xref: /netbsd-src/sys/dev/raidframe/rf_driver.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: rf_driver.c,v 1.37 2000/06/04 02:05:13 oster Exp $	*/
2 /*-
3  * Copyright (c) 1999 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Greg Oster
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *        This product includes software developed by the NetBSD
20  *        Foundation, Inc. and its contributors.
21  * 4. Neither the name of The NetBSD Foundation nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 1995 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Author: Mark Holland, Khalil Amiri, Claudson Bornstein, William V. Courtright II,
43  *         Robby Findler, Daniel Stodolsky, Rachad Youssef, Jim Zelenka
44  *
45  * Permission to use, copy, modify and distribute this software and
46  * its documentation is hereby granted, provided that both the copyright
47  * notice and this permission notice appear in all copies of the
48  * software, derivative works or modified versions, and any portions
49  * thereof, and that both notices appear in supporting documentation.
50  *
51  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
52  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
53  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
54  *
55  * Carnegie Mellon requests users of this software to return to
56  *
57  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
58  *  School of Computer Science
59  *  Carnegie Mellon University
60  *  Pittsburgh PA 15213-3890
61  *
62  * any improvements or extensions that they make and grant Carnegie the
63  * rights to redistribute these changes.
64  */
65 
66 /******************************************************************************
67  *
68  * rf_driver.c -- main setup, teardown, and access routines for the RAID driver
69  *
70  * all routines are prefixed with rf_ (raidframe), to avoid conficts.
71  *
72  ******************************************************************************/
73 
74 
75 #include <sys/types.h>
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/ioctl.h>
79 #include <sys/fcntl.h>
80 #include <sys/vnode.h>
81 
82 
83 #include "rf_archs.h"
84 #include "rf_threadstuff.h"
85 
86 #include <sys/errno.h>
87 
88 #include "rf_raid.h"
89 #include "rf_dag.h"
90 #include "rf_aselect.h"
91 #include "rf_diskqueue.h"
92 #include "rf_parityscan.h"
93 #include "rf_alloclist.h"
94 #include "rf_dagutils.h"
95 #include "rf_utils.h"
96 #include "rf_etimer.h"
97 #include "rf_acctrace.h"
98 #include "rf_configure.h"
99 #include "rf_general.h"
100 #include "rf_desc.h"
101 #include "rf_states.h"
102 #include "rf_freelist.h"
103 #include "rf_decluster.h"
104 #include "rf_map.h"
105 #include "rf_revent.h"
106 #include "rf_callback.h"
107 #include "rf_engine.h"
108 #include "rf_memchunk.h"
109 #include "rf_mcpair.h"
110 #include "rf_nwayxor.h"
111 #include "rf_debugprint.h"
112 #include "rf_copyback.h"
113 #include "rf_driver.h"
114 #include "rf_options.h"
115 #include "rf_shutdown.h"
116 #include "rf_kintf.h"
117 
118 #include <sys/buf.h>
119 
120 /* rad == RF_RaidAccessDesc_t */
121 static RF_FreeList_t *rf_rad_freelist;
122 #define RF_MAX_FREE_RAD 128
123 #define RF_RAD_INC       16
124 #define RF_RAD_INITIAL   32
125 
126 /* debug variables */
127 char    rf_panicbuf[2048];	/* a buffer to hold an error msg when we panic */
128 
129 /* main configuration routines */
130 static int raidframe_booted = 0;
131 
132 static void rf_ConfigureDebug(RF_Config_t * cfgPtr);
133 static void set_debug_option(char *name, long val);
134 static void rf_UnconfigureArray(void);
135 static int init_rad(RF_RaidAccessDesc_t *);
136 static void clean_rad(RF_RaidAccessDesc_t *);
137 static void rf_ShutdownRDFreeList(void *);
138 static int rf_ConfigureRDFreeList(RF_ShutdownList_t **);
139 
140 RF_DECLARE_MUTEX(rf_printf_mutex)	/* debug only:  avoids interleaved
141 					 * printfs by different stripes */
142 
143 #define SIGNAL_QUIESCENT_COND(_raid_)  wakeup(&((_raid_)->accesses_suspended))
144 #define WAIT_FOR_QUIESCENCE(_raid_) \
145 	tsleep(&((_raid_)->accesses_suspended),PRIBIO,"raidframe quiesce", 0);
146 
147 #define IO_BUF_ERR(bp, err) { \
148 	bp->b_flags |= B_ERROR; \
149 	bp->b_resid = bp->b_bcount; \
150 	bp->b_error = err; \
151 	biodone(bp); \
152 }
153 
154 static int configureCount = 0;	/* number of active configurations */
155 static int isconfigged = 0;	/* is basic raidframe (non per-array)
156 				 * stuff configged */
157 RF_DECLARE_STATIC_MUTEX(configureMutex)	/* used to lock the configuration
158 					 * stuff */
159 static RF_ShutdownList_t *globalShutdown;	/* non array-specific
160 						 * stuff */
161 
162 static int rf_ConfigureRDFreeList(RF_ShutdownList_t ** listp);
163 
164 /* called at system boot time */
165 int
166 rf_BootRaidframe()
167 {
168 	int     rc;
169 
170 	if (raidframe_booted)
171 		return (EBUSY);
172 	raidframe_booted = 1;
173 
174 	rc = rf_mutex_init(&configureMutex);
175 	if (rc) {
176 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
177 		    __LINE__, rc);
178 		RF_PANIC();
179 	}
180 	configureCount = 0;
181 	isconfigged = 0;
182 	globalShutdown = NULL;
183 	return (0);
184 }
185 /*
186  * This function is really just for debugging user-level stuff: it
187  * frees up all memory, other RAIDframe resources which might otherwise
188  * be kept around. This is used with systems like "sentinel" to detect
189  * memory leaks.
