xref: /illumos-gate/usr/src/uts/common/rpc/rpcmod.c (revision 03494a9880d80f834bec10a1e8f0a2f8f7c97bf4)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /* Copyright (c) 1990 Mentat Inc. */
26 
27 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 /*
31  * Kernel RPC filtering module
32  */
33 
34 #include <sys/param.h>
35 #include <sys/types.h>
36 #include <sys/stream.h>
37 #include <sys/stropts.h>
38 #include <sys/tihdr.h>
39 #include <sys/timod.h>
40 #include <sys/tiuser.h>
41 #include <sys/debug.h>
42 #include <sys/signal.h>
43 #include <sys/pcb.h>
44 #include <sys/user.h>
45 #include <sys/errno.h>
46 #include <sys/cred.h>
47 #include <sys/policy.h>
48 #include <sys/inline.h>
49 #include <sys/cmn_err.h>
50 #include <sys/kmem.h>
51 #include <sys/file.h>
52 #include <sys/sysmacros.h>
53 #include <sys/systm.h>
54 #include <sys/t_lock.h>
55 #include <sys/ddi.h>
56 #include <sys/vtrace.h>
57 #include <sys/callb.h>
58 #include <sys/strsun.h>
59 
60 #include <sys/strlog.h>
61 #include <rpc/rpc_com.h>
62 #include <inet/common.h>
63 #include <rpc/types.h>
64 #include <sys/time.h>
65 #include <rpc/xdr.h>
66 #include <rpc/auth.h>
67 #include <rpc/clnt.h>
68 #include <rpc/rpc_msg.h>
69 #include <rpc/clnt.h>
70 #include <rpc/svc.h>
71 #include <rpc/rpcsys.h>
72 #include <rpc/rpc_rdma.h>
73 
74 /*
75  * This is the loadable module wrapper.
76  */
77 #include <sys/conf.h>
78 #include <sys/modctl.h>
79 #include <sys/syscall.h>
80 
81 extern struct streamtab rpcinfo;
82 
83 static struct fmodsw fsw = {
84 	"rpcmod",
85 	&rpcinfo,
86 	D_NEW|D_MP,
87 };
88 
89 /*
90  * Module linkage information for the kernel.
91  */
92 
93 static struct modlstrmod modlstrmod = {
94 	&mod_strmodops, "rpc interface str mod", &fsw
95 };
96 
97 /*
98  * For the RPC system call.
99  */
100 static struct sysent rpcsysent = {
101 	2,
102 	SE_32RVAL1 | SE_ARGC | SE_NOUNLOAD,
103 	rpcsys
104 };
105 
106 static struct modlsys modlsys = {
107 	&mod_syscallops,
108 	"RPC syscall",
109 	&rpcsysent
110 };
111 
112 #ifdef _SYSCALL32_IMPL
113 static struct modlsys modlsys32 = {
114 	&mod_syscallops32,
115 	"32-bit RPC syscall",
116 	&rpcsysent
117 };
118 #endif /* _SYSCALL32_IMPL */
119 
120 static struct modlinkage modlinkage = {
121 	MODREV_1,
122 	{
123 		&modlsys,
124 #ifdef _SYSCALL32_IMPL
125 		&modlsys32,
126 #endif
127 		&modlstrmod,
128 		NULL
129 	}
130 };
131 
132 int
133 _init(void)
134 {
135 	int error = 0;
136 	callb_id_t cid;
137 	int status;
138 
139 	svc_init();
140 	clnt_init();
141 	cid = callb_add(connmgr_cpr_reset, 0, CB_CL_CPR_RPC, "rpc");
142 
143 	if (error = mod_install(&modlinkage)) {
144 		/*
145 		 * Could not install module, cleanup previous
146 		 * initialization work.
147 		 */
148 		clnt_fini();
149 		if (cid != NULL)
150 			(void) callb_delete(cid);
151 
152 		return (error);
153 	}
154 
155 	/*
156 	 * Load up the RDMA plugins and initialize the stats. Even if the
157 	 * plugins loadup fails, but rpcmod was successfully installed the
158 	 * counters still get initialized.
159 	 */
160 	rw_init(&rdma_lock, NULL, RW_DEFAULT, NULL);
161 	mutex_init(&rdma_modload_lock, NULL, MUTEX_DEFAULT, NULL);
162 	mt_kstat_init();
163 
164 	/*
165 	 * Get our identification into ldi.  This is used for loading
166 	 * other modules, e.g. rpcib.
167 	 */
168 	status = ldi_ident_from_mod(&modlinkage, &rpcmod_li);
169 	if (status != 0) {
170 		cmn_err(CE_WARN, "ldi_ident_from_mod fails with %d", status);
171 		rpcmod_li = NULL;
172 	}
173 
174 	return (error);
175 }
176 
177 /*
178  * The unload entry point fails, because we advertise entry points into
179  * rpcmod from the rest of kRPC: rpcmod_release().
180  */
181 int
182 _fini(void)
183 {
184 	return (EBUSY);
185 }
186 
187 int
188 _info(struct modinfo *modinfop)
189 {
190 	return (mod_info(&modlinkage, modinfop));
191 }
192 
193 extern int nulldev();
194 
195 #define	RPCMOD_ID	2049
196 
197 int rmm_open(), rmm_close();
198 
199 /*
200  * To save instructions, since STREAMS ignores the return value
201  * from these functions, they are defined as void here. Kind of icky, but...
202  */
203 void rmm_rput(queue_t *, mblk_t *);
204 void rmm_wput(queue_t *, mblk_t *);
205 void rmm_rsrv(queue_t *);
206 void rmm_wsrv(queue_t *);
207 
208 int rpcmodopen(), rpcmodclose();
209 void rpcmodrput(), rpcmodwput();
210 void rpcmodrsrv(), rpcmodwsrv();
211 
212 static	void	rpcmodwput_other(queue_t *, mblk_t *);
213 static	int	mir_close(queue_t *q);
214 static	int	mir_open(queue_t *q, dev_t *devp, int flag, int sflag,
215 		    cred_t *credp);
216 static	void	mir_rput(queue_t *q, mblk_t *mp);
217 static	void	mir_rsrv(queue_t *q);
218 static	void	mir_wput(queue_t *q, mblk_t *mp);
219 static	void	mir_wsrv(queue_t *q);
220 
221 static struct module_info rpcmod_info =
222 	{RPCMOD_ID, "rpcmod", 0, INFPSZ, 256*1024, 1024};
223 
224 /*
225  * Read side has no service procedure.
226  */
227 static struct qinit rpcmodrinit = {
228 	(int (*)())rmm_rput,
229 	(int (*)())rmm_rsrv,
230 	rmm_open,
231 	rmm_close,
232 	nulldev,
233 	&rpcmod_info,
234 	NULL
235 };
236 
237 /*
238  * The write put procedure is simply putnext to conserve stack space.
239  * The write service procedure is not used to queue data, but instead to
240  * synchronize with flow control.
241  */
242 static struct qinit rpcmodwinit = {
243 	(int (*)())rmm_wput,
244 	(int (*)())rmm_wsrv,
245 	rmm_open,
246 	rmm_close,
247 	nulldev,
248 	&rpcmod_info,
249 	NULL
250 };
251 struct streamtab rpcinfo = { &rpcmodrinit, &rpcmodwinit, NULL, NULL };
252 
253 struct xprt_style_ops {
254 	int (*xo_open)();
255 	int (*xo_close)();
256 	void (*xo_wput)();
257 	void (*xo_wsrv)();
258 	void (*xo_rput)();
259 	void (*xo_rsrv)();
260 };
261 
262 static struct xprt_style_ops xprt_clts_ops = {
263 	rpcmodopen,
264 	rpcmodclose,
265 	rpcmodwput,
266 	rpcmodwsrv,
267 	rpcmodrput,
268 	NULL
269 };
270 
271 static struct xprt_style_ops xprt_cots_ops = {
272 	mir_open,
273 	mir_close,
274 	mir_wput,
275 	mir_wsrv,
276 	mir_rput,
277 	mir_rsrv
278 };
279 
280 /*
281  * Per rpcmod "slot" data structure. q->q_ptr points to one of these.
282  */
283 struct rpcm {
284 	void		*rm_krpc_cell;	/* Reserved for use by KRPC */
285 	struct		xprt_style_ops	*rm_ops;
286 	int		rm_type;	/* Client or server side stream */
287 #define	RM_CLOSING	0x1		/* somebody is trying to close slot */
288 	uint_t		rm_state;	/* state of the slot. see above */
289 	uint_t		rm_ref;		/* cnt of external references to slot */
290 	kmutex_t	rm_lock;	/* mutex protecting above fields */
291 	kcondvar_t	rm_cwait;	/* condition for closing */
292 	zoneid_t	rm_zoneid;	/* zone which pushed rpcmod */
293 };
294 
295 struct temp_slot {
296 	void *cell;
297 	struct xprt_style_ops *ops;
298 	int type;
299 	mblk_t *info_ack;
300 	kmutex_t lock;
301 	kcondvar_t wait;
302 };
303 
304 typedef struct mir_s {
305 	void	*mir_krpc_cell;	/* Reserved for KRPC use. This field */
306 					/* must be first in the structure. */
307 	struct xprt_style_ops	*rm_ops;
308 	int	mir_type;		/* Client or server side stream */
309 
310 	mblk_t	*mir_head_mp;		/* RPC msg in progress */
311 		/*
312 		 * mir_head_mp points the first mblk being collected in
313 		 * the current RPC message.  Record headers are removed
314 		 * before data is linked into mir_head_mp.
315 		 */
316 	mblk_t	*mir_tail_mp;		/* Last mblk in mir_head_mp */
317 		/*
318 		 * mir_tail_mp points to the last mblk in the message
319 		 * chain starting at mir_head_mp.  It is only valid
320 		 * if mir_head_mp is non-NULL and is used to add new
321 		 * data blocks to the end of chain quickly.
322 		 */
323 
324 	int32_t	mir_frag_len;		/* Bytes seen in the current frag */
325 		/*
326 		 * mir_frag_len starts at -4 for beginning of each fragment.
327 		 * When this length is negative, it indicates the number of
328 		 * bytes that rpcmod needs to complete the record marker
329 		 * header.  When it is positive or zero, it holds the number
330 		 * of bytes that have arrived for the current fragment and
331 		 * are held in mir_header_mp.
332 		 */
333 
334 	int32_t	mir_frag_header;
335 		/*
336 		 * Fragment header as collected for the current fragment.
337 		 * It holds the last-fragment indicator and the number
338 		 * of bytes in the fragment.
339 		 */
340 
341 	unsigned int
342 		mir_ordrel_pending : 1,	/* Sent T_ORDREL_REQ */
343 		mir_hold_inbound : 1,	/* Hold inbound messages on server */
344 					/* side until outbound flow control */
345 					/* is relieved. */
346 		mir_closing : 1,	/* The stream is being closed */
347 		mir_inrservice : 1,	/* data queued or rd srv proc running */
348 		mir_inwservice : 1,	/* data queued or wr srv proc running */
349 		mir_inwflushdata : 1,	/* flush M_DATAs when srv runs */
350 		/*
351 		 * On client streams, mir_clntreq is 0 or 1; it is set
352 		 * to 1 whenever a new request is sent out (mir_wput)
353 		 * and cleared when the timer fires (mir_timer).  If
354 		 * the timer fires with this value equal to 0, then the
355 		 * stream is considered idle and KRPC is notified.
356 		 */
357 		mir_clntreq : 1,
358 		/*
359 		 * On server streams, stop accepting messages
360 		 */
361 		mir_svc_no_more_msgs : 1,
362 		mir_listen_stream : 1,	/* listen end point */
363 		mir_unused : 1,	/* no longer used */
364 		mir_timer_call : 1,
365 		mir_junk_fill_thru_bit_31 : 21;
366 
367 	int	mir_setup_complete;	/* server has initialized everything */
368 	timeout_id_t mir_timer_id;	/* Timer for idle checks */
369 	clock_t	mir_idle_timeout;	/* Allowed idle time before shutdown */
370 		/*
371 		 * This value is copied from clnt_idle_timeout or
372 		 * svc_idle_timeout during the appropriate ioctl.
373 		 * Kept in milliseconds
374 		 */
375 	clock_t	mir_use_timestamp;	/* updated on client with each use */
376 		/*
377 		 * This value is set to lbolt
378 		 * every time a client stream sends or receives data.
379 		 * Even if the timer message arrives, we don't shutdown
380 		 * client unless:
381 		 *    lbolt >= MSEC_TO_TICK(mir_idle_timeout)+mir_use_timestamp.
382 		 * This value is kept in HZ.
383 		 */
384 
385 	uint_t	*mir_max_msg_sizep;	/* Reference to sanity check size */
386 		/*
387 		 * This pointer is set to &clnt_max_msg_size or
388 		 * &svc_max_msg_size during the appropriate ioctl.
389 		 */
390 	zoneid_t mir_zoneid;	/* zone which pushed rpcmod */
391 	/* Server-side fields. */
392 	int	mir_ref_cnt;		/* Reference count: server side only */
393 					/* counts the number of references */
394 					/* that a kernel RPC server thread */
395 					/* (see svc_run()) has on this rpcmod */
396 					/* slot. Effectively, it is the */
397 					/* number * of unprocessed messages */
398 					/* that have been passed up to the */
399 					/* KRPC layer */
400 
401 	mblk_t	*mir_svc_pend_mp;	/* Pending T_ORDREL_IND or */
402 					/* T_DISCON_IND */
403 
404 	/*
405 	 * these fields are for both client and server, but for debugging,
406 	 * it is easier to have these last in the structure.
