xref: /netbsd-src/sys/dev/raidframe/rf_driver.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: rf_driver.c,v 1.133 2016/12/10 23:03:27 maya 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 1.133 2016/12/10 23:03:27 maya 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 /* main configuration routines */
125 static int raidframe_booted = 0;
126 
127 static void rf_ConfigureDebug(RF_Config_t * cfgPtr);
128 static void set_debug_option(char *name, long val);
129 static void rf_UnconfigureArray(void);
130 static void rf_ShutdownRDFreeList(void *);
131 static int rf_ConfigureRDFreeList(RF_ShutdownList_t **);
132 
133 rf_declare_mutex2(rf_printf_mutex);	/* debug only:  avoids interleaved
134 					 * printfs by different stripes */
135 
136 #define SIGNAL_QUIESCENT_COND(_raid_) \
137 	rf_broadcast_cond2((_raid_)->access_suspend_cv)
138 #define WAIT_FOR_QUIESCENCE(_raid_) \
139 	rf_wait_cond2((_raid_)->access_suspend_cv, \
140 		      (_raid_)->access_suspend_mutex)
141 
142 static int configureCount = 0;	/* number of active configurations */
143 static int isconfigged = 0;	/* is basic raidframe (non per-array)
144 				 * stuff configured */
145 static rf_declare_mutex2(configureMutex); /* used to lock the configuration
146 					   * stuff */
147 static RF_ShutdownList_t *globalShutdown;	/* non array-specific
148 						 * stuff */
149 
150 static int rf_ConfigureRDFreeList(RF_ShutdownList_t ** listp);
151 static int rf_AllocEmergBuffers(RF_Raid_t *);
152 static void rf_FreeEmergBuffers(RF_Raid_t *);
153 static void rf_destroy_mutex_cond(RF_Raid_t *);
154 static void rf_alloc_mutex_cond(RF_Raid_t *);
155 
156 /* called at system boot time */
157 int
158 rf_BootRaidframe(bool boot)
159 {
160 
161 	if (boot) {
162 		if (raidframe_booted)
163 			return (EBUSY);
164 		raidframe_booted = 1;
165 		rf_init_mutex2(configureMutex, IPL_NONE);
166  		configureCount = 0;
167 		isconfigged = 0;
168 		globalShutdown = NULL;
169 	} else {
170 		rf_destroy_mutex2(configureMutex);
171 		raidframe_booted = 0;
172 	}
173 	return (0);
174 }
175 
176 /*
177  * Called whenever an array is shutdown
178  */
179 static void
180 rf_UnconfigureArray(void)
181 {
182 
183 	rf_lock_mutex2(configureMutex);
184 	if (--configureCount == 0) {	/* if no active configurations, shut
185 					 * everything down */
186 		rf_destroy_mutex2(rf_printf_mutex);
187 		isconfigged = 0;
188 		rf_ShutdownList(&globalShutdown);
189 
190 		/*
191 	         * We must wait until now, because the AllocList module
192 	         * uses the DebugMem module.
193 	         */
194 #if RF_DEBUG_MEM
195 		if (rf_memDebug)
196 			rf_print_unfreed();
197 #endif
198 	}
199 	rf_unlock_mutex2(configureMutex);
200 }
201 
202 /*
203  * Called to shut down an array.
204  */
205 int
206 rf_Shutdown(RF_Raid_t *raidPtr)
207 {
208 
209 	if (!raidPtr->valid) {
210 		RF_ERRORMSG("Attempt to shut down unconfigured RAIDframe driver.  Aborting shutdown\n");
211 		return (EINVAL);
212 	}
213 	/*
214          * wait for outstanding IOs to land
215          * As described in rf_raid.h, we use the rad_freelist lock
216          * to protect the per-array info about outstanding descs
217          * since we need to do freelist locking anyway, and this
218          * cuts down on the amount of serialization we've got going
219          * on.