190  */
191 int
192 rf_UnbootRaidframe()
193 {
194 	int     rc;
195 
196 	RF_LOCK_MUTEX(configureMutex);
197 	if (configureCount) {
198 		RF_UNLOCK_MUTEX(configureMutex);
199 		return (EBUSY);
200 	}
201 	raidframe_booted = 0;
202 	RF_UNLOCK_MUTEX(configureMutex);
203 	rc = rf_mutex_destroy(&configureMutex);
204 	if (rc) {
205 		RF_ERRORMSG3("Unable to destroy mutex file %s line %d rc=%d\n", __FILE__,
206 		    __LINE__, rc);
207 		RF_PANIC();
208 	}
209 	return (0);
210 }
211 /*
212  * Called whenever an array is shutdown
213  */
214 static void
215 rf_UnconfigureArray()
216 {
217 	int     rc;
218 
219 	RF_LOCK_MUTEX(configureMutex);
220 	if (--configureCount == 0) {	/* if no active configurations, shut
221 					 * everything down */
222 		isconfigged = 0;
223 
224 		rc = rf_ShutdownList(&globalShutdown);
225 		if (rc) {
226 			RF_ERRORMSG1("RAIDFRAME: unable to do global shutdown, rc=%d\n", rc);
227 		}
228 
229 		/*
230 	         * We must wait until now, because the AllocList module
231 	         * uses the DebugMem module.
232 	         */
233 		if (rf_memDebug)
234 			rf_print_unfreed();
235 	}
236 	RF_UNLOCK_MUTEX(configureMutex);
237 }
238 
239 /*
240  * Called to shut down an array.
241  */
242 int
243 rf_Shutdown(raidPtr)
244 	RF_Raid_t *raidPtr;
245 {
246 
247 	if (!raidPtr->valid) {
248 		RF_ERRORMSG("Attempt to shut down unconfigured RAIDframe driver.  Aborting shutdown\n");
249 		return (EINVAL);
250 	}
251 	/*
252          * wait for outstanding IOs to land
253          * As described in rf_raid.h, we use the rad_freelist lock
254          * to protect the per-array info about outstanding descs
255          * since we need to do freelist locking anyway, and this
256          * cuts down on the amount of serialization we've got going
257          * on.
