xref: /netbsd-src/sys/dev/raidframe/rf_driver.c (revision ba65fde2d7fefa7d39838fa5fa855e62bd606b5e)
1 /*	$NetBSD: rf_driver.c,v 1.131 2012/12/10 08:36:03 msaitoh 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  *
18  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 /*
32  * Copyright (c) 1995 Carnegie-Mellon University.
33  * All rights reserved.
34  *
35  * Author: Mark Holland, Khalil Amiri, Claudson Bornstein, William V. Courtright II,
36  *         Robby Findler, Daniel Stodolsky, Rachad Youssef, Jim Zelenka
37  *
38  * Permission to use, copy, modify and distribute this software and
39  * its documentation is hereby granted, provided that both the copyright
40  * notice and this permission notice appear in all copies of the
41  * software, derivative works or modified versions, and any portions
42  * thereof, and that both notices appear in supporting documentation.
43  *
44  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
45  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
46  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
47  *
48  * Carnegie Mellon requests users of this software to return to
49  *
50  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
51  *  School of Computer Science
52  *  Carnegie Mellon University
53  *  Pittsburgh PA 15213-3890
54  *
55  * any improvements or extensions that they make and grant Carnegie the
56  * rights to redistribute these changes.
57  */
58 
59 /******************************************************************************
60  *
61  * rf_driver.c -- main setup, teardown, and access routines for the RAID driver
62  *
63  * all routines are prefixed with rf_ (raidframe), to avoid conficts.
64  *
65  ******************************************************************************/
66 
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: rf_driver.c,v 1.131 2012/12/10 08:36:03 msaitoh Exp $");
70 
71 #ifdef _KERNEL_OPT
72 #include "opt_raid_diagnostic.h"
73 #endif
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/ioctl.h>
78 #include <sys/fcntl.h>
79 #include <sys/vnode.h>
80 
81 
82 #include "rf_archs.h"
83 #include "rf_threadstuff.h"
84 
85 #include <sys/errno.h>
86 
87 #include "rf_raid.h"
88 #include "rf_dag.h"
89 #include "rf_aselect.h"
90 #include "rf_diskqueue.h"
91 #include "rf_parityscan.h"
92 #include "rf_alloclist.h"
93 #include "rf_dagutils.h"
94 #include "rf_utils.h"
95 #include "rf_etimer.h"
96 #include "rf_acctrace.h"
97 #include "rf_general.h"
98 #include "rf_desc.h"
99 #include "rf_states.h"
100 #include "rf_decluster.h"
101 #include "rf_map.h"
102 #include "rf_revent.h"
103 #include "rf_callback.h"
104 #include "rf_engine.h"
105 #include "rf_mcpair.h"
106 #include "rf_nwayxor.h"
107 #include "rf_copyback.h"
108 #include "rf_driver.h"
109 #include "rf_options.h"
110 #include "rf_shutdown.h"
111 #include "rf_kintf.h"
112 #include "rf_paritymap.h"
113 
114 #include <sys/buf.h>
115 
116 #ifndef RF_ACCESS_DEBUG
117 #define RF_ACCESS_DEBUG 0
118 #endif
119 
120 /* rad == RF_RaidAccessDesc_t */
121 #define RF_MAX_FREE_RAD 128
122 #define RF_MIN_FREE_RAD  32
123 
124 /* debug variables */
125 char    rf_panicbuf[2048];	/* a buffer to hold an error msg when we panic */
126 
127 /* main configuration routines */
128 static int raidframe_booted = 0;
129 
130 static void rf_ConfigureDebug(RF_Config_t * cfgPtr);
131 static void set_debug_option(char *name, long val);
132 static void rf_UnconfigureArray(void);
133 static void rf_ShutdownRDFreeList(void *);
134 static int rf_ConfigureRDFreeList(RF_ShutdownList_t **);
135 
136 rf_declare_mutex2(rf_printf_mutex);	/* debug only:  avoids interleaved
137 					 * printfs by different stripes */
138 
139 #define SIGNAL_QUIESCENT_COND(_raid_) \
140 	rf_broadcast_cond2((_raid_)->access_suspend_cv)
