xref: /titanic_52/usr/src/uts/common/rpc/svc_clts.c (revision 3c112a2b34403220c06c3e2fcac403358cfba168)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  *  Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
26 /*	  All Rights Reserved  	*/
27 
28 /*
29  * Portions of this source code were derived from Berkeley 4.3 BSD
30  * under license from the Regents of the University of California.
31  */
32 
33 /*
34  * svc_clts.c
35  * Server side for RPC in the kernel.
36  *
37  */
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/sysmacros.h>
42 #include <sys/file.h>
43 #include <sys/stream.h>
44 #include <sys/strsubr.h>
45 #include <sys/tihdr.h>
46 #include <sys/tiuser.h>
47 #include <sys/t_kuser.h>
48 #include <sys/fcntl.h>
49 #include <sys/errno.h>
50 #include <sys/kmem.h>
51 #include <sys/systm.h>
52 #include <sys/cmn_err.h>
53 #include <sys/kstat.h>
54 #include <sys/vtrace.h>
55 #include <sys/debug.h>
56 
57 #include <rpc/types.h>
58 #include <rpc/xdr.h>
59 #include <rpc/auth.h>
60 #include <rpc/clnt.h>
61 #include <rpc/rpc_msg.h>
62 #include <rpc/svc.h>
63 #include <inet/ip.h>
64 
65 /*
66  * Routines exported through ops vector.
67  */
68 static bool_t		svc_clts_krecv(SVCXPRT *, mblk_t *, struct rpc_msg *);
69 static bool_t		svc_clts_ksend(SVCXPRT *, struct rpc_msg *);
70 static bool_t		svc_clts_kgetargs(SVCXPRT *, xdrproc_t, caddr_t);
71 static bool_t		svc_clts_kfreeargs(SVCXPRT *, xdrproc_t, caddr_t);
72 static void		svc_clts_kdestroy(SVCMASTERXPRT *);
73 static int		svc_clts_kdup(struct svc_req *, caddr_t, int,
74 				struct dupreq **, bool_t *);
75 static void		svc_clts_kdupdone(struct dupreq *, caddr_t,
76 				void (*)(), int, int);
77 static int32_t		*svc_clts_kgetres(SVCXPRT *, int);
78 static void		svc_clts_kclone_destroy(SVCXPRT *);
79 static void		svc_clts_kfreeres(SVCXPRT *);
80 static void		svc_clts_kstart(SVCMASTERXPRT *);
81 static void		svc_clts_kclone_xprt(SVCXPRT *, SVCXPRT *);
82 static void		svc_clts_ktattrs(SVCXPRT *, int, void **);
83 
84 /*
85  * Server transport operations vector.
86  */
87 struct svc_ops svc_clts_op = {
88 	svc_clts_krecv,		/* Get requests */
89 	svc_clts_kgetargs,	/* Deserialize arguments */
90 	svc_clts_ksend,		/* Send reply */
91 	svc_clts_kfreeargs,	/* Free argument data space */
92 	svc_clts_kdestroy,	/* Destroy transport handle */
93 	svc_clts_kdup,		/* Check entry in dup req cache */
94 	svc_clts_kdupdone,	/* Mark entry in dup req cache as done */
95 	svc_clts_kgetres,	/* Get pointer to response buffer */
96 	svc_clts_kfreeres,	/* Destroy pre-serialized response header */
97 	svc_clts_kclone_destroy, /* Destroy a clone xprt */
98 	svc_clts_kstart,	/* Tell `ready-to-receive' to rpcmod */
99 	svc_clts_kclone_xprt,	/* transport specific clone xprt function */
100 	svc_clts_ktattrs	/* Transport specific attributes. */
101 };
102 
103 /*
104  * Transport private data.
105  * Kept in xprt->xp_p2buf.