258          */
259 	RF_FREELIST_DO_LOCK(rf_rad_freelist);
260 	if (raidPtr->waitShutdown) {
261 		RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
262 		return (EBUSY);
263 	}
264 	raidPtr->waitShutdown = 1;
265 	while (raidPtr->nAccOutstanding) {
266 		RF_WAIT_COND(raidPtr->outstandingCond, RF_FREELIST_MUTEX_OF(rf_rad_freelist));
267 	}
268 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
269 
270 	/* Wait for any parity re-writes to stop... */
271 	while (raidPtr->parity_rewrite_in_progress) {
272 		printf("Waiting for parity re-write to exit...\n");
273 		tsleep(&raidPtr->parity_rewrite_in_progress, PRIBIO,
274 		       "rfprwshutdown", 0);
275 	}
276 
277 	raidPtr->valid = 0;
278 
279 	rf_update_component_labels(raidPtr, RF_FINAL_COMPONENT_UPDATE);
280 
281 	rf_UnconfigureVnodes(raidPtr);
282 
283 	rf_ShutdownList(&raidPtr->shutdownList);
284 
285 	rf_UnconfigureArray();
286 
287 	return (0);
288 }
289 
290 
291 #define DO_INIT_CONFIGURE(f) { \
292 	rc = f (&globalShutdown); \
293 	if (rc) { \
294 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
295 		rf_ShutdownList(&globalShutdown); \
296 		configureCount--; \
297 		RF_UNLOCK_MUTEX(configureMutex); \
298 		return(rc); \
299 	} \
300 }
301 
302 #define DO_RAID_FAIL() { \
303 	rf_UnconfigureVnodes(raidPtr); \
304 	rf_ShutdownList(&raidPtr->shutdownList); \
305 	rf_UnconfigureArray(); \
306 }
307 
308 #define DO_RAID_INIT_CONFIGURE(f) { \
309 	rc = f (&raidPtr->shutdownList, raidPtr, cfgPtr); \
310 	if (rc) { \
311 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
312 		DO_RAID_FAIL(); \
313 		return(rc); \
314 	} \
315 }
316 
317 #define DO_RAID_MUTEX(_m_) { \
318 	rc = rf_create_managed_mutex(&raidPtr->shutdownList, (_m_)); \
319 	if (rc) { \
320 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", \
321 			__FILE__, __LINE__, rc); \
322 		DO_RAID_FAIL(); \
323 		return(rc); \
324 	} \
325 }
326 
327 #define DO_RAID_COND(_c_) { \
328 	rc = rf_create_managed_cond(&raidPtr->shutdownList, (_c_)); \
329 	if (rc) { \
330 		RF_ERRORMSG3("Unable to init cond file %s line %d rc=%d\n", \
331 			__FILE__, __LINE__, rc); \
332 		DO_RAID_FAIL(); \
333 		return(rc); \
334 	} \
335 }
336 
337 int
338 rf_Configure(raidPtr, cfgPtr, ac)
339 	RF_Raid_t *raidPtr;
340 	RF_Config_t *cfgPtr;
341 	RF_AutoConfig_t *ac;
342 {
343 	RF_RowCol_t row, col;
344 	int     i, rc;
345 
346 	/* XXX This check can probably be removed now, since
347 	   RAIDFRAME_CONFIGURRE now checks to make sure that the
348 	   RAID set is not already valid
349 	*/
350 	if (raidPtr->valid) {
351 		RF_ERRORMSG("RAIDframe configuration not shut down.  Aborting configure.\n");
352 		return (EINVAL);
353 	}
354 	RF_LOCK_MUTEX(configureMutex);
355 	configureCount++;
356 	if (isconfigged == 0) {
357 		rc = rf_create_managed_mutex(&globalShutdown, &rf_printf_mutex);
358 		if (rc) {
359 			RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
360 			    __LINE__, rc);
361 			rf_ShutdownList(&globalShutdown);
362 			return (rc);
363 		}
364 		/* initialize globals */
365 		printf("RAIDFRAME: protectedSectors is %ld\n",
366 		       rf_protectedSectors);
367 
368 		rf_clear_debug_print_buffer();
369 
370 		DO_INIT_CONFIGURE(rf_ConfigureAllocList);
371 
372 		/*
373 	         * Yes, this does make debugging general to the whole
374 	         * system instead of being array specific. Bummer, drag.
375 		 */
376 		rf_ConfigureDebug(cfgPtr);
377 		DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
378 		DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
379 		DO_INIT_CONFIGURE(rf_ConfigureMapModule);
380 		DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
381 		DO_INIT_CONFIGURE(rf_ConfigureCallback);
382 		DO_INIT_CONFIGURE(rf_ConfigureMemChunk);
383 		DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
384 		DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
385 		DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
