xref: /netbsd-src/sys/dev/raidframe/rf_driver.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: rf_driver.c,v 1.41 2001/07/27 03:30:07 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 	ltsleep(&((_raid_)->accesses_suspended), PRIBIO, \
146 		"raidframe quiesce", 0, &((_raid_)->access_suspend_mutex))
147 
148 #define IO_BUF_ERR(bp, err) { \
149 	bp->b_flags |= B_ERROR; \
150 	bp->b_resid = bp->b_bcount; \
151 	bp->b_error = err; \
152 	biodone(bp); \
153 }
154 
155 static int configureCount = 0;	/* number of active configurations */
156 static int isconfigged = 0;	/* is basic raidframe (non per-array)
157 				 * stuff configged */
158 RF_DECLARE_STATIC_MUTEX(configureMutex)	/* used to lock the configuration
159 					 * stuff */
160 static RF_ShutdownList_t *globalShutdown;	/* non array-specific
161 						 * stuff */
162 
163 static int rf_ConfigureRDFreeList(RF_ShutdownList_t ** listp);
164 
165 /* called at system boot time */
166 int
167 rf_BootRaidframe()
168 {
169 	int     rc;
170 
171 	if (raidframe_booted)
172 		return (EBUSY);
173 	raidframe_booted = 1;
174 
175 	rc = rf_mutex_init(&configureMutex);
176 	if (rc) {
177 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
178 		    __LINE__, rc);
179 		RF_PANIC();
180 	}
181 	configureCount = 0;
182 	isconfigged = 0;
183 	globalShutdown = NULL;
184 	return (0);
185 }
186 /*
187  * This function is really just for debugging user-level stuff: it
188  * frees up all memory, other RAIDframe resources which might otherwise
189  * be kept around. This is used with systems like "sentinel" to detect
190  * memory leaks.
191  */
192 int
193 rf_UnbootRaidframe()
194 {
195 	int     rc;
196 
197 	RF_LOCK_MUTEX(configureMutex);
198 	if (configureCount) {
199 		RF_UNLOCK_MUTEX(configureMutex);
200 		return (EBUSY);
201 	}
202 	raidframe_booted = 0;
203 	RF_UNLOCK_MUTEX(configureMutex);
204 	rc = rf_mutex_destroy(&configureMutex);
205 	if (rc) {
206 		RF_ERRORMSG3("Unable to destroy mutex file %s line %d rc=%d\n", __FILE__,
207 		    __LINE__, rc);
208 		RF_PANIC();
209 	}
210 	return (0);
211 }
212 /*
213  * Called whenever an array is shutdown
214  */
215 static void
216 rf_UnconfigureArray()
217 {
218 	int     rc;
219 
220 	RF_LOCK_MUTEX(configureMutex);
221 	if (--configureCount == 0) {	/* if no active configurations, shut
222 					 * everything down */
223 		isconfigged = 0;
224 
225 		rc = rf_ShutdownList(&globalShutdown);
226 		if (rc) {
227 			RF_ERRORMSG1("RAIDFRAME: unable to do global shutdown, rc=%d\n", rc);
228 		}
229 
230 		/*
231 	         * We must wait until now, because the AllocList module
232 	         * uses the DebugMem module.
233 	         */
234 		if (rf_memDebug)
235 			rf_print_unfreed();
236 	}
237 	RF_UNLOCK_MUTEX(configureMutex);
238 }
239 
240 /*
241  * Called to shut down an array.
242  */
243 int
244 rf_Shutdown(raidPtr)
245 	RF_Raid_t *raidPtr;
246 {
247 
248 	if (!raidPtr->valid) {
249 		RF_ERRORMSG("Attempt to shut down unconfigured RAIDframe driver.  Aborting shutdown\n");
250 		return (EINVAL);
251 	}
252 	/*
253          * wait for outstanding IOs to land
254          * As described in rf_raid.h, we use the rad_freelist lock
255          * to protect the per-array info about outstanding descs
256          * since we need to do freelist locking anyway, and this
257          * cuts down on the amount of serialization we've got going
258          * on.
