c revision 261048
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 261048 2014-01-22 23:47:29Z 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 * Add a service program to the callout list.
410 * The dispatch routine will be called when a rpc request for this
411 * program number comes in.
412 */
413bool_t
414svc_reg(SVCXPRT *xprt, const rpcprog_t prog, const rpcvers_t vers,
415    void (*dispatch)(struct svc_req *, SVCXPRT *),
416    const struct netconfig *nconf)
417{
418	SVCPOOL *pool = xprt->xp_pool;
419	struct svc_callout *s;
420	char *netid = NULL;
421	int flag = 0;
422
423/* VARIABLES PROTECTED BY svc_lock: s, svc_head */
424
425	if (xprt->xp_netid) {
426		netid = strdup(xprt->xp_netid, M_RPC);
427		flag = 1;
428	} else if (nconf && nconf->nc_netid) {
429		netid = strdup(nconf->nc_netid, M_RPC);
430		flag = 1;
431	} /* must have been created with svc_raw_create */
432	if ((netid == NULL) && (flag == 1)) {
433		return (FALSE);
434	}
435
436	mtx_lock(&pool->sp_lock);
437	if ((s = svc_find(pool, prog, vers, netid)) != NULL) {
438		if (netid)
439			free(netid, M_RPC);
440		if (s->sc_dispatch == dispatch)
441			goto rpcb_it; /* he is registering another xptr */
442		mtx_unlock(&pool->sp_lock);
443		return (FALSE);
444	}
445	s = malloc(sizeof (struct svc_callout), M_RPC, M_NOWAIT);
446	if (s == NULL) {
447		if (netid)
448			free(netid, M_RPC);
449		mtx_unlock(&pool->sp_lock);
450		return (FALSE);
451	}
452
453	s->sc_prog = prog;
454	s->sc_vers = vers;
455	s->sc_dispatch = dispatch;
456	s->sc_netid = netid;
457	TAILQ_INSERT_TAIL(&pool->sp_callouts, s, sc_link);
458
459	if ((xprt->xp_netid == NULL) && (flag == 1) && netid)
460		((SVCXPRT *) xprt)->xp_netid = strdup(netid, M_RPC);
461
462rpcb_it:
463	mtx_unlock(&pool->sp_lock);
464	/* now register the information with the local binder service */
465	if (nconf) {
466		bool_t dummy;
467		struct netconfig tnc;
468		struct netbuf nb;
469		tnc = *nconf;
470		nb.buf = &xprt->xp_ltaddr;
471		nb.len = xprt->xp_ltaddr.ss_len;
472		dummy = rpcb_set(prog, vers, &tnc, &nb);
473		return (dummy);
474	}
475	return (TRUE);
476}
477
478/*
479 * Remove a service program from the callout list.
480 */
481void
482svc_unreg(SVCPOOL *pool, const rpcprog_t prog, const rpcvers_t vers)
483{
484	struct svc_callout *s;
485
486	/* unregister the information anyway */
487	(void) rpcb_unset(prog, vers, NULL);
488	mtx_lock(&pool->sp_lock);
489	while ((s = svc_find(pool, prog, vers, NULL)) != NULL) {
490		TAILQ_REMOVE(&pool->sp_callouts, s, sc_link);
491		if (s->sc_netid)
492			mem_free(s->sc_netid, sizeof (s->sc_netid) + 1);
493		mem_free(s, sizeof (struct svc_callout));
494	}
495	mtx_unlock(&pool->sp_lock);
496}
497
498/* ********************** CALLOUT list related stuff ************* */
499
500/*
501 * Search the callout list for a program number, return the callout
502 * struct.