386 		DO_INIT_CONFIGURE(rf_ConfigureMCPair);
387 		DO_INIT_CONFIGURE(rf_ConfigureDAGs);
388 		DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
389 		DO_INIT_CONFIGURE(rf_ConfigureDebugPrint);
390 		DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
391 		DO_INIT_CONFIGURE(rf_ConfigureCopyback);
392 		DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
393 		isconfigged = 1;
394 	}
395 	RF_UNLOCK_MUTEX(configureMutex);
396 
397 	DO_RAID_MUTEX(&raidPtr->mutex);
398 	/* set up the cleanup list.  Do this after ConfigureDebug so that
399 	 * value of memDebug will be set */
400 
401 	rf_MakeAllocList(raidPtr->cleanupList);
402 	if (raidPtr->cleanupList == NULL) {
403 		DO_RAID_FAIL();
404 		return (ENOMEM);
405 	}
406 	rc = rf_ShutdownCreate(&raidPtr->shutdownList,
407 	    (void (*) (void *)) rf_FreeAllocList,
408 	    raidPtr->cleanupList);
409 	if (rc) {
410 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n",
411 		    __FILE__, __LINE__, rc);
412 		DO_RAID_FAIL();
413 		return (rc);
414 	}
415 	raidPtr->numRow = cfgPtr->numRow;
416 	raidPtr->numCol = cfgPtr->numCol;
417 	raidPtr->numSpare = cfgPtr->numSpare;
418 
419 	/* XXX we don't even pretend to support more than one row in the
420 	 * kernel... */
421 	if (raidPtr->numRow != 1) {
422 		RF_ERRORMSG("Only one row supported in kernel.\n");
423 		DO_RAID_FAIL();
424 		return (EINVAL);
425 	}
426 	RF_CallocAndAdd(raidPtr->status, raidPtr->numRow, sizeof(RF_RowStatus_t),
427 	    (RF_RowStatus_t *), raidPtr->cleanupList);
428 	if (raidPtr->status == NULL) {
429 		DO_RAID_FAIL();
430 		return (ENOMEM);
431 	}
432 	RF_CallocAndAdd(raidPtr->reconControl, raidPtr->numRow,
433 	    sizeof(RF_ReconCtrl_t *), (RF_ReconCtrl_t **), raidPtr->cleanupList);
434 	if (raidPtr->reconControl == NULL) {
435 		DO_RAID_FAIL();
436 		return (ENOMEM);
437 	}
438 	for (i = 0; i < raidPtr->numRow; i++) {
439 		raidPtr->status[i] = rf_rs_optimal;
440 		raidPtr->reconControl[i] = NULL;
441 	}
442 
443 	DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
444 	DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
445 
446 	DO_RAID_COND(&raidPtr->outstandingCond);
447 
448 	raidPtr->nAccOutstanding = 0;
449 	raidPtr->waitShutdown = 0;
450 
451 	DO_RAID_MUTEX(&raidPtr->access_suspend_mutex);
452 	DO_RAID_COND(&raidPtr->quiescent_cond);
453 
454 	DO_RAID_COND(&raidPtr->waitForReconCond);
455 
456 	DO_RAID_MUTEX(&raidPtr->recon_done_proc_mutex);
457 
458 	if (ac!=NULL) {
459 		/* We have an AutoConfig structure..  Don't do the
460 		   normal disk configuration... call the auto config
461 		   stuff */
462 		rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
463 	} else {
464 		DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
465 		DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
466 	}
467 	/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
468 	 * no. is set */
469 	DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
470 
471 	DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
472 
473 	DO_RAID_INIT_CONFIGURE(rf_ConfigurePSStatus);
474 
475 	for (row = 0; row < raidPtr->numRow; row++) {
476 		for (col = 0; col < raidPtr->numCol; col++) {
477 			/*
478 		         * XXX better distribution
479 		         */
480 			raidPtr->hist_diskreq[row][col] = 0;
481 		}
482 	}
483 
484 	raidPtr->numNewFailures = 0;
485 	raidPtr->copyback_in_progress = 0;
486 	raidPtr->parity_rewrite_in_progress = 0;
487 	raidPtr->recon_in_progress = 0;
488 	raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
489 
490 	/* autoconfigure and root_partition will actually get filled in
491 	   after the config is done */
492 	raidPtr->autoconfigure = 0;
493 	raidPtr->root_partition = 0;
494 	raidPtr->last_unit = raidPtr->raidid;
495 	raidPtr->config_order = 0;
496 
497 	if (rf_keepAccTotals) {
498 		raidPtr->keep_acc_totals = 1;
499 	}
500 	rf_StartUserStats(raidPtr);
501 
502 	raidPtr->valid = 1;
503 	return (0);