407 	 */
408 	kmutex_t	mir_mutex;	/* Mutex and condvar for close */
409 	kcondvar_t	mir_condvar;	/* synchronization. */
410 	kcondvar_t	mir_timer_cv;	/* Timer routine sync. */
411 } mir_t;
412 
413 void tmp_rput(queue_t *q, mblk_t *mp);
414 
415 struct xprt_style_ops tmpops = {
416 	NULL,
417 	NULL,
418 	putnext,
419 	NULL,
420 	tmp_rput,
421 	NULL
422 };
423 
424 void
425 tmp_rput(queue_t *q, mblk_t *mp)
426 {
427 	struct temp_slot *t = (struct temp_slot *)(q->q_ptr);
428 	struct T_info_ack *pptr;
429 
430 	switch (mp->b_datap->db_type) {
431 	case M_PCPROTO:
432 		pptr = (struct T_info_ack *)mp->b_rptr;
433 		switch (pptr->PRIM_type) {
434 		case T_INFO_ACK:
435 			mutex_enter(&t->lock);
436 			t->info_ack = mp;
437 			cv_signal(&t->wait);
438 			mutex_exit(&t->lock);
439 			return;
440 		default:
441 			break;
442 		}
443 	default:
444 		break;
445 	}
446 
447 	/*
448 	 * Not an info-ack, so free it. This is ok because we should
449 	 * not be receiving data until the open finishes: rpcmod
450 	 * is pushed well before the end-point is bound to an address.
451 	 */
452 	freemsg(mp);
453 }
454 
455 int
456 rmm_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *crp)
457 {
458 	mblk_t *bp;
459 	struct temp_slot ts, *t;
460 	struct T_info_ack *pptr;
461 	int error = 0;
462 
463 	ASSERT(q != NULL);
464 	/*
465 	 * Check for re-opens.
466 	 */
467 	if (q->q_ptr) {
468 		TRACE_1(TR_FAC_KRPC, TR_RPCMODOPEN_END,
469 		    "rpcmodopen_end:(%s)", "q->qptr");
470 		return (0);
471 	}
472 
473 	t = &ts;
474 	bzero(t, sizeof (*t));
475 	q->q_ptr = (void *)t;
476 	WR(q)->q_ptr = (void *)t;
477 
478 	/*
479 	 * Allocate the required messages upfront.
480 	 */
481 	if ((bp = allocb(sizeof (struct T_info_req) +
482 	    sizeof (struct T_info_ack), BPRI_LO)) == (mblk_t *)NULL) {
483 		return (ENOBUFS);
484 	}
485 
486 	mutex_init(&t->lock, NULL, MUTEX_DEFAULT, NULL);
487 	cv_init(&t->wait, NULL, CV_DEFAULT, NULL);
488 
489 	t->ops = &tmpops;
490 
491 	qprocson(q);
492 	bp->b_datap->db_type = M_PCPROTO;
493 	*(int32_t *)bp->b_wptr = (int32_t)T_INFO_REQ;
494 	bp->b_wptr += sizeof (struct T_info_req);
495 	putnext(WR(q), bp);
496 
497 	mutex_enter(&t->lock);
498 	while (t->info_ack == NULL) {
499 		if (cv_wait_sig(&t->wait, &t->lock) == 0) {
500 			error = EINTR;
501 			break;
502 		}
503 	}
504 	mutex_exit(&t->lock);
505 
506 	if (error)
507 		goto out;
508 
509 	pptr = (struct T_info_ack *)t->info_ack->b_rptr;
510 
511 	if (pptr->SERV_type == T_CLTS) {
512 		if ((error = rpcmodopen(q, devp, flag, sflag, crp)) == 0)
513 			((struct rpcm *)q->q_ptr)->rm_ops = &xprt_clts_ops;
514 	} else {
515 		if ((error = mir_open(q, devp, flag, sflag, crp)) == 0)
516 			((mir_t *)q->q_ptr)->rm_ops = &xprt_cots_ops;
517 	}
518 
519 out:
520 	if (error)
521 		qprocsoff(q);
522 
523 	freemsg(t->info_ack);
524 	mutex_destroy(&t->lock);
525 	cv_destroy(&t->wait);
526 
527 	return (error);
528 }
529 
530 void
531 rmm_rput(queue_t *q, mblk_t  *mp)
532 {
533 	(*((struct temp_slot *)q->q_ptr)->ops->xo_rput)(q, mp);
534 }
535 
536 void
537 rmm_rsrv(queue_t *q)
538 {
539 	(*((struct temp_slot *)q->q_ptr)->ops->xo_rsrv)(q);
540 }
541 
542 void
543 rmm_wput(queue_t *q, mblk_t *mp)
544 {
545 	(*((struct temp_slot *)q->q_ptr)->ops->xo_wput)(q, mp);
546 }
547 
548 void
549 rmm_wsrv(queue_t *q)
550 {
551 	(*((struct temp_slot *)q->q_ptr)->ops->xo_wsrv)(q);
552 }
553 
554 int
555 rmm_close(queue_t *q, int flag, cred_t *crp)
556 {
557 	return ((*((struct temp_slot *)q->q_ptr)->ops->xo_close)(q, flag, crp));
558 }
559 
560 /*
561  * rpcmodopen -	open routine gets called when the module gets pushed
562  *		onto the stream.
563  */
564 /*ARGSUSED*/
565 int
566 rpcmodopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *crp)
567 {
568 	struct rpcm *rmp;
569 
570 	extern void (*rpc_rele)(queue_t *, mblk_t *);
571 	static void rpcmod_release(queue_t *, mblk_t *);
572 
573 	TRACE_0(TR_FAC_KRPC, TR_RPCMODOPEN_START, "rpcmodopen_start:");
574 
575 	/*
576 	 * Initialize entry points to release a rpcmod slot (and an input
577 	 * message if supplied) and to send an output message to the module
578 	 * below rpcmod.
579 	 */
580 	if (rpc_rele == NULL)
581 		rpc_rele = rpcmod_release;
582 
583 	/*
584 	 * Only sufficiently privileged users can use this module, and it
585 	 * is assumed that they will use this module properly, and NOT send
586 	 * bulk data from downstream.
587 	 */
588 	if (secpolicy_rpcmod_open(crp) != 0)
589 		return (EPERM);
590 
591 	/*
592 	 * Allocate slot data structure.
593 	 */
594 	rmp = kmem_zalloc(sizeof (*rmp), KM_SLEEP);
595 
596 	mutex_init(&rmp->rm_lock, NULL, MUTEX_DEFAULT, NULL);
597 	cv_init(&rmp->rm_cwait, NULL, CV_DEFAULT, NULL);
598 	rmp->rm_zoneid = rpc_zoneid();
599 	/*
600 	 * slot type will be set by kRPC client and server ioctl's
601 	 */
602 	rmp->rm_type = 0;
603 
604 	q->q_ptr = (void *)rmp;
605 	WR(q)->q_ptr = (void *)rmp;
606 
607 	TRACE_1(TR_FAC_KRPC, TR_RPCMODOPEN_END, "rpcmodopen_end:(%s)", "end");
608 	return (0);
609 }
610 
611 /*
612  * rpcmodclose - This routine gets called when the module gets popped
613  * off of the stream.
614  */
615 /*ARGSUSED*/
616 int
617 rpcmodclose(queue_t *q, int flag, cred_t *crp)
618 {
619 	struct rpcm *rmp;
620 
621 	ASSERT(q != NULL);
622 	rmp = (struct rpcm *)q->q_ptr;
623 
624 	/*
625 	 * Mark our state as closing.
626 	 */
627 	mutex_enter(&rmp->rm_lock);
628 	rmp->rm_state |= RM_CLOSING;
629 
630 	/*
631 	 * Check and see if there are any messages on the queue.  If so, send
632 	 * the messages, regardless whether the downstream module is ready to
633 	 * accept data.
634 	 */
635 	if (rmp->rm_type == RPC_SERVER) {
636 		flushq(q, FLUSHDATA);
637 
638 		qenable(WR(q));
639 
640 		if (rmp->rm_ref) {
641 			mutex_exit(&rmp->rm_lock);
642 			/*
643 			 * call into SVC to clean the queue
644 			 */
645 			svc_queueclean(q);
646 			mutex_enter(&rmp->rm_lock);
647 
648 			/*
649 			 * Block while there are kRPC threads with a reference
650 			 * to this message.
651 			 */
652 			while (rmp->rm_ref)
653 				cv_wait(&rmp->rm_cwait, &rmp->rm_lock);
654 		}
655 
656 		mutex_exit(&rmp->rm_lock);
657 
658 		/*
659 		 * It is now safe to remove this queue from the stream. No kRPC
660 		 * threads have a reference to the stream, and none ever will,
661 		 * because RM_CLOSING is set.
662 		 */
663 		qprocsoff(q);
664 
665 		/* Notify kRPC that this stream is going away. */
666 		svc_queueclose(q);
667 	} else {
668 		mutex_exit(&rmp->rm_lock);
669 		qprocsoff(q);
670 	}
671 
672 	q->q_ptr = NULL;
673 	WR(q)->q_ptr = NULL;
674 	mutex_destroy(&rmp->rm_lock);
675 	cv_destroy(&rmp->rm_cwait);
676 	kmem_free(rmp, sizeof (*rmp));
677 	return (0);
678 }
679 
680 #ifdef	DEBUG
681 int	rpcmod_send_msg_up = 0;
682 int	rpcmod_send_uderr = 0;
683 int	rpcmod_send_dup = 0;
684 int	rpcmod_send_dup_cnt = 0;
685 #endif
686 
687 /*
688  * rpcmodrput -	Module read put procedure.  This is called from
689  *		the module, driver, or stream head downstream.
690  */
691 void
692 rpcmodrput(queue_t *q, mblk_t *mp)
693 {
694 	struct rpcm *rmp;
695 	union T_primitives *pptr;
696 	int hdrsz;
697 
698 	TRACE_0(TR_FAC_KRPC, TR_RPCMODRPUT_START, "rpcmodrput_start:");
699 
700 	ASSERT(q != NULL);
701 	rmp = (struct rpcm *)q->q_ptr;
702 
703 	if (rmp->rm_type == 0) {
704 		freemsg(mp);
705 		return;
706 	}
707 
708 #ifdef DEBUG
709 	if (rpcmod_send_msg_up > 0) {
710 		mblk_t *nmp = copymsg(mp);
711 		if (nmp) {
712 			putnext(q, nmp);
713 			rpcmod_send_msg_up--;
714 		}
715 	}
716 	if ((rpcmod_send_uderr > 0) && mp->b_datap->db_type == M_PROTO) {
717 		mblk_t *nmp;
718 		struct T_unitdata_ind *data;
719 		struct T_uderror_ind *ud;
720 		int d;
721 		data = (struct T_unitdata_ind *)mp->b_rptr;
722 		if (data->PRIM_type == T_UNITDATA_IND) {
723 			d = sizeof (*ud) - sizeof (*data);
724 			nmp = allocb(mp->b_wptr - mp->b_rptr + d, BPRI_HI);
725 			if (nmp) {
726 				ud = (struct T_uderror_ind *)nmp->b_rptr;
727 				ud->PRIM_type = T_UDERROR_IND;
728 				ud->DEST_length = data->SRC_length;
729 				ud->DEST_offset = data->SRC_offset + d;
730 				ud->OPT_length = data->OPT_length;
731 				ud->OPT_offset = data->OPT_offset + d;
732 				ud->ERROR_type = ENETDOWN;
733 				if (data->SRC_length) {
734 					bcopy(mp->b_rptr +
735 					    data->SRC_offset,
736 					    nmp->b_rptr +
737 					    ud->DEST_offset,
738 					    data->SRC_length);
739 				}
740 				if (data->OPT_length) {
741 					bcopy(mp->b_rptr +
742 					    data->OPT_offset,
743 					    nmp->b_rptr +
744 					    ud->OPT_offset,
745 					    data->OPT_length);
746 				}
747 				nmp->b_wptr += d;
748 				nmp->b_wptr += (mp->b_wptr - mp->b_rptr);
749 				nmp->b_datap->db_type = M_PROTO;
750 				putnext(q, nmp);
751 				rpcmod_send_uderr--;
752 			}
753 		}
754 	}
755 #endif
756 	switch (mp->b_datap->db_type) {
757 	default:
758 		putnext(q, mp);
759 		break;
760 
761 	case M_PROTO:
762 	case M_PCPROTO:
763 		ASSERT((mp->b_wptr - mp->b_rptr) >= sizeof (int32_t));
764 		pptr = (union T_primitives *)mp->b_rptr;
765 
766 		/*
767 		 * Forward this message to krpc if it is data.
768 		 */
769 		if (pptr->type == T_UNITDATA_IND) {
770 			mblk_t *nmp;
771 
772 		/*
773 		 * Check if the module is being popped.
774 		 */
775 			mutex_enter(&rmp->rm_lock);
776 			if (rmp->rm_state & RM_CLOSING) {
777 				mutex_exit(&rmp->rm_lock);
778 				putnext(q, mp);
779 				break;
780 			}
781 
782 			switch (rmp->rm_type) {
783 			case RPC_CLIENT:
784 				mutex_exit(&rmp->rm_lock);
785 				hdrsz = mp->b_wptr - mp->b_rptr;
786 
787 				/*
788 				 * Make sure the header is sane.
789 				 */
790 				if (hdrsz < TUNITDATAINDSZ ||
791 				    hdrsz < (pptr->unitdata_ind.OPT_length +
792 				    pptr->unitdata_ind.OPT_offset) ||
793 				    hdrsz < (pptr->unitdata_ind.SRC_length +
794 				    pptr->unitdata_ind.SRC_offset)) {
795 					freemsg(mp);
796 					return;
797 				}
798 
799 				/*
800 				 * Call clnt_clts_dispatch_notify, so that it
801 				 * can pass the message to the proper caller.
802 				 * Don't discard the header just yet since the
803 				 * client may need the sender's address.