259          */
260 	RF_FREELIST_DO_LOCK(rf_rad_freelist);
261 	if (raidPtr->waitShutdown) {
262 		RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
263 		return (EBUSY);
264 	}
265 	raidPtr->waitShutdown = 1;
266 	while (raidPtr->nAccOutstanding) {
267 		RF_WAIT_COND(raidPtr->outstandingCond, RF_FREELIST_MUTEX_OF(rf_rad_freelist));
268 	}
269 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
270 
271 	/* Wait for any parity re-writes to stop... */
272 	while (raidPtr->parity_rewrite_in_progress) {
273 		printf("Waiting for parity re-write to exit...\n");
274 		tsleep(&raidPtr->parity_rewrite_in_progress, PRIBIO,
275 		       "rfprwshutdown", 0);
276 	}
277 
278 	raidPtr->valid = 0;
279 
280 	rf_update_component_labels(raidPtr, RF_FINAL_COMPONENT_UPDATE);
281 
282 	rf_UnconfigureVnodes(raidPtr);
283 
284 	rf_ShutdownList(&raidPtr->shutdownList);
285 
286 	rf_UnconfigureArray();
287 
288 	return (0);
289 }
290 
291 
292 #define DO_INIT_CONFIGURE(f) { \
293 	rc = f (&globalShutdown); \
294 	if (rc) { \
295 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
296 		rf_ShutdownList(&globalShutdown); \
297 		configureCount--; \
298 		RF_UNLOCK_MUTEX(configureMutex); \
299 		return(rc); \
300 	} \
301 }
302 
303 #define DO_RAID_FAIL() { \
304 	rf_UnconfigureVnodes(raidPtr); \
305 	rf_ShutdownList(&raidPtr->shutdownList); \
306 	rf_UnconfigureArray(); \
307 }
308 
309 #define DO_RAID_INIT_CONFIGURE(f) { \
310 	rc = f (&raidPtr->shutdownList, raidPtr, cfgPtr); \
311 	if (rc) { \
312 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
313 		DO_RAID_FAIL(); \
314 		return(rc); \
315 	} \
316 }
317 
318 #define DO_RAID_MUTEX(_m_) { \
319 	rc = rf_create_managed_mutex(&raidPtr->shutdownList, (_m_)); \
320 	if (rc) { \
321 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", \
322 			__FILE__, __LINE__, rc); \
323 		DO_RAID_FAIL(); \
324 		return(rc); \
325 	} \
326 }
327 
328 #define DO_RAID_COND(_c_) { \
329 	rc = rf_create_managed_cond(&raidPtr->shutdownList, (_c_)); \
330 	if (rc) { \
331 		RF_ERRORMSG3("Unable to init cond file %s line %d rc=%d\n", \
332 			__FILE__, __LINE__, rc); \
333 		DO_RAID_FAIL(); \
334 		return(rc); \
335 	} \
336 }
337 
338 int
339 rf_Configure(raidPtr, cfgPtr, ac)
340 	RF_Raid_t *raidPtr;
341 	RF_Config_t *cfgPtr;
342 	RF_AutoConfig_t *ac;
343 {
344 	RF_RowCol_t row, col;
345 	int     i, rc;
346 
347 	/* XXX This check can probably be removed now, since
348 	   RAIDFRAME_CONFIGURRE now checks to make sure that the
349 	   RAID set is not already valid
350 	*/
351 	if (raidPtr->valid) {
352 		RF_ERRORMSG("RAIDframe configuration not shut down.  Aborting configure.\n");
353 		return (EINVAL);
354 	}
355 	RF_LOCK_MUTEX(configureMutex);
356 	configureCount++;
357 	if (isconfigged == 0) {
358 		rc = rf_create_managed_mutex(&globalShutdown, &rf_printf_mutex);
359 		if (rc) {
360 			RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
361 			    __LINE__, rc);
362 			rf_ShutdownList(&globalShutdown);
363 			return (rc);
364 		}
365 		/* initialize globals */
366 		printf("RAIDFRAME: protectedSectors is %ld\n",
367 		       rf_protectedSectors);
368 
369 		rf_clear_debug_print_buffer();
370 
371 		DO_INIT_CONFIGURE(rf_ConfigureAllocList);
372 
373 		/*
374 	         * Yes, this does make debugging general to the whole
375 	         * system instead of being array specific. Bummer, drag.