504 }
505 
506 static int
507 init_rad(desc)
508 	RF_RaidAccessDesc_t *desc;
509 {
510 	int     rc;
511 
512 	rc = rf_mutex_init(&desc->mutex);
513 	if (rc) {
514 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
515 		    __LINE__, rc);
516 		return (rc);
517 	}
518 	rc = rf_cond_init(&desc->cond);
519 	if (rc) {
520 		RF_ERRORMSG3("Unable to init cond file %s line %d rc=%d\n", __FILE__,
521 		    __LINE__, rc);
522 		rf_mutex_destroy(&desc->mutex);
523 		return (rc);
524 	}
525 	return (0);
526 }
527 
528 static void
529 clean_rad(desc)
530 	RF_RaidAccessDesc_t *desc;
531 {
532 	rf_mutex_destroy(&desc->mutex);
533 	rf_cond_destroy(&desc->cond);
534 }
535 
536 static void
537 rf_ShutdownRDFreeList(ignored)
538 	void   *ignored;
539 {
540 	RF_FREELIST_DESTROY_CLEAN(rf_rad_freelist, next, (RF_RaidAccessDesc_t *), clean_rad);
541 }
542 
543 static int
544 rf_ConfigureRDFreeList(listp)
545 	RF_ShutdownList_t **listp;
546 {
547 	int     rc;
548 
549 	RF_FREELIST_CREATE(rf_rad_freelist, RF_MAX_FREE_RAD,
550 	    RF_RAD_INC, sizeof(RF_RaidAccessDesc_t));
551 	if (rf_rad_freelist == NULL) {
552 		return (ENOMEM);
553 	}
554 	rc = rf_ShutdownCreate(listp, rf_ShutdownRDFreeList, NULL);
555 	if (rc) {
556 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
557 		    __LINE__, rc);
558 		rf_ShutdownRDFreeList(NULL);
559 		return (rc);
560 	}
561 	RF_FREELIST_PRIME_INIT(rf_rad_freelist, RF_RAD_INITIAL, next,
562 	    (RF_RaidAccessDesc_t *), init_rad);
563 	return (0);
564 }
565 
566 RF_RaidAccessDesc_t *
567 rf_AllocRaidAccDesc(
568     RF_Raid_t * raidPtr,
569     RF_IoType_t type,
570     RF_RaidAddr_t raidAddress,
571     RF_SectorCount_t numBlocks,
572     caddr_t bufPtr,
573     void *bp,
574     RF_DagHeader_t ** paramDAG,
575     RF_AccessStripeMapHeader_t ** paramASM,
576     RF_RaidAccessFlags_t flags,
577     void (*cbF) (struct buf *),
578     void *cbA,
579     RF_AccessState_t * states)
580 {
581 	RF_RaidAccessDesc_t *desc;
582 
583 	RF_FREELIST_GET_INIT_NOUNLOCK(rf_rad_freelist, desc, next, (RF_RaidAccessDesc_t *), init_rad);
584 	if (raidPtr->waitShutdown) {
585 		/*
586 	         * Actually, we're shutting the array down. Free the desc
587 	         * and return NULL.
588 	         */
589 		RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
590 		RF_FREELIST_FREE_CLEAN(rf_rad_freelist, desc, next, clean_rad);
591 		return (NULL);
592 	}
593 	raidPtr->nAccOutstanding++;
594 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
595 
596 	desc->raidPtr = (void *) raidPtr;
597 	desc->type = type;
598 	desc->raidAddress = raidAddress;
599 	desc->numBlocks = numBlocks;
600 	desc->bufPtr = bufPtr;
601 	desc->bp = bp;
602 	desc->paramDAG = paramDAG;
603 	desc->paramASM = paramASM;
604 	desc->flags = flags;
605 	desc->states = states;
606 	desc->state = 0;
607 
608 	desc->status = 0;
609 	bzero((char *) &desc->tracerec, sizeof(RF_AccTraceEntry_t));
610 	desc->callbackFunc = (void (*) (RF_CBParam_t)) cbF;	/* XXX */
611 	desc->callbackArg = cbA;
612 	desc->next = NULL;
613 	desc->head = desc;
614 	desc->numPending = 0;
615 	desc->cleanupList = NULL;
616 	rf_MakeAllocList(desc->cleanupList);
617 	return (desc);
618 }
619 
620 void
621 rf_FreeRaidAccDesc(RF_RaidAccessDesc_t * desc)
622 {
623 	RF_Raid_t *raidPtr = desc->raidPtr;
624 
625 	RF_ASSERT(desc);
626 
627 	rf_FreeAllocList(desc->cleanupList);
628 	RF_FREELIST_FREE_CLEAN_NOUNLOCK(rf_rad_freelist, desc, next, clean_rad);
629 	raidPtr->nAccOutstanding--;
630 	if (raidPtr->waitShutdown) {
631 		RF_SIGNAL_COND(raidPtr->outstandingCond);
632 	}
633 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
634 }
635 /*********************************************************************
636  * Main routine for performing an access.
637  * Accesses are retried until a DAG can not be selected.  This occurs
638  * when either the DAG library is incomplete or there are too many
639  * failures in a parity group.