804 				 */
805 				clnt_clts_dispatch_notify(mp, hdrsz,
806 				    rmp->rm_zoneid);
807 				return;
808 			case RPC_SERVER:
809 				/*
810 				 * rm_krpc_cell is exclusively used by the kRPC
811 				 * CLTS server
812 				 */
813 				if (rmp->rm_krpc_cell) {
814 #ifdef DEBUG
815 					/*
816 					 * Test duplicate request cache and
817 					 * rm_ref count handling by sending a
818 					 * duplicate every so often, if
819 					 * desired.
820 					 */
821 					if (rpcmod_send_dup &&
822 					    rpcmod_send_dup_cnt++ %
823 					    rpcmod_send_dup)
824 						nmp = copymsg(mp);
825 					else
826 						nmp = NULL;
827 #endif
828 					/*
829 					 * Raise the reference count on this
830 					 * module to prevent it from being
831 					 * popped before krpc generates the
832 					 * reply.
833 					 */
834 					rmp->rm_ref++;
835 					mutex_exit(&rmp->rm_lock);
836 
837 					/*
838 					 * Submit the message to krpc.
839 					 */
840 					svc_queuereq(q, mp);
841 #ifdef DEBUG
842 					/*
843 					 * Send duplicate if we created one.
844 					 */
845 					if (nmp) {
846 						mutex_enter(&rmp->rm_lock);
847 						rmp->rm_ref++;
848 						mutex_exit(&rmp->rm_lock);
849 						svc_queuereq(q, nmp);
850 					}
851 #endif
852 				} else {
853 					mutex_exit(&rmp->rm_lock);
854 					freemsg(mp);
855 				}
856 				return;
857 			default:
858 				mutex_exit(&rmp->rm_lock);
859 				freemsg(mp);
860 				return;
861 			} /* end switch(rmp->rm_type) */
862 		} else if (pptr->type == T_UDERROR_IND) {
863 			mutex_enter(&rmp->rm_lock);
864 			hdrsz = mp->b_wptr - mp->b_rptr;
865 
866 			/*
867 			 * Make sure the header is sane
868 			 */
869 			if (hdrsz < TUDERRORINDSZ ||
870 			    hdrsz < (pptr->uderror_ind.OPT_length +
871 			    pptr->uderror_ind.OPT_offset) ||
872 			    hdrsz < (pptr->uderror_ind.DEST_length +
873 			    pptr->uderror_ind.DEST_offset)) {
874 				mutex_exit(&rmp->rm_lock);
875 				freemsg(mp);
876 				return;
877 			}
878 
879 			/*
880 			 * In the case where a unit data error has been
881 			 * received, all we need to do is clear the message from
882 			 * the queue.
883 			 */
884 			mutex_exit(&rmp->rm_lock);
885 			freemsg(mp);
886 			RPCLOG(32, "rpcmodrput: unitdata error received at "
887 			    "%ld\n", gethrestime_sec());
888 			return;
889 		} /* end else if (pptr->type == T_UDERROR_IND) */
890 
891 		putnext(q, mp);
892 		break;
893 	} /* end switch (mp->b_datap->db_type) */
894 
895 	TRACE_0(TR_FAC_KRPC, TR_RPCMODRPUT_END,
896 	    "rpcmodrput_end:");
897 	/*
898 	 * Return codes are not looked at by the STREAMS framework.
899 	 */
900 }
901 
902 /*
903  * write put procedure
904  */
905 void
906 rpcmodwput(queue_t *q, mblk_t *mp)
907 {
908 	struct rpcm	*rmp;
909 
910 	ASSERT(q != NULL);
911 
912 	switch (mp->b_datap->db_type) {
913 		case M_PROTO:
914 		case M_PCPROTO:
915 			break;
916 		default:
917 			rpcmodwput_other(q, mp);
918 			return;
919 	}
920 
921 	/*
922 	 * Check to see if we can send the message downstream.
923 	 */
924 	if (canputnext(q)) {
925 		putnext(q, mp);
926 		return;
927 	}
928 
929 	rmp = (struct rpcm *)q->q_ptr;
930 	ASSERT(rmp != NULL);
931 
932 	/*
933 	 * The first canputnext failed.  Try again except this time with the
934 	 * lock held, so that we can check the state of the stream to see if
935 	 * it is closing.  If either of these conditions evaluate to true
936 	 * then send the meesage.
937 	 */
938 	mutex_enter(&rmp->rm_lock);
939 	if (canputnext(q) || (rmp->rm_state & RM_CLOSING)) {
940 		mutex_exit(&rmp->rm_lock);
941 		putnext(q, mp);
942 	} else {
943 		/*
944 		 * canputnext failed again and the stream is not closing.
945 		 * Place the message on the queue and let the service
946 		 * procedure handle the message.
947 		 */
948 		mutex_exit(&rmp->rm_lock);
949 		(void) putq(q, mp);
950 	}
951 }
952 
953 static void
954 rpcmodwput_other(queue_t *q, mblk_t *mp)
955 {
956 	struct rpcm	*rmp;
957 	struct iocblk	*iocp;
958 
959 	rmp = (struct rpcm *)q->q_ptr;
960 	ASSERT(rmp != NULL);
961 
962 	switch (mp->b_datap->db_type) {
963 		case M_IOCTL:
964 			iocp = (struct iocblk *)mp->b_rptr;
965 			ASSERT(iocp != NULL);
966 			switch (iocp->ioc_cmd) {
967 				case RPC_CLIENT:
968 				case RPC_SERVER:
969 					mutex_enter(&rmp->rm_lock);
970 					rmp->rm_type = iocp->ioc_cmd;
971 					mutex_exit(&rmp->rm_lock);
972 					mp->b_datap->db_type = M_IOCACK;
973 					qreply(q, mp);
974 					return;
975 				default:
976 				/*
977 				 * pass the ioctl downstream and hope someone
978 				 * down there knows how to handle it.
979 				 */
980 					putnext(q, mp);
981 					return;
982 			}
983 		default:
984 			break;
985 	}
986 	/*
987 	 * This is something we definitely do not know how to handle, just
988 	 * pass the message downstream
989 	 */
990 	putnext(q, mp);
991 }
992 
993 /*
994  * Module write service procedure. This is called by downstream modules
995  * for back enabling during flow control.
996  */
997 void
998 rpcmodwsrv(queue_t *q)
999 {
1000 	struct rpcm	*rmp;
1001 	mblk_t		*mp = NULL;
1002 
1003 	rmp = (struct rpcm *)q->q_ptr;
1004 	ASSERT(rmp != NULL);
1005 
1006 	/*
1007 	 * Get messages that may be queued and send them down stream
1008 	 */
1009 	while ((mp = getq(q)) != NULL) {
1010 		/*
1011 		 * Optimize the service procedure for the server-side, by
1012 		 * avoiding a call to canputnext().
1013 		 */
1014 		if (rmp->rm_type == RPC_SERVER || canputnext(q)) {
1015 			putnext(q, mp);
1016 			continue;
1017 		}
1018 		(void) putbq(q, mp);
1019 		return;
1020 	}
1021 }
1022 
1023 static void
1024 rpcmod_release(queue_t *q, mblk_t *bp)
1025 {
1026 	struct rpcm *rmp;
1027 
1028 	/*
1029 	 * For now, just free the message.
1030 	 */
1031 	if (bp)
1032 		freemsg(bp);
1033 	rmp = (struct rpcm *)q->q_ptr;
1034 
1035 	mutex_enter(&rmp->rm_lock);
1036 	rmp->rm_ref--;
1037 
1038 	if (rmp->rm_ref == 0 && (rmp->rm_state & RM_CLOSING)) {
1039 		cv_broadcast(&rmp->rm_cwait);
1040 	}
1041 
1042 	mutex_exit(&rmp->rm_lock);
1043 }
1044 
1045 /*
1046  * This part of rpcmod is pushed on a connection-oriented transport for use
1047  * by RPC.  It serves to bypass the Stream head, implements
1048  * the record marking protocol, and dispatches incoming RPC messages.
1049  */
1050 
1051 /* Default idle timer values */
1052 #define	MIR_CLNT_IDLE_TIMEOUT	(5 * (60 * 1000L))	/* 5 minutes */
1053 #define	MIR_SVC_IDLE_TIMEOUT	(6 * (60 * 1000L))	/* 6 minutes */
1054 #define	MIR_SVC_ORDREL_TIMEOUT	(10 * (60 * 1000L))	/* 10 minutes */
1055 #define	MIR_LASTFRAG	0x80000000	/* Record marker */
1056 
1057 #define	DLEN(mp) (mp->b_cont ? msgdsize(mp) : (mp->b_wptr - mp->b_rptr))
1058 
1059 #define	MIR_SVC_QUIESCED(mir)	\
1060 	(mir->mir_ref_cnt == 0 && mir->mir_inrservice == 0)
1061 
1062 #define	MIR_CLEAR_INRSRV(mir_ptr)	{	\
1063 	(mir_ptr)->mir_inrservice = 0;	\
1064 	if ((mir_ptr)->mir_type == RPC_SERVER &&	\
1065 		(mir_ptr)->mir_closing)	\
1066 		cv_signal(&(mir_ptr)->mir_condvar);	\
1067 }
1068 
1069 /*
1070  * Don't block service procedure (and mir_close) if
1071  * we are in the process of closing.
1072  */
1073 #define	MIR_WCANPUTNEXT(mir_ptr, write_q)	\
1074 	(canputnext(write_q) || ((mir_ptr)->mir_svc_no_more_msgs == 1))
1075 
1076 static int	mir_clnt_dup_request(queue_t *q, mblk_t *mp);
1077 static void	mir_rput_proto(queue_t *q, mblk_t *mp);
1078 static int	mir_svc_policy_notify(queue_t *q, int event);
1079 static void	mir_svc_release(queue_t *wq, mblk_t *mp);
1080 static void	mir_svc_start(queue_t *wq);
1081 static void	mir_svc_idle_start(queue_t *, mir_t *);
1082 static void	mir_svc_idle_stop(queue_t *, mir_t *);
1083 static void	mir_svc_start_close(queue_t *, mir_t *);
1084 static void	mir_clnt_idle_do_stop(queue_t *);
1085 static void	mir_clnt_idle_stop(queue_t *, mir_t *);
1086 static void	mir_clnt_idle_start(queue_t *, mir_t *);
1087 static void	mir_wput(queue_t *q, mblk_t *mp);
1088 static void	mir_wput_other(queue_t *q, mblk_t *mp);
1089 static void	mir_wsrv(queue_t *q);
1090 static	void	mir_disconnect(queue_t *, mir_t *ir);
1091 static	int	mir_check_len(queue_t *, int32_t, mblk_t *);
1092 static	void	mir_timer(void *);
1093 
1094 extern void	(*mir_rele)(queue_t *, mblk_t *);
1095 extern void	(*mir_start)(queue_t *);
1096 extern void	(*clnt_stop_idle)(queue_t *);
1097 
1098 clock_t	clnt_idle_timeout = MIR_CLNT_IDLE_TIMEOUT;
1099 clock_t	svc_idle_timeout = MIR_SVC_IDLE_TIMEOUT;
1100 
1101 /*
1102  * Timeout for subsequent notifications of idle connection.  This is
1103  * typically used to clean up after a wedged orderly release.
1104  */
1105 clock_t	svc_ordrel_timeout = MIR_SVC_ORDREL_TIMEOUT; /* milliseconds */
1106 
1107 extern	uint_t	*clnt_max_msg_sizep;
1108 extern	uint_t	*svc_max_msg_sizep;
1109 uint_t	clnt_max_msg_size = RPC_MAXDATASIZE;
1110 uint_t	svc_max_msg_size = RPC_MAXDATASIZE;
1111 uint_t	mir_krpc_cell_null;
1112 
1113 static void
1114 mir_timer_stop(mir_t *mir)
1115 {
1116 	timeout_id_t tid;
1117 
1118 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
1119 
1120 	/*
1121 	 * Since the mir_mutex lock needs to be released to call
1122 	 * untimeout(), we need to make sure that no other thread
1123 	 * can start/stop the timer (changing mir_timer_id) during
1124 	 * that time.  The mir_timer_call bit and the mir_timer_cv
1125 	 * condition variable are used to synchronize this.  Setting
1126 	 * mir_timer_call also tells mir_timer() (refer to the comments
1127 	 * in mir_timer()) that it does not need to do anything.
1128 	 */
1129 	while (mir->mir_timer_call)
1130 		cv_wait(&mir->mir_timer_cv, &mir->mir_mutex);
1131 	mir->mir_timer_call = B_TRUE;
1132 
1133 	if ((tid = mir->mir_timer_id) != 0) {
1134 		mir->mir_timer_id = 0;
1135 		mutex_exit(&mir->mir_mutex);
1136 		(void) untimeout(tid);
1137 		mutex_enter(&mir->mir_mutex);
1138 	}
1139 	mir->mir_timer_call = B_FALSE;
1140 	cv_broadcast(&mir->mir_timer_cv);
1141 }
1142 
1143 static void
1144 mir_timer_start(queue_t *q, mir_t *mir, clock_t intrvl)
1145 {
1146 	timeout_id_t tid;
1147 
1148 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
1149 
1150 	while (mir->mir_timer_call)
1151 		cv_wait(&mir->mir_timer_cv, &mir->mir_mutex);
1152 	mir->mir_timer_call = B_TRUE;
1153 
1154 	if ((tid = mir->mir_timer_id) != 0) {
1155 		mutex_exit(&mir->mir_mutex);
1156 		(void) untimeout(tid);
1157 		mutex_enter(&mir->mir_mutex);
1158 	}
1159 	/* Only start the timer when it is not closing. */
1160 	if (!mir->mir_closing) {
1161 		mir->mir_timer_id = timeout(mir_timer, q,
1162 		    MSEC_TO_TICK(intrvl));
1163 	}
1164 	mir->mir_timer_call = B_FALSE;
1165 	cv_broadcast(&mir->mir_timer_cv);
1166 }
1167 
1168 static int
1169 mir_clnt_dup_request(queue_t *q, mblk_t *mp)
1170 {
1171 	mblk_t  *mp1;
1172 	uint32_t  new_xid;
1173 	uint32_t  old_xid;
1174 
1175 	ASSERT(MUTEX_HELD(&((mir_t *)q->q_ptr)->mir_mutex));
1176 	new_xid = BE32_TO_U32(&mp->b_rptr[4]);
1177 	/*
1178 	 * This loop is a bit tacky -- it walks the STREAMS list of
1179 	 * flow-controlled messages.