141 #define WAIT_FOR_QUIESCENCE(_raid_) \
142 	rf_wait_cond2((_raid_)->access_suspend_cv, \
143 		      (_raid_)->access_suspend_mutex)
144 
145 static int configureCount = 0;	/* number of active configurations */
146 static int isconfigged = 0;	/* is basic raidframe (non per-array)
147 				 * stuff configured */
148 static rf_declare_mutex2(configureMutex); /* used to lock the configuration
149 					   * stuff */
150 static RF_ShutdownList_t *globalShutdown;	/* non array-specific
151 						 * stuff */
152 
153 static int rf_ConfigureRDFreeList(RF_ShutdownList_t ** listp);
154 static int rf_AllocEmergBuffers(RF_Raid_t *);
155 static void rf_FreeEmergBuffers(RF_Raid_t *);
156 static void rf_destroy_mutex_cond(RF_Raid_t *);
157 static void rf_alloc_mutex_cond(RF_Raid_t *);
158 
159 /* called at system boot time */
160 int
161 rf_BootRaidframe(void)
162 {
163 
164 	if (raidframe_booted)
165 		return (EBUSY);
166 	raidframe_booted = 1;
167 	rf_init_mutex2(configureMutex, IPL_NONE);
168  	configureCount = 0;
169 	isconfigged = 0;
170 	globalShutdown = NULL;
171 	return (0);
172 }
173 
174 /*
175  * Called whenever an array is shutdown
176  */
177 static void
178 rf_UnconfigureArray(void)
179 {
180 
181 	rf_lock_mutex2(configureMutex);
182 	if (--configureCount == 0) {	/* if no active configurations, shut
183 					 * everything down */
184 		rf_destroy_mutex2(rf_printf_mutex);
185 		isconfigged = 0;
186 		rf_ShutdownList(&globalShutdown);
187 
188 		/*
189 	         * We must wait until now, because the AllocList module
190 	         * uses the DebugMem module.
191 	         */
192 #if RF_DEBUG_MEM
193 		if (rf_memDebug)
194 			rf_print_unfreed();
195 #endif
196 	}
197 	rf_unlock_mutex2(configureMutex);
198 }
199 
200 /*
201  * Called to shut down an array.
202  */
203 int
204 rf_Shutdown(RF_Raid_t *raidPtr)
205 {
206 
207 	if (!raidPtr->valid) {
208 		RF_ERRORMSG("Attempt to shut down unconfigured RAIDframe driver.  Aborting shutdown\n");
209 		return (EINVAL);
210 	}
211 	/*
212          * wait for outstanding IOs to land
213          * As described in rf_raid.h, we use the rad_freelist lock
214          * to protect the per-array info about outstanding descs
215          * since we need to do freelist locking anyway, and this
216          * cuts down on the amount of serialization we've got going
217          * on.
218          */
219 	rf_lock_mutex2(raidPtr->rad_lock);
220 	if (raidPtr->waitShutdown) {
221 		rf_unlock_mutex2(raidPtr->rad_lock);
222 		return (EBUSY);
223 	}
224 	raidPtr->waitShutdown = 1;
225 	while (raidPtr->nAccOutstanding) {
226 		rf_wait_cond2(raidPtr->outstandingCond, raidPtr->rad_lock);
227 	}
228 	rf_unlock_mutex2(raidPtr->rad_lock);
229 
230 	/* Wait for any parity re-writes to stop... */
231 	while (raidPtr->parity_rewrite_in_progress) {
232 		printf("raid%d: Waiting for parity re-write to exit...\n",
233 		       raidPtr->raidid);
234 		tsleep(&raidPtr->parity_rewrite_in_progress, PRIBIO,
235 		       "rfprwshutdown", 0);
236 	}
237 
238 	/* Wait for any reconstruction to stop... */
239 	rf_lock_mutex2(raidPtr->mutex);
240 	while (raidPtr->reconInProgress) {
241 		printf("raid%d: Waiting for reconstruction to stop...\n",
242 		       raidPtr->raidid);
243 		rf_wait_cond2(raidPtr->waitForReconCond, raidPtr->mutex);
244 	}
245 	rf_unlock_mutex2(raidPtr->mutex);
246 
247 	raidPtr->valid = 0;
248 
249 	if (raidPtr->parity_map != NULL)
250 		rf_paritymap_detach(raidPtr);
251 
252 	rf_update_component_labels(raidPtr, RF_FINAL_COMPONENT_UPDATE);
253 
254 	rf_UnconfigureVnodes(raidPtr);
255 
256 	rf_FreeEmergBuffers(raidPtr);
257 
258 	rf_ShutdownList(&raidPtr->shutdownList);
259 
260 	rf_destroy_mutex_cond(raidPtr);
261 
262 	rf_UnconfigureArray();
263 