640  ********************************************************************/
641 int
642 rf_DoAccess(
643     RF_Raid_t * raidPtr,
644     RF_IoType_t type,
645     int async_flag,
646     RF_RaidAddr_t raidAddress,
647     RF_SectorCount_t numBlocks,
648     caddr_t bufPtr,
649     void *bp_in,
650     RF_DagHeader_t ** paramDAG,
651     RF_AccessStripeMapHeader_t ** paramASM,
652     RF_RaidAccessFlags_t flags,
653     RF_RaidAccessDesc_t ** paramDesc,
654     void (*cbF) (struct buf *),
655     void *cbA)
656 /*
657 type should be read or write
658 async_flag should be RF_TRUE or RF_FALSE
659 bp_in is a buf pointer.  void * to facilitate ignoring it outside the kernel
660 */
661 {
662 	RF_RaidAccessDesc_t *desc;
663 	caddr_t lbufPtr = bufPtr;
664 	struct buf *bp = (struct buf *) bp_in;
665 
666 	raidAddress += rf_raidSectorOffset;
667 
668 	if (!raidPtr->valid) {
669 		RF_ERRORMSG("RAIDframe driver not successfully configured.  Rejecting access.\n");
670 		IO_BUF_ERR(bp, EINVAL);
671 		return (EINVAL);
672 	}
673 
674 	if (rf_accessDebug) {
675 
676 		printf("logBytes is: %d %d %d\n", raidPtr->raidid,
677 		    raidPtr->logBytesPerSector,
678 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks));
679 		printf("raid%d: %s raidAddr %d (stripeid %d-%d) numBlocks %d (%d bytes) buf 0x%lx\n", raidPtr->raidid,
680 		    (type == RF_IO_TYPE_READ) ? "READ" : "WRITE", (int) raidAddress,
681 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress),
682 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress + numBlocks - 1),
683 		    (int) numBlocks,
684 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks),
685 		    (long) bufPtr);
686 	}
687 	if (raidAddress + numBlocks > raidPtr->totalSectors) {
688 
689 		printf("DoAccess: raid addr %lu too large to access %lu sectors.  Max legal addr is %lu\n",
690 		    (u_long) raidAddress, (u_long) numBlocks, (u_long) raidPtr->totalSectors);
691 
692 		IO_BUF_ERR(bp, ENOSPC);
693 		return (ENOSPC);
694 	}
695 	desc = rf_AllocRaidAccDesc(raidPtr, type, raidAddress,
696 	    numBlocks, lbufPtr, bp, paramDAG, paramASM,
697 	    flags, cbF, cbA, raidPtr->Layout.map->states);
698 
699 	if (desc == NULL) {
700 		return (ENOMEM);
701 	}
702 	RF_ETIMER_START(desc->tracerec.tot_timer);
703 
704 	desc->async_flag = async_flag;
705 
706 	rf_ContinueRaidAccess(desc);
707 
708 	return (0);
709 }
710 /* force the array into reconfigured mode without doing reconstruction */
711 int
712 rf_SetReconfiguredMode(raidPtr, row, col)
713 	RF_Raid_t *raidPtr;
714 	int     row;
715 	int     col;
716 {
717 	if (!(raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
718 		printf("Can't set reconfigured mode in dedicated-spare array\n");
719 		RF_PANIC();
720 	}
721 	RF_LOCK_MUTEX(raidPtr->mutex);
722 	raidPtr->numFailures++;
723 	raidPtr->Disks[row][col].status = rf_ds_dist_spared;
724 	raidPtr->status[row] = rf_rs_reconfigured;
725 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
726 	/* install spare table only if declustering + distributed sparing
727 	 * architecture. */
728 	if (raidPtr->Layout.map->flags & RF_BD_DECLUSTERED)
729 		rf_InstallSpareTable(raidPtr, row, col);
730 	RF_UNLOCK_MUTEX(raidPtr->mutex);
731 	return (0);
732 }
733 
734 extern int fail_row, fail_col, fail_time;
735 extern int delayed_recon;
736 
737 int
738 rf_FailDisk(
739     RF_Raid_t * raidPtr,
740     int frow,
741     int fcol,
742     int initRecon)
743 {
744 	printf("raid%d: Failing disk r%d c%d\n", raidPtr->raidid, frow, fcol);
745 	RF_LOCK_MUTEX(raidPtr->mutex);
746 	raidPtr->numFailures++;
747 	raidPtr->Disks[frow][fcol].status = rf_ds_failed;
748 	raidPtr->status[frow] = rf_rs_degraded;
749 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
750 	RF_UNLOCK_MUTEX(raidPtr->mutex);
751 	if (initRecon)
752 		rf_ReconstructFailedDisk(raidPtr, frow, fcol);
753 	return (0);
754 }
755 /* releases a thread that is waiting for the array to become quiesced.