1180 	 */
1181 	if ((mp1 = q->q_first) != NULL) {
1182 		do {
1183 			old_xid = BE32_TO_U32(&mp1->b_rptr[4]);
1184 			if (new_xid == old_xid)
1185 				return (1);
1186 		} while ((mp1 = mp1->b_next) != NULL);
1187 	}
1188 	return (0);
1189 }
1190 
1191 static int
1192 mir_close(queue_t *q)
1193 {
1194 	mir_t	*mir = q->q_ptr;
1195 	mblk_t	*mp;
1196 	bool_t queue_cleaned = FALSE;
1197 
1198 	RPCLOG(32, "rpcmod: mir_close of q 0x%p\n", (void *)q);
1199 	ASSERT(MUTEX_NOT_HELD(&mir->mir_mutex));
1200 	mutex_enter(&mir->mir_mutex);
1201 	if ((mp = mir->mir_head_mp) != NULL) {
1202 		mir->mir_head_mp = NULL;
1203 		mir->mir_tail_mp = NULL;
1204 		freemsg(mp);
1205 	}
1206 	/*
1207 	 * Set mir_closing so we get notified when MIR_SVC_QUIESCED()
1208 	 * is TRUE.  And mir_timer_start() won't start the timer again.
1209 	 */
1210 	mir->mir_closing = B_TRUE;
1211 	mir_timer_stop(mir);
1212 
1213 	if (mir->mir_type == RPC_SERVER) {
1214 		flushq(q, FLUSHDATA);	/* Ditch anything waiting on read q */
1215 
1216 		/*
1217 		 * This will prevent more requests from arriving and
1218 		 * will force rpcmod to ignore flow control.
1219 		 */
1220 		mir_svc_start_close(WR(q), mir);
1221 
1222 		while ((!MIR_SVC_QUIESCED(mir)) || mir->mir_inwservice == 1) {
1223 
1224 			if (mir->mir_ref_cnt && !mir->mir_inrservice &&
1225 			    (queue_cleaned == FALSE)) {
1226 				/*
1227 				 * call into SVC to clean the queue
1228 				 */
1229 				mutex_exit(&mir->mir_mutex);
1230 				svc_queueclean(q);
1231 				queue_cleaned = TRUE;
1232 				mutex_enter(&mir->mir_mutex);
1233 				continue;
1234 			}
1235 
1236 			/*
1237 			 * Bugid 1253810 - Force the write service
1238 			 * procedure to send its messages, regardless
1239 			 * whether the downstream  module is ready
1240 			 * to accept data.
1241 			 */
1242 			if (mir->mir_inwservice == 1)
1243 				qenable(WR(q));
1244 
1245 			cv_wait(&mir->mir_condvar, &mir->mir_mutex);
1246 		}
1247 
1248 		mutex_exit(&mir->mir_mutex);
1249 		qprocsoff(q);
1250 
1251 		/* Notify KRPC that this stream is going away. */
1252 		svc_queueclose(q);
1253 	} else {
1254 		mutex_exit(&mir->mir_mutex);
1255 		qprocsoff(q);
1256 	}
1257 
1258 	mutex_destroy(&mir->mir_mutex);
1259 	cv_destroy(&mir->mir_condvar);
1260 	cv_destroy(&mir->mir_timer_cv);
1261 	kmem_free(mir, sizeof (mir_t));
1262 	return (0);
1263 }
1264 
1265 /*
1266  * This is server side only (RPC_SERVER).
1267  *
1268  * Exit idle mode.
1269  */
1270 static void
1271 mir_svc_idle_stop(queue_t *q, mir_t *mir)
1272 {
1273 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
1274 	ASSERT((q->q_flag & QREADR) == 0);
1275 	ASSERT(mir->mir_type == RPC_SERVER);
1276 	RPCLOG(16, "rpcmod: mir_svc_idle_stop of q 0x%p\n", (void *)q);
1277 
1278 	mir_timer_stop(mir);
1279 }
1280 
1281 /*
1282  * This is server side only (RPC_SERVER).
1283  *
1284  * Start idle processing, which will include setting idle timer if the
1285  * stream is not being closed.
1286  */
1287 static void
1288 mir_svc_idle_start(queue_t *q, mir_t *mir)
1289 {
1290 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
1291 	ASSERT((q->q_flag & QREADR) == 0);
1292 	ASSERT(mir->mir_type == RPC_SERVER);
1293 	RPCLOG(16, "rpcmod: mir_svc_idle_start q 0x%p\n", (void *)q);
1294 
1295 	/*
1296 	 * Don't re-start idle timer if we are closing queues.
1297 	 */
1298 	if (mir->mir_closing) {
1299 		RPCLOG(16, "mir_svc_idle_start - closing: 0x%p\n",
1300 		    (void *)q);
1301 
1302 		/*
1303 		 * We will call mir_svc_idle_start() whenever MIR_SVC_QUIESCED()
1304 		 * is true.  When it is true, and we are in the process of
1305 		 * closing the stream, signal any thread waiting in
1306 		 * mir_close().
1307 		 */
1308 		if (mir->mir_inwservice == 0)
1309 			cv_signal(&mir->mir_condvar);
1310 
1311 	} else {
1312 		RPCLOG(16, "mir_svc_idle_start - reset %s timer\n",
1313 		    mir->mir_ordrel_pending ? "ordrel" : "normal");
1314 		/*
1315 		 * Normal condition, start the idle timer.  If an orderly
1316 		 * release has been sent, set the timeout to wait for the
1317 		 * client to close its side of the connection.  Otherwise,
1318 		 * use the normal idle timeout.
1319 		 */
1320 		mir_timer_start(q, mir, mir->mir_ordrel_pending ?
1321 		    svc_ordrel_timeout : mir->mir_idle_timeout);
1322 	}
1323 }
1324 
1325 /* ARGSUSED */
1326 static int
1327 mir_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1328 {
1329 	mir_t	*mir;
1330 
1331 	RPCLOG(32, "rpcmod: mir_open of q 0x%p\n", (void *)q);
1332 	/* Set variables used directly by KRPC. */
1333 	if (!mir_rele)
1334 		mir_rele = mir_svc_release;
1335 	if (!mir_start)
1336 		mir_start = mir_svc_start;
1337 	if (!clnt_stop_idle)
1338 		clnt_stop_idle = mir_clnt_idle_do_stop;
1339 	if (!clnt_max_msg_sizep)
1340 		clnt_max_msg_sizep = &clnt_max_msg_size;
1341 	if (!svc_max_msg_sizep)
1342 		svc_max_msg_sizep = &svc_max_msg_size;
1343 
1344 	/* Allocate a zero'ed out mir structure for this stream. */
1345 	mir = kmem_zalloc(sizeof (mir_t), KM_SLEEP);
1346 
1347 	/*
1348 	 * We set hold inbound here so that incoming messages will
1349 	 * be held on the read-side queue until the stream is completely
1350 	 * initialized with a RPC_CLIENT or RPC_SERVER ioctl.  During
1351 	 * the ioctl processing, the flag is cleared and any messages that
1352 	 * arrived between the open and the ioctl are delivered to KRPC.
1353 	 *
1354 	 * Early data should never arrive on a client stream since
1355 	 * servers only respond to our requests and we do not send any.
1356 	 * until after the stream is initialized.  Early data is
1357 	 * very common on a server stream where the client will start
1358 	 * sending data as soon as the connection is made (and this
1359 	 * is especially true with TCP where the protocol accepts the
1360 	 * connection before nfsd or KRPC is notified about it).
1361 	 */
1362 
1363 	mir->mir_hold_inbound = 1;
1364 
1365 	/*
1366 	 * Start the record marker looking for a 4-byte header.  When
1367 	 * this length is negative, it indicates that rpcmod is looking
1368 	 * for bytes to consume for the record marker header.  When it
1369 	 * is positive, it holds the number of bytes that have arrived
1370 	 * for the current fragment and are being held in mir_header_mp.
1371 	 */
1372 
1373 	mir->mir_frag_len = -(int32_t)sizeof (uint32_t);
1374 
1375 	mir->mir_zoneid = rpc_zoneid();
1376 	mutex_init(&mir->mir_mutex, NULL, MUTEX_DEFAULT, NULL);
1377 	cv_init(&mir->mir_condvar, NULL, CV_DRIVER, NULL);
1378 	cv_init(&mir->mir_timer_cv, NULL, CV_DRIVER, NULL);
1379 
1380 	q->q_ptr = (char *)mir;
1381 	WR(q)->q_ptr = (char *)mir;
1382 
1383 	/*
1384 	 * We noenable the read-side queue because we don't want it
1385 	 * automatically enabled by putq.  We enable it explicitly
1386 	 * in mir_wsrv when appropriate. (See additional comments on
1387 	 * flow control at the beginning of mir_rsrv.)
1388 	 */
1389 	noenable(q);
1390 
1391 	qprocson(q);
1392 	return (0);
1393 }
1394 
1395 /*
1396  * Read-side put routine for both the client and server side.  Does the
1397  * record marking for incoming RPC messages, and when complete, dispatches
1398  * the message to either the client or server.
1399  */
1400 static void
1401 mir_rput(queue_t *q, mblk_t *mp)
1402 {
1403 	int	excess;
1404 	int32_t	frag_len, frag_header;
1405 	mblk_t	*cont_mp, *head_mp, *tail_mp, *mp1;
1406 	mir_t	*mir = q->q_ptr;
1407 	boolean_t stop_timer = B_FALSE;
1408 
1409 	ASSERT(mir != NULL);
1410 
1411 	/*
1412 	 * If the stream has not been set up as a RPC_CLIENT or RPC_SERVER
1413 	 * with the corresponding ioctl, then don't accept
1414 	 * any inbound data.  This should never happen for streams
1415 	 * created by nfsd or client-side KRPC because they are careful
1416 	 * to set the mode of the stream before doing anything else.
1417 	 */
1418 	if (mir->mir_type == 0) {
1419 		freemsg(mp);
1420 		return;
1421 	}
1422 
1423 	ASSERT(MUTEX_NOT_HELD(&mir->mir_mutex));
1424 
1425 	switch (mp->b_datap->db_type) {
1426 	case M_DATA:
1427 		break;
1428 	case M_PROTO:
1429 	case M_PCPROTO:
1430 		if (MBLKL(mp) < sizeof (t_scalar_t)) {
1431 			RPCLOG(1, "mir_rput: runt TPI message (%d bytes)\n",
1432 			    (int)MBLKL(mp));
1433 			freemsg(mp);
1434 			return;
1435 		}
1436 		if (((union T_primitives *)mp->b_rptr)->type != T_DATA_IND) {
1437 			mir_rput_proto(q, mp);
1438 			return;
1439 		}
1440 
1441 		/* Throw away the T_DATA_IND block and continue with data. */
1442 		mp1 = mp;
1443 		mp = mp->b_cont;
1444 		freeb(mp1);
1445 		break;
1446 	case M_SETOPTS:
1447 		/*
1448 		 * If a module on the stream is trying set the Stream head's
1449 		 * high water mark, then set our hiwater to the requested
1450 		 * value.  We are the "stream head" for all inbound
1451 		 * data messages since messages are passed directly to KRPC.
1452 		 */
1453 		if (MBLKL(mp) >= sizeof (struct stroptions)) {
1454 			struct stroptions	*stropts;
1455 
1456 			stropts = (struct stroptions *)mp->b_rptr;
1457 			if ((stropts->so_flags & SO_HIWAT) &&
1458 			    !(stropts->so_flags & SO_BAND)) {
1459 				(void) strqset(q, QHIWAT, 0, stropts->so_hiwat);
1460 			}
1461 		}
1462 		putnext(q, mp);
1463 		return;
1464 	case M_FLUSH:
1465 		RPCLOG(32, "mir_rput: ignoring M_FLUSH %x ", *mp->b_rptr);
1466 		RPCLOG(32, "on q 0x%p\n", (void *)q);
1467 		putnext(q, mp);
1468 		return;
1469 	default:
1470 		putnext(q, mp);
1471 		return;
1472 	}
1473 
1474 	mutex_enter(&mir->mir_mutex);
1475 
1476 	/*
1477 	 * If this connection is closing, don't accept any new messages.
1478 	 */
1479 	if (mir->mir_svc_no_more_msgs) {
1480 		ASSERT(mir->mir_type == RPC_SERVER);
1481 		mutex_exit(&mir->mir_mutex);
1482 		freemsg(mp);
1483 		return;
1484 	}
1485 
1486 	/* Get local copies for quicker access. */
1487 	frag_len = mir->mir_frag_len;
1488 	frag_header = mir->mir_frag_header;
1489 	head_mp = mir->mir_head_mp;
1490 	tail_mp = mir->mir_tail_mp;
1491 
1492 	/* Loop, processing each message block in the mp chain separately. */
1493 	do {
1494 		cont_mp = mp->b_cont;
1495 		mp->b_cont = NULL;
1496 
1497 		/*
1498 		 * Drop zero-length mblks to prevent unbounded kernel memory
1499 		 * consumption.
1500 		 */
1501 		if (MBLKL(mp) == 0) {
1502 			freeb(mp);
1503 			continue;
1504 		}
1505 
1506 		/*
1507 		 * If frag_len is negative, we're still in the process of
1508 		 * building frag_header -- try to complete it with this mblk.