106  */
107 struct udp_data {
108 	mblk_t	*ud_resp;			/* buffer for response */
109 	mblk_t	*ud_inmp;			/* mblk chain of request */
110 };
111 
112 #define	UD_MAXSIZE	8800
113 #define	UD_INITSIZE	2048
114 
115 /*
116  * Connectionless server statistics
117  */
118 static const struct rpc_clts_server {
119 	kstat_named_t	rscalls;
120 	kstat_named_t	rsbadcalls;
121 	kstat_named_t	rsnullrecv;
122 	kstat_named_t	rsbadlen;
123 	kstat_named_t	rsxdrcall;
124 	kstat_named_t	rsdupchecks;
125 	kstat_named_t	rsdupreqs;
126 } clts_rsstat_tmpl = {
127 	{ "calls",	KSTAT_DATA_UINT64 },
128 	{ "badcalls",	KSTAT_DATA_UINT64 },
129 	{ "nullrecv",	KSTAT_DATA_UINT64 },
130 	{ "badlen",	KSTAT_DATA_UINT64 },
131 	{ "xdrcall",	KSTAT_DATA_UINT64 },
132 	{ "dupchecks",	KSTAT_DATA_UINT64 },
133 	{ "dupreqs",	KSTAT_DATA_UINT64 }
134 };
135 
136 static uint_t clts_rsstat_ndata =
137 	sizeof (clts_rsstat_tmpl) / sizeof (kstat_named_t);
138 
139 #define	CLONE2STATS(clone_xprt)	\
140 	(struct rpc_clts_server *)(clone_xprt)->xp_master->xp_p2
141 
142 #define	RSSTAT_INCR(stats, x)	\
143 	atomic_add_64(&(stats)->x.value.ui64, 1)
144 
145 /*
146  * Create a transport record.
147  * The transport record, output buffer, and private data structure
148  * are allocated.  The output buffer is serialized into using xdrmem.
149  * There is one transport record per user process which implements a
150  * set of services.
151  */
152 /* ARGSUSED */
153 int
154 svc_clts_kcreate(file_t *fp, uint_t sendsz, struct T_info_ack *tinfo,
155     SVCMASTERXPRT **nxprt)
156 {
157 	SVCMASTERXPRT *xprt;
158 	struct rpcstat *rpcstat;
159 
160 	if (nxprt == NULL)
161 		return (EINVAL);
162 
163 	rpcstat = zone_getspecific(rpcstat_zone_key, curproc->p_zone);
164 	ASSERT(rpcstat != NULL);
165 
166 	xprt = kmem_zalloc(sizeof (*xprt), KM_SLEEP);
167 	xprt->xp_lcladdr.buf = kmem_zalloc(sizeof (sin6_t), KM_SLEEP);
168 	xprt->xp_p2 = (caddr_t)rpcstat->rpc_clts_server;
169 	xprt->xp_ops = &svc_clts_op;
170 	xprt->xp_msg_size = tinfo->TSDU_size;
171 
172 	xprt->xp_rtaddr.buf = NULL;
173 	xprt->xp_rtaddr.maxlen = tinfo->ADDR_size;
174 	xprt->xp_rtaddr.len = 0;
175 
176 	*nxprt = xprt;
177 
178 	return (0);
179 }
180 
181 /*
182  * Destroy a transport record.
183  * Frees the space allocated for a transport record.
184  */
185 static void
186 svc_clts_kdestroy(SVCMASTERXPRT *xprt)
187 {
188 	if (xprt->xp_netid)
189 		kmem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
190 	if (xprt->xp_addrmask.maxlen)
191 		kmem_free(xprt->xp_addrmask.buf, xprt->xp_addrmask.maxlen);
192 
193 	mutex_destroy(&xprt->xp_req_lock);
194 	mutex_destroy(&xprt->xp_thread_lock);
195 
196 	kmem_free(xprt->xp_lcladdr.buf, sizeof (sin6_t));
197 	kmem_free(xprt, sizeof (SVCMASTERXPRT));
198 }
199 
200 /*
201  * Transport-type specific part of svc_xprt_cleanup().
202  * Frees the message buffer space allocated for a clone of a transport record
203  */
204 static void
205 svc_clts_kclone_destroy(SVCXPRT *clone_xprt)
206 {
207 	/* LINTED pointer alignment */
208 	struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
209 
210 	if (ud->ud_resp) {
211 		/*
212 		 * There should not be any left over results buffer.
213 		 */
214 		ASSERT(ud->ud_resp->b_cont == NULL);
215 
216 		/*
217 		 * Free the T_UNITDATA_{REQ/IND} that svc_clts_krecv
218 		 * saved.
219 		 */
220 		freeb(ud->ud_resp);
221 	}
222 	if (ud->ud_inmp)
223 		freemsg(ud->ud_inmp);
224 }
225 
226 /*
227  * svc_tli_kcreate() calls this function at the end to tell
228  * rpcmod that the transport is ready to receive requests.
229  */
230 /* ARGSUSED */
231 static void
232 svc_clts_kstart(SVCMASTERXPRT *xprt)
233 {
234 }
235 
236 static void
237 svc_clts_kclone_xprt(SVCXPRT *src_xprt, SVCXPRT *dst_xprt)
238 {
239 	struct udp_data *ud_src =
240 	    (struct udp_data *)src_xprt->xp_p2buf;
241 	struct udp_data *ud_dst =
242 	    (struct udp_data *)dst_xprt->xp_p2buf;
243 
244 	if (ud_src->ud_resp)
245 		ud_dst->ud_resp = dupb(ud_src->ud_resp);
246 
247 }
248 
249 static void
250 svc_clts_ktattrs(SVCXPRT *clone_xprt, int attrflag, void **tattr)
251 {
252 	*tattr = NULL;
253 
254 	switch (attrflag) {
255 	case SVC_TATTR_ADDRMASK:
256 		*tattr = (void *)&clone_xprt->xp_master->xp_addrmask;
257 	}
258 }
259 
260 /*
261  * Receive rpc requests.