376 		 */
377 		rf_ConfigureDebug(cfgPtr);
378 		DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
379 		DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
380 		DO_INIT_CONFIGURE(rf_ConfigureMapModule);
381 		DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
382 		DO_INIT_CONFIGURE(rf_ConfigureCallback);
383 		DO_INIT_CONFIGURE(rf_ConfigureMemChunk);
384 		DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
385 		DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
386 		DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
387 		DO_INIT_CONFIGURE(rf_ConfigureMCPair);
388 		DO_INIT_CONFIGURE(rf_ConfigureDAGs);
389 		DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
390 		DO_INIT_CONFIGURE(rf_ConfigureDebugPrint);
391 		DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
392 		DO_INIT_CONFIGURE(rf_ConfigureCopyback);
393 		DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
394 		isconfigged = 1;
395 	}
396 	RF_UNLOCK_MUTEX(configureMutex);
397 
398 	DO_RAID_MUTEX(&raidPtr->mutex);
399 	/* set up the cleanup list.  Do this after ConfigureDebug so that
400 	 * value of memDebug will be set */
401 
402 	rf_MakeAllocList(raidPtr->cleanupList);
403 	if (raidPtr->cleanupList == NULL) {
404 		DO_RAID_FAIL();
405 		return (ENOMEM);
406 	}
407 	rc = rf_ShutdownCreate(&raidPtr->shutdownList,
408 	    (void (*) (void *)) rf_FreeAllocList,
409 	    raidPtr->cleanupList);
410 	if (rc) {
411 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n",
412 		    __FILE__, __LINE__, rc);
413 		DO_RAID_FAIL();
414 		return (rc);
415 	}
416 	raidPtr->numRow = cfgPtr->numRow;
417 	raidPtr->numCol = cfgPtr->numCol;
418 	raidPtr->numSpare = cfgPtr->numSpare;
419 
420 	/* XXX we don't even pretend to support more than one row in the
421 	 * kernel... */
422 	if (raidPtr->numRow != 1) {
423 		RF_ERRORMSG("Only one row supported in kernel.\n");
424 		DO_RAID_FAIL();
425 		return (EINVAL);
426 	}
427 	RF_CallocAndAdd(raidPtr->status, raidPtr->numRow, sizeof(RF_RowStatus_t),
428 	    (RF_RowStatus_t *), raidPtr->cleanupList);
429 	if (raidPtr->status == NULL) {
430 		DO_RAID_FAIL();
431 		return (ENOMEM);
432 	}
433 	RF_CallocAndAdd(raidPtr->reconControl, raidPtr->numRow,
434 	    sizeof(RF_ReconCtrl_t *), (RF_ReconCtrl_t **), raidPtr->cleanupList);
435 	if (raidPtr->reconControl == NULL) {
436 		DO_RAID_FAIL();
437 		return (ENOMEM);
438 	}
439 	for (i = 0; i < raidPtr->numRow; i++) {
440 		raidPtr->status[i] = rf_rs_optimal;
441 		raidPtr->reconControl[i] = NULL;
442 	}
443 
444 	DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
445 	DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
446 
447 	DO_RAID_COND(&raidPtr->outstandingCond);
448 
449 	raidPtr->nAccOutstanding = 0;
450 	raidPtr->waitShutdown = 0;
451 
452 	DO_RAID_MUTEX(&raidPtr->access_suspend_mutex);
453 	DO_RAID_COND(&raidPtr->quiescent_cond);
454 
455 	DO_RAID_COND(&raidPtr->waitForReconCond);
456 
457 	DO_RAID_MUTEX(&raidPtr->recon_done_proc_mutex);
458 
459 	if (ac!=NULL) {
460 		/* We have an AutoConfig structure..  Don't do the
461 		   normal disk configuration... call the auto config
462 		   stuff */
463 		rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
464 	} else {
465 		DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