1509 		 */
1510 		while (frag_len < 0 && mp->b_rptr < mp->b_wptr) {
1511 			frag_len++;
1512 			frag_header <<= 8;
1513 			frag_header += *mp->b_rptr++;
1514 		}
1515 
1516 		if (MBLKL(mp) == 0 && frag_len < 0) {
1517 			/*
1518 			 * We consumed this mblk while trying to complete the
1519 			 * fragment header.  Free it and move on.
1520 			 */
1521 			freeb(mp);
1522 			continue;
1523 		}
1524 
1525 		ASSERT(frag_len >= 0);
1526 
1527 		/*
1528 		 * Now frag_header has the number of bytes in this fragment
1529 		 * and we're just waiting to collect them all.  Chain our
1530 		 * latest mblk onto the list and see if we now have enough
1531 		 * bytes to complete the fragment.
1532 		 */
1533 		if (head_mp == NULL) {
1534 			ASSERT(tail_mp == NULL);
1535 			head_mp = tail_mp = mp;
1536 		} else {
1537 			tail_mp->b_cont = mp;
1538 			tail_mp = mp;
1539 		}
1540 
1541 		frag_len += MBLKL(mp);
1542 		excess = frag_len - (frag_header & ~MIR_LASTFRAG);
1543 		if (excess < 0) {
1544 			/*
1545 			 * We still haven't received enough data to complete
1546 			 * the fragment, so continue on to the next mblk.
1547 			 */
1548 			continue;
1549 		}
1550 
1551 		/*
1552 		 * We've got a complete fragment.  If there are excess bytes,
1553 		 * then they're part of the next fragment's header (of either
1554 		 * this RPC message or the next RPC message).  Split that part
1555 		 * into its own mblk so that we can safely freeb() it when
1556 		 * building frag_header above.
1557 		 */
1558 		if (excess > 0) {
1559 			if ((mp1 = dupb(mp)) == NULL &&
1560 			    (mp1 = copyb(mp)) == NULL) {
1561 				freemsg(head_mp);
1562 				freemsg(cont_mp);
1563 				RPCLOG0(1, "mir_rput: dupb/copyb failed\n");
1564 				mir->mir_frag_header = 0;
1565 				mir->mir_frag_len = -(int32_t)sizeof (uint32_t);
1566 				mir->mir_head_mp = NULL;
1567 				mir->mir_tail_mp = NULL;
1568 				mir_disconnect(q, mir);	/* drops mir_mutex */
1569 				return;
1570 			}
1571 
1572 			/*
1573 			 * Relink the message chain so that the next mblk is
1574 			 * the next fragment header, followed by the rest of
1575 			 * the message chain.
1576 			 */
1577 			mp1->b_cont = cont_mp;
1578 			cont_mp = mp1;
1579 
1580 			/*
1581 			 * Data in the new mblk begins at the next fragment,
1582 			 * and data in the old mblk ends at the next fragment.
1583 			 */
1584 			mp1->b_rptr = mp1->b_wptr - excess;
1585 			mp->b_wptr -= excess;
1586 		}
1587 
1588 		/*
1589 		 * Reset frag_len and frag_header for the next fragment.
1590 		 */
1591 		frag_len = -(int32_t)sizeof (uint32_t);
1592 		if (!(frag_header & MIR_LASTFRAG)) {
1593 			/*
1594 			 * The current fragment is complete, but more
1595 			 * fragments need to be processed before we can
1596 			 * pass along the RPC message headed at head_mp.
1597 			 */
1598 			frag_header = 0;
1599 			continue;
1600 		}
1601 		frag_header = 0;
1602 
1603 		/*
1604 		 * We've got a complete RPC message; pass it to the
1605 		 * appropriate consumer.
1606 		 */
1607 		switch (mir->mir_type) {
1608 		case RPC_CLIENT:
1609 			if (clnt_dispatch_notify(head_mp, mir->mir_zoneid)) {
1610 				/*
1611 				 * Mark this stream as active.  This marker
1612 				 * is used in mir_timer().
1613 				 */
1614 				mir->mir_clntreq = 1;
1615 				mir->mir_use_timestamp = lbolt;
1616 			} else {
1617 				freemsg(head_mp);
1618 			}
1619 			break;
1620 
1621 		case RPC_SERVER:
1622 			/*
1623 			 * Check for flow control before passing the
1624 			 * message to KRPC.
1625 			 */
1626 			if (!mir->mir_hold_inbound) {
1627 				if (mir->mir_krpc_cell) {
1628 					/*
1629 					 * If the reference count is 0
1630 					 * (not including this request),
1631 					 * then the stream is transitioning
1632 					 * from idle to non-idle.  In this case,
1633 					 * we cancel the idle timer.
1634 					 */
1635 					if (mir->mir_ref_cnt++ == 0)
1636 						stop_timer = B_TRUE;
1637 					if (mir_check_len(q,
1638 					    (int32_t)msgdsize(mp), mp))
1639 						return;
1640 					svc_queuereq(q, head_mp); /* to KRPC */
1641 				} else {
1642 					/*
1643 					 * Count # of times this happens. Should
1644 					 * be never, but experience shows
1645 					 * otherwise.
1646 					 */
1647 					mir_krpc_cell_null++;
1648 					freemsg(head_mp);
1649 				}
1650 			} else {
1651 				/*
1652 				 * If the outbound side of the stream is
1653 				 * flow controlled, then hold this message
1654 				 * until client catches up. mir_hold_inbound
1655 				 * is set in mir_wput and cleared in mir_wsrv.
1656 				 */
1657 				(void) putq(q, head_mp);
1658 				mir->mir_inrservice = B_TRUE;
1659 			}
1660 			break;
1661 		default:
1662 			RPCLOG(1, "mir_rput: unknown mir_type %d\n",
1663 			    mir->mir_type);
1664 			freemsg(head_mp);
1665 			break;
1666 		}
1667 
1668 		/*
1669 		 * Reset the chain since we're starting on a new RPC message.
1670 		 */
1671 		head_mp = tail_mp = NULL;
1672 	} while ((mp = cont_mp) != NULL);
1673 
1674 	/*
1675 	 * Sanity check the message length; if it's too large mir_check_len()
1676 	 * will shutdown the connection, drop mir_mutex, and return non-zero.
1677 	 */
1678 	if (head_mp != NULL && mir->mir_setup_complete &&
1679 	    mir_check_len(q, frag_len, head_mp))
1680 		return;
1681 
1682 	/* Save our local copies back in the mir structure. */
1683 	mir->mir_frag_header = frag_header;
1684 	mir->mir_frag_len = frag_len;
1685 	mir->mir_head_mp = head_mp;
1686 	mir->mir_tail_mp = tail_mp;
1687 
1688 	/*
1689 	 * The timer is stopped after the whole message chain is processed.
1690 	 * The reason is that stopping the timer releases the mir_mutex
1691 	 * lock temporarily.  This means that the request can be serviced
1692 	 * while we are still processing the message chain.  This is not
1693 	 * good.  So we stop the timer here instead.
1694 	 *
1695 	 * Note that if the timer fires before we stop it, it will not
1696 	 * do any harm as MIR_SVC_QUIESCED() is false and mir_timer()
1697 	 * will just return.
1698 	 */
1699 	if (stop_timer) {
1700 		RPCLOG(16, "mir_rput: stopping idle timer on 0x%p because "
1701 		    "ref cnt going to non zero\n", (void *)WR(q));
1702 		mir_svc_idle_stop(WR(q), mir);
1703 	}
1704 	mutex_exit(&mir->mir_mutex);
1705 }
1706 
1707 static void
1708 mir_rput_proto(queue_t *q, mblk_t *mp)
1709 {
1710 	mir_t	*mir = (mir_t *)q->q_ptr;
1711 	uint32_t	type;
1712 	uint32_t reason = 0;
1713 
1714 	ASSERT(MUTEX_NOT_HELD(&mir->mir_mutex));
1715 
1716 	type = ((union T_primitives *)mp->b_rptr)->type;
1717 	switch (mir->mir_type) {
1718 	case RPC_CLIENT:
1719 		switch (type) {
1720 		case T_DISCON_IND:
1721 			reason = ((struct T_discon_ind *)
1722 			    (mp->b_rptr))->DISCON_reason;
1723 			/*FALLTHROUGH*/
1724 		case T_ORDREL_IND:
1725 			mutex_enter(&mir->mir_mutex);
1726 			if (mir->mir_head_mp) {
1727 				freemsg(mir->mir_head_mp);
1728 				mir->mir_head_mp = (mblk_t *)0;
1729 				mir->mir_tail_mp = (mblk_t *)0;
1730 			}
1731 			/*
1732 			 * We are disconnecting, but not necessarily
1733 			 * closing. By not closing, we will fail to
1734 			 * pick up a possibly changed global timeout value,
1735 			 * unless we store it now.
1736 			 */
1737 			mir->mir_idle_timeout = clnt_idle_timeout;
1738 			mir_clnt_idle_stop(WR(q), mir);
1739 
1740 			/*
1741 			 * Even though we are unconnected, we still
1742 			 * leave the idle timer going on the client. The
1743 			 * reason for is that if we've disconnected due
1744 			 * to a server-side disconnect, reset, or connection
1745 			 * timeout, there is a possibility the client may
1746 			 * retry the RPC request. This retry needs to done on
1747 			 * the same bound address for the server to interpret
1748 			 * it as such. However, we don't want
1749 			 * to wait forever for that possibility. If the
1750 			 * end-point stays unconnected for mir_idle_timeout
1751 			 * units of time, then that is a signal to the
1752 			 * connection manager to give up waiting for the
1753 			 * application (eg. NFS) to send a retry.
1754 			 */
1755 			mir_clnt_idle_start(WR(q), mir);
1756 			mutex_exit(&mir->mir_mutex);
1757 			clnt_dispatch_notifyall(WR(q), type, reason);
1758 			freemsg(mp);
1759 			return;
1760 		case T_ERROR_ACK:
1761 		{
1762 			struct T_error_ack	*terror;
1763 
1764 			terror = (struct T_error_ack *)mp->b_rptr;
1765 			RPCLOG(1, "mir_rput_proto T_ERROR_ACK for queue 0x%p",
1766 			    (void *)q);
1767 			RPCLOG(1, " ERROR_prim: %s,",
1768 			    rpc_tpiprim2name(terror->ERROR_prim));
1769 			RPCLOG(1, " TLI_error: %s,",
1770 			    rpc_tpierr2name(terror->TLI_error));
1771 			RPCLOG(1, " UNIX_error: %d\n", terror->UNIX_error);
1772 			if (terror->ERROR_prim == T_DISCON_REQ)  {
1773 				clnt_dispatch_notifyall(WR(q), type, reason);
1774 				freemsg(mp);
1775 				return;
1776 			} else {
1777 				if (clnt_dispatch_notifyconn(WR(q), mp))
1778 					return;
1779 			}
1780 			break;
1781 		}
1782 		case T_OK_ACK:
1783 		{
1784 			struct T_ok_ack	*tok = (struct T_ok_ack *)mp->b_rptr;
1785 
1786 			if (tok->CORRECT_prim == T_DISCON_REQ) {
1787 				clnt_dispatch_notifyall(WR(q), type, reason);
1788 				freemsg(mp);
1789 				return;
1790 			} else {
1791 				if (clnt_dispatch_notifyconn(WR(q), mp))
1792 					return;
1793 			}
1794 			break;
1795 		}
1796 		case T_CONN_CON:
1797 		case T_INFO_ACK:
1798 		case T_OPTMGMT_ACK:
1799 			if (clnt_dispatch_notifyconn(WR(q), mp))
1800 				return;
1801 			break;
1802 		case T_BIND_ACK:
1803 			break;
1804 		default:
1805 			RPCLOG(1, "mir_rput: unexpected message %d "
1806 			    "for KRPC client\n",
1807 			    ((union T_primitives *)mp->b_rptr)->type);
1808 			break;
1809 		}
1810 		break;
1811 
1812 	case RPC_SERVER:
1813 		switch (type) {
1814 		case T_BIND_ACK:
1815 		{
1816 			struct T_bind_ack	*tbind;
1817 
1818 			/*
1819 			 * If this is a listening stream, then shut
1820 			 * off the idle timer.
1821 			 */
1822 			tbind = (struct T_bind_ack *)mp->b_rptr;
1823 			if (tbind->CONIND_number > 0) {
1824 				mutex_enter(&mir->mir_mutex);
1825 				mir_svc_idle_stop(WR(q), mir);
1826 
1827 				/*
1828 				 * mark this as a listen endpoint
1829 				 * for special handling.
1830 				 */
1831 
1832 				mir->mir_listen_stream = 1;
1833 				mutex_exit(&mir->mir_mutex);
1834 			}
1835 			break;
1836 		}
1837 		case T_DISCON_IND:
1838 		case T_ORDREL_IND:
1839 			RPCLOG(16, "mir_rput_proto: got %s indication\n",
1840 			    type == T_DISCON_IND ? "disconnect"
1841 			    : "orderly release");
1842 
1843 			/*
1844 			 * For listen endpoint just pass
1845 			 * on the message.
1846 			 */
1847 
1848 			if (mir->mir_listen_stream)
1849 				break;
1850 
1851 			mutex_enter(&mir->mir_mutex);
1852 
1853 			/*
1854 			 * If client wants to break off connection, record
1855 			 * that fact.
1856 			 */
1857 			mir_svc_start_close(WR(q), mir);
1858 
1859 			/*
1860 			 * If we are idle, then send the orderly release
1861 			 * or disconnect indication to nfsd.
1862 			 */
1863 			if (MIR_SVC_QUIESCED(mir)) {
1864 				mutex_exit(&mir->mir_mutex);
1865 				break;
1866 			}
1867 
1868 			RPCLOG(16, "mir_rput_proto: not idle, so "
1869 			    "disconnect/ord rel indication not passed "
1870 			    "upstream on 0x%p\n", (void *)q);
1871 
1872 			/*
1873 			 * Hold the indication until we get idle
1874 			 * If there already is an indication stored,
1875 			 * replace it if the new one is a disconnect. The
1876 			 * reasoning is that disconnection takes less time
1877 			 * to process, and once a client decides to
1878 			 * disconnect, we should do that.