264 	return (0);
265 }
266 
267 
268 #define DO_INIT_CONFIGURE(f) { \
269 	rc = f (&globalShutdown); \
270 	if (rc) { \
271 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
272 		rf_ShutdownList(&globalShutdown); \
273 		configureCount--; \
274 		rf_unlock_mutex2(configureMutex); \
275 		rf_destroy_mutex2(rf_printf_mutex); \
276 		return(rc); \
277 	} \
278 }
279 
280 #define DO_RAID_FAIL() { \
281 	rf_UnconfigureVnodes(raidPtr); \
282 	rf_FreeEmergBuffers(raidPtr); \
283 	rf_ShutdownList(&raidPtr->shutdownList); \
284 	rf_UnconfigureArray(); \
285 	rf_destroy_mutex_cond(raidPtr); \
286 }
287 
288 #define DO_RAID_INIT_CONFIGURE(f) { \
289 	rc = f (&raidPtr->shutdownList, raidPtr, cfgPtr); \
290 	if (rc) { \
291 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
292 		DO_RAID_FAIL(); \
293 		return(rc); \
294 	} \
295 }
296 
297 int
298 rf_Configure(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr, RF_AutoConfig_t *ac)
299 {
300 	RF_RowCol_t col;
301 	int rc;
302 
303 	rf_lock_mutex2(configureMutex);
304 	configureCount++;
305 	if (isconfigged == 0) {
306 		rf_init_mutex2(rf_printf_mutex, IPL_VM);
307 
308 		/* initialize globals */
309 
310 		DO_INIT_CONFIGURE(rf_ConfigureAllocList);
311 
312 		/*
313 	         * Yes, this does make debugging general to the whole
314 	         * system instead of being array specific. Bummer, drag.
315 		 */
316 		rf_ConfigureDebug(cfgPtr);
317 		DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
318 #if RF_ACC_TRACE > 0
319 		DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
320 #endif
321 		DO_INIT_CONFIGURE(rf_ConfigureMapModule);
322 		DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
323 		DO_INIT_CONFIGURE(rf_ConfigureCallback);
324 		DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
325 		DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
326 		DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
327 		DO_INIT_CONFIGURE(rf_ConfigureMCPair);
328 		DO_INIT_CONFIGURE(rf_ConfigureDAGs);
329 		DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
330 		DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
331 		DO_INIT_CONFIGURE(rf_ConfigureCopyback);
332 		DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
333 		DO_INIT_CONFIGURE(rf_ConfigurePSStatus);
334 		isconfigged = 1;
335 	}
336 	rf_unlock_mutex2(configureMutex);
337 
338 	rf_alloc_mutex_cond(raidPtr);
339 
340 	/* set up the cleanup list.  Do this after ConfigureDebug so that
341 	 * value of memDebug will be set */
342 
343 	rf_MakeAllocList(raidPtr->cleanupList);
344 	if (raidPtr->cleanupList == NULL) {
345 		DO_RAID_FAIL();
346 		return (ENOMEM);
347 	}
348 	rf_ShutdownCreate(&raidPtr->shutdownList,
349 			  (void (*) (void *)) rf_FreeAllocList,
350 			  raidPtr->cleanupList);
351 
352 	raidPtr->numCol = cfgPtr->numCol;
353 	raidPtr->numSpare = cfgPtr->numSpare;
354 
355 	raidPtr->status = rf_rs_optimal;
356 	raidPtr->reconControl = NULL;
357 
358 	DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
359 	DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
360 
361 	raidPtr->nAccOutstanding = 0;
362 	raidPtr->waitShutdown = 0;
363 
364 	if (ac!=NULL) {
365 		/* We have an AutoConfig structure..  Don't do the
366 		   normal disk configuration... call the auto config
367 		   stuff */
368 		rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
369 	} else {
370 		DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
371 		DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
372 	}
373 	/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
374 	 * no. is set */
375 	DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
376 
377 	DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
378 
379 	/* Initialize per-RAID PSS bits */
380 	rf_InitPSStatus(raidPtr);