466 		DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
467 	}
468 	/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
469 	 * no. is set */
470 	DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
471 
472 	DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
473 
474 	DO_RAID_INIT_CONFIGURE(rf_ConfigurePSStatus);
475 
476 	for (row = 0; row < raidPtr->numRow; row++) {
477 		for (col = 0; col < raidPtr->numCol; col++) {
478 			/*
479 		         * XXX better distribution
480 		         */
481 			raidPtr->hist_diskreq[row][col] = 0;
482 		}
483 	}
484 
485 	raidPtr->numNewFailures = 0;
486 	raidPtr->copyback_in_progress = 0;
487 	raidPtr->parity_rewrite_in_progress = 0;
488 	raidPtr->recon_in_progress = 0;
489 	raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
490 
491 	/* autoconfigure and root_partition will actually get filled in
492 	   after the config is done */
493 	raidPtr->autoconfigure = 0;
494 	raidPtr->root_partition = 0;
495 	raidPtr->last_unit = raidPtr->raidid;
496 	raidPtr->config_order = 0;
497 
498 	if (rf_keepAccTotals) {
499 		raidPtr->keep_acc_totals = 1;
500 	}
501 	rf_StartUserStats(raidPtr);
502 
503 	raidPtr->valid = 1;
504 	return (0);
505 }
506 
507 static int
508 init_rad(desc)
509 	RF_RaidAccessDesc_t *desc;
510 {
511 	int     rc;
512 
513 	rc = rf_mutex_init(&desc->mutex);
514 	if (rc) {
515 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
516 		    __LINE__, rc);
517 		return (rc);
518 	}
519 	rc = rf_cond_init(&desc->cond);
520 	if (rc) {
521 		RF_ERRORMSG3("Unable to init cond file %s line %d rc=%d\n", __FILE__,
522 		    __LINE__, rc);
523 		rf_mutex_destroy(&desc->mutex);
524 		return (rc);
525 	}
526 	return (0);
527 }
528 
529 static void
530 clean_rad(desc)
531 	RF_RaidAccessDesc_t *desc;
532 {
533 	rf_mutex_destroy(&desc->mutex);
534 	rf_cond_destroy(&desc->cond);
535 }
536 
537 static void
538 rf_ShutdownRDFreeList(ignored)
539 	void   *ignored;
540 {
541 	RF_FREELIST_DESTROY_CLEAN(rf_rad_freelist, next, (RF_RaidAccessDesc_t *), clean_rad);
542 }
543 
544 static int
545 rf_ConfigureRDFreeList(listp)
546 	RF_ShutdownList_t **listp;
547 {
548 	int     rc;
549 
550 	RF_FREELIST_CREATE(rf_rad_freelist, RF_MAX_FREE_RAD,
551 	    RF_RAD_INC, sizeof(RF_RaidAccessDesc_t));
552 	if (rf_rad_freelist == NULL) {
553 		return (ENOMEM);
554 	}
555 	rc = rf_ShutdownCreate(listp, rf_ShutdownRDFreeList, NULL);
556 	if (rc) {
557 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
558 		    __LINE__, rc);
559 		rf_ShutdownRDFreeList(NULL);
560 		return (rc);
561 	}
562 	RF_FREELIST_PRIME_INIT(rf_rad_freelist, RF_RAD_INITIAL, next,
563 	    (RF_RaidAccessDesc_t *), init_rad);
564 	return (0);
565 }
566 
567 RF_RaidAccessDesc_t *
568 rf_AllocRaidAccDesc(
569     RF_Raid_t * raidPtr,
570     RF_IoType_t type,
571     RF_RaidAddr_t raidAddress,
572     RF_SectorCount_t numBlocks,
573     caddr_t bufPtr,
574     void *bp,
575     RF_RaidAccessFlags_t flags,
576     RF_AccessState_t * states)
577 {
578 	RF_RaidAccessDesc_t *desc;
579 
580 	RF_FREELIST_GET_INIT_NOUNLOCK(rf_rad_freelist, desc, next, (RF_RaidAccessDesc_t *), init_rad);
581 	if (raidPtr->waitShutdown) {
582 		/*
583 	         * Actually, we're shutting the array down. Free the desc
584 	         * and return NULL.