262  * Pulls a request in off the socket, checks if the packet is intact,
263  * and deserializes the call packet.
264  */
265 static bool_t
266 svc_clts_krecv(SVCXPRT *clone_xprt, mblk_t *mp, struct rpc_msg *msg)
267 {
268 	/* LINTED pointer alignment */
269 	struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
270 	XDR *xdrs = &clone_xprt->xp_xdrin;
271 	struct rpc_clts_server *stats = CLONE2STATS(clone_xprt);
272 	union T_primitives *pptr;
273 	int hdrsz;
274 	cred_t *cr;
275 
276 	TRACE_0(TR_FAC_KRPC, TR_SVC_CLTS_KRECV_START,
277 	    "svc_clts_krecv_start:");
278 
279 	RSSTAT_INCR(stats, rscalls);
280 
281 	/*
282 	 * The incoming request should start with an M_PROTO message.
283 	 */
284 	if (mp->b_datap->db_type != M_PROTO) {
285 		goto bad;
286 	}
287 
288 	/*
289 	 * The incoming request should be an T_UNITDTA_IND.  There
290 	 * might be other messages coming up the stream, but we can
291 	 * ignore them.
292 	 */
293 	pptr = (union T_primitives *)mp->b_rptr;
294 	if (pptr->type != T_UNITDATA_IND) {
295 		goto bad;
296 	}
297 	/*
298 	 * Do some checking to make sure that the header at least looks okay.
299 	 */
300 	hdrsz = (int)(mp->b_wptr - mp->b_rptr);
301 	if (hdrsz < TUNITDATAINDSZ ||
302 	    hdrsz < (pptr->unitdata_ind.OPT_offset +
303 	    pptr->unitdata_ind.OPT_length) ||
304 	    hdrsz < (pptr->unitdata_ind.SRC_offset +
305 	    pptr->unitdata_ind.SRC_length)) {
306 		goto bad;
307 	}
308 
309 	/*
310 	 * Make sure that the transport provided a usable address.
311 	 */
312 	if (pptr->unitdata_ind.SRC_length <= 0) {
313 		goto bad;
314 	}
315 	/*
316 	 * Point the remote transport address in the service_transport
317 	 * handle at the address in the request.
318 	 */
319 	clone_xprt->xp_rtaddr.buf = (char *)mp->b_rptr +
320 	    pptr->unitdata_ind.SRC_offset;
321 	clone_xprt->xp_rtaddr.len = pptr->unitdata_ind.SRC_length;
322 
323 	/*
324 	 * Copy the local transport address in the service_transport
325 	 * handle at the address in the request. We will have only
326 	 * the local IP address in options.
327 	 */
328 	if (pptr->unitdata_ind.OPT_length && pptr->unitdata_ind.OPT_offset) {
329 		char *dstopt = (char *)mp->b_rptr +
330 		    pptr->unitdata_ind.OPT_offset;
331 		struct T_opthdr *toh = (struct T_opthdr *)dstopt;
332 
333 		if (toh->level == IPPROTO_IPV6 && toh->status == 0 &&
334 		    toh->name == IPV6_PKTINFO) {
335 			struct in6_pktinfo *pkti;
336 
337 			dstopt += sizeof (struct T_opthdr);
338 			pkti = (struct in6_pktinfo *)dstopt;
339 			((sin6_t *)(clone_xprt->xp_lcladdr.buf))->sin6_addr
340 			    = pkti->ipi6_addr;
341 		} else if (toh->level == IPPROTO_IP && toh->status == 0 &&
342 		    toh->name == IP_RECVDSTADDR) {
343 			dstopt += sizeof (struct T_opthdr);
344 			((sin_t *)(clone_xprt->xp_lcladdr.buf))->sin_addr
345 			    = *(struct in_addr *)dstopt;
346 		}
347 	}
348 
349 	/*
350 	 * Save the first mblk which contains the T_unidata_ind in
351 	 * ud_resp.  It will be used to generate the T_unitdata_req
352 	 * during the reply.