381 
382 #if RF_INCLUDE_CHAINDECLUSTER > 0
383 	for (col = 0; col < raidPtr->numCol; col++) {
384 		/*
385 		 * XXX better distribution
386 		 */
387 		raidPtr->hist_diskreq[col] = 0;
388 	}
389 #endif
390 	raidPtr->numNewFailures = 0;
391 	raidPtr->copyback_in_progress = 0;
392 	raidPtr->parity_rewrite_in_progress = 0;
393 	raidPtr->adding_hot_spare = 0;
394 	raidPtr->recon_in_progress = 0;
395 
396 	raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
397 
398 	/* autoconfigure and root_partition will actually get filled in
399 	   after the config is done */
400 	raidPtr->autoconfigure = 0;
401 	raidPtr->root_partition = 0;
402 	raidPtr->last_unit = raidPtr->raidid;
403 	raidPtr->config_order = 0;
404 
405 	if (rf_keepAccTotals) {
406 		raidPtr->keep_acc_totals = 1;
407 	}
408 
409 	/* Allocate a bunch of buffers to be used in low-memory conditions */
410 	raidPtr->iobuf = NULL;
411 
412 	rc = rf_AllocEmergBuffers(raidPtr);
413 	if (rc) {
414 		printf("raid%d: Unable to allocate emergency buffers.\n",
415 		       raidPtr->raidid);
416 		DO_RAID_FAIL();
417 		return(rc);
418 	}
419 
420 	/* Set up parity map stuff, if applicable. */
421 #ifndef RF_NO_PARITY_MAP
422 	rf_paritymap_attach(raidPtr, cfgPtr->force);
423 #endif
424 
425 	raidPtr->valid = 1;
426 
427 	printf("raid%d: %s\n", raidPtr->raidid,
428 	       raidPtr->Layout.map->configName);
429 	printf("raid%d: Components:", raidPtr->raidid);
430 
431 	for (col = 0; col < raidPtr->numCol; col++) {
432 		printf(" %s", raidPtr->Disks[col].devname);
433 		if (RF_DEAD_DISK(raidPtr->Disks[col].status)) {
434 			printf("[**FAILED**]");
435 		}
436 	}
437 	printf("\n");
438 	printf("raid%d: Total Sectors: %" PRIu64 " (%" PRIu64 " MB)\n",
439 	       raidPtr->raidid,
440 	       raidPtr->totalSectors,
441 	       (raidPtr->totalSectors / 1024 *
442 				(1 << raidPtr->logBytesPerSector) / 1024));
443 
444 	return (0);
445 }
446 
447 
448 /*
449 
450   Routines to allocate and free the "emergency buffers" for a given
451   RAID set.  These emergency buffers will be used when the kernel runs
452   out of kernel memory.
453 
454  */
455 
456 static int
457 rf_AllocEmergBuffers(RF_Raid_t *raidPtr)
458 {
459 	void *tmpbuf;
460 	RF_VoidPointerListElem_t *vple;
461 	int i;
462 
463 	/* XXX next line needs tuning... */
464 	raidPtr->numEmergencyBuffers = 10 * raidPtr->numCol;
465 #if DEBUG
466 	printf("raid%d: allocating %d buffers of %d bytes.\n",
467 	       raidPtr->raidid,
468 	       raidPtr->numEmergencyBuffers,
469 	       (int)(raidPtr->Layout.sectorsPerStripeUnit <<
470 	       raidPtr->logBytesPerSector));
471 #endif
472 	for (i = 0; i < raidPtr->numEmergencyBuffers; i++) {
473 		tmpbuf = malloc( raidPtr->Layout.sectorsPerStripeUnit <<
474 				 raidPtr->logBytesPerSector,
475 				 M_RAIDFRAME, M_WAITOK);
476 		if (tmpbuf) {
477 			vple = rf_AllocVPListElem();
478 			vple->p= tmpbuf;
479 			vple->next = raidPtr->iobuf;
480 			raidPtr->iobuf = vple;
481 			raidPtr->iobuf_count++;
482 		} else {
483 			printf("raid%d: failed to allocate emergency buffer!\n",
484 			       raidPtr->raidid);
485 			return 1;
486 		}
487 	}
488 
489 	/* XXX next line needs tuning too... */
490 	raidPtr->numEmergencyStripeBuffers = 10;
491         for (i = 0; i < raidPtr->numEmergencyStripeBuffers; i++) {
492                 tmpbuf = malloc( raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit <<
493                                  raidPtr->logBytesPerSector),
494                                  M_RAIDFRAME, M_WAITOK);
495                 if (tmpbuf) {
496                         vple = rf_AllocVPListElem();
497                         vple->p= tmpbuf;
498                         vple->next = raidPtr->stripebuf;
499                         raidPtr->stripebuf = vple;