503 */
504static struct svc_callout *
505svc_find(SVCPOOL *pool, rpcprog_t prog, rpcvers_t vers, char *netid)
506{
507	struct svc_callout *s;
508
509	mtx_assert(&pool->sp_lock, MA_OWNED);
510	TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
511		if (s->sc_prog == prog && s->sc_vers == vers
512		    && (netid == NULL || s->sc_netid == NULL ||
513			strcmp(netid, s->sc_netid) == 0))
514			break;
515	}
516
517	return (s);
518}
519
520/* ******************* REPLY GENERATION ROUTINES  ************ */
521
522static bool_t
523svc_sendreply_common(struct svc_req *rqstp, struct rpc_msg *rply,
524    struct mbuf *body)
525{
526	SVCXPRT *xprt = rqstp->rq_xprt;
527	bool_t ok;
528
529	if (rqstp->rq_args) {
530		m_freem(rqstp->rq_args);
531		rqstp->rq_args = NULL;
532	}
533
534	if (xprt->xp_pool->sp_rcache)
535		replay_setreply(xprt->xp_pool->sp_rcache,
536		    rply, svc_getrpccaller(rqstp), body);
537
538	if (!SVCAUTH_WRAP(&rqstp->rq_auth, &body))
539		return (FALSE);
540
541	ok = SVC_REPLY(xprt, rply, rqstp->rq_addr, body);
542	if (rqstp->rq_addr) {
543		free(rqstp->rq_addr, M_SONAME);
544		rqstp->rq_addr = NULL;
545	}
546
547	return (ok);
548}
549
550/*
551 * Send a reply to an rpc request
552 */
553bool_t
554svc_sendreply(struct svc_req *rqstp, xdrproc_t xdr_results, void * xdr_location)
555{
556	struct rpc_msg rply;
557	struct mbuf *m;
558	XDR xdrs;
559	bool_t ok;
560
561	rply.rm_xid = rqstp->rq_xid;
562	rply.rm_direction = REPLY;
563	rply.rm_reply.rp_stat = MSG_ACCEPTED;
564	rply.acpted_rply.ar_verf = rqstp->rq_verf;
565	rply.acpted_rply.ar_stat = SUCCESS;
566	rply.acpted_rply.ar_results.where = NULL;
567	rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
568
569	m = m_getcl(M_WAITOK, MT_DATA, 0);
570	xdrmbuf_create(&xdrs, m, XDR_ENCODE);
571	ok = xdr_results(&xdrs, xdr_location);
572	XDR_DESTROY(&xdrs);
573
574	if (ok) {
575		return (svc_sendreply_common(rqstp, &rply, m));
576	} else {
577		m_freem(m);
578		return (FALSE);
579	}
580}
581
582bool_t
583svc_sendreply_mbuf(struct svc_req *rqstp, struct mbuf *m)
584{
585	struct rpc_msg rply;
586
587	rply.rm_xid = rqstp->rq_xid;
588	rply.rm_direction = REPLY;
589	rply.rm_reply.rp_stat = MSG_ACCEPTED;
590	rply.acpted_rply.ar_verf = rqstp->rq_verf;
591	rply.acpted_rply.ar_stat = SUCCESS;
592	rply.acpted_rply.ar_results.where = NULL;
593	rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
594
595	return (svc_sendreply_common(rqstp, &rply, m));
596}
597
598/*
599 * No procedure error reply
600 */
601void
602svcerr_noproc(struct svc_req *rqstp)
603{
604	SVCXPRT *xprt = rqstp->rq_xprt;
605	struct rpc_msg rply;
606
607	rply.rm_xid = rqstp->rq_xid;
608	rply.rm_direction = REPLY;
609	rply.rm_reply.rp_stat = MSG_ACCEPTED;
610	rply.acpted_rply.ar_verf = rqstp->rq_verf;
611	rply.acpted_rply.ar_stat = PROC_UNAVAIL;
612
613	if (xprt->xp_pool->sp_rcache)
614		replay_setreply(xprt->xp_pool->sp_rcache,
615		    &rply, svc_getrpccaller(rqstp), NULL);
616
617	svc_sendreply_common(rqstp, &rply, NULL);
618}
619
620/*
621 * Can't decode args error reply
622 */
623void
624svcerr_decode(struct svc_req *rqstp)
625{
626	SVCXPRT *xprt = rqstp->rq_xprt;
627	struct rpc_msg rply;