220          */
221 	rf_lock_mutex2(raidPtr->rad_lock);
222 	if (raidPtr->waitShutdown) {
223 		rf_unlock_mutex2(raidPtr->rad_lock);
224 		return (EBUSY);
225 	}
226 	raidPtr->waitShutdown = 1;
227 	while (raidPtr->nAccOutstanding) {
228 		rf_wait_cond2(raidPtr->outstandingCond, raidPtr->rad_lock);
229 	}
230 	rf_unlock_mutex2(raidPtr->rad_lock);
231 
232 	/* Wait for any parity re-writes to stop... */
233 	while (raidPtr->parity_rewrite_in_progress) {
234 		printf("raid%d: Waiting for parity re-write to exit...\n",
235 		       raidPtr->raidid);
236 		tsleep(&raidPtr->parity_rewrite_in_progress, PRIBIO,
237 		       "rfprwshutdown", 0);
238 	}
239 
240 	/* Wait for any reconstruction to stop... */
241 	rf_lock_mutex2(raidPtr->mutex);
242 	while (raidPtr->reconInProgress) {
243 		printf("raid%d: Waiting for reconstruction to stop...\n",
244 		       raidPtr->raidid);
245 		rf_wait_cond2(raidPtr->waitForReconCond, raidPtr->mutex);
246 	}
247 	rf_unlock_mutex2(raidPtr->mutex);
248 
249 	raidPtr->valid = 0;
250 
251 	if (raidPtr->parity_map != NULL)
252 		rf_paritymap_detach(raidPtr);
253 
254 	rf_update_component_labels(raidPtr, RF_FINAL_COMPONENT_UPDATE);
255 
256 	rf_UnconfigureVnodes(raidPtr);
257 
258 	rf_FreeEmergBuffers(raidPtr);
259 
260 	rf_ShutdownList(&raidPtr->shutdownList);
261 
262 	rf_destroy_mutex_cond(raidPtr);
263 
264 	rf_UnconfigureArray();
265 
266 	return (0);
267 }
268 
269 
270 #define DO_INIT_CONFIGURE(f) { \
271 	rc = f (&globalShutdown); \
272 	if (rc) { \
273 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
274 		rf_ShutdownList(&globalShutdown); \
275 		configureCount--; \
276 		rf_unlock_mutex2(configureMutex); \
277 		rf_destroy_mutex2(rf_printf_mutex); \
278 		return(rc); \
279 	} \
280 }
281 
282 #define DO_RAID_FAIL() { \
283 	rf_UnconfigureVnodes(raidPtr); \
284 	rf_FreeEmergBuffers(raidPtr); \
285 	rf_ShutdownList(&raidPtr->shutdownList); \
286 	rf_UnconfigureArray(); \
287 	rf_destroy_mutex_cond(raidPtr); \
288 }
289 
290 #define DO_RAID_INIT_CONFIGURE(f) { \
291 	rc = f (&raidPtr->shutdownList, raidPtr, cfgPtr); \
292 	if (rc) { \
293 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
294 		DO_RAID_FAIL(); \
295 		return(rc); \
296 	} \
297 }
298 
299 int
300 rf_Configure(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr, RF_AutoConfig_t *ac)
301 {
302 	RF_RowCol_t col;
303 	int rc;
304 
305 	rf_lock_mutex2(configureMutex);
306 	configureCount++;
307 	if (isconfigged == 0) {
308 		rf_init_mutex2(rf_printf_mutex, IPL_VM);
309 
310 		/* initialize globals */
311 
312 		DO_INIT_CONFIGURE(rf_ConfigureAllocList);
313 
314 		/*
315 	         * Yes, this does make debugging general to the whole
316 	         * system instead of being array specific. Bummer, drag.