500                         raidPtr->stripebuf_count++;
501                 } else {
502                         printf("raid%d: failed to allocate emergency stripe buffer!\n",
503                                raidPtr->raidid);
504 			return 1;
505                 }
506         }
507 
508 	return (0);
509 }
510 
511 static void
512 rf_FreeEmergBuffers(RF_Raid_t *raidPtr)
513 {
514 	RF_VoidPointerListElem_t *tmp;
515 
516 	/* Free the emergency IO buffers */
517 	while (raidPtr->iobuf != NULL) {
518 		tmp = raidPtr->iobuf;
519 		raidPtr->iobuf = raidPtr->iobuf->next;
520 		free(tmp->p, M_RAIDFRAME);
521 		rf_FreeVPListElem(tmp);
522 	}
523 
524 	/* Free the emergency stripe buffers */
525 	while (raidPtr->stripebuf != NULL) {
526 		tmp = raidPtr->stripebuf;
527 		raidPtr->stripebuf = raidPtr->stripebuf->next;
528 		free(tmp->p, M_RAIDFRAME);
529 		rf_FreeVPListElem(tmp);
530 	}
531 }
532 
533 
534 static void
535 rf_ShutdownRDFreeList(void *ignored)
536 {
537 	pool_destroy(&rf_pools.rad);
538 }
539 
540 static int
541 rf_ConfigureRDFreeList(RF_ShutdownList_t **listp)
542 {
543 
544 	rf_pool_init(&rf_pools.rad, sizeof(RF_RaidAccessDesc_t),
545 		     "rf_rad_pl", RF_MIN_FREE_RAD, RF_MAX_FREE_RAD);
546 	rf_ShutdownCreate(listp, rf_ShutdownRDFreeList, NULL);
547 	return (0);
548 }
549 
550 RF_RaidAccessDesc_t *
551 rf_AllocRaidAccDesc(RF_Raid_t *raidPtr, RF_IoType_t type,
552 		    RF_RaidAddr_t raidAddress, RF_SectorCount_t numBlocks,
553 		    void *bufPtr, void *bp, RF_RaidAccessFlags_t flags,
554 		    const RF_AccessState_t *states)
555 {
556 	RF_RaidAccessDesc_t *desc;
557 
558 	desc = pool_get(&rf_pools.rad, PR_WAITOK);
559 
560 	rf_lock_mutex2(raidPtr->rad_lock);
561 	if (raidPtr->waitShutdown) {
562 		/*
563 	         * Actually, we're shutting the array down. Free the desc
564 	         * and return NULL.
565 	         */
566 
567 		rf_unlock_mutex2(raidPtr->rad_lock);
568 		pool_put(&rf_pools.rad, desc);
569 		return (NULL);
570 	}
571 	raidPtr->nAccOutstanding++;
572 
573 	rf_unlock_mutex2(raidPtr->rad_lock);
574 
575 	desc->raidPtr = (void *) raidPtr;
576 	desc->type = type;
577 	desc->raidAddress = raidAddress;
578 	desc->numBlocks = numBlocks;
579 	desc->bufPtr = bufPtr;
580 	desc->bp = bp;
581 	desc->flags = flags;
582 	desc->states = states;
583 	desc->state = 0;
584 	desc->dagList = NULL;
585 
586 	desc->status = 0;
587 	desc->numRetries = 0;
588 #if RF_ACC_TRACE > 0
589 	memset((char *) &desc->tracerec, 0, sizeof(RF_AccTraceEntry_t));
590 #endif
591 	desc->callbackFunc = NULL;
592 	desc->callbackArg = NULL;
593 	desc->next = NULL;
594 	desc->iobufs = NULL;
595 	desc->stripebufs = NULL;
596 
597 	return (desc);
598 }
599 
600 void
601 rf_FreeRaidAccDesc(RF_RaidAccessDesc_t *desc)
602 {
603 	RF_Raid_t *raidPtr = desc->raidPtr;
604 	RF_DagList_t *dagList, *temp;
605 	RF_VoidPointerListElem_t *tmp;
606 
607 	RF_ASSERT(desc);
608 
609 	/* Cleanup the dagList(s) */
610 	dagList = desc->dagList;
611 	while(dagList != NULL) {
612 		temp = dagList;
613 		dagList = dagList->next;
614 		rf_FreeDAGList(temp);
615 	}
616 
617 	while (desc->iobufs) {
618 		tmp = desc->iobufs;
619 		desc->iobufs = desc->iobufs->next;
620 		rf_FreeIOBuffer(raidPtr, tmp);
621 	}
622 
623 	while (desc->stripebufs) {
624 		tmp = desc->stripebufs;
625 		desc->stripebufs = desc->stripebufs->next;
626 		rf_FreeStripeBuffer(raidPtr, tmp);
627 	}
628 
629 	pool_put(&rf_pools.rad, desc);
630 	rf_lock_mutex2(raidPtr->rad_lock);
631 	raidPtr->nAccOutstanding--;
632 	if (raidPtr->waitShutdown) {
633 		rf_signal_cond2(raidPtr->outstandingCond);
634 	}
635 	rf_unlock_mutex2(raidPtr->rad_lock);
636 }
637 /*********************************************************************
638  * Main routine for performing an access.