628
629	rply.rm_xid = rqstp->rq_xid;
630	rply.rm_direction = REPLY;
631	rply.rm_reply.rp_stat = MSG_ACCEPTED;
632	rply.acpted_rply.ar_verf = rqstp->rq_verf;
633	rply.acpted_rply.ar_stat = GARBAGE_ARGS;
634
635	if (xprt->xp_pool->sp_rcache)
636		replay_setreply(xprt->xp_pool->sp_rcache,
637		    &rply, (struct sockaddr *) &xprt->xp_rtaddr, NULL);
638
639	svc_sendreply_common(rqstp, &rply, NULL);
640}
641
642/*
643 * Some system error
644 */
645void
646svcerr_systemerr(struct svc_req *rqstp)
647{
648	SVCXPRT *xprt = rqstp->rq_xprt;
649	struct rpc_msg rply;
650
651	rply.rm_xid = rqstp->rq_xid;
652	rply.rm_direction = REPLY;
653	rply.rm_reply.rp_stat = MSG_ACCEPTED;
654	rply.acpted_rply.ar_verf = rqstp->rq_verf;
655	rply.acpted_rply.ar_stat = SYSTEM_ERR;
656
657	if (xprt->xp_pool->sp_rcache)
658		replay_setreply(xprt->xp_pool->sp_rcache,
659		    &rply, svc_getrpccaller(rqstp), NULL);
660
661	svc_sendreply_common(rqstp, &rply, NULL);
662}
663
664/*
665 * Authentication error reply
666 */
667void
668svcerr_auth(struct svc_req *rqstp, enum auth_stat why)
669{
670	SVCXPRT *xprt = rqstp->rq_xprt;
671	struct rpc_msg rply;
672
673	rply.rm_xid = rqstp->rq_xid;
674	rply.rm_direction = REPLY;
675	rply.rm_reply.rp_stat = MSG_DENIED;
676	rply.rjcted_rply.rj_stat = AUTH_ERROR;
677	rply.rjcted_rply.rj_why = why;
678
679	if (xprt->xp_pool->sp_rcache)
680		replay_setreply(xprt->xp_pool->sp_rcache,
681		    &rply, svc_getrpccaller(rqstp), NULL);
682
683	svc_sendreply_common(rqstp, &rply, NULL);
684}
685
686/*
687 * Auth too weak error reply
688 */
689void
690svcerr_weakauth(struct svc_req *rqstp)
691{
692
693	svcerr_auth(rqstp, AUTH_TOOWEAK);
694}
695
696/*
697 * Program unavailable error reply
698 */
699void
700svcerr_noprog(struct svc_req *rqstp)
701{
702	SVCXPRT *xprt = rqstp->rq_xprt;
703	struct rpc_msg rply;
704
705	rply.rm_xid = rqstp->rq_xid;
706	rply.rm_direction = REPLY;
707	rply.rm_reply.rp_stat = MSG_ACCEPTED;
708	rply.acpted_rply.ar_verf = rqstp->rq_verf;
709	rply.acpted_rply.ar_stat = PROG_UNAVAIL;
710
711	if (xprt->xp_pool->sp_rcache)
712		replay_setreply(xprt->xp_pool->sp_rcache,
713		    &rply, svc_getrpccaller(rqstp), NULL);
714
715	svc_sendreply_common(rqstp, &rply, NULL);
716}
717
718/*
719 * Program version mismatch error reply
720 */
721void
722svcerr_progvers(struct svc_req *rqstp, rpcvers_t low_vers, rpcvers_t high_vers)
723{
724	SVCXPRT *xprt = rqstp->rq_xprt;
725	struct rpc_msg rply;
726
727	rply.rm_xid = rqstp->rq_xid;
728	rply.rm_direction = REPLY;
729	rply.rm_reply.rp_stat = MSG_ACCEPTED;
730	rply.acpted_rply.ar_verf = rqstp->rq_verf;
731	rply.acpted_rply.ar_stat = PROG_MISMATCH;
732	rply.acpted_rply.ar_vers.low = (uint32_t)low_vers;
733	rply.acpted_rply.ar_vers.high = (uint32_t)high_vers;
734
735	if (xprt->xp_pool->sp_rcache)
736		replay_setreply(xprt->xp_pool->sp_rcache,
737		    &rply, svc_getrpccaller(rqstp), NULL);
738
739	svc_sendreply_common(rqstp, &rply, NULL);
740}
741
742/*
743 * Allocate a new server transport structure. All fields are
744 * initialized to zero and xp_p3 is initialized to point at an
745 * extension structure to hold various flags and authentication
746 * parameters.