353 	 * We reuse any options in the T_unitdata_ind for the T_unitdata_req
354 	 * since we must pass any SCM_UCRED across in order for TX to
355 	 * work. We also make sure any cred_t is carried across.
356 	 */
357 	if (ud->ud_resp) {
358 		if (ud->ud_resp->b_cont != NULL) {
359 			cmn_err(CE_WARN, "svc_clts_krecv: ud_resp %p, "
360 			    "b_cont %p", (void *)ud->ud_resp,
361 			    (void *)ud->ud_resp->b_cont);
362 		}
363 		freeb(ud->ud_resp);
364 	}
365 	/* Move any cred_t to the first mblk in the message */
366 	cr = msg_getcred(mp, NULL);
367 	if (cr != NULL)
368 		mblk_setcred(mp, cr, NOPID);
369 
370 	ud->ud_resp = mp;
371 	mp = mp->b_cont;
372 	ud->ud_resp->b_cont = NULL;
373 
374 	xdrmblk_init(xdrs, mp, XDR_DECODE, 0);
375 
376 	TRACE_0(TR_FAC_KRPC, TR_XDR_CALLMSG_START,
377 	    "xdr_callmsg_start:");
378 	if (! xdr_callmsg(xdrs, msg)) {
379 		TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
380 		    "xdr_callmsg_end:(%S)", "bad");
381 		RSSTAT_INCR(stats, rsxdrcall);
382 		goto bad;
383 	}
384 	TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
385 	    "xdr_callmsg_end:(%S)", "good");
386 
387 	clone_xprt->xp_xid = msg->rm_xid;
388 	ud->ud_inmp = mp;
389 
390 	TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KRECV_END,
391 	    "svc_clts_krecv_end:(%S)", "good");
392 	return (TRUE);
393 
394 bad:
395 	if (mp)
396 		freemsg(mp);
397 	if (ud->ud_resp) {
398 		/*
399 		 * There should not be any left over results buffer.
400 		 */
401 		ASSERT(ud->ud_resp->b_cont == NULL);
402 		freeb(ud->ud_resp);
403 		ud->ud_resp = NULL;
404 	}
405 
406 	RSSTAT_INCR(stats, rsbadcalls);
407 	TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KRECV_END,
408 	    "svc_clts_krecv_end:(%S)", "bad");
409 	return (FALSE);
410 }
411 
412 /*
413  * Send rpc reply.
414  * Serialize the reply packet into the output buffer then
415  * call t_ksndudata to send it.
416  */
417 static bool_t
418 svc_clts_ksend(SVCXPRT *clone_xprt, struct rpc_msg *msg)
419 {
420 	/* LINTED pointer alignment */
421 	struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
422 	XDR *xdrs = &clone_xprt->xp_xdrout;
423 	int stat = FALSE;
424 	mblk_t *mp;
425 	int msgsz;
426 	struct T_unitdata_req *udreq;
427 	xdrproc_t xdr_results;
428 	caddr_t xdr_location;
429 	bool_t has_args;
430 
431 	TRACE_0(TR_FAC_KRPC, TR_SVC_CLTS_KSEND_START,
432 	    "svc_clts_ksend_start:");
433 
434 	ASSERT(ud->ud_resp != NULL);
435 
436 	/*
437 	 * If there is a result procedure specified in the reply message,
438 	 * it will be processed in the xdr_replymsg and SVCAUTH_WRAP.
439 	 * We need to make sure it won't be processed twice, so we null
440 	 * it for xdr_replymsg here.
441 	 */
442 	has_args = FALSE;
443 	if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
444 	    msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
445 		if ((xdr_results = msg->acpted_rply.ar_results.proc) != NULL) {
446 			has_args = TRUE;
447 			xdr_location = msg->acpted_rply.ar_results.where;
448 			msg->acpted_rply.ar_results.proc = xdr_void;
449 			msg->acpted_rply.ar_results.where = NULL;
450 		}
451 	}
452 
453 	if (ud->ud_resp->b_cont == NULL) {
454 		/*
455 		 * Allocate an initial mblk for the response data.
456 		 */
457 		while ((mp = allocb(UD_INITSIZE, BPRI_LO)) == NULL) {
458 			if (strwaitbuf(UD_INITSIZE, BPRI_LO)) {
459 				TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KSEND_END,
460 				    "svc_clts_ksend_end:(%S)", "strwaitbuf");
461 				return (FALSE);
462 			}
463 		}
464 
465 		/*
466 		 * Initialize the XDR decode stream.  Additional mblks
467 		 * will be allocated if necessary.  They will be UD_MAXSIZE
468 		 * sized.
469 		 */
470 		xdrmblk_init(xdrs, mp, XDR_ENCODE, UD_MAXSIZE);
471 
472 		/*
473 		 * Leave some space for protocol headers.