639  * Accesses are retried until a DAG can not be selected.  This occurs
640  * when either the DAG library is incomplete or there are too many
641  * failures in a parity group.
642  *
643  * type should be read or write async_flag should be RF_TRUE or
644  * RF_FALSE bp_in is a buf pointer.  void *to facilitate ignoring it
645  * outside the kernel
646  ********************************************************************/
647 int
648 rf_DoAccess(RF_Raid_t * raidPtr, RF_IoType_t type, int async_flag,
649 	    RF_RaidAddr_t raidAddress, RF_SectorCount_t numBlocks,
650 	    void *bufPtr, struct buf *bp, RF_RaidAccessFlags_t flags)
651 {
652 	RF_RaidAccessDesc_t *desc;
653 	void *lbufPtr = bufPtr;
654 
655 	raidAddress += rf_raidSectorOffset;
656 
657 #if RF_ACCESS_DEBUG
658 	if (rf_accessDebug) {
659 
660 		printf("logBytes is: %d %d %d\n", raidPtr->raidid,
661 		    raidPtr->logBytesPerSector,
662 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks));
663 		printf("raid%d: %s raidAddr %d (stripeid %d-%d) numBlocks %d (%d bytes) buf 0x%lx\n", raidPtr->raidid,
664 		    (type == RF_IO_TYPE_READ) ? "READ" : "WRITE", (int) raidAddress,
665 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress),
666 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress + numBlocks - 1),
667 		    (int) numBlocks,
668 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks),
669 		    (long) bufPtr);
670 	}
671 #endif
672 
673 	desc = rf_AllocRaidAccDesc(raidPtr, type, raidAddress,
674 	    numBlocks, lbufPtr, bp, flags, raidPtr->Layout.map->states);
675 
676 	if (desc == NULL) {
677 		return (ENOMEM);
678 	}
679 #if RF_ACC_TRACE > 0
680 	RF_ETIMER_START(desc->tracerec.tot_timer);
681 #endif
682 	desc->async_flag = async_flag;
683 
684 	if (raidPtr->parity_map != NULL &&
685 	    type == RF_IO_TYPE_WRITE)
686 		rf_paritymap_begin(raidPtr->parity_map, raidAddress,
687 		    numBlocks);
688 
689 	rf_ContinueRaidAccess(desc);
690 
691 	return (0);
692 }
693 #if 0
694 /* force the array into reconfigured mode without doing reconstruction */
695 int
696 rf_SetReconfiguredMode(RF_Raid_t *raidPtr, int col)
697 {
698 	if (!(raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
699 		printf("Can't set reconfigured mode in dedicated-spare array\n");
700 		RF_PANIC();
701 	}
702 	rf_lock_mutex2(raidPtr->mutex);
703 	raidPtr->numFailures++;
704 	raidPtr->Disks[col].status = rf_ds_dist_spared;
705 	raidPtr->status = rf_rs_reconfigured;
706 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
707 	/* install spare table only if declustering + distributed sparing
708 	 * architecture. */
709 	if (raidPtr->Layout.map->flags & RF_BD_DECLUSTERED)
710 		rf_InstallSpareTable(raidPtr, col);
711 	rf_unlock_mutex2(raidPtr->mutex);
712 	return (0);
713 }
714 #endif
715 
716 int
717 rf_FailDisk(RF_Raid_t *raidPtr, int fcol, int initRecon)
718 {
719 
720 	/* need to suspend IO's here -- if there are DAGs in flight
721 	   and we pull the rug out from under ci_vp, Bad Things
722 	   can happen.  */
723 
724 	rf_SuspendNewRequestsAndWait(raidPtr);
725 
726 	rf_lock_mutex2(raidPtr->mutex);
727 	if (raidPtr->Disks[fcol].status != rf_ds_failed) {
728 		/* must be failing something that is valid, or else it's