747 */
748SVCXPRT *
749svc_xprt_alloc()
750{
751	SVCXPRT *xprt;
752	SVCXPRT_EXT *ext;
753
754	xprt = mem_alloc(sizeof(SVCXPRT));
755	memset(xprt, 0, sizeof(SVCXPRT));
756	ext = mem_alloc(sizeof(SVCXPRT_EXT));
757	memset(ext, 0, sizeof(SVCXPRT_EXT));
758	xprt->xp_p3 = ext;
759	refcount_init(&xprt->xp_refs, 1);
760
761	return (xprt);
762}
763
764/*
765 * Free a server transport structure.
766 */
767void
768svc_xprt_free(xprt)
769	SVCXPRT *xprt;
770{
771
772	mem_free(xprt->xp_p3, sizeof(SVCXPRT_EXT));
773	mem_free(xprt, sizeof(SVCXPRT));
774}
775
776/* ******************* SERVER INPUT STUFF ******************* */
777
778/*
779 * Read RPC requests from a transport and queue them to be
780 * executed. We handle authentication and replay cache replies here.
781 * Actually dispatching the RPC is deferred till svc_executereq.
782 */
783static enum xprt_stat
784svc_getreq(SVCXPRT *xprt, struct svc_req **rqstp_ret)
785{
786	SVCPOOL *pool = xprt->xp_pool;
787	struct svc_req *r;
788	struct rpc_msg msg;
789	struct mbuf *args;
790	enum xprt_stat stat;
791
792	/* now receive msgs from xprtprt (support batch calls) */
793	r = malloc(sizeof(*r), M_RPC, M_WAITOK|M_ZERO);
794
795	msg.rm_call.cb_cred.oa_base = r->rq_credarea;
796	msg.rm_call.cb_verf.oa_base = &r->rq_credarea[MAX_AUTH_BYTES];
797	r->rq_clntcred = &r->rq_credarea[2*MAX_AUTH_BYTES];
798	if (SVC_RECV(xprt, &msg, &r->rq_addr, &args)) {
799		enum auth_stat why;
800
801		/*
802		 * Handle replays and authenticate before queuing the
803		 * request to be executed.
804		 */
805		SVC_ACQUIRE(xprt);
806		r->rq_xprt = xprt;
807		if (pool->sp_rcache) {
808			struct rpc_msg repmsg;
809			struct mbuf *repbody;
810			enum replay_state rs;
811			rs = replay_find(pool->sp_rcache, &msg,
812			    svc_getrpccaller(r), &repmsg, &repbody);
813			switch (rs) {
814			case RS_NEW:
815				break;
816			case RS_DONE:
817				SVC_REPLY(xprt, &repmsg, r->rq_addr,
818				    repbody);
819				if (r->rq_addr) {
820					free(r->rq_addr, M_SONAME);
821					r->rq_addr = NULL;
822				}
823				m_freem(args);
824				goto call_done;
825
826			default:
827				m_freem(args);
828				goto call_done;
829			}
830		}
831
832		r->rq_xid = msg.rm_xid;
833		r->rq_prog = msg.rm_call.cb_prog;
834		r->rq_vers = msg.rm_call.cb_vers;
835		r->rq_proc = msg.rm_call.cb_proc;
836		r->rq_size = sizeof(*r) + m_length(args, NULL);
837		r->rq_args = args;
838		if ((why = _authenticate(r, &msg)) != AUTH_OK) {
839			/*
840			 * RPCSEC_GSS uses this return code
841			 * for requests that form part of its
842			 * context establishment protocol and
843			 * should not be dispatched to the
844			 * application.