317 		 */
318 		rf_ConfigureDebug(cfgPtr);
319 		DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
320 #if RF_ACC_TRACE > 0
321 		DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
322 #endif
323 		DO_INIT_CONFIGURE(rf_ConfigureMapModule);
324 		DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
325 		DO_INIT_CONFIGURE(rf_ConfigureCallback);
326 		DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
327 		DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
328 		DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
329 		DO_INIT_CONFIGURE(rf_ConfigureMCPair);
330 		DO_INIT_CONFIGURE(rf_ConfigureDAGs);
331 		DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
332 		DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
333 		DO_INIT_CONFIGURE(rf_ConfigureCopyback);
334 		DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
335 		DO_INIT_CONFIGURE(rf_ConfigurePSStatus);
336 		isconfigged = 1;
337 	}
338 	rf_unlock_mutex2(configureMutex);
339 
340 	rf_alloc_mutex_cond(raidPtr);
341 
342 	/* set up the cleanup list.  Do this after ConfigureDebug so that
343 	 * value of memDebug will be set */
344 
345 	rf_MakeAllocList(raidPtr->cleanupList);
346 	if (raidPtr->cleanupList == NULL) {
347 		DO_RAID_FAIL();
348 		return (ENOMEM);
349 	}
350 	rf_ShutdownCreate(&raidPtr->shutdownList,
351 			  (void (*) (void *)) rf_FreeAllocList,
352 			  raidPtr->cleanupList);
353 
354 	raidPtr->numCol = cfgPtr->numCol;
355 	raidPtr->numSpare = cfgPtr->numSpare;
356 
357 	raidPtr->status = rf_rs_optimal;
358 	raidPtr->reconControl = NULL;
359 
360 	DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
361 	DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
362 
363 	raidPtr->nAccOutstanding = 0;
364 	raidPtr->waitShutdown = 0;
365 
366 	if (ac!=NULL) {
367 		/* We have an AutoConfig structure..  Don't do the
368 		   normal disk configuration... call the auto config
369 		   stuff */
370 		rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
371 	} else {
372 		DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
373 		DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
374 	}
375 	/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
376 	 * no. is set */
377 	DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
378 
379 	DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
380 
381 	/* Initialize per-RAID PSS bits */
382 	rf_InitPSStatus(raidPtr);
383 
384 #if RF_INCLUDE_CHAINDECLUSTER > 0
385 	for (col = 0; col < raidPtr->numCol; col++) {
386 		/*
387 		 * XXX better distribution
388 		 */
389 		raidPtr->hist_diskreq[col] = 0;
390 	}
391 #endif
392 	raidPtr->numNewFailures = 0;
393 	raidPtr->copyback_in_progress = 0;
394 	raidPtr->parity_rewrite_in_progress = 0;
395 	raidPtr->adding_hot_spare = 0;
396 	raidPtr->recon_in_progress = 0;
397 
398 	raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
399 
400 	/* autoconfigure and root_partition will actually get filled in
401 	   after the config is done */
402 	raidPtr->autoconfigure = 0;
403 	raidPtr->root_partition = 0;
404 	raidPtr->last_unit = raidPtr->raidid;
405 	raidPtr->config_order = 0;
406 
407 	if (rf_keepAccTotals) {
408 		raidPtr->keep_acc_totals = 1;
409 	}
410 
411 	/* Allocate a bunch of buffers to be used in low-memory conditions */
412 	raidPtr->iobuf = NULL;
413 
414 	rc = rf_AllocEmergBuffers(raidPtr);
415 	if (rc) {
416 		printf("raid%d: Unable to allocate emergency buffers.\n",
417 		       raidPtr->raidid);
418 		DO_RAID_FAIL();
419 		return(rc);
420 	}
421 
422 	/* Set up parity map stuff, if applicable. */
423 #ifndef RF_NO_PARITY_MAP
424 	rf_paritymap_attach(raidPtr, cfgPtr->force);
425 #endif
426 
427 	raidPtr->valid = 1;
428 
429 	printf("raid%d: %s\n", raidPtr->raidid,
430 	       raidPtr->Layout.map->configName);
431 	printf("raid%d: Components:", raidPtr->raidid);
432 
433 	for (col = 0; col < raidPtr->numCol; col++) {
434 		printf(" %s", raidPtr->Disks[col].devname);
435 		if (RF_DEAD_DISK(raidPtr->Disks[col].status)) {
436 			printf("[**FAILED**]");
437 		}
438 	}
439 	printf("\n");
440 	printf("raid%d: Total Sectors: %" PRIu64 " (%" PRIu64 " MB)\n",
441 	       raidPtr->raidid,
442 	       raidPtr->totalSectors,
443 	       (raidPtr->totalSectors / 1024 *
444 				(1 << raidPtr->logBytesPerSector) / 1024));
445 
446 	return (0);
447 }
448 
449 
450 /*
451 
452   Routines to allocate and free the "emergency buffers" for a given
453   RAID set.  These emergency buffers will be used when the kernel runs
454   out of kernel memory.