474 		 */
475 		(void) XDR_SETPOS(xdrs, 512);
476 		mp->b_rptr += 512;
477 
478 		msg->rm_xid = clone_xprt->xp_xid;
479 
480 		ud->ud_resp->b_cont = mp;
481 
482 		TRACE_0(TR_FAC_KRPC, TR_XDR_REPLYMSG_START,
483 		    "xdr_replymsg_start:");
484 		if (!(xdr_replymsg(xdrs, msg) &&
485 		    (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
486 		    xdr_results, xdr_location)))) {
487 			TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
488 			    "xdr_replymsg_end:(%S)", "bad");
489 			RPCLOG0(1, "xdr_replymsg/SVCAUTH_WRAP failed\n");
490 			goto out;
491 		}
492 		TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
493 		    "xdr_replymsg_end:(%S)", "good");
494 
495 	} else if (!(xdr_replymsg_body(xdrs, msg) &&
496 	    (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
497 	    xdr_results, xdr_location)))) {
498 		RPCLOG0(1, "xdr_replymsg_body/SVCAUTH_WRAP failed\n");
499 		goto out;
500 	}
501 
502 	msgsz = (int)xmsgsize(ud->ud_resp->b_cont);
503 
504 	if (msgsz <= 0 || (clone_xprt->xp_msg_size != -1 &&
505 	    msgsz > clone_xprt->xp_msg_size)) {
506 #ifdef	DEBUG
507 		cmn_err(CE_NOTE,
508 "KRPC: server response message of %d bytes; transport limits are [0, %d]",
509 		    msgsz, clone_xprt->xp_msg_size);
510 #endif
511 		goto out;
512 	}
513 
514 	/*
515 	 * Construct the T_unitdata_req.  We take advantage
516 	 * of the fact that T_unitdata_ind looks just like
517 	 * T_unitdata_req, except for the primitive type.
518 	 * Reusing it means we preserve any options, and we must preserve
519 	 * the SCM_UCRED option for TX to work.
520 	 * This has the side effect of also passing certain receive-side
521 	 * options like IP_RECVDSTADDR back down the send side. This
522 	 * implies that we can not ASSERT on a non-NULL db_credp when
523 	 * we have send-side options in UDP.
524 	 */
525 	udreq = (struct T_unitdata_req *)ud->ud_resp->b_rptr;
526 	udreq->PRIM_type = T_UNITDATA_REQ;
527 	put(clone_xprt->xp_wq, ud->ud_resp);
528 	stat = TRUE;
529 	ud->ud_resp = NULL;
530 
531 out:
532 	if (stat == FALSE) {
533 		freemsg(ud->ud_resp);
534 		ud->ud_resp = NULL;
535 	}
536 
537 	/*
538 	 * This is completely disgusting.  If public is set it is
539 	 * a pointer to a structure whose first field is the address
540 	 * of the function to free that structure and any related
541 	 * stuff.  (see rrokfree in nfs_xdr.c).
542 	 */
543 	if (xdrs->x_public) {
544 		/* LINTED pointer alignment */
545 		(**((int (**)())xdrs->x_public))(xdrs->x_public);
546 	}
547 
548 	TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KSEND_END,
549 	    "svc_clts_ksend_end:(%S)", "done");
550 	return (stat);
551 }
552 
553 /*
554  * Deserialize arguments.
555  */
556 static bool_t
557 svc_clts_kgetargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
558     caddr_t args_ptr)
559 {
560 
561 	/* LINTED pointer alignment */
562 	return (SVCAUTH_UNWRAP(&clone_xprt->xp_auth, &clone_xprt->xp_xdrin,
563 	    xdr_args, args_ptr));
564 
565 }
566 
567 static bool_t
568 svc_clts_kfreeargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
569     caddr_t args_ptr)
570 {
571 	/* LINTED pointer alignment */
572 	struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
573 	XDR *xdrs = &clone_xprt->xp_xdrin;
574 	bool_t retval;
575 
576 	if (args_ptr) {
577 		xdrs->x_op = XDR_FREE;
578 		retval = (*xdr_args)(xdrs, args_ptr);
579 	} else
580 		retval = TRUE;
581 
582 	if (ud->ud_inmp) {
583 		freemsg(ud->ud_inmp);
584 		ud->ud_inmp = NULL;
585 	}
586 
587 	return (retval);
588 }
589 
590 static int32_t *
591 svc_clts_kgetres(SVCXPRT *clone_xprt, int size)
592 {
593 	/* LINTED pointer alignment */
594 	struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
595 	XDR *xdrs = &clone_xprt->xp_xdrout;
596 	mblk_t *mp;
597 	int32_t *buf;
598 	struct rpc_msg rply;
599 
600 	/*
601 	 * Allocate an initial mblk for the response data.