756  * access_suspend_mutex should be locked upon calling this
757  */
758 void
759 rf_SignalQuiescenceLock(raidPtr, reconDesc)
760 	RF_Raid_t *raidPtr;
761 	RF_RaidReconDesc_t *reconDesc;
762 {
763 	if (rf_quiesceDebug) {
764 		printf("raid%d: Signalling quiescence lock\n",
765 		       raidPtr->raidid);
766 	}
767 	raidPtr->access_suspend_release = 1;
768 
769 	if (raidPtr->waiting_for_quiescence) {
770 		SIGNAL_QUIESCENT_COND(raidPtr);
771 	}
772 }
773 /* suspends all new requests to the array.  No effect on accesses that are in flight.  */
774 int
775 rf_SuspendNewRequestsAndWait(raidPtr)
776 	RF_Raid_t *raidPtr;
777 {
778 	if (rf_quiesceDebug)
779 		printf("Suspending new reqs\n");
780 
781 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
782 	raidPtr->accesses_suspended++;
783 	raidPtr->waiting_for_quiescence = (raidPtr->accs_in_flight == 0) ? 0 : 1;
784 
785 	if (raidPtr->waiting_for_quiescence) {
786 		raidPtr->access_suspend_release = 0;
787 		while (!raidPtr->access_suspend_release) {
788 			printf("Suspending: Waiting for Quiescence\n");
789 			WAIT_FOR_QUIESCENCE(raidPtr);
790 			raidPtr->waiting_for_quiescence = 0;
791 		}
792 	}
793 	printf("Quiescence reached..\n");
794 
795 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
796 	return (raidPtr->waiting_for_quiescence);
797 }
798 /* wake up everyone waiting for quiescence to be released */
799 void
800 rf_ResumeNewRequests(raidPtr)
801 	RF_Raid_t *raidPtr;
802 {
803 	RF_CallbackDesc_t *t, *cb;
804 
805 	if (rf_quiesceDebug)
806 		printf("Resuming new reqs\n");
807 
808 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
809 	raidPtr->accesses_suspended--;
810 	if (raidPtr->accesses_suspended == 0)
811 		cb = raidPtr->quiesce_wait_list;
812 	else
813 		cb = NULL;
814 	raidPtr->quiesce_wait_list = NULL;
815 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
816 
817 	while (cb) {
818 		t = cb;
819 		cb = cb->next;
820 		(t->callbackFunc) (t->callbackArg);
821 		rf_FreeCallbackDesc(t);
822 	}
823 }
824 /*****************************************************************************************
825  *
826  * debug routines
827  *
828  ****************************************************************************************/
829 
830 static void
831 set_debug_option(name, val)
832 	char   *name;
833 	long    val;
834 {
835 	RF_DebugName_t *p;
836 
837 	for (p = rf_debugNames; p->name; p++) {
838 		if (!strcmp(p->name, name)) {
839 			*(p->ptr) = val;
840 			printf("[Set debug variable %s to %ld]\n", name, val);
841 			return;
842 		}
843 	}
844 	RF_ERRORMSG1("Unknown debug string \"%s\"\n", name);
845 }
846 
847 
848 /* would like to use sscanf here, but apparently not available in kernel */
849 /*ARGSUSED*/
850 static void
851 rf_ConfigureDebug(cfgPtr)
852 	RF_Config_t *cfgPtr;
853 {
854 	char   *val_p, *name_p, *white_p;
855 	long    val;
856 	int     i;
857 
858 	rf_ResetDebugOptions();
859 	for (i = 0; cfgPtr->debugVars[i][0] && i < RF_MAXDBGV; i++) {
860 		name_p = rf_find_non_white(&cfgPtr->debugVars[i][0]);
861 		white_p = rf_find_white(name_p);	/* skip to start of 2nd
862 							 * word */
863 		val_p = rf_find_non_white(white_p);
864 		if (*val_p == '0' && *(val_p + 1) == 'x')
865 			val = rf_htoi(val_p + 2);
866 		else
867 			val = rf_atoi(val_p);
868 		*white_p = '\0';
869 		set_debug_option(name_p, val);
870 	}
871 }
872 /* performance monitoring stuff */
873 
874 #define TIMEVAL_TO_US(t) (((long) t.tv_sec) * 1000000L + (long) t.tv_usec)
875 
876 #if !defined(_KERNEL) && !defined(SIMULATE)
877 
878 /*
879  * Throughput stats currently only used in user-level RAIDframe
880  */
881 
882 static int
883 rf_InitThroughputStats(
884     RF_ShutdownList_t ** listp,
885     RF_Raid_t * raidPtr,
886     RF_Config_t * cfgPtr)
887 {