455 
456  */
457 
458 static int
459 rf_AllocEmergBuffers(RF_Raid_t *raidPtr)
460 {
461 	void *tmpbuf;
462 	RF_VoidPointerListElem_t *vple;
463 	int i;
464 
465 	/* XXX next line needs tuning... */
466 	raidPtr->numEmergencyBuffers = 10 * raidPtr->numCol;
467 #if DEBUG
468 	printf("raid%d: allocating %d buffers of %d bytes.\n",
469 	       raidPtr->raidid,
470 	       raidPtr->numEmergencyBuffers,
471 	       (int)(raidPtr->Layout.sectorsPerStripeUnit <<
472 	       raidPtr->logBytesPerSector));
473 #endif
474 	for (i = 0; i < raidPtr->numEmergencyBuffers; i++) {
475 		tmpbuf = malloc( raidPtr->Layout.sectorsPerStripeUnit <<
476 				 raidPtr->logBytesPerSector,
477 				 M_RAIDFRAME, M_WAITOK);
478 		if (tmpbuf) {
479 			vple = rf_AllocVPListElem();
480 			vple->p= tmpbuf;
481 			vple->next = raidPtr->iobuf;
482 			raidPtr->iobuf = vple;
483 			raidPtr->iobuf_count++;
484 		} else {
485 			printf("raid%d: failed to allocate emergency buffer!\n",
486 			       raidPtr->raidid);
487 			return 1;
488 		}
489 	}
490 
491 	/* XXX next line needs tuning too... */
492 	raidPtr->numEmergencyStripeBuffers = 10;
493         for (i = 0; i < raidPtr->numEmergencyStripeBuffers; i++) {
494                 tmpbuf = malloc( raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit <<
495                                  raidPtr->logBytesPerSector),
496                                  M_RAIDFRAME, M_WAITOK);
497                 if (tmpbuf) {
498                         vple = rf_AllocVPListElem();
499                         vple->p= tmpbuf;
500                         vple->next = raidPtr->stripebuf;
501                         raidPtr->stripebuf = vple;
502                         raidPtr->stripebuf_count++;
503                 } else {
504                         printf("raid%d: failed to allocate emergency stripe buffer!\n",
505                                raidPtr->raidid);
506 			return 1;
507                 }
508         }
509 
510 	return (0);
511 }
512 
513 static void
514 rf_FreeEmergBuffers(RF_Raid_t *raidPtr)
515 {
516 	RF_VoidPointerListElem_t *tmp;
517 
518 	/* Free the emergency IO buffers */
519 	while (raidPtr->iobuf != NULL) {
520 		tmp = raidPtr->iobuf;
521 		raidPtr->iobuf = raidPtr->iobuf->next;
522 		free(tmp->p, M_RAIDFRAME);
523 		rf_FreeVPListElem(tmp);
524 	}
525 
526 	/* Free the emergency stripe buffers */
527 	while (raidPtr->stripebuf != NULL) {
528 		tmp = raidPtr->stripebuf;
529 		raidPtr->stripebuf = raidPtr->stripebuf->next;
530 		free(tmp->p, M_RAIDFRAME);
531 		rf_FreeVPListElem(tmp);
532 	}
533 }
534 
535 
536 static void
537 rf_ShutdownRDFreeList(void *ignored)
538 {
539 	pool_destroy(&rf_pools.rad);
540 }
541 
542 static int
543 rf_ConfigureRDFreeList(RF_ShutdownList_t **listp)
544 {
545 
546 	rf_pool_init(&rf_pools.rad, sizeof(RF_RaidAccessDesc_t),
547 		     "rf_rad_pl", RF_MIN_FREE_RAD, RF_MAX_FREE_RAD);
548 	rf_ShutdownCreate(listp, rf_ShutdownRDFreeList, NULL);
549 	return (0);
550 }
551 
552 RF_RaidAccessDesc_t *
553 rf_AllocRaidAccDesc(RF_Raid_t *raidPtr, RF_IoType_t type,
554 		    RF_RaidAddr_t raidAddress, RF_SectorCount_t numBlocks,
555 		    void *bufPtr, void *bp, RF_RaidAccessFlags_t flags,
556 		    const RF_AccessState_t *states)
557 {
558 	RF_RaidAccessDesc_t *desc;
559 
560 	desc = pool_get(&rf_pools.rad, PR_WAITOK);
561 
562 	rf_lock_mutex2(raidPtr->rad_lock);
563 	if (raidPtr->waitShutdown) {
564 		/*
565 	         * Actually, we're shutting the array down. Free the desc
566 	         * and return NULL.