602 	 */
603 	while ((mp = allocb(UD_INITSIZE, BPRI_LO)) == NULL) {
604 		if (strwaitbuf(UD_INITSIZE, BPRI_LO)) {
605 			return (FALSE);
606 		}
607 	}
608 
609 	mp->b_cont = NULL;
610 
611 	/*
612 	 * Initialize the XDR decode stream.  Additional mblks
613 	 * will be allocated if necessary.  They will be UD_MAXSIZE
614 	 * sized.
615 	 */
616 	xdrmblk_init(xdrs, mp, XDR_ENCODE, UD_MAXSIZE);
617 
618 	/*
619 	 * Leave some space for protocol headers.
620 	 */
621 	(void) XDR_SETPOS(xdrs, 512);
622 	mp->b_rptr += 512;
623 
624 	/*
625 	 * Assume a successful RPC since most of them are.
626 	 */
627 	rply.rm_xid = clone_xprt->xp_xid;
628 	rply.rm_direction = REPLY;
629 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
630 	rply.acpted_rply.ar_verf = clone_xprt->xp_verf;
631 	rply.acpted_rply.ar_stat = SUCCESS;
632 
633 	if (!xdr_replymsg_hdr(xdrs, &rply)) {
634 		freeb(mp);
635 		return (NULL);
636 	}
637 
638 	buf = XDR_INLINE(xdrs, size);
639 
640 	if (buf == NULL)
641 		freeb(mp);
642 	else
643 		ud->ud_resp->b_cont = mp;
644 
645 	return (buf);
646 }
647 
648 static void
649 svc_clts_kfreeres(SVCXPRT *clone_xprt)
650 {
651 	/* LINTED pointer alignment */
652 	struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
653 
654 	if (ud->ud_resp == NULL || ud->ud_resp->b_cont == NULL)
655 		return;
656 
657 	/*
658 	 * SVC_FREERES() is called whenever the server decides not to
659 	 * send normal reply. Thus, we expect only one mblk to be allocated,
660 	 * because we have not attempted any XDR encoding.
661 	 * If we do any XDR encoding and we get an error, then SVC_REPLY()
662 	 * will freemsg(ud->ud_resp);
663 	 */
664 	ASSERT(ud->ud_resp->b_cont->b_cont == NULL);
665 	freeb(ud->ud_resp->b_cont);
666 	ud->ud_resp->b_cont = NULL;
667 }
668 
669 /*
670  * the dup cacheing routines below provide a cache of non-failure
671  * transaction id's.  rpc service routines can use this to detect
672  * retransmissions and re-send a non-failure response.
673  */
674 
675 /*
676  * MAXDUPREQS is the number of cached items.  It should be adjusted
677  * to the service load so that there is likely to be a response entry
678  * when the first retransmission comes in.
679  */
680 #define	MAXDUPREQS	1024
681 
682 /*
683  * This should be appropriately scaled to MAXDUPREQS.
684  */
685 #define	DRHASHSZ	257
686 
687 #if ((DRHASHSZ & (DRHASHSZ - 1)) == 0)
688 #define	XIDHASH(xid)	((xid) & (DRHASHSZ - 1))
689 #else
690 #define	XIDHASH(xid)	((xid) % DRHASHSZ)
691 #endif
692 #define	DRHASH(dr)	XIDHASH((dr)->dr_xid)
693 #define	REQTOXID(req)	((req)->rq_xprt->xp_xid)
694 
695 static int	ndupreqs = 0;
696 int	maxdupreqs = MAXDUPREQS;
697 static kmutex_t dupreq_lock;
698 static struct dupreq *drhashtbl[DRHASHSZ];
699 static int	drhashstat[DRHASHSZ];
700 
701 static void unhash(struct dupreq *);
702 
703 /*
704  * drmru points to the head of a circular linked list in lru order.
705  * drmru->dr_next == drlru
706  */
707 struct dupreq *drmru;
708 
709 /*
710  * PSARC 2003/523 Contract Private Interface
711  * svc_clts_kdup
712  * Changes must be reviewed by Solaris File Sharing
713  * Changes must be communicated to contract-2003-523@sun.com
714  *
715  * svc_clts_kdup searches the request cache and returns 0 if the
716  * request is not found in the cache.  If it is found, then it
717  * returns the state of the request (in progress or done) and
718  * the status or attributes that were part of the original reply.
