xref: /netbsd-src/sys/dev/raidframe/rf_driver.c (revision 627f7eb200a4419d89b531d55fccd2ee3ffdcde0)
1 /*	$NetBSD: rf_driver.c,v 1.136 2019/10/10 03:43:59 christos 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 1.136 2019/10/10 03:43:59 christos 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 
231 	/* Wait for any parity re-writes to stop... */
232 	while (raidPtr->parity_rewrite_in_progress) {
233 		printf("raid%d: Waiting for parity re-write to exit...\n",
234 		       raidPtr->raidid);
235 		rf_wait_cond2(raidPtr->parity_rewrite_cv, raidPtr->rad_lock);
236 	}
237 	rf_unlock_mutex2(raidPtr->rad_lock);
238 
239 	/* Wait for any reconstruction to stop... */
240 	rf_lock_mutex2(raidPtr->mutex);
241 	while (raidPtr->reconInProgress) {
242 		printf("raid%d: Waiting for reconstruction to stop...\n",
243 		       raidPtr->raidid);
244 		rf_wait_cond2(raidPtr->waitForReconCond, raidPtr->mutex);
245 	}
246 	rf_unlock_mutex2(raidPtr->mutex);
247 
248 	raidPtr->valid = 0;
249 
250 	if (raidPtr->parity_map != NULL)
251 		rf_paritymap_detach(raidPtr);
252 
253 	rf_update_component_labels(raidPtr, RF_FINAL_COMPONENT_UPDATE);
254 
255 	rf_UnconfigureVnodes(raidPtr);
256 
257 	rf_FreeEmergBuffers(raidPtr);
258 
259 	rf_ShutdownList(&raidPtr->shutdownList);
260 
261 	rf_destroy_mutex_cond(raidPtr);
262 
263 	rf_UnconfigureArray();
264 
265 	return (0);
266 }
267 
268 
269 #define DO_INIT_CONFIGURE(f) { \
270 	rc = f (&globalShutdown); \
271 	if (rc) { \
272 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
273 		rf_ShutdownList(&globalShutdown); \
274 		configureCount--; \
275 		rf_unlock_mutex2(configureMutex); \
276 		rf_destroy_mutex2(rf_printf_mutex); \
277 		return(rc); \
278 	} \
279 }
280 
281 #define DO_RAID_FAIL() { \
282 	rf_UnconfigureVnodes(raidPtr); \
283 	rf_FreeEmergBuffers(raidPtr); \
284 	rf_ShutdownList(&raidPtr->shutdownList); \
285 	rf_UnconfigureArray(); \
286 	rf_destroy_mutex_cond(raidPtr); \
287 }
288 
289 #define DO_RAID_INIT_CONFIGURE(f) { \
290 	rc = f (&raidPtr->shutdownList, raidPtr, cfgPtr); \
291 	if (rc) { \
292 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
293 		DO_RAID_FAIL(); \
294 		return(rc); \
295 	} \
296 }
297 
298 int
299 rf_Configure(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr, RF_AutoConfig_t *ac)
300 {
301 	RF_RowCol_t col;
302 	int rc;
303 
304 	rf_lock_mutex2(configureMutex);
305 	configureCount++;
306 	if (isconfigged == 0) {
307 		rf_init_mutex2(rf_printf_mutex, IPL_VM);
308 
309 		/* initialize globals */
310 
311 		DO_INIT_CONFIGURE(rf_ConfigureAllocList);
312 
313 		/*
314 	         * Yes, this does make debugging general to the whole
315 	         * system instead of being array specific. Bummer, drag.