567 	         */
568 
569 		rf_unlock_mutex2(raidPtr->rad_lock);
570 		pool_put(&rf_pools.rad, desc);
571 		return (NULL);
572 	}
573 	raidPtr->nAccOutstanding++;
574 
575 	rf_unlock_mutex2(raidPtr->rad_lock);
576 
577 	desc->raidPtr = (void *) raidPtr;
578 	desc->type = type;
579 	desc->raidAddress = raidAddress;
580 	desc->numBlocks = numBlocks;
581 	desc->bufPtr = bufPtr;
582 	desc->bp = bp;
583 	desc->flags = flags;
584 	desc->states = states;
585 	desc->state = 0;
586 	desc->dagList = NULL;
587 
588 	desc->status = 0;
589 	desc->numRetries = 0;
590 #if RF_ACC_TRACE > 0
591 	memset((char *) &desc->tracerec, 0, sizeof(RF_AccTraceEntry_t));
592 #endif
593 	desc->callbackFunc = NULL;
594 	desc->callbackArg = NULL;
595 	desc->next = NULL;
596 	desc->iobufs = NULL;
597 	desc->stripebufs = NULL;
598 
599 	return (desc);
600 }
601 
602 void
603 rf_FreeRaidAccDesc(RF_RaidAccessDesc_t *desc)
604 {
605 	RF_Raid_t *raidPtr = desc->raidPtr;
606 	RF_DagList_t *dagList, *temp;
607 	RF_VoidPointerListElem_t *tmp;
608 
609 	RF_ASSERT(desc);
610 
611 	/* Cleanup the dagList(s) */
612 	dagList = desc->dagList;
613 	while(dagList != NULL) {
614 		temp = dagList;
615 		dagList = dagList->next;
616 		rf_FreeDAGList(temp);
617 	}
618 
619 	while (desc->iobufs) {
620 		tmp = desc->iobufs;
621 		desc->iobufs = desc->iobufs->next;
622 		rf_FreeIOBuffer(raidPtr, tmp);
623 	}
624 
625 	while (desc->stripebufs) {
626 		tmp = desc->stripebufs;
627 		desc->stripebufs = desc->stripebufs->next;
628 		rf_FreeStripeBuffer(raidPtr, tmp);
629 	}
630 
631 	pool_put(&rf_pools.rad, desc);
632 	rf_lock_mutex2(raidPtr->rad_lock);
633 	raidPtr->nAccOutstanding--;
634 	if (raidPtr->waitShutdown) {
635 		rf_signal_cond2(raidPtr->outstandingCond);
636 	}
637 	rf_unlock_mutex2(raidPtr->rad_lock);
638 }
639 /*********************************************************************
640  * Main routine for performing an access.
641  * Accesses are retried until a DAG can not be selected.  This occurs
642  * when either the DAG library is incomplete or there are too many
643  * failures in a parity group.
644  *
645  * type should be read or write async_flag should be RF_TRUE or
646  * RF_FALSE bp_in is a buf pointer.  void *to facilitate ignoring it
647  * outside the kernel
648  ********************************************************************/
649 int
650 rf_DoAccess(RF_Raid_t * raidPtr, RF_IoType_t type, int async_flag,
651 	    RF_RaidAddr_t raidAddress, RF_SectorCount_t numBlocks,
652 	    void *bufPtr, struct buf *bp, RF_RaidAccessFlags_t flags)
653 {
654 	RF_RaidAccessDesc_t *desc;
655 	void *lbufPtr = bufPtr;
656 
657 	raidAddress += rf_raidSectorOffset;
658 
659 #if RF_ACCESS_DEBUG
660 	if (rf_accessDebug) {
661 
662 		printf("logBytes is: %d %d %d\n", raidPtr->raidid,
663 		    raidPtr->logBytesPerSector,
664 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks));
665 		printf("raid%d: %s raidAddr %d (stripeid %d-%d) numBlocks %d (%d bytes) buf 0x%lx\n", raidPtr->raidid,
666 		    (type == RF_IO_TYPE_READ) ? "READ" : "WRITE", (int) raidAddress,
667 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress),