719  *
720  * If DUP_DONE (there is a duplicate) svc_clts_kdup copies over the
721  * value of the response. In that case, also return in *dupcachedp
722  * whether the response free routine is cached in the dupreq - in which case
723  * the caller should not be freeing it, because it will be done later
724  * in the svc_clts_kdup code when the dupreq is reused.
725  */
726 static int
727 svc_clts_kdup(struct svc_req *req, caddr_t res, int size, struct dupreq **drpp,
728 	bool_t *dupcachedp)
729 {
730 	struct rpc_clts_server *stats = CLONE2STATS(req->rq_xprt);
731 	struct dupreq *dr;
732 	uint32_t xid;
733 	uint32_t drhash;
734 	int status;
735 
736 	xid = REQTOXID(req);
737 	mutex_enter(&dupreq_lock);
738 	RSSTAT_INCR(stats, rsdupchecks);
739 	/*
740 	 * Check to see whether an entry already exists in the cache.
741 	 */
742 	dr = drhashtbl[XIDHASH(xid)];
743 	while (dr != NULL) {
744 		if (dr->dr_xid == xid &&
745 		    dr->dr_proc == req->rq_proc &&
746 		    dr->dr_prog == req->rq_prog &&
747 		    dr->dr_vers == req->rq_vers &&
748 		    dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
749 		    bcmp(dr->dr_addr.buf, req->rq_xprt->xp_rtaddr.buf,
750 		    dr->dr_addr.len) == 0) {
751 			status = dr->dr_status;
752 			if (status == DUP_DONE) {
753 				bcopy(dr->dr_resp.buf, res, size);
754 				if (dupcachedp != NULL)
755 					*dupcachedp = (dr->dr_resfree != NULL);
756 			} else {
757 				dr->dr_status = DUP_INPROGRESS;
758 				*drpp = dr;
759 			}
760 			RSSTAT_INCR(stats, rsdupreqs);
761 			mutex_exit(&dupreq_lock);
762 			return (status);
763 		}
764 		dr = dr->dr_chain;
765 	}
766 
767 	/*
768 	 * There wasn't an entry, either allocate a new one or recycle
769 	 * an old one.
770 	 */
771 	if (ndupreqs < maxdupreqs) {
772 		dr = kmem_alloc(sizeof (*dr), KM_NOSLEEP);
773 		if (dr == NULL) {
774 			mutex_exit(&dupreq_lock);
775 			return (DUP_ERROR);
776 		}
777 		dr->dr_resp.buf = NULL;
778 		dr->dr_resp.maxlen = 0;
779 		dr->dr_addr.buf = NULL;
780 		dr->dr_addr.maxlen = 0;
781 		if (drmru) {
782 			dr->dr_next = drmru->dr_next;
783 			drmru->dr_next = dr;
784 		} else {
785 			dr->dr_next = dr;
786 		}
787 		ndupreqs++;
788 	} else {
789 		dr = drmru->dr_next;
790 		while (dr->dr_status == DUP_INPROGRESS) {
791 			dr = dr->dr_next;
792 			if (dr == drmru->dr_next) {
793 				cmn_err(CE_WARN, "svc_clts_kdup no slots free");
794 				mutex_exit(&dupreq_lock);
795 				return (DUP_ERROR);
796 			}
797 		}
798 		unhash(dr);
799 		if (dr->dr_resfree) {
800 			(*dr->dr_resfree)(dr->dr_resp.buf);
801 		}
802 	}
803 	dr->dr_resfree = NULL;
804 	drmru = dr;
805 
806 	dr->dr_xid = REQTOXID(req);
807 	dr->dr_prog = req->rq_prog;
808 	dr->dr_vers = req->rq_vers;
809 	dr->dr_proc = req->rq_proc;
810 	if (dr->dr_addr.maxlen < req->rq_xprt->xp_rtaddr.len) {
811 		if (dr->dr_addr.buf != NULL)
812 			kmem_free(dr->dr_addr.buf, dr->dr_addr.maxlen);
813 		dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
814 		dr->dr_addr.buf = kmem_alloc(dr->dr_addr.maxlen,
815 		    KM_NOSLEEP);
816 		if (dr->dr_addr.buf == NULL) {
817 			dr->dr_addr.maxlen = 0;
818 			dr->dr_status = DUP_DROP;
819 			mutex_exit(&dupreq_lock);
820 			return (DUP_ERROR);
821 		}
822 	}
823 	dr->dr_addr.len = req->rq_xprt->xp_rtaddr.len;
824 	bcopy(req->rq_xprt->xp_rtaddr.buf, dr->dr_addr.buf, dr->dr_addr.len);
825 	if (dr->dr_resp.maxlen < size) {
826 		if (dr->dr_resp.buf != NULL)
827 			kmem_free(dr->dr_resp.buf, dr->dr_resp.maxlen);
828 		dr->dr_resp.maxlen = (unsigned int)size;
829 		dr->dr_resp.buf = kmem_alloc(size, KM_NOSLEEP);
830 		if (dr->dr_resp.buf == NULL) {
831 			dr->dr_resp.maxlen = 0;
832 			dr->dr_status = DUP_DROP;
833 			mutex_exit(&dupreq_lock);
834 			return (DUP_ERROR);
835 		}
836 	}
837 	dr->dr_status = DUP_INPROGRESS;
838 
839 	drhash = (uint32_t)DRHASH(dr);
840 	dr->dr_chain = drhashtbl[drhash];
841 	drhashtbl[drhash] = dr;
842 	drhashstat[drhash]++;
843 	mutex_exit(&dupreq_lock);
844 	*drpp = dr;
845 	return (DUP_NEW);
846 }
847 
848 /*
849  * PSARC 2003/523 Contract Private Interface
850  * svc_clts_kdupdone
851  * Changes must be reviewed by Solaris File Sharing
852  * Changes must be communicated to contract-2003-523@sun.com
853  *
854  * svc_clts_kdupdone marks the request done (DUP_DONE or DUP_DROP)
855  * and stores the response.