316 		 */
317 		rf_ConfigureDebug(cfgPtr);
318 		DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
319 #if RF_ACC_TRACE > 0
320 		DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
321 #endif
322 		DO_INIT_CONFIGURE(rf_ConfigureMapModule);
323 		DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
324 		DO_INIT_CONFIGURE(rf_ConfigureCallback);
325 		DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
326 		DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
327 		DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
328 		DO_INIT_CONFIGURE(rf_ConfigureMCPair);
329 		DO_INIT_CONFIGURE(rf_ConfigureDAGs);
330 		DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
331 		DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
332 		DO_INIT_CONFIGURE(rf_ConfigureCopyback);
333 		DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
334 		DO_INIT_CONFIGURE(rf_ConfigurePSStatus);
335 		isconfigged = 1;
336 	}
337 	rf_unlock_mutex2(configureMutex);
338 
339 	rf_alloc_mutex_cond(raidPtr);
340 
341 	/* set up the cleanup list.  Do this after ConfigureDebug so that
342 	 * value of memDebug will be set */
343 
344 	rf_MakeAllocList(raidPtr->cleanupList);
345 	if (raidPtr->cleanupList == NULL) {
346 		DO_RAID_FAIL();
347 		return (ENOMEM);
348 	}
349 	rf_ShutdownCreate(&raidPtr->shutdownList,
350 			  (void (*) (void *)) rf_FreeAllocList,
351 			  raidPtr->cleanupList);
352 
353 	raidPtr->numCol = cfgPtr->numCol;
354 	raidPtr->numSpare = cfgPtr->numSpare;
355 
356 	raidPtr->status = rf_rs_optimal;
357 	raidPtr->reconControl = NULL;
358 
359 	DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
360 	DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
361 
362 	raidPtr->nAccOutstanding = 0;
363 	raidPtr->waitShutdown = 0;
364 
365 	if (ac!=NULL) {
366 		/* We have an AutoConfig structure..  Don't do the
367 		   normal disk configuration... call the auto config
368 		   stuff */
369 		rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
370 	} else {
371 		DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
372 		DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
373 	}
374 	/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
375 	 * no. is set */
376 	DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
377 
378 	DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
379 
380 	/* Initialize per-RAID PSS bits */
381 	rf_InitPSStatus(raidPtr);
382 
383 #if RF_INCLUDE_CHAINDECLUSTER > 0
384 	for (col = 0; col < raidPtr->numCol; col++) {
385 		/*
386 		 * XXX better distribution
387 		 */
388 		raidPtr->hist_diskreq[col] = 0;
389 	}
390 #endif
391 	raidPtr->numNewFailures = 0;
392 	raidPtr->copyback_in_progress = 0;
393 	raidPtr->parity_rewrite_in_progress = 0;
394 	raidPtr->adding_hot_spare = 0;
395 	raidPtr->recon_in_progress = 0;
396 
397 	raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
398 
399 	/* autoconfigure and root_partition will actually get filled in
400 	   after the config is done */
401 	raidPtr->autoconfigure = 0;
402 	raidPtr->root_partition = 0;
403 	raidPtr->last_unit = raidPtr->raidid;
404 	raidPtr->config_order = 0;
405 
406 	if (rf_keepAccTotals) {
407 		raidPtr->keep_acc_totals = 1;
408 	}
409 
410 	/* Allocate a bunch of buffers to be used in low-memory conditions */
411 	raidPtr->iobuf = NULL;
412 
413 	rc = rf_AllocEmergBuffers(raidPtr);
414 	if (rc) {
415 		printf("raid%d: Unable to allocate emergency buffers.\n",
416 		       raidPtr->raidid);
417 		DO_RAID_FAIL();
418 		return(rc);
419 	}
420 
421 	/* Set up parity map stuff, if applicable. */
422 #ifndef RF_NO_PARITY_MAP
423 	rf_paritymap_attach(raidPtr, cfgPtr->force);
424 #endif
425 
426 	raidPtr->valid = 1;
427 
428 	printf("raid%d: %s\n", raidPtr->raidid,
429 	       raidPtr->Layout.map->configName);
430 	printf("raid%d: Components:", raidPtr->raidid);
431 
432 	for (col = 0; col < raidPtr->numCol; col++) {
433 		printf(" %s", raidPtr->Disks[col].devname);
434 		if (RF_DEAD_DISK(raidPtr->Disks[col].status)) {
435 			printf("[**FAILED**]");
436 		}
437 	}
438 	printf("\n");
439 	printf("raid%d: Total Sectors: %" PRIu64 " (%" PRIu64 " MB)\n",
440 	       raidPtr->raidid,
441 	       raidPtr->totalSectors,
442 	       (raidPtr->totalSectors / 1024 *
443 				(1 << raidPtr->logBytesPerSector) / 1024));
444 
445 	return (0);
446 }
447 
448 
449 /*
450 
451   Routines to allocate and free the "emergency buffers" for a given
452   RAID set.  These emergency buffers will be used when the kernel runs
453   out of kernel memory.