845			 */
846			if (why != RPCSEC_GSS_NODISPATCH)
847				svcerr_auth(r, why);
848			goto call_done;
849		}
850
851		if (!SVCAUTH_UNWRAP(&r->rq_auth, &r->rq_args)) {
852			svcerr_decode(r);
853			goto call_done;
854		}
855
856		/*
857		 * Everything checks out, return request to caller.
858		 */
859		*rqstp_ret = r;
860		r = NULL;
861	}
862call_done:
863	if (r) {
864		svc_freereq(r);
865		r = NULL;
866	}
867	if ((stat = SVC_STAT(xprt)) == XPRT_DIED) {
868		xprt_unregister(xprt);
869	}
870
871	return (stat);
872}
873
874static void
875svc_executereq(struct svc_req *rqstp)
876{
877	SVCXPRT *xprt = rqstp->rq_xprt;
878	SVCPOOL *pool = xprt->xp_pool;
879	int prog_found;
880	rpcvers_t low_vers;
881	rpcvers_t high_vers;
882	struct svc_callout *s;
883
884	/* now match message with a registered service*/
885	prog_found = FALSE;
886	low_vers = (rpcvers_t) -1L;
887	high_vers = (rpcvers_t) 0L;
888	TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
889		if (s->sc_prog == rqstp->rq_prog) {
890			if (s->sc_vers == rqstp->rq_vers) {
891				/*
892				 * We hand ownership of r to the
893				 * dispatch method - they must call
894				 * svc_freereq.
895				 */
896				(*s->sc_dispatch)(rqstp, xprt);
897				return;
898			}  /* found correct version */
899			prog_found = TRUE;
900			if (s->sc_vers < low_vers)
901				low_vers = s->sc_vers;
902			if (s->sc_vers > high_vers)
903				high_vers = s->sc_vers;
904		}   /* found correct program */
905	}
906
907	/*
908	 * if we got here, the program or version
909	 * is not served ...
910	 */
911	if (prog_found)
912		svcerr_progvers(rqstp, low_vers, high_vers);
913	else
914		svcerr_noprog(rqstp);
915
916	svc_freereq(rqstp);
917}
918
919static void
920svc_checkidle(SVCPOOL *pool)
921{
922	SVCXPRT *xprt, *nxprt;
923	time_t timo;
924	struct svcxprt_list cleanup;
925
926	TAILQ_INIT(&cleanup);
927	TAILQ_FOREACH_SAFE(xprt, &pool->sp_xlist, xp_link, nxprt) {
928		/*
929		 * Only some transports have idle timers. Don't time
930		 * something out which is just waking up.
931		 */
932		if (!xprt->xp_idletimeout || xprt->xp_thread)
933			continue;
934
935		timo = xprt->xp_lastactive + xprt->xp_idletimeout;
936		if (time_uptime > timo) {
937			xprt_unregister_locked(xprt);
938			TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
939		}
940	}
941
942	mtx_unlock(&pool->sp_lock);
943	TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
944		SVC_RELEASE(xprt);
945	}
946	mtx_lock(&pool->sp_lock);
947
948}
949
950static void
951svc_assign_waiting_sockets(SVCPOOL *pool)
952{
953	SVCXPRT *xprt;
954
955	while ((xprt = TAILQ_FIRST(&pool->sp_active)) != NULL) {
956		if (xprt_assignthread(xprt))
957			TAILQ_REMOVE(&pool->sp_active, xprt, xp_alink);
958		else
959			break;
960	}
961}
962
963static bool_t
964svc_request_space_available(SVCPOOL *pool)
965{
966
967	mtx_assert(&pool->sp_lock, MA_OWNED);
968
969	if (pool->sp_space_throttled) {
970		/*
971		 * Below the low-water yet? If so, assign any waiting sockets.