856  */
857 static void
858 svc_clts_kdupdone(struct dupreq *dr, caddr_t res, void (*dis_resfree)(),
859 	int size, int status)
860 {
861 
862 	ASSERT(dr->dr_resfree == NULL);
863 	if (status == DUP_DONE) {
864 		bcopy(res, dr->dr_resp.buf, size);
865 		dr->dr_resfree = dis_resfree;
866 	}
867 	dr->dr_status = status;
868 }
869 
870 /*
871  * This routine expects that the mutex, dupreq_lock, is already held.
872  */
873 static void
874 unhash(struct dupreq *dr)
875 {
876 	struct dupreq *drt;
877 	struct dupreq *drtprev = NULL;
878 	uint32_t drhash;
879 
880 	ASSERT(MUTEX_HELD(&dupreq_lock));
881 
882 	drhash = (uint32_t)DRHASH(dr);
883 	drt = drhashtbl[drhash];
884 	while (drt != NULL) {
885 		if (drt == dr) {
886 			drhashstat[drhash]--;
887 			if (drtprev == NULL) {
888 				drhashtbl[drhash] = drt->dr_chain;
889 			} else {
890 				drtprev->dr_chain = drt->dr_chain;
891 			}
892 			return;
893 		}
894 		drtprev = drt;
895 		drt = drt->dr_chain;
896 	}
897 }
898 
899 void
900 svc_clts_stats_init(zoneid_t zoneid, struct rpc_clts_server **statsp)
901 {
902 	kstat_t *ksp;
903 	kstat_named_t *knp;
904 
905 	knp = rpcstat_zone_init_common(zoneid, "unix", "rpc_clts_server",
906 	    (const kstat_named_t *)&clts_rsstat_tmpl,
907 	    sizeof (clts_rsstat_tmpl));
908 	/*
909 	 * Backwards compatibility for old kstat clients
910 	 */
911 	ksp = kstat_create_zone("unix", 0, "rpc_server", "rpc",
912 	    KSTAT_TYPE_NAMED, clts_rsstat_ndata,
913 	    KSTAT_FLAG_VIRTUAL | KSTAT_FLAG_WRITABLE, zoneid);
914 	if (ksp) {
915 		ksp->ks_data = knp;
916 		kstat_install(ksp);
917 	}
918 	*statsp = (struct rpc_clts_server *)knp;
919 }
920 
921 void
922 svc_clts_stats_fini(zoneid_t zoneid, struct rpc_clts_server **statsp)
923 {
924 	rpcstat_zone_fini_common(zoneid, "unix", "rpc_clts_server");
925 	kstat_delete_byname_zone("unix", 0, "rpc_server", zoneid);
926 	kmem_free(*statsp, sizeof (clts_rsstat_tmpl));
927 }
928 
929 void
930 svc_clts_init()
931 {
932 	/*
933 	 * Check to make sure that the clts private data will fit into
934 	 * the stack buffer allocated by svc_run.  The compiler should
935 	 * remove this check, but it's a safety net if the udp_data
936 	 * structure ever changes.
937 	 */
938 	/*CONSTANTCONDITION*/
939 	ASSERT(sizeof (struct udp_data) <= SVC_P2LEN);
940 
941 	mutex_init(&dupreq_lock, NULL, MUTEX_DEFAULT, NULL);
942 }
943