xref: /netbsd-src/sys/dev/raidframe/rf_driver.c (revision 89c5a767f8fc7a4633b2d409966e2becbb98ff92)
1 /*	$NetBSD: rf_driver.c,v 1.34 2000/03/07 02:28:05 oster Exp $	*/
2 /*-
3  * Copyright (c) 1999 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Greg Oster
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *        This product includes software developed by the NetBSD
20  *        Foundation, Inc. and its contributors.
21  * 4. Neither the name of The NetBSD Foundation nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 1995 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Author: Mark Holland, Khalil Amiri, Claudson Bornstein, William V. Courtright II,
43  *         Robby Findler, Daniel Stodolsky, Rachad Youssef, Jim Zelenka
44  *
45  * Permission to use, copy, modify and distribute this software and
46  * its documentation is hereby granted, provided that both the copyright
47  * notice and this permission notice appear in all copies of the
48  * software, derivative works or modified versions, and any portions
49  * thereof, and that both notices appear in supporting documentation.
50  *
51  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
52  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
53  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
54  *
55  * Carnegie Mellon requests users of this software to return to
56  *
57  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
58  *  School of Computer Science
59  *  Carnegie Mellon University
60  *  Pittsburgh PA 15213-3890
61  *
62  * any improvements or extensions that they make and grant Carnegie the
63  * rights to redistribute these changes.
64  */
65 
66 /******************************************************************************
67  *
68  * rf_driver.c -- main setup, teardown, and access routines for the RAID driver
69  *
70  * all routines are prefixed with rf_ (raidframe), to avoid conficts.
71  *
72  ******************************************************************************/
73 
74 
75 #include <sys/types.h>
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/ioctl.h>
79 #include <sys/fcntl.h>
80 #include <sys/vnode.h>
81 
82 
83 #include "rf_archs.h"
84 #include "rf_threadstuff.h"
85 
86 #include <sys/errno.h>
87 
88 #include "rf_raid.h"
89 #include "rf_dag.h"
90 #include "rf_aselect.h"
91 #include "rf_diskqueue.h"
92 #include "rf_parityscan.h"
93 #include "rf_alloclist.h"
94 #include "rf_dagutils.h"
95 #include "rf_utils.h"
96 #include "rf_etimer.h"
97 #include "rf_acctrace.h"
98 #include "rf_configure.h"
99 #include "rf_general.h"
100 #include "rf_desc.h"
101 #include "rf_states.h"
102 #include "rf_freelist.h"
103 #include "rf_decluster.h"
104 #include "rf_map.h"
105 #include "rf_revent.h"
106 #include "rf_callback.h"
107 #include "rf_engine.h"
108 #include "rf_memchunk.h"
109 #include "rf_mcpair.h"
110 #include "rf_nwayxor.h"
111 #include "rf_debugprint.h"
112 #include "rf_copyback.h"
113 #include "rf_driver.h"
114 #include "rf_options.h"
115 #include "rf_shutdown.h"
116 #include "rf_kintf.h"
117 
118 #include <sys/buf.h>
119 
120 /* rad == RF_RaidAccessDesc_t */
121 static RF_FreeList_t *rf_rad_freelist;
122 #define RF_MAX_FREE_RAD 128
123 #define RF_RAD_INC       16
124 #define RF_RAD_INITIAL   32
125 
126 /* debug variables */
127 char    rf_panicbuf[2048];	/* a buffer to hold an error msg when we panic */
128 
129 /* main configuration routines */
130 static int raidframe_booted = 0;
131 
132 static void rf_ConfigureDebug(RF_Config_t * cfgPtr);
133 static void set_debug_option(char *name, long val);
134 static void rf_UnconfigureArray(void);
135 static int init_rad(RF_RaidAccessDesc_t *);
136 static void clean_rad(RF_RaidAccessDesc_t *);
137 static void rf_ShutdownRDFreeList(void *);
138 static int rf_ConfigureRDFreeList(RF_ShutdownList_t **);
139 
140 RF_DECLARE_MUTEX(rf_printf_mutex)	/* debug only:  avoids interleaved
141 					 * printfs by different stripes */
142 
143 #define SIGNAL_QUIESCENT_COND(_raid_)  wakeup(&((_raid_)->accesses_suspended))
144 #define WAIT_FOR_QUIESCENCE(_raid_) \
145 	tsleep(&((_raid_)->accesses_suspended),PRIBIO,"raidframe quiesce", 0);
146 
147 #define IO_BUF_ERR(bp, err) { \
148 	bp->b_flags |= B_ERROR; \
149 	bp->b_resid = bp->b_bcount; \
150 	bp->b_error = err; \
151 	biodone(bp); \
152 }
153 
154 static int configureCount = 0;	/* number of active configurations */
155 static int isconfigged = 0;	/* is basic raidframe (non per-array)
156 				 * stuff configged */
157 RF_DECLARE_STATIC_MUTEX(configureMutex)	/* used to lock the configuration
158 					 * stuff */
159 static RF_ShutdownList_t *globalShutdown;	/* non array-specific
160 						 * stuff */
161 
162 static int rf_ConfigureRDFreeList(RF_ShutdownList_t ** listp);
163 
164 /* called at system boot time */
165 int
166 rf_BootRaidframe()
167 {
168 	int     rc;
169 
170 	if (raidframe_booted)
171 		return (EBUSY);
172 	raidframe_booted = 1;
173 
174 	rc = rf_mutex_init(&configureMutex);
175 	if (rc) {
176 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
177 		    __LINE__, rc);
178 		RF_PANIC();
179 	}
180 	configureCount = 0;
181 	isconfigged = 0;
182 	globalShutdown = NULL;
183 	return (0);
184 }
185 /*
186  * This function is really just for debugging user-level stuff: it
187  * frees up all memory, other RAIDframe resources which might otherwise
188  * be kept around. This is used with systems like "sentinel" to detect
189  * memory leaks.