1879 			 */
1880 			if (mir->mir_svc_pend_mp) {
1881 				if (type == T_DISCON_IND) {
1882 					RPCLOG(16, "mir_rput_proto: replacing"
1883 					    " held disconnect/ord rel"
1884 					    " indication with disconnect on"
1885 					    " 0x%p\n", (void *)q);
1886 
1887 					freemsg(mir->mir_svc_pend_mp);
1888 					mir->mir_svc_pend_mp = mp;
1889 				} else {
1890 					RPCLOG(16, "mir_rput_proto: already "
1891 					    "held a disconnect/ord rel "
1892 					    "indication. freeing ord rel "
1893 					    "ind on 0x%p\n", (void *)q);
1894 					freemsg(mp);
1895 				}
1896 			} else
1897 				mir->mir_svc_pend_mp = mp;
1898 
1899 			mutex_exit(&mir->mir_mutex);
1900 			return;
1901 
1902 		default:
1903 			/* nfsd handles server-side non-data messages. */
1904 			break;
1905 		}
1906 		break;
1907 
1908 	default:
1909 		break;
1910 	}
1911 
1912 	putnext(q, mp);
1913 }
1914 
1915 /*
1916  * The server-side read queues are used to hold inbound messages while
1917  * outbound flow control is exerted.  When outbound flow control is
1918  * relieved, mir_wsrv qenables the read-side queue.  Read-side queues
1919  * are not enabled by STREAMS and are explicitly noenable'ed in mir_open.
1920  *
1921  * For the server side,  we have two types of messages queued. The first type
1922  * are messages that are ready to be XDR decoded and and then sent to the
1923  * RPC program's dispatch routine. The second type are "raw" messages that
1924  * haven't been processed, i.e. assembled from rpc record fragements into
1925  * full requests. The only time we will see the second type of message
1926  * queued is if we have a memory allocation failure while processing a
1927  * a raw message. The field mir_first_non_processed_mblk will mark the
1928  * first such raw message. So the flow for server side is:
1929  *
1930  *	- send processed queued messages to kRPC until we run out or find
1931  *	  one that needs additional processing because we were short on memory
1932  *	  earlier
1933  *	- process a message that was deferred because of lack of
1934  *	  memory
1935  *	- continue processing messages until the queue empties or we
1936  *	  have to stop because of lack of memory
1937  *	- during each of the above phase, if the queue is empty and
1938  *	  there are no pending messages that were passed to the RPC
1939  *	  layer, send upstream the pending disconnect/ordrel indication if
1940  *	  there is one
1941  *
1942  * The read-side queue is also enabled by a bufcall callback if dupmsg
1943  * fails in mir_rput.
1944  */
1945 static void
1946 mir_rsrv(queue_t *q)
1947 {
1948 	mir_t	*mir;
1949 	mblk_t	*mp;
1950 	mblk_t	*cmp = NULL;
1951 	boolean_t stop_timer = B_FALSE;
1952 
1953 	mir = (mir_t *)q->q_ptr;
1954 	mutex_enter(&mir->mir_mutex);
1955 
1956 	mp = NULL;
1957 	switch (mir->mir_type) {
1958 	case RPC_SERVER:
1959 		if (mir->mir_ref_cnt == 0)
1960 			mir->mir_hold_inbound = 0;
1961 		if (mir->mir_hold_inbound) {
1962 
1963 			ASSERT(cmp == NULL);
1964 			if (q->q_first == NULL) {
1965 
1966 				MIR_CLEAR_INRSRV(mir);
1967 
1968 				if (MIR_SVC_QUIESCED(mir)) {
1969 					cmp = mir->mir_svc_pend_mp;
1970 					mir->mir_svc_pend_mp = NULL;
1971 				}
1972 			}
1973 
1974 			mutex_exit(&mir->mir_mutex);
1975 
1976 			if (cmp != NULL) {
1977 				RPCLOG(16, "mir_rsrv: line %d: sending a held "
1978 				    "disconnect/ord rel indication upstream\n",
1979 				    __LINE__);
1980 				putnext(q, cmp);
1981 			}
1982 
1983 			return;
1984 		}
1985 		while (mp = getq(q)) {
1986 			if (mir->mir_krpc_cell &&
1987 			    (mir->mir_svc_no_more_msgs == 0)) {
1988 				/*
1989 				 * If we were idle, turn off idle timer since
1990 				 * we aren't idle any more.
1991 				 */
1992 				if (mir->mir_ref_cnt++ == 0)
1993 					stop_timer = B_TRUE;
1994 				if (mir_check_len(q,
1995 				    (int32_t)msgdsize(mp), mp))
1996 					return;
1997 				svc_queuereq(q, mp);
1998 			} else {
1999 				/*
2000 				 * Count # of times this happens. Should be
2001 				 * never, but experience shows otherwise.
2002 				 */
2003 				if (mir->mir_krpc_cell == NULL)
2004 					mir_krpc_cell_null++;
2005 				freemsg(mp);
2006 			}
2007 		}
2008 		break;
2009 	case RPC_CLIENT:
2010 		break;
2011 	default:
2012 		RPCLOG(1, "mir_rsrv: unexpected mir_type %d\n", mir->mir_type);
2013 
2014 		if (q->q_first == NULL)
2015 			MIR_CLEAR_INRSRV(mir);
2016 
2017 		mutex_exit(&mir->mir_mutex);
2018 
2019 		return;
2020 	}
2021 
2022 	/*
2023 	 * The timer is stopped after all the messages are processed.
2024 	 * The reason is that stopping the timer releases the mir_mutex
2025 	 * lock temporarily.  This means that the request can be serviced
2026 	 * while we are still processing the message queue.  This is not
2027 	 * good.  So we stop the timer here instead.
2028 	 */
2029 	if (stop_timer)  {
2030 		RPCLOG(16, "mir_rsrv stopping idle timer on 0x%p because ref "
2031 		    "cnt going to non zero\n", (void *)WR(q));
2032 		mir_svc_idle_stop(WR(q), mir);
2033 	}
2034 
2035 	if (q->q_first == NULL) {
2036 
2037 		MIR_CLEAR_INRSRV(mir);
2038 
2039 		ASSERT(cmp == NULL);
2040 		if (mir->mir_type == RPC_SERVER && MIR_SVC_QUIESCED(mir)) {
2041 			cmp = mir->mir_svc_pend_mp;
2042 			mir->mir_svc_pend_mp = NULL;
2043 		}
2044 
2045 		mutex_exit(&mir->mir_mutex);
2046 
2047 		if (cmp != NULL) {
2048 			RPCLOG(16, "mir_rsrv: line %d: sending a held "
2049 			    "disconnect/ord rel indication upstream\n",
2050 			    __LINE__);
2051 			putnext(q, cmp);
2052 		}
2053 
2054 		return;
2055 	}
2056 	mutex_exit(&mir->mir_mutex);
2057 }
2058 
2059 static int mir_svc_policy_fails;
2060 
2061 /*
2062  * Called to send an event code to nfsd/lockd so that it initiates
2063  * connection close.
2064  */
2065 static int
2066 mir_svc_policy_notify(queue_t *q, int event)
2067 {
2068 	mblk_t	*mp;
2069 #ifdef DEBUG
2070 	mir_t *mir = (mir_t *)q->q_ptr;
2071 	ASSERT(MUTEX_NOT_HELD(&mir->mir_mutex));
2072 #endif
2073 	ASSERT(q->q_flag & QREADR);
2074 
2075 	/*
2076 	 * Create an M_DATA message with the event code and pass it to the
2077 	 * Stream head (nfsd or whoever created the stream will consume it).
2078 	 */
2079 	mp = allocb(sizeof (int), BPRI_HI);
2080 
2081 	if (!mp) {
2082 
2083 		mir_svc_policy_fails++;
2084 		RPCLOG(16, "mir_svc_policy_notify: could not allocate event "
2085 		    "%d\n", event);
2086 		return (ENOMEM);
2087 	}
2088 
2089 	U32_TO_BE32(event, mp->b_rptr);
2090 	mp->b_wptr = mp->b_rptr + sizeof (int);
2091 	putnext(q, mp);
2092 	return (0);
2093 }
2094 
2095 /*
2096  * Server side: start the close phase. We want to get this rpcmod slot in an
2097  * idle state before mir_close() is called.
2098  */
2099 static void
2100 mir_svc_start_close(queue_t *wq, mir_t *mir)
2101 {
2102 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
2103 	ASSERT((wq->q_flag & QREADR) == 0);
2104 	ASSERT(mir->mir_type == RPC_SERVER);
2105 
2106 
2107 	/*
2108 	 * Do not accept any more messages.
2109 	 */
2110 	mir->mir_svc_no_more_msgs = 1;
2111 
2112 	/*
2113 	 * Next two statements will make the read service procedure invoke
2114 	 * svc_queuereq() on everything stuck in the streams read queue.
2115 	 * It's not necessary because enabling the write queue will
2116 	 * have the same effect, but why not speed the process along?
2117 	 */
2118 	mir->mir_hold_inbound = 0;
2119 	qenable(RD(wq));
2120 
2121 	/*
2122 	 * Meanwhile force the write service procedure to send the
2123 	 * responses downstream, regardless of flow control.
2124 	 */
2125 	qenable(wq);
2126 }
2127 
2128 /*
2129  * This routine is called directly by KRPC after a request is completed,
2130  * whether a reply was sent or the request was dropped.
2131  */
2132 static void
2133 mir_svc_release(queue_t *wq, mblk_t *mp)
2134 {
2135 	mir_t   *mir = (mir_t *)wq->q_ptr;
2136 	mblk_t	*cmp = NULL;
2137 
2138 	ASSERT((wq->q_flag & QREADR) == 0);
2139 	if (mp)
2140 		freemsg(mp);
2141 
2142 	mutex_enter(&mir->mir_mutex);
2143 
2144 	/*
2145 	 * Start idle processing if this is the last reference.
2146 	 */
2147 	if ((mir->mir_ref_cnt == 1) && (mir->mir_inrservice == 0)) {
2148 		cmp = mir->mir_svc_pend_mp;
2149 		mir->mir_svc_pend_mp = NULL;
2150 	}
2151 
2152 	if (cmp) {
2153 		RPCLOG(16, "mir_svc_release: sending a held "
2154 		    "disconnect/ord rel indication upstream on queue 0x%p\n",
2155 		    (void *)RD(wq));
2156 
2157 		mutex_exit(&mir->mir_mutex);
2158 
2159 		putnext(RD(wq), cmp);
2160 
2161 		mutex_enter(&mir->mir_mutex);
2162 	}
2163 
2164 	/*
2165 	 * Start idle processing if this is the last reference.
2166 	 */
2167 	if (mir->mir_ref_cnt == 1 && mir->mir_inrservice == 0) {
2168 
2169 		RPCLOG(16, "mir_svc_release starting idle timer on 0x%p "
2170 		    "because ref cnt is zero\n", (void *) wq);
2171 
2172 		mir_svc_idle_start(wq, mir);
2173 	}
2174 
2175 	mir->mir_ref_cnt--;
2176 	ASSERT(mir->mir_ref_cnt >= 0);
2177 
2178 	/*
2179 	 * Wake up the thread waiting to close.
2180 	 */
2181 
2182 	if ((mir->mir_ref_cnt == 0) && mir->mir_closing)
2183 		cv_signal(&mir->mir_condvar);
2184 
2185 	mutex_exit(&mir->mir_mutex);
2186 }
2187 
2188 /*
2189  * This routine is called by server-side KRPC when it is ready to
2190  * handle inbound messages on the stream.
2191  */
2192 static void
2193 mir_svc_start(queue_t *wq)
2194 {
2195 	mir_t   *mir = (mir_t *)wq->q_ptr;
2196 
2197 	/*
2198 	 * no longer need to take the mir_mutex because the
2199 	 * mir_setup_complete field has been moved out of
2200 	 * the binary field protected by the mir_mutex.
2201 	 */
2202 
2203 	mir->mir_setup_complete = 1;
2204 	qenable(RD(wq));
2205 }
2206 
2207 /*
2208  * client side wrapper for stopping timer with normal idle timeout.
2209  */
2210 static void
2211 mir_clnt_idle_stop(queue_t *wq, mir_t *mir)
2212 {
2213 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
2214 	ASSERT((wq->q_flag & QREADR) == 0);
2215 	ASSERT(mir->mir_type == RPC_CLIENT);
2216 
2217 	mir_timer_stop(mir);
2218 }
2219 
2220 /*
2221  * client side wrapper for stopping timer with normal idle timeout.
2222  */
2223 static void
2224 mir_clnt_idle_start(queue_t *wq, mir_t *mir)
2225 {
2226 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
2227 	ASSERT((wq->q_flag & QREADR) == 0);
2228 	ASSERT(mir->mir_type == RPC_CLIENT);
2229 
2230 	mir_timer_start(wq, mir, mir->mir_idle_timeout);
2231 }
2232 
2233 /*
2234  * client side only. Forces rpcmod to stop sending T_ORDREL_REQs on
2235  * end-points that aren't connected.
2236  */
2237 static void
2238 mir_clnt_idle_do_stop(queue_t *wq)
2239 {
2240 	mir_t   *mir = (mir_t *)wq->q_ptr;
2241 
2242 	RPCLOG(1, "mir_clnt_idle_do_stop: wq 0x%p\n", (void *)wq);
2243 	ASSERT(MUTEX_NOT_HELD(&mir->mir_mutex));
2244 	mutex_enter(&mir->mir_mutex);
2245 	mir_clnt_idle_stop(wq, mir);
2246 	mutex_exit(&mir->mir_mutex);
2247 }
2248 
2249 /*
2250  * Timer handler.  It handles idle timeout and memory shortage problem.