454 
455  */
456 
457 static int
458 rf_AllocEmergBuffers(RF_Raid_t *raidPtr)
459 {
460 	void *tmpbuf;
461 	RF_VoidPointerListElem_t *vple;
462 	int i;
463 
464 	/* XXX next line needs tuning... */
465 	raidPtr->numEmergencyBuffers = 10 * raidPtr->numCol;
466 #if DEBUG
467 	printf("raid%d: allocating %d buffers of %d bytes.\n",
468 	       raidPtr->raidid,
469 	       raidPtr->numEmergencyBuffers,
470 	       (int)(raidPtr->Layout.sectorsPerStripeUnit <<
471 	       raidPtr->logBytesPerSector));
472 #endif
473 	for (i = 0; i < raidPtr->numEmergencyBuffers; i++) {
474 		tmpbuf = malloc( raidPtr->Layout.sectorsPerStripeUnit <<
475 				 raidPtr->logBytesPerSector,
476 				 M_RAIDFRAME, M_WAITOK);
477 		if (tmpbuf) {
478 			vple = rf_AllocVPListElem();
479 			vple->p= tmpbuf;
480 			vple->next = raidPtr->iobuf;
481 			raidPtr->iobuf = vple;
482 			raidPtr->iobuf_count++;
483 		} else {
484 			printf("raid%d: failed to allocate emergency buffer!\n",
485 			       raidPtr->raidid);
486 			return 1;
487 		}
488 	}
489 
490 	/* XXX next line needs tuning too... */
491 	raidPtr->numEmergencyStripeBuffers = 10;
492         for (i = 0; i < raidPtr->numEmergencyStripeBuffers; i++) {
493                 tmpbuf = malloc( raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit <<
494                                  raidPtr->logBytesPerSector),
495                                  M_RAIDFRAME, M_WAITOK);
496                 if (tmpbuf) {
497                         vple = rf_AllocVPListElem();
498                         vple->p= tmpbuf;
499                         vple->next = raidPtr->stripebuf;
500                         raidPtr->stripebuf = vple;
501                         raidPtr->stripebuf_count++;
502                 } else {
503                         printf("raid%d: failed to allocate emergency stripe buffer!\n",
504                                raidPtr->raidid);
505 			return 1;
506                 }
507         }
508 
509 	return (0);
510 }
511 
512 static void
513 rf_FreeEmergBuffers(RF_Raid_t *raidPtr)
514 {
515 	RF_VoidPointerListElem_t *tmp;
516 
517 	/* Free the emergency IO buffers */
518 	while (raidPtr->iobuf != NULL) {
519 		tmp = raidPtr->iobuf;
520 		raidPtr->iobuf = raidPtr->iobuf->next;
521 		free(tmp->p, M_RAIDFRAME);
522 		rf_FreeVPListElem(tmp);
523 	}
524 
525 	/* Free the emergency stripe buffers */
526 	while (raidPtr->stripebuf != NULL) {
527 		tmp = raidPtr->stripebuf;
528 		raidPtr->stripebuf = raidPtr->stripebuf->next;
529 		free(tmp->p, M_RAIDFRAME);
530 		rf_FreeVPListElem(tmp);
531 	}
532 }
533 
534 
535 static void
536 rf_ShutdownRDFreeList(void *ignored)
537 {
538 	pool_destroy(&rf_pools.rad);
539 }
540 
541 static int
542 rf_ConfigureRDFreeList(RF_ShutdownList_t **listp)
543 {
544 
545 	rf_pool_init(&rf_pools.rad, sizeof(RF_RaidAccessDesc_t),
546 		     "rf_rad_pl", RF_MIN_FREE_RAD, RF_MAX_FREE_RAD);
547 	rf_ShutdownCreate(listp, rf_ShutdownRDFreeList, NULL);
548 	return (0);
549 }
550 
551 RF_RaidAccessDesc_t *
552 rf_AllocRaidAccDesc(RF_Raid_t *raidPtr, RF_IoType_t type,
553 		    RF_RaidAddr_t raidAddress, RF_SectorCount_t numBlocks,
554 		    void *bufPtr, void *bp, RF_RaidAccessFlags_t flags,
555 		    const RF_AccessState_t *states)
556 {
557 	RF_RaidAccessDesc_t *desc;
558 
559 	desc = pool_get(&rf_pools.rad, PR_WAITOK);
560 
561 	rf_lock_mutex2(raidPtr->rad_lock);
562 	if (raidPtr->waitShutdown) {
563 		/*
564 	         * Actually, we're shutting the array down. Free the desc
565 	         * and return NULL.