585 	         */
586 		RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
587 		RF_FREELIST_FREE_CLEAN(rf_rad_freelist, desc, next, clean_rad);
588 		return (NULL);
589 	}
590 	raidPtr->nAccOutstanding++;
591 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
592 
593 	desc->raidPtr = (void *) raidPtr;
594 	desc->type = type;
595 	desc->raidAddress = raidAddress;
596 	desc->numBlocks = numBlocks;
597 	desc->bufPtr = bufPtr;
598 	desc->bp = bp;
599 	desc->paramDAG = NULL;
600 	desc->paramASM = NULL;
601 	desc->flags = flags;
602 	desc->states = states;
603 	desc->state = 0;
604 
605 	desc->status = 0;
606 	memset((char *) &desc->tracerec, 0, sizeof(RF_AccTraceEntry_t));
607 	desc->callbackFunc = NULL;
608 	desc->callbackArg = NULL;
609 	desc->next = NULL;
610 	desc->head = desc;
611 	desc->numPending = 0;
612 	desc->cleanupList = NULL;
613 	rf_MakeAllocList(desc->cleanupList);
614 	return (desc);
615 }
616 
617 void
618 rf_FreeRaidAccDesc(RF_RaidAccessDesc_t * desc)
619 {
620 	RF_Raid_t *raidPtr = desc->raidPtr;
621 
622 	RF_ASSERT(desc);
623 
624 	rf_FreeAllocList(desc->cleanupList);
625 	RF_FREELIST_FREE_CLEAN_NOUNLOCK(rf_rad_freelist, desc, next, clean_rad);
626 	raidPtr->nAccOutstanding--;
627 	if (raidPtr->waitShutdown) {
628 		RF_SIGNAL_COND(raidPtr->outstandingCond);
629 	}
630 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
631 }
632 /*********************************************************************
633  * Main routine for performing an access.
634  * Accesses are retried until a DAG can not be selected.  This occurs
635  * when either the DAG library is incomplete or there are too many
636  * failures in a parity group.
637  ********************************************************************/
638 int
639 rf_DoAccess(
640     RF_Raid_t * raidPtr,
641     RF_IoType_t type,
642     int async_flag,
643     RF_RaidAddr_t raidAddress,
644     RF_SectorCount_t numBlocks,
645     caddr_t bufPtr,
646     void *bp_in,
647     RF_RaidAccessFlags_t flags)
648 /*
649 type should be read or write
650 async_flag should be RF_TRUE or RF_FALSE
651 bp_in is a buf pointer.  void * to facilitate ignoring it outside the kernel
652 */
653 {
654 	RF_RaidAccessDesc_t *desc;
655 	caddr_t lbufPtr = bufPtr;
656 	struct buf *bp = (struct buf *) bp_in;
657 
658 	raidAddress += rf_raidSectorOffset;
659 
660 	if (!raidPtr->valid) {
661 		RF_ERRORMSG("RAIDframe driver not successfully configured.  Rejecting access.\n");
662 		IO_BUF_ERR(bp, EINVAL);
663 		return (EINVAL);
664 	}
665 
666 	if (rf_accessDebug) {
667 
668 		printf("logBytes is: %d %d %d\n", raidPtr->raidid,
669 		    raidPtr->logBytesPerSector,
670 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks));
671 		printf("raid%d: %s raidAddr %d (stripeid %d-%d) numBlocks %d (%d bytes) buf 0x%lx\n", raidPtr->raidid,
672 		    (type == RF_IO_TYPE_READ) ? "READ" : "WRITE", (int) raidAddress,
673 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress),
674 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress + numBlocks - 1),
675 		    (int) numBlocks,
676 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks),
677 		    (long) bufPtr);
678 	}
679 	if (raidAddress + numBlocks > raidPtr->totalSectors) {
680 
681 		printf("DoAccess: raid addr %lu too large to access %lu sectors.  Max legal addr is %lu\n",