729 		   already marked as failed (in which case we don't
730 		   want to mark it failed again!) */
731 		raidPtr->numFailures++;
732 		raidPtr->Disks[fcol].status = rf_ds_failed;
733 		raidPtr->status = rf_rs_degraded;
734 	}
735 	rf_unlock_mutex2(raidPtr->mutex);
736 
737 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
738 
739 	/* Close the component, so that it's not "locked" if someone
740 	   else want's to use it! */
741 
742 	rf_close_component(raidPtr, raidPtr->raid_cinfo[fcol].ci_vp,
743 			   raidPtr->Disks[fcol].auto_configured);
744 
745 	rf_lock_mutex2(raidPtr->mutex);
746 	raidPtr->raid_cinfo[fcol].ci_vp = NULL;
747 
748 	/* Need to mark the component as not being auto_configured
749 	   (in case it was previously). */
750 
751 	raidPtr->Disks[fcol].auto_configured = 0;
752 	rf_unlock_mutex2(raidPtr->mutex);
753 	/* now we can allow IO to continue -- we'll be suspending it
754 	   again in rf_ReconstructFailedDisk() if we have to.. */
755 
756 	rf_ResumeNewRequests(raidPtr);
757 
758 	if (initRecon)
759 		rf_ReconstructFailedDisk(raidPtr, fcol);
760 	return (0);
761 }
762 /* releases a thread that is waiting for the array to become quiesced.
763  * access_suspend_mutex should be locked upon calling this
764  */
765 void
766 rf_SignalQuiescenceLock(RF_Raid_t *raidPtr)
767 {
768 #if RF_DEBUG_QUIESCE
769 	if (rf_quiesceDebug) {
770 		printf("raid%d: Signalling quiescence lock\n",
771 		       raidPtr->raidid);
772 	}
773 #endif
774 	raidPtr->access_suspend_release = 1;
775 
776 	if (raidPtr->waiting_for_quiescence) {
777 		SIGNAL_QUIESCENT_COND(raidPtr);
778 	}
779 }
780 /* suspends all new requests to the array.  No effect on accesses that are in flight.  */
781 int
782 rf_SuspendNewRequestsAndWait(RF_Raid_t *raidPtr)
783 {
784 #if RF_DEBUG_QUIESCE
785 	if (rf_quiesceDebug)
786 		printf("raid%d: Suspending new reqs\n", raidPtr->raidid);
787 #endif
788 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
789 	raidPtr->accesses_suspended++;
790 	raidPtr->waiting_for_quiescence = (raidPtr->accs_in_flight == 0) ? 0 : 1;
791 
792 	if (raidPtr->waiting_for_quiescence) {
793 		raidPtr->access_suspend_release = 0;
794 		while (!raidPtr->access_suspend_release) {
795 #if RF_DEBUG_QUIESCE
796 			printf("raid%d: Suspending: Waiting for Quiescence\n",
797 			       raidPtr->raidid);
798 #endif
799 			WAIT_FOR_QUIESCENCE(raidPtr);
800 			raidPtr->waiting_for_quiescence = 0;
801 		}
802 	}
803 #if RF_DEBUG_QUIESCE
804 	printf("raid%d: Quiescence reached..\n", raidPtr->raidid);
805 #endif
806 
807 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
808 	return (raidPtr->waiting_for_quiescence);
809 }
810 /* wake up everyone waiting for quiescence to be released */
811 void
812 rf_ResumeNewRequests(RF_Raid_t *raidPtr)
813 {
814 	RF_CallbackDesc_t *t, *cb;
815 
816 #if RF_DEBUG_QUIESCE
817 	if (rf_quiesceDebug)
818 		printf("raid%d: Resuming new requests\n", raidPtr->raidid);
819 #endif
820 
821 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
822 	raidPtr->accesses_suspended--;
823 	if (raidPtr->accesses_suspended == 0)
824 		cb = raidPtr->quiesce_wait_list;
825 	else
826 		cb = NULL;
827 	raidPtr->quiesce_wait_list = NULL;
828 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
829 