566 	         */
567 
568 		rf_unlock_mutex2(raidPtr->rad_lock);
569 		pool_put(&rf_pools.rad, desc);
570 		return (NULL);
571 	}
572 	raidPtr->nAccOutstanding++;
573 
574 	rf_unlock_mutex2(raidPtr->rad_lock);
575 
576 	desc->raidPtr = (void *) raidPtr;
577 	desc->type = type;
578 	desc->raidAddress = raidAddress;
579 	desc->numBlocks = numBlocks;
580 	desc->bufPtr = bufPtr;
581 	desc->bp = bp;
582 	desc->flags = flags;
583 	desc->states = states;
584 	desc->state = 0;
585 	desc->dagList = NULL;
586 
587 	desc->status = 0;
588 	desc->numRetries = 0;
589 #if RF_ACC_TRACE > 0
590 	memset(&desc->tracerec, 0, sizeof(desc->tracerec));
591 #endif
592 	desc->callbackFunc = NULL;
593 	desc->callbackArg = NULL;
594 	desc->next = NULL;
595 	desc->iobufs = NULL;
596 	desc->stripebufs = NULL;
597 
598 	return (desc);
599 }
600 
601 void
602 rf_FreeRaidAccDesc(RF_RaidAccessDesc_t *desc)
603 {
604 	RF_Raid_t *raidPtr = desc->raidPtr;
605 	RF_DagList_t *dagList, *temp;
606 	RF_VoidPointerListElem_t *tmp;
607 
608 	RF_ASSERT(desc);
609 
610 	/* Cleanup the dagList(s) */
611 	dagList = desc->dagList;
612 	while(dagList != NULL) {
613 		temp = dagList;
614 		dagList = dagList->next;
615 		rf_FreeDAGList(temp);
616 	}
617 
618 	while (desc->iobufs) {
619 		tmp = desc->iobufs;
620 		desc->iobufs = desc->iobufs->next;
621 		rf_FreeIOBuffer(raidPtr, tmp);
622 	}
623 
624 	while (desc->stripebufs) {
625 		tmp = desc->stripebufs;
626 		desc->stripebufs = desc->stripebufs->next;
627 		rf_FreeStripeBuffer(raidPtr, tmp);
628 	}
629 
630 	pool_put(&rf_pools.rad, desc);
631 	rf_lock_mutex2(raidPtr->rad_lock);
632 	raidPtr->nAccOutstanding--;
633 	if (raidPtr->waitShutdown) {
634 		rf_signal_cond2(raidPtr->outstandingCond);
635 	}
636 	rf_unlock_mutex2(raidPtr->rad_lock);
637 }
638 /*********************************************************************
639  * Main routine for performing an access.
640  * Accesses are retried until a DAG can not be selected.  This occurs
641  * when either the DAG library is incomplete or there are too many
642  * failures in a parity group.
643  *
644  * type should be read or write async_flag should be RF_TRUE or
645  * RF_FALSE bp_in is a buf pointer.  void *to facilitate ignoring it
646  * outside the kernel
647  ********************************************************************/
648 int
649 rf_DoAccess(RF_Raid_t * raidPtr, RF_IoType_t type, int async_flag,
650 	    RF_RaidAddr_t raidAddress, RF_SectorCount_t numBlocks,
651 	    void *bufPtr, struct buf *bp, RF_RaidAccessFlags_t flags)
652 {
653 	RF_RaidAccessDesc_t *desc;
654 	void *lbufPtr = bufPtr;
655 
656 	raidAddress += rf_raidSectorOffset;
657 
658 #if RF_ACCESS_DEBUG
659 	if (rf_accessDebug) {
660 
661 		printf("logBytes is: %d %d %d\n", raidPtr->raidid,
662 		    raidPtr->logBytesPerSector,
663 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks));
664 		printf("raid%d: %s raidAddr %d (stripeid %d-%d) numBlocks %d (%d bytes) buf 0x%lx\n", raidPtr->raidid,
665 		    (type == RF_IO_TYPE_READ) ? "READ" : "WRITE", (int) raidAddress,
666 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress),