682 		    (u_long) raidAddress, (u_long) numBlocks, (u_long) raidPtr->totalSectors);
683 
684 		IO_BUF_ERR(bp, ENOSPC);
685 		return (ENOSPC);
686 	}
687 	desc = rf_AllocRaidAccDesc(raidPtr, type, raidAddress,
688 	    numBlocks, lbufPtr, bp, flags, raidPtr->Layout.map->states);
689 
690 	if (desc == NULL) {
691 		return (ENOMEM);
692 	}
693 	RF_ETIMER_START(desc->tracerec.tot_timer);
694 
695 	desc->async_flag = async_flag;
696 
697 	rf_ContinueRaidAccess(desc);
698 
699 	return (0);
700 }
701 /* force the array into reconfigured mode without doing reconstruction */
702 int
703 rf_SetReconfiguredMode(raidPtr, row, col)
704 	RF_Raid_t *raidPtr;
705 	int     row;
706 	int     col;
707 {
708 	if (!(raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
709 		printf("Can't set reconfigured mode in dedicated-spare array\n");
710 		RF_PANIC();
711 	}
712 	RF_LOCK_MUTEX(raidPtr->mutex);
713 	raidPtr->numFailures++;
714 	raidPtr->Disks[row][col].status = rf_ds_dist_spared;
715 	raidPtr->status[row] = rf_rs_reconfigured;
716 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
717 	/* install spare table only if declustering + distributed sparing
718 	 * architecture. */
719 	if (raidPtr->Layout.map->flags & RF_BD_DECLUSTERED)
720 		rf_InstallSpareTable(raidPtr, row, col);
721 	RF_UNLOCK_MUTEX(raidPtr->mutex);
722 	return (0);
723 }
724 
725 extern int fail_row, fail_col, fail_time;
726 extern int delayed_recon;
727 
728 int
729 rf_FailDisk(
730     RF_Raid_t * raidPtr,
731     int frow,
732     int fcol,
733     int initRecon)
734 {
735 	printf("raid%d: Failing disk r%d c%d\n", raidPtr->raidid, frow, fcol);
736 	RF_LOCK_MUTEX(raidPtr->mutex);
737 	raidPtr->numFailures++;
738 	raidPtr->Disks[frow][fcol].status = rf_ds_failed;
739 	raidPtr->status[frow] = rf_rs_degraded;
740 	rf_update_component_labels(raidPtr, RF_NORMAL_COMPONENT_UPDATE);
741 	RF_UNLOCK_MUTEX(raidPtr->mutex);
742 	if (initRecon)
743 		rf_ReconstructFailedDisk(raidPtr, frow, fcol);
744 	return (0);
745 }
746 /* releases a thread that is waiting for the array to become quiesced.
747  * access_suspend_mutex should be locked upon calling this
748  */
749 void
750 rf_SignalQuiescenceLock(raidPtr, reconDesc)
751 	RF_Raid_t *raidPtr;
752 	RF_RaidReconDesc_t *reconDesc;
753 {
754 	if (rf_quiesceDebug) {
755 		printf("raid%d: Signalling quiescence lock\n",
756 		       raidPtr->raidid);
757 	}
758 	raidPtr->access_suspend_release = 1;
759 
760 	if (raidPtr->waiting_for_quiescence) {
761 		SIGNAL_QUIESCENT_COND(raidPtr);
762 	}
763 }
764 /* suspends all new requests to the array.  No effect on accesses that are in flight.  */
765 int
766 rf_SuspendNewRequestsAndWait(raidPtr)
767 	RF_Raid_t *raidPtr;
768 {
769 	if (rf_quiesceDebug)
770 		printf("Suspending new reqs\n");
771 
772 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
773 	raidPtr->accesses_suspended++;
774 	raidPtr->waiting_for_quiescence = (raidPtr->accs_in_flight == 0) ? 0 : 1;
775 
776 	if (raidPtr->waiting_for_quiescence) {
777 		raidPtr->access_suspend_release = 0;
778 		while (!raidPtr->access_suspend_release) {
779 			printf("Suspending: Waiting for Quiescence\n");
780 			WAIT_FOR_QUIESCENCE(raidPtr);
781 			raidPtr->waiting_for_quiescence = 0;
782 		}
783 	}
784 	printf("Quiescence reached..\n");
785 
786 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
787 	return (raidPtr->waiting_for_quiescence);