668 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress + numBlocks - 1),
669 		    (int) numBlocks,
670 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks),
671 		    (long) bufPtr);
672 	}
673 #endif
674 
675 	desc = rf_AllocRaidAccDesc(raidPtr, type, raidAddress,
676 	    numBlocks, lbufPtr, bp, flags, raidPtr->Layout.map->states);
677 
678 	if (desc == NULL) {
679 		return (ENOMEM);
680 	}
681 #if RF_ACC_TRACE > 0
682 	RF_ETIMER_START(desc->tracerec.tot_timer);
683 #endif
684 	desc->async_flag = async_flag;
685 
686 	if (raidPtr->parity_map != NULL &&
687 	    type == RF_IO_TYPE_WRITE)
688 		rf_paritymap_begin(raidPtr->parity_map, raidAddress,
689 		    numBlocks);
690 
691 	rf_ContinueRaidAccess(desc);
692 
693 	return (0);
694 }
695 #if 0
696 /* force the array into reconfigured mode without doing reconstruction */
697 int
698 rf_SetReconfiguredMode(RF_Raid_t *raidPtr, int col)
699 {
700 	if (!(raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
701 		printf("Can't set reconfigured mode in dedicated-spare array\n");
702 		RF_PANIC();
703 	}
704 	rf_lock_mutex2(raidPtr->mutex);
705 	raidPtr->numFailures++;
706 	raidPtr->Disks[col].status = rf_ds_dist_spared;
707 	raidPtr->status = rf_rs_reconfigured;
708 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
709 	/* install spare table only if declustering + distributed sparing
710 	 * architecture. */
711 	if (raidPtr->Layout.map->flags & RF_BD_DECLUSTERED)
712 		rf_InstallSpareTable(raidPtr, col);
713 	rf_unlock_mutex2(raidPtr->mutex);
714 	return (0);
715 }
716 #endif
717 
718 int
719 rf_FailDisk(RF_Raid_t *raidPtr, int fcol, int initRecon)
720 {
721 
722 	/* need to suspend IO's here -- if there are DAGs in flight
723 	   and we pull the rug out from under ci_vp, Bad Things
724 	   can happen.  */
725 
726 	rf_SuspendNewRequestsAndWait(raidPtr);
727 
728 	rf_lock_mutex2(raidPtr->mutex);
729 	if (raidPtr->Disks[fcol].status != rf_ds_failed) {
730 		/* must be failing something that is valid, or else it's
731 		   already marked as failed (in which case we don't
732 		   want to mark it failed again!) */
733 		raidPtr->numFailures++;
734 		raidPtr->Disks[fcol].status = rf_ds_failed;
735 		raidPtr->status = rf_rs_degraded;
736 	}
737 	rf_unlock_mutex2(raidPtr->mutex);
738 
739 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
740 
741 	/* Close the component, so that it's not "locked" if someone
742 	   else want's to use it! */
743 
744 	rf_close_component(raidPtr, raidPtr->raid_cinfo[fcol].ci_vp,
745 			   raidPtr->Disks[fcol].auto_configured);
746 
747 	rf_lock_mutex2(raidPtr->mutex);
748 	raidPtr->raid_cinfo[fcol].ci_vp = NULL;
749 
750 	/* Need to mark the component as not being auto_configured
751 	   (in case it was previously). */
752 
753 	raidPtr->Disks[fcol].auto_configured = 0;
754 	rf_unlock_mutex2(raidPtr->mutex);
755 	/* now we can allow IO to continue -- we'll be suspending it
756 	   again in rf_ReconstructFailedDisk() if we have to.. */
757 
758 	rf_ResumeNewRequests(raidPtr);
759 
760 	if (initRecon)
761 		rf_ReconstructFailedDisk(raidPtr, fcol);
762 	return (0);
763 }
764 /* releases a thread that is waiting for the array to become quiesced.