972		 */
973		if (pool->sp_space_used < pool->sp_space_low) {
974			pool->sp_space_throttled = FALSE;
975			svc_assign_waiting_sockets(pool);
976			return TRUE;
977		}
978
979		return FALSE;
980	} else {
981		if (pool->sp_space_used
982		    >= pool->sp_space_high) {
983			pool->sp_space_throttled = TRUE;
984			pool->sp_space_throttle_count++;
985			return FALSE;
986		}
987
988		return TRUE;
989	}
990}
991
992static void
993svc_run_internal(SVCPOOL *pool, bool_t ismaster)
994{
995	SVCTHREAD *st, *stpref;
996	SVCXPRT *xprt;
997	enum xprt_stat stat;
998	struct svc_req *rqstp;
999	int error;
1000
1001	st = mem_alloc(sizeof(*st));
1002	st->st_xprt = NULL;
1003	STAILQ_INIT(&st->st_reqs);
1004	cv_init(&st->st_cond, "rpcsvc");
1005
1006	mtx_lock(&pool->sp_lock);
1007	LIST_INSERT_HEAD(&pool->sp_threads, st, st_link);
1008
1009	/*
1010	 * If we are a new thread which was spawned to cope with
1011	 * increased load, set the state back to SVCPOOL_ACTIVE.
1012	 */
1013	if (pool->sp_state == SVCPOOL_THREADSTARTING)
1014		pool->sp_state = SVCPOOL_ACTIVE;
1015
1016	while (pool->sp_state != SVCPOOL_CLOSING) {
1017		/*
1018		 * Create new thread if requested.
1019		 */
1020		if (pool->sp_state == SVCPOOL_THREADWANTED) {
1021			pool->sp_state = SVCPOOL_THREADSTARTING;
1022			pool->sp_lastcreatetime = time_uptime;
1023			mtx_unlock(&pool->sp_lock);
1024			svc_new_thread(pool);
1025			mtx_lock(&pool->sp_lock);
1026			continue;
1027		}
1028
1029		/*
1030		 * Check for idle transports once per second.
1031		 */
1032		if (time_uptime > pool->sp_lastidlecheck) {
1033			pool->sp_lastidlecheck = time_uptime;
1034			svc_checkidle(pool);
1035		}
1036
1037		xprt = st->st_xprt;
1038		if (!xprt && STAILQ_EMPTY(&st->st_reqs)) {
1039			/*
1040			 * Enforce maxthreads count.
1041			 */
1042			if (pool->sp_threadcount > pool->sp_maxthreads)
1043				break;
1044
1045			/*
1046			 * Before sleeping, see if we can find an
1047			 * active transport which isn't being serviced
1048			 * by a thread.
1049			 */
1050			if (svc_request_space_available(pool) &&
1051			    (xprt = TAILQ_FIRST(&pool->sp_active)) != NULL) {
1052				TAILQ_REMOVE(&pool->sp_active, xprt, xp_alink);
1053				SVC_ACQUIRE(xprt);
1054				xprt->xp_thread = st;
1055				st->st_xprt = xprt;
1056				continue;
1057			}
1058
1059			LIST_INSERT_HEAD(&pool->sp_idlethreads, st, st_ilink);
1060			st->st_idle = TRUE;
1061			if (ismaster || (!ismaster &&
1062			    pool->sp_threadcount > pool->sp_minthreads))
1063				error = cv_timedwait_sig(&st->st_cond,
1064				    &pool->sp_lock, 5 * hz);
1065			else
1066				error = cv_wait_sig(&st->st_cond,
1067				    &pool->sp_lock);
1068			if (st->st_idle) {
1069				LIST_REMOVE(st, st_ilink);
1070				st->st_idle = FALSE;
1071			}
1072
1073			/*
1074			 * Reduce worker thread count when idle.
1075			 */
1076			if (error == EWOULDBLOCK) {
1077				if (!ismaster
1078				    && (pool->sp_threadcount
1079					> pool->sp_minthreads)
1080					&& !st->st_xprt
1081					&& STAILQ_EMPTY(&st->st_reqs))
1082					break;
1083			} else if (error) {
1084				mtx_unlock(&pool->sp_lock);
1085				svc_exit(pool);
1086				mtx_lock(&pool->sp_lock);
1087				break;
1088			}
1089			continue;
1090		}
1091
1092		if (xprt) {
1093			/*
1094			 * Drain the transport socket and queue up any
1095			 * RPCs.