888 	int     rc;
889 
890 	/* these used by user-level raidframe only */
891 	rc = rf_create_managed_mutex(listp, &raidPtr->throughputstats.mutex);
892 	if (rc) {
893 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
894 		    __LINE__, rc);
895 		return (rc);
896 	}
897 	raidPtr->throughputstats.sum_io_us = 0;
898 	raidPtr->throughputstats.num_ios = 0;
899 	raidPtr->throughputstats.num_out_ios = 0;
900 	return (0);
901 }
902 
903 void
904 rf_StartThroughputStats(RF_Raid_t * raidPtr)
905 {
906 	RF_LOCK_MUTEX(raidPtr->throughputstats.mutex);
907 	raidPtr->throughputstats.num_ios++;
908 	raidPtr->throughputstats.num_out_ios++;
909 	if (raidPtr->throughputstats.num_out_ios == 1)
910 		RF_GETTIME(raidPtr->throughputstats.start);
911 	RF_UNLOCK_MUTEX(raidPtr->throughputstats.mutex);
912 }
913 
914 static void
915 rf_StopThroughputStats(RF_Raid_t * raidPtr)
916 {
917 	struct timeval diff;
918 
919 	RF_LOCK_MUTEX(raidPtr->throughputstats.mutex);
920 	raidPtr->throughputstats.num_out_ios--;
921 	if (raidPtr->throughputstats.num_out_ios == 0) {
922 		RF_GETTIME(raidPtr->throughputstats.stop);
923 		RF_TIMEVAL_DIFF(&raidPtr->throughputstats.start, &raidPtr->throughputstats.stop, &diff);
924 		raidPtr->throughputstats.sum_io_us += TIMEVAL_TO_US(diff);
925 	}
926 	RF_UNLOCK_MUTEX(raidPtr->throughputstats.mutex);
927 }
928 
929 static void
930 rf_PrintThroughputStats(RF_Raid_t * raidPtr)
931 {
932 	RF_ASSERT(raidPtr->throughputstats.num_out_ios == 0);
933 	if (raidPtr->throughputstats.sum_io_us != 0) {
934 		printf("[Througphut: %8.2f IOs/second]\n", raidPtr->throughputstats.num_ios
935 		    / (raidPtr->throughputstats.sum_io_us / 1000000.0));
936 	}
937 }
938 #endif				/* !KERNEL && !SIMULATE */
939 
940 void
941 rf_StartUserStats(RF_Raid_t * raidPtr)
942 {
943 	RF_GETTIME(raidPtr->userstats.start);
944 	raidPtr->userstats.sum_io_us = 0;
945 	raidPtr->userstats.num_ios = 0;
946 	raidPtr->userstats.num_sect_moved = 0;
947 }
948 
949 void
950 rf_StopUserStats(RF_Raid_t * raidPtr)
951 {
952 	RF_GETTIME(raidPtr->userstats.stop);
953 }
954 
955 void
956 rf_UpdateUserStats(raidPtr, rt, numsect)
957 	RF_Raid_t *raidPtr;
958 	int     rt;		/* resp time in us */
959 	int     numsect;	/* number of sectors for this access */
960 {
961 	raidPtr->userstats.sum_io_us += rt;
962 	raidPtr->userstats.num_ios++;
963 	raidPtr->userstats.num_sect_moved += numsect;
964 }
965 
966 void
967 rf_PrintUserStats(RF_Raid_t * raidPtr)
968 {
969 	long    elapsed_us, mbs, mbs_frac;
970 	struct timeval diff;
971 
972 	RF_TIMEVAL_DIFF(&raidPtr->userstats.start, &raidPtr->userstats.stop, &diff);
973 	elapsed_us = TIMEVAL_TO_US(diff);
974 
975 	/* 2000 sectors per megabyte, 10000000 microseconds per second */
976 	if (elapsed_us)
977 		mbs = (raidPtr->userstats.num_sect_moved / 2000) / (elapsed_us / 1000000);
978 	else
979 		mbs = 0;
980 
981 	/* this computes only the first digit of the fractional mb/s moved */
982 	if (elapsed_us) {
983 		mbs_frac = ((raidPtr->userstats.num_sect_moved / 200) / (elapsed_us / 1000000))
984 		    - (mbs * 10);
985 	} else {
986 		mbs_frac = 0;
987 	}
988 
989 	printf("Number of I/Os:             %ld\n", raidPtr->userstats.num_ios);
990 	printf("Elapsed time (us):          %ld\n", elapsed_us);
991 	printf("User I/Os per second:       %ld\n", RF_DB0_CHECK(raidPtr->userstats.num_ios, (elapsed_us / 1000000)));
992 	printf("Average user response time: %ld us\n", RF_DB0_CHECK(raidPtr->userstats.sum_io_us, raidPtr->userstats.num_ios));
993 	printf("Total sectors moved:        %ld\n", raidPtr->userstats.num_sect_moved);
994 	printf("Average access size (sect): %ld\n", RF_DB0_CHECK(raidPtr->userstats.num_sect_moved, raidPtr->userstats.num_ios));
995 	printf("Achieved data rate:         %ld.%ld MB/sec\n", mbs, mbs_frac);
996 }
997