765  * access_suspend_mutex should be locked upon calling this
766  */
767 void
768 rf_SignalQuiescenceLock(RF_Raid_t *raidPtr)
769 {
770 #if RF_DEBUG_QUIESCE
771 	if (rf_quiesceDebug) {
772 		printf("raid%d: Signalling quiescence lock\n",
773 		       raidPtr->raidid);
774 	}
775 #endif
776 	raidPtr->access_suspend_release = 1;
777 
778 	if (raidPtr->waiting_for_quiescence) {
779 		SIGNAL_QUIESCENT_COND(raidPtr);
780 	}
781 }
782 /* suspends all new requests to the array.  No effect on accesses that are in flight.  */
783 int
784 rf_SuspendNewRequestsAndWait(RF_Raid_t *raidPtr)
785 {
786 #if RF_DEBUG_QUIESCE
787 	if (rf_quiesceDebug)
788 		printf("raid%d: Suspending new reqs\n", raidPtr->raidid);
789 #endif
790 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
791 	raidPtr->accesses_suspended++;
792 	raidPtr->waiting_for_quiescence = (raidPtr->accs_in_flight == 0) ? 0 : 1;
793 
794 	if (raidPtr->waiting_for_quiescence) {
795 		raidPtr->access_suspend_release = 0;
796 		while (!raidPtr->access_suspend_release) {
797 #if RF_DEBUG_QUIESCE
798 			printf("raid%d: Suspending: Waiting for Quiescence\n",
799 			       raidPtr->raidid);
800 #endif
801 			WAIT_FOR_QUIESCENCE(raidPtr);
802 			raidPtr->waiting_for_quiescence = 0;
803 		}
804 	}
805 #if RF_DEBUG_QUIESCE
806 	printf("raid%d: Quiescence reached..\n", raidPtr->raidid);
807 #endif
808 
809 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
810 	return (raidPtr->waiting_for_quiescence);
811 }
812 /* wake up everyone waiting for quiescence to be released */
813 void
814 rf_ResumeNewRequests(RF_Raid_t *raidPtr)
815 {
816 	RF_CallbackDesc_t *t, *cb;
817 
818 #if RF_DEBUG_QUIESCE
819 	if (rf_quiesceDebug)
820 		printf("raid%d: Resuming new requests\n", raidPtr->raidid);
821 #endif
822 
823 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
824 	raidPtr->accesses_suspended--;
825 	if (raidPtr->accesses_suspended == 0)
826 		cb = raidPtr->quiesce_wait_list;
827 	else
828 		cb = NULL;
829 	raidPtr->quiesce_wait_list = NULL;
830 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
831 
832 	while (cb) {
833 		t = cb;
834 		cb = cb->next;
835 		(t->callbackFunc) (t->callbackArg);
836 		rf_FreeCallbackDesc(t);
837 	}
838 }
839 /*****************************************************************************************
840  *
841  * debug routines
842  *
843  ****************************************************************************************/
844 
845 static void
846 set_debug_option(char *name, long val)
847 {
848 	RF_DebugName_t *p;
849 
850 	for (p = rf_debugNames; p->name; p++) {
851 		if (!strcmp(p->name, name)) {
852 			*(p->ptr) = val;
853 			printf("[Set debug variable %s to %ld]\n", name, val);
854 			return;
855 		}
856 	}
857 	RF_ERRORMSG1("Unknown debug string \"%s\"\n", name);
858 }
859 
860 
861 /* would like to use sscanf here, but apparently not available in kernel */
862 /*ARGSUSED*/
863 static void
864 rf_ConfigureDebug(RF_Config_t *cfgPtr)
865 {
866 	char   *val_p, *name_p, *white_p;
867 	long    val;
868 	int     i;
869 
870 	rf_ResetDebugOptions();
871 	for (i = 0; i < RF_MAXDBGV && cfgPtr->debugVars[i][0]; i++) {
872 		name_p = rf_find_non_white(&cfgPtr->debugVars[i][0]);
873 		white_p = rf_find_white(name_p);	/* skip to start of 2nd
874 							 * word */
875 		val_p = rf_find_non_white(white_p);
876 		if (*val_p == '0' && *(val_p + 1) == 'x')
877 			val = rf_htoi(val_p + 2);
878 		else
879 			val = rf_atoi(val_p);
880 		*white_p = '\0';
881 		set_debug_option(name_p, val);
882 	}
883 }
884 
885 void
886 rf_print_panic_message(int line, const char *file)
887 {
888 	kern_assert("raidframe error at line %d file %s", line, file);
889 }
890 
891 #ifdef RAID_DIAGNOSTIC
892 void
893 rf_print_assert_panic_message(int line,	const char *file, const char *condition)
894 {
895 	kern_assert("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