1096			 */
1097			xprt->xp_lastactive = time_uptime;
1098			stat = XPRT_IDLE;
1099			do {
1100				if (!svc_request_space_available(pool))
1101					break;
1102				rqstp = NULL;
1103				mtx_unlock(&pool->sp_lock);
1104				stat = svc_getreq(xprt, &rqstp);
1105				mtx_lock(&pool->sp_lock);
1106				if (rqstp) {
1107					/*
1108					 * See if the application has
1109					 * a preference for some other
1110					 * thread.
1111					 */
1112					stpref = st;
1113					if (pool->sp_assign)
1114						stpref = pool->sp_assign(st,
1115						    rqstp);
1116
1117					pool->sp_space_used +=
1118						rqstp->rq_size;
1119					if (pool->sp_space_used
1120					    > pool->sp_space_used_highest)
1121						pool->sp_space_used_highest =
1122							pool->sp_space_used;
1123					rqstp->rq_thread = stpref;
1124					STAILQ_INSERT_TAIL(&stpref->st_reqs,
1125					    rqstp, rq_link);
1126					stpref->st_reqcount++;
1127
1128					/*
1129					 * If we assigned the request
1130					 * to another thread, make
1131					 * sure its awake and continue
1132					 * reading from the
1133					 * socket. Otherwise, try to
1134					 * find some other thread to
1135					 * read from the socket and
1136					 * execute the request
1137					 * immediately.
1138					 */
1139					if (stpref == st)
1140						break;
1141					if (stpref->st_idle) {
1142						LIST_REMOVE(stpref, st_ilink);
1143						stpref->st_idle = FALSE;
1144						cv_signal(&stpref->st_cond);
1145					}
1146				}
1147			} while (stat == XPRT_MOREREQS
1148			    && pool->sp_state != SVCPOOL_CLOSING);
1149
1150			/*
1151			 * Move this transport to the end of the
1152			 * active list to ensure fairness when
1153			 * multiple transports are active. If this was
1154			 * the last queued request, svc_getreq will
1155			 * end up calling xprt_inactive to remove from
1156			 * the active list.
1157			 */
1158			xprt->xp_thread = NULL;
1159			st->st_xprt = NULL;
1160			if (xprt->xp_active) {
1161				if (!xprt_assignthread(xprt))
1162					TAILQ_INSERT_TAIL(&pool->sp_active,
1163					    xprt, xp_alink);
1164			}
1165			mtx_unlock(&pool->sp_lock);
1166			SVC_RELEASE(xprt);
1167			mtx_lock(&pool->sp_lock);
1168		}
1169
1170		/*
1171		 * Execute what we have queued.
1172		 */
1173		while ((rqstp = STAILQ_FIRST(&st->st_reqs)) != NULL) {
1174			size_t sz = rqstp->rq_size;
1175			mtx_unlock(&pool->sp_lock);
1176			svc_executereq(rqstp);
1177			mtx_lock(&pool->sp_lock);
1178			pool->sp_space_used -= sz;
1179		}
1180	}
1181
1182	if (st->st_xprt) {
1183		xprt = st->st_xprt;
1184		st->st_xprt = NULL;
1185		SVC_RELEASE(xprt);
1186	}
1187
1188	KASSERT(STAILQ_EMPTY(&st->st_reqs), ("stray reqs on exit"));
1189	LIST_REMOVE(st, st_link);
1190	pool->sp_threadcount--;
1191
1192	mtx_unlock(&pool->sp_lock);
1193
1194	cv_destroy(&st->st_cond);
1195	mem_free(st, sizeof(*st));
1196
1197	if (!ismaster)
1198		wakeup(pool);
1199}
1200