788 }
789 /* wake up everyone waiting for quiescence to be released */
790 void
791 rf_ResumeNewRequests(raidPtr)
792 	RF_Raid_t *raidPtr;
793 {
794 	RF_CallbackDesc_t *t, *cb;
795 
796 	if (rf_quiesceDebug)
797 		printf("Resuming new reqs\n");
798 
799 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
800 	raidPtr->accesses_suspended--;
801 	if (raidPtr->accesses_suspended == 0)
802 		cb = raidPtr->quiesce_wait_list;
803 	else
804 		cb = NULL;
805 	raidPtr->quiesce_wait_list = NULL;
806 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
807 
808 	while (cb) {
809 		t = cb;
810 		cb = cb->next;
811 		(t->callbackFunc) (t->callbackArg);
812 		rf_FreeCallbackDesc(t);
813 	}
814 }
815 /*****************************************************************************************
816  *
817  * debug routines
818  *
819  ****************************************************************************************/
820 
821 static void
822 set_debug_option(name, val)
823 	char   *name;
824 	long    val;
825 {
826 	RF_DebugName_t *p;
827 
828 	for (p = rf_debugNames; p->name; p++) {
829 		if (!strcmp(p->name, name)) {
830 			*(p->ptr) = val;
831 			printf("[Set debug variable %s to %ld]\n", name, val);
832 			return;
833 		}
834 	}
835 	RF_ERRORMSG1("Unknown debug string \"%s\"\n", name);
836 }
837 
838 
839 /* would like to use sscanf here, but apparently not available in kernel */
840 /*ARGSUSED*/
841 static void
842 rf_ConfigureDebug(cfgPtr)
843 	RF_Config_t *cfgPtr;
844 {
845 	char   *val_p, *name_p, *white_p;
846 	long    val;
847 	int     i;
848 
849 	rf_ResetDebugOptions();
850 	for (i = 0; cfgPtr->debugVars[i][0] && i < RF_MAXDBGV; i++) {
851 		name_p = rf_find_non_white(&cfgPtr->debugVars[i][0]);
852 		white_p = rf_find_white(name_p);	/* skip to start of 2nd
853 							 * word */
854 		val_p = rf_find_non_white(white_p);
855 		if (*val_p == '0' && *(val_p + 1) == 'x')
856 			val = rf_htoi(val_p + 2);
857 		else
858 			val = rf_atoi(val_p);
859 		*white_p = '\0';
860 		set_debug_option(name_p, val);
861 	}
862 }
863 /* performance monitoring stuff */
864 
865 #define TIMEVAL_TO_US(t) (((long) t.tv_sec) * 1000000L + (long) t.tv_usec)
866 
867 #if !defined(_KERNEL) && !defined(SIMULATE)
868 
869 /*
870  * Throughput stats currently only used in user-level RAIDframe
871  */
872 
873 static int
874 rf_InitThroughputStats(
875     RF_ShutdownList_t ** listp,
876     RF_Raid_t * raidPtr,
877     RF_Config_t * cfgPtr)
878 {
879 	int     rc;
880 
881 	/* these used by user-level raidframe only */
882 	rc = rf_create_managed_mutex(listp, &raidPtr->throughputstats.mutex);
883 	if (rc) {
884 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
885 		    __LINE__, rc);
886 		return (rc);
887 	}
888 	raidPtr->throughputstats.sum_io_us = 0;
889 	raidPtr->throughputstats.num_ios = 0;
890 	raidPtr->throughputstats.num_out_ios = 0;
891 	return (0);
892 }
893 
894 void
895 rf_StartThroughputStats(RF_Raid_t * raidPtr)
896 {
897 	RF_LOCK_MUTEX(raidPtr->throughputstats.mutex);
898 	raidPtr->throughputstats.num_ios++;
899 	raidPtr->throughputstats.num_out_ios++;
900 	if (raidPtr->throughputstats.num_out_ios == 1)
901 		RF_GETTIME(raidPtr->throughputstats.start);
902 	RF_UNLOCK_MUTEX(raidPtr->throughputstats.mutex);
903 }
904 
905 static void
906 rf_StopThroughputStats(RF_Raid_t * raidPtr)
907 {