830 	while (cb) {
831 		t = cb;
832 		cb = cb->next;
833 		(t->callbackFunc) (t->callbackArg);
834 		rf_FreeCallbackDesc(t);
835 	}
836 }
837 /*****************************************************************************************
838  *
839  * debug routines
840  *
841  ****************************************************************************************/
842 
843 static void
844 set_debug_option(char *name, long val)
845 {
846 	RF_DebugName_t *p;
847 
848 	for (p = rf_debugNames; p->name; p++) {
849 		if (!strcmp(p->name, name)) {
850 			*(p->ptr) = val;
851 			printf("[Set debug variable %s to %ld]\n", name, val);
852 			return;
853 		}
854 	}
855 	RF_ERRORMSG1("Unknown debug string \"%s\"\n", name);
856 }
857 
858 
859 /* would like to use sscanf here, but apparently not available in kernel */
860 /*ARGSUSED*/
861 static void
862 rf_ConfigureDebug(RF_Config_t *cfgPtr)
863 {
864 	char   *val_p, *name_p, *white_p;
865 	long    val;
866 	int     i;
867 
868 	rf_ResetDebugOptions();
869 	for (i = 0; i < RF_MAXDBGV && cfgPtr->debugVars[i][0]; i++) {
870 		name_p = rf_find_non_white(&cfgPtr->debugVars[i][0]);
871 		white_p = rf_find_white(name_p);	/* skip to start of 2nd
872 							 * word */
873 		val_p = rf_find_non_white(white_p);
874 		if (*val_p == '0' && *(val_p + 1) == 'x')
875 			val = rf_htoi(val_p + 2);
876 		else
877 			val = rf_atoi(val_p);
878 		*white_p = '\0';
879 		set_debug_option(name_p, val);
880 	}
881 }
882 
883 void
884 rf_print_panic_message(int line, const char *file)
885 {
886 	snprintf(rf_panicbuf, sizeof(rf_panicbuf),
887 	    "raidframe error at line %d file %s", line, file);
888 }
889 
890 #ifdef RAID_DIAGNOSTIC
891 void
892 rf_print_assert_panic_message(int line,	const char *file, const char *condition)
893 {
894 	snprintf(rf_panicbuf, sizeof(rf_panicbuf),
895 		"raidframe error at line %d file %s (failed asserting %s)\n",
896 		line, file, condition);
897 }
898 #endif
899 
900 void
901 rf_print_unable_to_init_mutex(const char *file, int line, int rc)
902 {
903 	RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n",
904 		     file, line, rc);
905 }
906 
907 void
908 rf_print_unable_to_add_shutdown(const char *file, int line, int rc)
909 {
910 	RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n",
911 		     file, line, rc);
912 }
913 
914 static void
915 rf_alloc_mutex_cond(RF_Raid_t *raidPtr)
916 {
917 
918 	rf_init_mutex2(raidPtr->mutex, IPL_VM);
919 
920 	rf_init_cond2(raidPtr->outstandingCond, "rfocond");
921 	rf_init_mutex2(raidPtr->rad_lock, IPL_VM);
922 
923 	rf_init_mutex2(raidPtr->access_suspend_mutex, IPL_VM);
924 	rf_init_cond2(raidPtr->access_suspend_cv, "rfquiesce");
925 
926 	rf_init_cond2(raidPtr->waitForReconCond, "rfrcnw");
927 
928 	rf_init_cond2(raidPtr->adding_hot_spare_cv, "raidhs");
929 }
930 
931 static void
932 rf_destroy_mutex_cond(RF_Raid_t *raidPtr)
933 {
934 
935 	rf_destroy_cond2(raidPtr->waitForReconCond);
936 	rf_destroy_cond2(raidPtr->adding_hot_spare_cv);
937 
938 	rf_destroy_mutex2(raidPtr->access_suspend_mutex);
939 	rf_destroy_cond2(raidPtr->access_suspend_cv);
940 
941 	rf_destroy_cond2(raidPtr->outstandingCond);
942 	rf_destroy_mutex2(raidPtr->rad_lock);
943 
944 	rf_destroy_mutex2(raidPtr->mutex);
945 }
946