2251  */
2252 static void
2253 mir_timer(void *arg)
2254 {
2255 	queue_t *wq = (queue_t *)arg;
2256 	mir_t *mir = (mir_t *)wq->q_ptr;
2257 	boolean_t notify;
2258 
2259 	mutex_enter(&mir->mir_mutex);
2260 
2261 	/*
2262 	 * mir_timer_call is set only when either mir_timer_[start|stop]
2263 	 * is progressing.  And mir_timer() can only be run while they
2264 	 * are progressing if the timer is being stopped.  So just
2265 	 * return.
2266 	 */
2267 	if (mir->mir_timer_call) {
2268 		mutex_exit(&mir->mir_mutex);
2269 		return;
2270 	}
2271 	mir->mir_timer_id = 0;
2272 
2273 	switch (mir->mir_type) {
2274 	case RPC_CLIENT:
2275 
2276 		/*
2277 		 * For clients, the timer fires at clnt_idle_timeout
2278 		 * intervals.  If the activity marker (mir_clntreq) is
2279 		 * zero, then the stream has been idle since the last
2280 		 * timer event and we notify KRPC.  If mir_clntreq is
2281 		 * non-zero, then the stream is active and we just
2282 		 * restart the timer for another interval.  mir_clntreq
2283 		 * is set to 1 in mir_wput for every request passed
2284 		 * downstream.
2285 		 *
2286 		 * If this was a memory shortage timer reset the idle
2287 		 * timeout regardless; the mir_clntreq will not be a
2288 		 * valid indicator.
2289 		 *
2290 		 * The timer is initially started in mir_wput during
2291 		 * RPC_CLIENT ioctl processing.
2292 		 *
2293 		 * The timer interval can be changed for individual
2294 		 * streams with the ND variable "mir_idle_timeout".
2295 		 */
2296 		if (mir->mir_clntreq > 0 && mir->mir_use_timestamp +
2297 		    MSEC_TO_TICK(mir->mir_idle_timeout) - lbolt >= 0) {
2298 			clock_t tout;
2299 
2300 			tout = mir->mir_idle_timeout -
2301 			    TICK_TO_MSEC(lbolt - mir->mir_use_timestamp);
2302 			if (tout < 0)
2303 				tout = 1000;
2304 #if 0
2305 			printf("mir_timer[%d < %d + %d]: reset client timer "
2306 			    "to %d (ms)\n", TICK_TO_MSEC(lbolt),
2307 			    TICK_TO_MSEC(mir->mir_use_timestamp),
2308 			    mir->mir_idle_timeout, tout);
2309 #endif
2310 			mir->mir_clntreq = 0;
2311 			mir_timer_start(wq, mir, tout);
2312 			mutex_exit(&mir->mir_mutex);
2313 			return;
2314 		}
2315 #if 0
2316 printf("mir_timer[%d]: doing client timeout\n", lbolt / hz);
2317 #endif
2318 		/*
2319 		 * We are disconnecting, but not necessarily
2320 		 * closing. By not closing, we will fail to
2321 		 * pick up a possibly changed global timeout value,
2322 		 * unless we store it now.
2323 		 */
2324 		mir->mir_idle_timeout = clnt_idle_timeout;
2325 		mir_clnt_idle_start(wq, mir);
2326 
2327 		mutex_exit(&mir->mir_mutex);
2328 		/*
2329 		 * We pass T_ORDREL_REQ as an integer value
2330 		 * to KRPC as the indication that the stream
2331 		 * is idle.  This is not a T_ORDREL_REQ message,
2332 		 * it is just a convenient value since we call
2333 		 * the same KRPC routine for T_ORDREL_INDs and
2334 		 * T_DISCON_INDs.
2335 		 */
2336 		clnt_dispatch_notifyall(wq, T_ORDREL_REQ, 0);
2337 		return;
2338 
2339 	case RPC_SERVER:
2340 
2341 		/*
2342 		 * For servers, the timer is only running when the stream
2343 		 * is really idle or memory is short.  The timer is started
2344 		 * by mir_wput when mir_type is set to RPC_SERVER and
2345 		 * by mir_svc_idle_start whenever the stream goes idle
2346 		 * (mir_ref_cnt == 0).  The timer is cancelled in
2347 		 * mir_rput whenever a new inbound request is passed to KRPC
2348 		 * and the stream was previously idle.
2349 		 *
2350 		 * The timer interval can be changed for individual
2351 		 * streams with the ND variable "mir_idle_timeout".
2352 		 *
2353 		 * If the stream is not idle do nothing.
2354 		 */
2355 		if (!MIR_SVC_QUIESCED(mir)) {
2356 			mutex_exit(&mir->mir_mutex);
2357 			return;
2358 		}
2359 
2360 		notify = !mir->mir_inrservice;
2361 		mutex_exit(&mir->mir_mutex);
2362 
2363 		/*
2364 		 * If there is no packet queued up in read queue, the stream
2365 		 * is really idle so notify nfsd to close it.
2366 		 */
2367 		if (notify) {
2368 			RPCLOG(16, "mir_timer: telling stream head listener "
2369 			    "to close stream (0x%p)\n", (void *) RD(wq));
2370 			(void) mir_svc_policy_notify(RD(wq), 1);
2371 		}
2372 		return;
2373 	default:
2374 		RPCLOG(1, "mir_timer: unexpected mir_type %d\n",
2375 		    mir->mir_type);
2376 		mutex_exit(&mir->mir_mutex);
2377 		return;
2378 	}
2379 }
2380 
2381 /*
2382  * Called by the RPC package to send either a call or a return, or a
2383  * transport connection request.  Adds the record marking header.
2384  */
2385 static void
2386 mir_wput(queue_t *q, mblk_t *mp)
2387 {
2388 	uint_t	frag_header;
2389 	mir_t	*mir = (mir_t *)q->q_ptr;
2390 	uchar_t	*rptr = mp->b_rptr;
2391 
2392 	if (!mir) {
2393 		freemsg(mp);
2394 		return;
2395 	}
2396 
2397 	if (mp->b_datap->db_type != M_DATA) {
2398 		mir_wput_other(q, mp);
2399 		return;
2400 	}
2401 
2402 	if (mir->mir_ordrel_pending == 1) {
2403 		freemsg(mp);
2404 		RPCLOG(16, "mir_wput wq 0x%p: got data after T_ORDREL_REQ\n",
2405 		    (void *)q);
2406 		return;
2407 	}
2408 
2409 	frag_header = (uint_t)DLEN(mp);
2410 	frag_header |= MIR_LASTFRAG;
2411 
2412 	/* Stick in the 4 byte record marking header. */
2413 	if ((rptr - mp->b_datap->db_base) < sizeof (uint32_t) ||
2414 	    !IS_P2ALIGNED(mp->b_rptr, sizeof (uint32_t))) {
2415 		/*
2416 		 * Since we know that M_DATA messages are created exclusively
2417 		 * by KRPC, we expect that KRPC will leave room for our header
2418 		 * and 4 byte align which is normal for XDR.
2419 		 * If KRPC (or someone else) does not cooperate, then we
2420 		 * just throw away the message.
2421 		 */
2422 		RPCLOG(1, "mir_wput: KRPC did not leave space for record "
2423 		    "fragment header (%d bytes left)\n",
2424 		    (int)(rptr - mp->b_datap->db_base));
2425 		freemsg(mp);
2426 		return;
2427 	}
2428 	rptr -= sizeof (uint32_t);
2429 	*(uint32_t *)rptr = htonl(frag_header);
2430 	mp->b_rptr = rptr;
2431 
2432 	mutex_enter(&mir->mir_mutex);
2433 	if (mir->mir_type == RPC_CLIENT) {
2434 		/*
2435 		 * For the client, set mir_clntreq to indicate that the
2436 		 * connection is active.
2437 		 */
2438 		mir->mir_clntreq = 1;
2439 		mir->mir_use_timestamp = lbolt;
2440 	}
2441 
2442 	/*
2443 	 * If we haven't already queued some data and the downstream module
2444 	 * can accept more data, send it on, otherwise we queue the message
2445 	 * and take other actions depending on mir_type.
2446 	 */
2447 	if (!mir->mir_inwservice && MIR_WCANPUTNEXT(mir, q)) {
2448 		mutex_exit(&mir->mir_mutex);
2449 
2450 		/*
2451 		 * Now we pass the RPC message downstream.
2452 		 */
2453 		putnext(q, mp);
2454 		return;
2455 	}
2456 
2457 	switch (mir->mir_type) {
2458 	case RPC_CLIENT:
2459 		/*
2460 		 * Check for a previous duplicate request on the
2461 		 * queue.  If there is one, then we throw away
2462 		 * the current message and let the previous one
2463 		 * go through.  If we can't find a duplicate, then
2464 		 * send this one.  This tap dance is an effort
2465 		 * to reduce traffic and processing requirements
2466 		 * under load conditions.
2467 		 */
2468 		if (mir_clnt_dup_request(q, mp)) {
2469 			mutex_exit(&mir->mir_mutex);
2470 			freemsg(mp);
2471 			return;
2472 		}
2473 		break;
2474 	case RPC_SERVER:
2475 		/*
2476 		 * Set mir_hold_inbound so that new inbound RPC
2477 		 * messages will be held until the client catches
2478 		 * up on the earlier replies.  This flag is cleared
2479 		 * in mir_wsrv after flow control is relieved;
2480 		 * the read-side queue is also enabled at that time.
2481 		 */
2482 		mir->mir_hold_inbound = 1;
2483 		break;
2484 	default:
2485 		RPCLOG(1, "mir_wput: unexpected mir_type %d\n", mir->mir_type);
2486 		break;
2487 	}
2488 	mir->mir_inwservice = 1;
2489 	(void) putq(q, mp);
2490 	mutex_exit(&mir->mir_mutex);
2491 }
2492 
2493 static void
2494 mir_wput_other(queue_t *q, mblk_t *mp)
2495 {
2496 	mir_t	*mir = (mir_t *)q->q_ptr;
2497 	struct iocblk	*iocp;
2498 	uchar_t	*rptr = mp->b_rptr;
2499 	bool_t	flush_in_svc = FALSE;
2500 
2501 	ASSERT(MUTEX_NOT_HELD(&mir->mir_mutex));
2502 	switch (mp->b_datap->db_type) {
2503 	case M_IOCTL:
2504 		iocp = (struct iocblk *)rptr;
2505 		switch (iocp->ioc_cmd) {
2506 		case RPC_CLIENT:
2507 			mutex_enter(&mir->mir_mutex);
2508 			if (mir->mir_type != 0 &&
2509 			    mir->mir_type != iocp->ioc_cmd) {
2510 ioc_eperm:
2511 				mutex_exit(&mir->mir_mutex);
2512 				iocp->ioc_error = EPERM;
2513 				iocp->ioc_count = 0;
2514 				mp->b_datap->db_type = M_IOCACK;
2515 				qreply(q, mp);
2516 				return;
2517 			}
2518 
2519 			mir->mir_type = iocp->ioc_cmd;
2520 
2521 			/*
2522 			 * Clear mir_hold_inbound which was set to 1 by
2523 			 * mir_open.  This flag is not used on client
2524 			 * streams.
2525 			 */
2526 			mir->mir_hold_inbound = 0;
2527 			mir->mir_max_msg_sizep = &clnt_max_msg_size;
2528 
2529 			/*
2530 			 * Start the idle timer.  See mir_timer() for more
2531 			 * information on how client timers work.
2532 			 */
2533 			mir->mir_idle_timeout = clnt_idle_timeout;
2534 			mir_clnt_idle_start(q, mir);
2535 			mutex_exit(&mir->mir_mutex);
2536 
2537 			mp->b_datap->db_type = M_IOCACK;
2538 			qreply(q, mp);
2539 			return;
2540 		case RPC_SERVER:
2541 			mutex_enter(&mir->mir_mutex);
2542 			if (mir->mir_type != 0 &&
2543 			    mir->mir_type != iocp->ioc_cmd)
2544 				goto ioc_eperm;
2545 
2546 			/*
2547 			 * We don't clear mir_hold_inbound here because
2548 			 * mir_hold_inbound is used in the flow control
2549 			 * model. If we cleared it here, then we'd commit
2550 			 * a small violation to the model where the transport
2551 			 * might immediately block downstream flow.
2552 			 */
2553 
2554 			mir->mir_type = iocp->ioc_cmd;
2555 			mir->mir_max_msg_sizep = &svc_max_msg_size;
2556 
2557 			/*
2558 			 * Start the idle timer.  See mir_timer() for more
2559 			 * information on how server timers work.
2560 			 *
2561 			 * Note that it is important to start the idle timer
2562 			 * here so that connections time out even if we
2563 			 * never receive any data on them.
2564 			 */
2565 			mir->mir_idle_timeout = svc_idle_timeout;
2566 			RPCLOG(16, "mir_wput_other starting idle timer on 0x%p "
2567 			    "because we got RPC_SERVER ioctl\n", (void *)q);
2568 			mir_svc_idle_start(q, mir);
2569 			mutex_exit(&mir->mir_mutex);
2570 
2571 			mp->b_datap->db_type = M_IOCACK;
2572 			qreply(q, mp);
2573 			return;
2574 		default:
2575 			break;
2576 		}
2577 		break;
2578 
2579 	case M_PROTO:
2580 		if (mir->mir_type == RPC_CLIENT) {
2581 			/*
2582 			 * We are likely being called from the context of a
2583 			 * service procedure. So we need to enqueue. However
2584 			 * enqueing may put our message behind data messages.
2585 			 * So flush the data first.