667 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress + numBlocks - 1),
668 		    (int) numBlocks,
669 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks),
670 		    (long) bufPtr);
671 	}
672 #endif
673 
674 	desc = rf_AllocRaidAccDesc(raidPtr, type, raidAddress,
675 	    numBlocks, lbufPtr, bp, flags, raidPtr->Layout.map->states);
676 
677 	if (desc == NULL) {
678 		return (ENOMEM);
679 	}
680 #if RF_ACC_TRACE > 0
681 	RF_ETIMER_START(desc->tracerec.tot_timer);
682 #endif
683 	desc->async_flag = async_flag;
684 
685 	if (raidPtr->parity_map != NULL &&
686 	    type == RF_IO_TYPE_WRITE)
687 		rf_paritymap_begin(raidPtr->parity_map, raidAddress,
688 		    numBlocks);
689 
690 	rf_ContinueRaidAccess(desc);
691 
692 	return (0);
693 }
694 #if 0
695 /* force the array into reconfigured mode without doing reconstruction */
696 int
697 rf_SetReconfiguredMode(RF_Raid_t *raidPtr, int col)
698 {
699 	if (!(raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
700 		printf("Can't set reconfigured mode in dedicated-spare array\n");
701 		RF_PANIC();
702 	}
703 	rf_lock_mutex2(raidPtr->mutex);
704 	raidPtr->numFailures++;
705 	raidPtr->Disks[col].status = rf_ds_dist_spared;
706 	raidPtr->status = rf_rs_reconfigured;
707 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
708 	/* install spare table only if declustering + distributed sparing
709 	 * architecture. */
710 	if (raidPtr->Layout.map->flags & RF_BD_DECLUSTERED)
711 		rf_InstallSpareTable(raidPtr, col);
712 	rf_unlock_mutex2(raidPtr->mutex);
713 	return (0);
714 }
715 #endif
716 
717 int
718 rf_FailDisk(RF_Raid_t *raidPtr, int fcol, int initRecon)
719 {
720 
721 	/* need to suspend IO's here -- if there are DAGs in flight
722 	   and we pull the rug out from under ci_vp, Bad Things
723 	   can happen.  */
724 
725 	rf_SuspendNewRequestsAndWait(raidPtr);
726 
727 	rf_lock_mutex2(raidPtr->mutex);
728 	if (raidPtr->Disks[fcol].status != rf_ds_failed) {
729 		/* must be failing something that is valid, or else it's
730 		   already marked as failed (in which case we don't
731 		   want to mark it failed again!) */
732 		raidPtr->numFailures++;
733 		raidPtr->Disks[fcol].status = rf_ds_failed;
734 		raidPtr->status = rf_rs_degraded;
735 	}
736 	rf_unlock_mutex2(raidPtr->mutex);
737 
738 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
739 
740 	/* Close the component, so that it's not "locked" if someone
741 	   else want's to use it! */
742 
743 	rf_close_component(raidPtr, raidPtr->raid_cinfo[fcol].ci_vp,
744 			   raidPtr->Disks[fcol].auto_configured);
745 
746 	rf_lock_mutex2(raidPtr->mutex);
747 	raidPtr->raid_cinfo[fcol].ci_vp = NULL;
748 
749 	/* Need to mark the component as not being auto_configured
750 	   (in case it was previously). */
751 
752 	raidPtr->Disks[fcol].auto_configured = 0;
753 	rf_unlock_mutex2(raidPtr->mutex);
754 	/* now we can allow IO to continue -- we'll be suspending it
755 	   again in rf_ReconstructFailedDisk() if we have to.. */
756 
757 	rf_ResumeNewRequests(raidPtr);
758 
759 	if (initRecon)
760 		rf_ReconstructFailedDisk(raidPtr, fcol);
761 	return (0);
762 }
763 /* releases a thread that is waiting for the array to become quiesced.