908 	struct timeval diff;
909 
910 	RF_LOCK_MUTEX(raidPtr->throughputstats.mutex);
911 	raidPtr->throughputstats.num_out_ios--;
912 	if (raidPtr->throughputstats.num_out_ios == 0) {
913 		RF_GETTIME(raidPtr->throughputstats.stop);
914 		RF_TIMEVAL_DIFF(&raidPtr->throughputstats.start, &raidPtr->throughputstats.stop, &diff);
915 		raidPtr->throughputstats.sum_io_us += TIMEVAL_TO_US(diff);
916 	}
917 	RF_UNLOCK_MUTEX(raidPtr->throughputstats.mutex);
918 }
919 
920 static void
921 rf_PrintThroughputStats(RF_Raid_t * raidPtr)
922 {
923 	RF_ASSERT(raidPtr->throughputstats.num_out_ios == 0);
924 	if (raidPtr->throughputstats.sum_io_us != 0) {
925 		printf("[Througphut: %8.2f IOs/second]\n", raidPtr->throughputstats.num_ios
926 		    / (raidPtr->throughputstats.sum_io_us / 1000000.0));
927 	}
928 }
929 #endif				/* !KERNEL && !SIMULATE */
930 
931 void
932 rf_StartUserStats(RF_Raid_t * raidPtr)
933 {
934 	RF_GETTIME(raidPtr->userstats.start);
935 	raidPtr->userstats.sum_io_us = 0;
936 	raidPtr->userstats.num_ios = 0;
937 	raidPtr->userstats.num_sect_moved = 0;
938 }
939 
940 void
941 rf_StopUserStats(RF_Raid_t * raidPtr)
942 {
943 	RF_GETTIME(raidPtr->userstats.stop);
944 }
945 
946 void
947 rf_UpdateUserStats(raidPtr, rt, numsect)
948 	RF_Raid_t *raidPtr;
949 	int     rt;		/* resp time in us */
950 	int     numsect;	/* number of sectors for this access */
951 {
952 	raidPtr->userstats.sum_io_us += rt;
953 	raidPtr->userstats.num_ios++;
954 	raidPtr->userstats.num_sect_moved += numsect;
955 }
956 
957 void
958 rf_PrintUserStats(RF_Raid_t * raidPtr)
959 {
960 	long    elapsed_us, mbs, mbs_frac;
961 	struct timeval diff;
962 
963 	RF_TIMEVAL_DIFF(&raidPtr->userstats.start, &raidPtr->userstats.stop, &diff);
964 	elapsed_us = TIMEVAL_TO_US(diff);
965 
966 	/* 2000 sectors per megabyte, 10000000 microseconds per second */
967 	if (elapsed_us)
968 		mbs = (raidPtr->userstats.num_sect_moved / 2000) / (elapsed_us / 1000000);
969 	else
970 		mbs = 0;
971 
972 	/* this computes only the first digit of the fractional mb/s moved */
973 	if (elapsed_us) {
974 		mbs_frac = ((raidPtr->userstats.num_sect_moved / 200) / (elapsed_us / 1000000))
975 		    - (mbs * 10);
976 	} else {
977 		mbs_frac = 0;
978 	}
979 
980 	printf("Number of I/Os:             %ld\n", raidPtr->userstats.num_ios);
981 	printf("Elapsed time (us):          %ld\n", elapsed_us);
982 	printf("User I/Os per second:       %ld\n", RF_DB0_CHECK(raidPtr->userstats.num_ios, (elapsed_us / 1000000)));
983 	printf("Average user response time: %ld us\n", RF_DB0_CHECK(raidPtr->userstats.sum_io_us, raidPtr->userstats.num_ios));
984 	printf("Total sectors moved:        %ld\n", raidPtr->userstats.num_sect_moved);
985 	printf("Average access size (sect): %ld\n", RF_DB0_CHECK(raidPtr->userstats.num_sect_moved, raidPtr->userstats.num_ios));
986 	printf("Achieved data rate:         %ld.%ld MB/sec\n", mbs, mbs_frac);
987 }
988 
989 
990 void
991 rf_print_panic_message(line,file)
992 	int line;
993 	char *file;
994 {
995 	sprintf(rf_panicbuf,"raidframe error at line %d file %s",
996 		line, file);
997 }
998 
999 void
1000 rf_print_assert_panic_message(line,file,condition)
1001 	int line;
1002 	char *file;
1003 	char *condition;
1004 {
1005 	sprintf(rf_panicbuf,
1006 		"raidframe error at line %d file %s (failed asserting %s)\n",
1007 		line, file, condition);
1008 }
1009