1201static void
1202svc_thread_start(void *arg)
1203{
1204
1205	svc_run_internal((SVCPOOL *) arg, FALSE);
1206	kthread_exit();
1207}
1208
1209static void
1210svc_new_thread(SVCPOOL *pool)
1211{
1212	struct thread *td;
1213
1214	pool->sp_threadcount++;
1215	kthread_add(svc_thread_start, pool,
1216	    pool->sp_proc, &td, 0, 0,
1217	    "%s: service", pool->sp_name);
1218}
1219
1220void
1221svc_run(SVCPOOL *pool)
1222{
1223	int i;
1224	struct proc *p;
1225	struct thread *td;
1226
1227	p = curproc;
1228	td = curthread;
1229	snprintf(td->td_name, sizeof(td->td_name),
1230	    "%s: master", pool->sp_name);
1231	pool->sp_state = SVCPOOL_ACTIVE;
1232	pool->sp_proc = p;
1233	pool->sp_lastcreatetime = time_uptime;
1234	pool->sp_threadcount = 1;
1235
1236	for (i = 1; i < pool->sp_minthreads; i++) {
1237		svc_new_thread(pool);
1238	}
1239
1240	svc_run_internal(pool, TRUE);
1241
1242	mtx_lock(&pool->sp_lock);
1243	while (pool->sp_threadcount > 0)
1244		msleep(pool, &pool->sp_lock, 0, "svcexit", 0);
1245	mtx_unlock(&pool->sp_lock);
1246}
1247
1248void
1249svc_exit(SVCPOOL *pool)
1250{
1251	SVCTHREAD *st;
1252
1253	mtx_lock(&pool->sp_lock);
1254
1255	if (pool->sp_state != SVCPOOL_CLOSING) {
1256		pool->sp_state = SVCPOOL_CLOSING;
1257		LIST_FOREACH(st, &pool->sp_idlethreads, st_ilink)
1258			cv_signal(&st->st_cond);
1259	}
1260
1261	mtx_unlock(&pool->sp_lock);
1262}
1263
1264bool_t
1265svc_getargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
1266{
1267	struct mbuf *m;
1268	XDR xdrs;
1269	bool_t stat;
1270
1271	m = rqstp->rq_args;
1272	rqstp->rq_args = NULL;
1273
1274	xdrmbuf_create(&xdrs, m, XDR_DECODE);
1275	stat = xargs(&xdrs, args);
1276	XDR_DESTROY(&xdrs);
1277
1278	return (stat);
1279}
1280
1281bool_t
1282svc_freeargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
1283{
1284	XDR xdrs;
1285
1286	if (rqstp->rq_addr) {
1287		free(rqstp->rq_addr, M_SONAME);
1288		rqstp->rq_addr = NULL;
1289	}
1290
1291	xdrs.x_op = XDR_FREE;
1292	return (xargs(&xdrs, args));
1293}
1294
1295void
1296svc_freereq(struct svc_req *rqstp)
1297{
1298	SVCTHREAD *st;
1299	SVCXPRT *xprt;
1300	SVCPOOL *pool;
1301
1302	st = rqstp->rq_thread;
1303	xprt = rqstp->rq_xprt;
1304	if (xprt)
1305		pool = xprt->xp_pool;
1306	else
1307		pool = NULL;
1308	if (st) {
1309		mtx_lock(&pool->sp_lock);
1310		KASSERT(rqstp == STAILQ_FIRST(&st->st_reqs),
1311		    ("Freeing request out of order"));
1312		STAILQ_REMOVE_HEAD(&st->st_reqs, rq_link);
1313		st->st_reqcount--;
1314		if (pool->sp_done)
1315			pool->sp_done(st, rqstp);
1316		mtx_unlock(&pool->sp_lock);
1317	}
1318
1319	if (rqstp->rq_auth.svc_ah_ops)
1320		SVCAUTH_RELEASE(&rqstp->rq_auth);
1321
1322	if (rqstp->rq_xprt) {
1323		SVC_RELEASE(rqstp->rq_xprt);
1324	}
1325
1326	if (rqstp->rq_addr)
1327		free(rqstp->rq_addr, M_SONAME);
1328
1329	if (rqstp->rq_args)
1330		m_freem(rqstp->rq_args);
1331
1332	free(rqstp, M_RPC);
1333}
1334