764  * access_suspend_mutex should be locked upon calling this
765  */
766 void
767 rf_SignalQuiescenceLock(RF_Raid_t *raidPtr)
768 {
769 #if RF_DEBUG_QUIESCE
770 	if (rf_quiesceDebug) {
771 		printf("raid%d: Signalling quiescence lock\n",
772 		       raidPtr->raidid);
773 	}
774 #endif
775 	raidPtr->access_suspend_release = 1;
776 
777 	if (raidPtr->waiting_for_quiescence) {
778 		SIGNAL_QUIESCENT_COND(raidPtr);
779 	}
780 }
781 /* suspends all new requests to the array.  No effect on accesses that are in flight.  */
782 int
783 rf_SuspendNewRequestsAndWait(RF_Raid_t *raidPtr)
784 {
785 #if RF_DEBUG_QUIESCE
786 	if (rf_quiesceDebug)
787 		printf("raid%d: Suspending new reqs\n", raidPtr->raidid);
788 #endif
789 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
790 	raidPtr->accesses_suspended++;
791 	raidPtr->waiting_for_quiescence = (raidPtr->accs_in_flight == 0) ? 0 : 1;
792 
793 	if (raidPtr->waiting_for_quiescence) {
794 		raidPtr->access_suspend_release = 0;
795 		while (!raidPtr->access_suspend_release) {
796 #if RF_DEBUG_QUIESCE
797 			printf("raid%d: Suspending: Waiting for Quiescence\n",
798 			       raidPtr->raidid);
799 #endif
800 			WAIT_FOR_QUIESCENCE(raidPtr);
801 			raidPtr->waiting_for_quiescence = 0;
802 		}
803 	}
804 #if RF_DEBUG_QUIESCE
805 	printf("raid%d: Quiescence reached..\n", raidPtr->raidid);
806 #endif
807 
808 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
809 	return (raidPtr->waiting_for_quiescence);
810 }
811 /* wake up everyone waiting for quiescence to be released */
812 void
813 rf_ResumeNewRequests(RF_Raid_t *raidPtr)
814 {
815 	RF_CallbackFuncDesc_t *t, *cb;
816 
817 #if RF_DEBUG_QUIESCE
818 	if (rf_quiesceDebug)
819 		printf("raid%d: Resuming new requests\n", raidPtr->raidid);
820 #endif
821 
822 	rf_lock_mutex2(raidPtr->access_suspend_mutex);
823 	raidPtr->accesses_suspended--;
824 	if (raidPtr->accesses_suspended == 0)
825 		cb = raidPtr->quiesce_wait_list;
826 	else
827 		cb = NULL;
828 	raidPtr->quiesce_wait_list = NULL;
829 	rf_unlock_mutex2(raidPtr->access_suspend_mutex);
830 
831 	while (cb) {
832 		t = cb;
833 		cb = cb->next;
834 		(t->callbackFunc) (t->callbackArg);
835 		rf_FreeCallbackFuncDesc(t);
836 	}
837 }
838 /*****************************************************************************************
839  *
840  * debug routines
841  *
842  ****************************************************************************************/
843 
844 static void
845 set_debug_option(char *name, long val)
846 {
847 	RF_DebugName_t *p;
848 
849 	for (p = rf_debugNames; p->name; p++) {
850 		if (!strcmp(p->name, name)) {
851 			*(p->ptr) = val;
852 			printf("[Set debug variable %s to %ld]\n", name, val);
853 			return;
854 		}
855 	}
856 	RF_ERRORMSG1("Unknown debug string \"%s\"\n", name);
857 }
858 
859 
860 /* would like to use sscanf here, but apparently not available in kernel */
861 /*ARGSUSED*/
862 static void
863 rf_ConfigureDebug(RF_Config_t *cfgPtr)
864 {
865 	char   *val_p, *name_p, *white_p;
866 	long    val;
867 	int     i;
868 
869 	rf_ResetDebugOptions();
870 	for (i = 0; i < RF_MAXDBGV && cfgPtr->debugVars[i][0]; i++) {
871 		name_p = rf_find_non_white(&cfgPtr->debugVars[i][0]);
872 		white_p = rf_find_white(name_p);	/* skip to start of 2nd
873 							 * word */
874 		val_p = rf_find_non_white(white_p);
875 		if (*val_p == '0' && *(val_p + 1) == 'x')
876 			val = rf_htoi(val_p + 2);
877 		else
878 			val = rf_atoi(val_p);
879 		*white_p = '\0';
880 		set_debug_option(name_p, val);
881 	}
882 }
883 
884 void
885 rf_print_panic_message(int line, const char *file)
886 {
887 	kern_assert("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 	kern_assert("raidframe error at line %d file %s (failed asserting %s)\n",
895 	    line, file, condition);
896 }
897 #endif
898 
899 void
900 rf_print_unable_to_init_mutex(const char *file, int line, int rc)
901 {
902 	RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n",
903 		     file, line, rc);
904 }
905 
906 void
907 rf_print_unable_to_add_shutdown(const char *file, int line, int rc)
908 {
909 	RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n",
910 		     file, line, rc);
911 }
912 
913 static void
914 rf_alloc_mutex_cond(RF_Raid_t *raidPtr)
915 {
916 
917 	rf_init_mutex2(raidPtr->mutex, IPL_VM);
918 
919 	rf_init_cond2(raidPtr->outstandingCond, "rfocond");
920 	rf_init_cond2(raidPtr->parity_rewrite_cv, "rfprwshutdown");
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->parity_rewrite_cv);
942 	rf_destroy_cond2(raidPtr->outstandingCond);
943 	rf_destroy_mutex2(raidPtr->rad_lock);
944 
945 	rf_destroy_mutex2(raidPtr->mutex);
946 }
947