2586 			 */
2587 			flush_in_svc = TRUE;
2588 		}
2589 		if ((mp->b_wptr - rptr) < sizeof (uint32_t) ||
2590 		    !IS_P2ALIGNED(rptr, sizeof (uint32_t)))
2591 			break;
2592 
2593 		switch (((union T_primitives *)rptr)->type) {
2594 		case T_DATA_REQ:
2595 			/* Don't pass T_DATA_REQ messages downstream. */
2596 			freemsg(mp);
2597 			return;
2598 		case T_ORDREL_REQ:
2599 			RPCLOG(8, "mir_wput_other wq 0x%p: got T_ORDREL_REQ\n",
2600 			    (void *)q);
2601 			mutex_enter(&mir->mir_mutex);
2602 			if (mir->mir_type != RPC_SERVER) {
2603 				/*
2604 				 * We are likely being called from
2605 				 * clnt_dispatch_notifyall(). Sending
2606 				 * a T_ORDREL_REQ will result in
2607 				 * a some kind of _IND message being sent,
2608 				 * will be another call to
2609 				 * clnt_dispatch_notifyall(). To keep the stack
2610 				 * lean, queue this message.
2611 				 */
2612 				mir->mir_inwservice = 1;
2613 				(void) putq(q, mp);
2614 				mutex_exit(&mir->mir_mutex);
2615 				return;
2616 			}
2617 
2618 			/*
2619 			 * Mark the structure such that we don't accept any
2620 			 * more requests from client. We could defer this
2621 			 * until we actually send the orderly release
2622 			 * request downstream, but all that does is delay
2623 			 * the closing of this stream.
2624 			 */
2625 			RPCLOG(16, "mir_wput_other wq 0x%p: got T_ORDREL_REQ "
2626 			    " so calling mir_svc_start_close\n", (void *)q);
2627 
2628 			mir_svc_start_close(q, mir);
2629 
2630 			/*
2631 			 * If we have sent down a T_ORDREL_REQ, don't send
2632 			 * any more.
2633 			 */
2634 			if (mir->mir_ordrel_pending) {
2635 				freemsg(mp);
2636 				mutex_exit(&mir->mir_mutex);
2637 				return;
2638 			}
2639 
2640 			/*
2641 			 * If the stream is not idle, then we hold the
2642 			 * orderly release until it becomes idle.  This
2643 			 * ensures that KRPC will be able to reply to
2644 			 * all requests that we have passed to it.
2645 			 *
2646 			 * We also queue the request if there is data already
2647 			 * queued, because we cannot allow the T_ORDREL_REQ
2648 			 * to go before data. When we had a separate reply
2649 			 * count, this was not a problem, because the
2650 			 * reply count was reconciled when mir_wsrv()
2651 			 * completed.
2652 			 */
2653 			if (!MIR_SVC_QUIESCED(mir) ||
2654 			    mir->mir_inwservice == 1) {
2655 				mir->mir_inwservice = 1;
2656 				(void) putq(q, mp);
2657 
2658 				RPCLOG(16, "mir_wput_other: queuing "
2659 				    "T_ORDREL_REQ on 0x%p\n", (void *)q);
2660 
2661 				mutex_exit(&mir->mir_mutex);
2662 				return;
2663 			}
2664 
2665 			/*
2666 			 * Mark the structure so that we know we sent
2667 			 * an orderly release request, and reset the idle timer.
2668 			 */
2669 			mir->mir_ordrel_pending = 1;
2670 
2671 			RPCLOG(16, "mir_wput_other: calling mir_svc_idle_start"
2672 			    " on 0x%p because we got T_ORDREL_REQ\n",
2673 			    (void *)q);
2674 
2675 			mir_svc_idle_start(q, mir);
2676 			mutex_exit(&mir->mir_mutex);
2677 
2678 			/*
2679 			 * When we break, we will putnext the T_ORDREL_REQ.
2680 			 */
2681 			break;
2682 
2683 		case T_CONN_REQ:
2684 			mutex_enter(&mir->mir_mutex);
2685 			if (mir->mir_head_mp != NULL) {
2686 				freemsg(mir->mir_head_mp);
2687 				mir->mir_head_mp = NULL;
2688 				mir->mir_tail_mp = NULL;
2689 			}
2690 			mir->mir_frag_len = -(int32_t)sizeof (uint32_t);
2691 			/*
2692 			 * Restart timer in case mir_clnt_idle_do_stop() was
2693 			 * called.
2694 			 */
2695 			mir->mir_idle_timeout = clnt_idle_timeout;
2696 			mir_clnt_idle_stop(q, mir);
2697 			mir_clnt_idle_start(q, mir);
2698 			mutex_exit(&mir->mir_mutex);
2699 			break;
2700 
2701 		default:
2702 			/*
2703 			 * T_DISCON_REQ is one of the interesting default
2704 			 * cases here. Ideally, an M_FLUSH is done before
2705 			 * T_DISCON_REQ is done. However, that is somewhat
2706 			 * cumbersome for clnt_cots.c to do. So we queue
2707 			 * T_DISCON_REQ, and let the service procedure
2708 			 * flush all M_DATA.
2709 			 */
2710 			break;
2711 		}
2712 		/* fallthru */;
2713 	default:
2714 		if (mp->b_datap->db_type >= QPCTL) {
2715 			if (mp->b_datap->db_type == M_FLUSH) {
2716 				if (mir->mir_type == RPC_CLIENT &&
2717 				    *mp->b_rptr & FLUSHW) {
2718 					RPCLOG(32, "mir_wput_other: flushing "
2719 					    "wq 0x%p\n", (void *)q);
2720 					if (*mp->b_rptr & FLUSHBAND) {
2721 						flushband(q, *(mp->b_rptr + 1),
2722 						    FLUSHDATA);
2723 					} else {
2724 						flushq(q, FLUSHDATA);
2725 					}
2726 				} else {
2727 					RPCLOG(32, "mir_wput_other: ignoring "
2728 					    "M_FLUSH on wq 0x%p\n", (void *)q);
2729 				}
2730 			}
2731 			break;
2732 		}
2733 
2734 		mutex_enter(&mir->mir_mutex);
2735 		if (mir->mir_inwservice == 0 && MIR_WCANPUTNEXT(mir, q)) {
2736 			mutex_exit(&mir->mir_mutex);
2737 			break;
2738 		}
2739 		mir->mir_inwservice = 1;
2740 		mir->mir_inwflushdata = flush_in_svc;
2741 		(void) putq(q, mp);
2742 		mutex_exit(&mir->mir_mutex);
2743 		qenable(q);
2744 
2745 		return;
2746 	}
2747 	putnext(q, mp);
2748 }
2749 
2750 static void
2751 mir_wsrv(queue_t *q)
2752 {
2753 	mblk_t	*mp;
2754 	mir_t	*mir;
2755 	bool_t flushdata;
2756 
2757 	mir = (mir_t *)q->q_ptr;
2758 	mutex_enter(&mir->mir_mutex);
2759 
2760 	flushdata = mir->mir_inwflushdata;
2761 	mir->mir_inwflushdata = 0;
2762 
2763 	while (mp = getq(q)) {
2764 		if (mp->b_datap->db_type == M_DATA) {
2765 			/*
2766 			 * Do not send any more data if we have sent
2767 			 * a T_ORDREL_REQ.
2768 			 */
2769 			if (flushdata || mir->mir_ordrel_pending == 1) {
2770 				freemsg(mp);
2771 				continue;
2772 			}
2773 
2774 			/*
2775 			 * Make sure that the stream can really handle more
2776 			 * data.
2777 			 */
2778 			if (!MIR_WCANPUTNEXT(mir, q)) {
2779 				(void) putbq(q, mp);
2780 				mutex_exit(&mir->mir_mutex);
2781 				return;
2782 			}
2783 
2784 			/*
2785 			 * Now we pass the RPC message downstream.
2786 			 */
2787 			mutex_exit(&mir->mir_mutex);
2788 			putnext(q, mp);
2789 			mutex_enter(&mir->mir_mutex);
2790 			continue;
2791 		}
2792 
2793 		/*
2794 		 * This is not an RPC message, pass it downstream
2795 		 * (ignoring flow control) if the server side is not sending a
2796 		 * T_ORDREL_REQ downstream.
2797 		 */
2798 		if (mir->mir_type != RPC_SERVER ||
2799 		    ((union T_primitives *)mp->b_rptr)->type !=
2800 		    T_ORDREL_REQ) {
2801 			mutex_exit(&mir->mir_mutex);
2802 			putnext(q, mp);
2803 			mutex_enter(&mir->mir_mutex);
2804 			continue;
2805 		}
2806 
2807 		if (mir->mir_ordrel_pending == 1) {
2808 			/*
2809 			 * Don't send two T_ORDRELs
2810 			 */
2811 			freemsg(mp);
2812 			continue;
2813 		}
2814 
2815 		/*
2816 		 * Mark the structure so that we know we sent an orderly
2817 		 * release request.  We will check to see slot is idle at the
2818 		 * end of this routine, and if so, reset the idle timer to
2819 		 * handle orderly release timeouts.
2820 		 */
2821 		mir->mir_ordrel_pending = 1;
2822 		RPCLOG(16, "mir_wsrv: sending ordrel req on q 0x%p\n",
2823 		    (void *)q);
2824 		/*
2825 		 * Send the orderly release downstream. If there are other
2826 		 * pending replies we won't be able to send them.  However,
2827 		 * the only reason we should send the orderly release is if
2828 		 * we were idle, or if an unusual event occurred.
2829 		 */
2830 		mutex_exit(&mir->mir_mutex);
2831 		putnext(q, mp);
2832 		mutex_enter(&mir->mir_mutex);
2833 	}
2834 
2835 	if (q->q_first == NULL)
2836 		/*
2837 		 * If we call mir_svc_idle_start() below, then
2838 		 * clearing mir_inwservice here will also result in
2839 		 * any thread waiting in mir_close() to be signaled.
2840 		 */
2841 		mir->mir_inwservice = 0;
2842 
2843 	if (mir->mir_type != RPC_SERVER) {
2844 		mutex_exit(&mir->mir_mutex);
2845 		return;
2846 	}
2847 
2848 	/*
2849 	 * If idle we call mir_svc_idle_start to start the timer (or wakeup
2850 	 * a close). Also make sure not to start the idle timer on the
2851 	 * listener stream. This can cause nfsd to send an orderly release
2852 	 * command on the listener stream.
2853 	 */
2854 	if (MIR_SVC_QUIESCED(mir) && !(mir->mir_listen_stream)) {
2855 		RPCLOG(16, "mir_wsrv: calling mir_svc_idle_start on 0x%p "
2856 		    "because mir slot is idle\n", (void *)q);
2857 		mir_svc_idle_start(q, mir);
2858 	}
2859 
2860 	/*
2861 	 * If outbound flow control has been relieved, then allow new
2862 	 * inbound requests to be processed.
2863 	 */
2864 	if (mir->mir_hold_inbound) {
2865 		mir->mir_hold_inbound = 0;
2866 		qenable(RD(q));
2867 	}
2868 	mutex_exit(&mir->mir_mutex);
2869 }
2870 
2871 static void
2872 mir_disconnect(queue_t *q, mir_t *mir)
2873 {
2874 	ASSERT(MUTEX_HELD(&mir->mir_mutex));
2875 
2876 	switch (mir->mir_type) {
2877 	case RPC_CLIENT:
2878 		/*
2879 		 * We are disconnecting, but not necessarily
2880 		 * closing. By not closing, we will fail to
2881 		 * pick up a possibly changed global timeout value,
2882 		 * unless we store it now.
2883 		 */
2884 		mir->mir_idle_timeout = clnt_idle_timeout;
2885 		mir_clnt_idle_start(WR(q), mir);
2886 		mutex_exit(&mir->mir_mutex);
2887 
2888 		/*
2889 		 * T_DISCON_REQ is passed to KRPC as an integer value
2890 		 * (this is not a TPI message).  It is used as a
2891 		 * convenient value to indicate a sanity check
2892 		 * failure -- the same KRPC routine is also called
2893 		 * for T_DISCON_INDs and T_ORDREL_INDs.
2894 		 */
2895 		clnt_dispatch_notifyall(WR(q), T_DISCON_REQ, 0);
2896 		break;
2897 
2898 	case RPC_SERVER:
2899 		mir->mir_svc_no_more_msgs = 1;
2900 		mir_svc_idle_stop(WR(q), mir);
2901 		mutex_exit(&mir->mir_mutex);
2902 		RPCLOG(16, "mir_disconnect: telling "
2903 		    "stream head listener to disconnect stream "
2904 		    "(0x%p)\n", (void *) q);
2905 		(void) mir_svc_policy_notify(q, 2);
2906 		break;
2907 
2908 	default:
2909 		mutex_exit(&mir->mir_mutex);
2910 		break;
2911 	}
2912 }
2913 
2914 /*
2915  * Sanity check the message length, and if it's too large, shutdown the
2916  * connection.  Returns 1 if the connection is shutdown; 0 otherwise.
2917  */
2918 static int
2919 mir_check_len(queue_t *q, int32_t frag_len, mblk_t *head_mp)
2920 {
2921 	mir_t *mir = q->q_ptr;
2922 	uint_t maxsize = 0;
2923 
2924 	if (mir->mir_max_msg_sizep != NULL)
2925 		maxsize = *mir->mir_max_msg_sizep;
2926 
2927 	if (maxsize == 0 || frag_len <= (int)maxsize)
2928 		return (0);
2929 
2930 	freemsg(head_mp);
2931 	mir->mir_head_mp = NULL;
2932 	mir->mir_tail_mp = NULL;
2933 	mir->mir_frag_header = 0;
2934 	mir->mir_frag_len = -(int32_t)sizeof (uint32_t);
2935 	if (mir->mir_type != RPC_SERVER || mir->mir_setup_complete) {
2936 		cmn_err(CE_NOTE,
2937 		    "KRPC: record fragment from %s of size(%d) exceeds "
2938 		    "maximum (%u). Disconnecting",
2939 		    (mir->mir_type == RPC_CLIENT) ? "server" :
2940 		    (mir->mir_type == RPC_SERVER) ? "client" :
2941 		    "test tool", frag_len, maxsize);
2942 	}
2943 
2944 	mir_disconnect(q, mir);
2945 	return (1);
2946 }
2947