190  */
191 int
192 rf_UnbootRaidframe()
193 {
194 	int     rc;
195 
196 	RF_LOCK_MUTEX(configureMutex);
197 	if (configureCount) {
198 		RF_UNLOCK_MUTEX(configureMutex);
199 		return (EBUSY);
200 	}
201 	raidframe_booted = 0;
202 	RF_UNLOCK_MUTEX(configureMutex);
203 	rc = rf_mutex_destroy(&configureMutex);
204 	if (rc) {
205 		RF_ERRORMSG3("Unable to destroy mutex file %s line %d rc=%d\n", __FILE__,
206 		    __LINE__, rc);
207 		RF_PANIC();
208 	}
209 	return (0);
210 }
211 /*
212  * Called whenever an array is shutdown
213  */
214 static void
215 rf_UnconfigureArray()
216 {
217 	int     rc;
218 
219 	RF_LOCK_MUTEX(configureMutex);
220 	if (--configureCount == 0) {	/* if no active configurations, shut
221 					 * everything down */
222 		isconfigged = 0;
223 
224 		rc = rf_ShutdownList(&globalShutdown);
225 		if (rc) {
226 			RF_ERRORMSG1("RAIDFRAME: unable to do global shutdown, rc=%d\n", rc);
227 		}
228 
229 		/*
230 	         * We must wait until now, because the AllocList module
231 	         * uses the DebugMem module.
232 	         */
233 		if (rf_memDebug)
234 			rf_print_unfreed();
235 	}
236 	RF_UNLOCK_MUTEX(configureMutex);
237 }
238 
239 /*
240  * Called to shut down an array.
241  */
242 int
243 rf_Shutdown(raidPtr)
244 	RF_Raid_t *raidPtr;
245 {
246 
247 	if (!raidPtr->valid) {
248 		RF_ERRORMSG("Attempt to shut down unconfigured RAIDframe driver.  Aborting shutdown\n");
249 		return (EINVAL);
250 	}
251 	/*
252          * wait for outstanding IOs to land
253          * As described in rf_raid.h, we use the rad_freelist lock
254          * to protect the per-array info about outstanding descs
255          * since we need to do freelist locking anyway, and this
256          * cuts down on the amount of serialization we've got going
257          * on.
258          */
259 	RF_FREELIST_DO_LOCK(rf_rad_freelist);
260 	if (raidPtr->waitShutdown) {
261 		RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
262 		return (EBUSY);
263 	}
264 	raidPtr->waitShutdown = 1;
265 	while (raidPtr->nAccOutstanding) {
266 		RF_WAIT_COND(raidPtr->outstandingCond, RF_FREELIST_MUTEX_OF(rf_rad_freelist));
267 	}
268 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
269 
270 	raidPtr->valid = 0;
271 
272 	rf_final_update_component_labels(raidPtr);
273 
274 	rf_UnconfigureVnodes(raidPtr);
275 
276 	rf_ShutdownList(&raidPtr->shutdownList);
277 
278 	rf_UnconfigureArray();
279 
280 	return (0);
281 }
282 
283 
284 #define DO_INIT_CONFIGURE(f) { \
285 	rc = f (&globalShutdown); \
286 	if (rc) { \
287 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
288 		rf_ShutdownList(&globalShutdown); \
289 		configureCount--; \
290 		RF_UNLOCK_MUTEX(configureMutex); \
291 		return(rc); \
292 	} \
293 }
294 
295 #define DO_RAID_FAIL() { \
296 	rf_UnconfigureVnodes(raidPtr); \
297 	rf_ShutdownList(&raidPtr->shutdownList); \
298 	rf_UnconfigureArray(); \
299 }
300 
301 #define DO_RAID_INIT_CONFIGURE(f) { \
302 	rc = f (&raidPtr->shutdownList, raidPtr, cfgPtr); \
303 	if (rc) { \
304 		RF_ERRORMSG2("RAIDFRAME: failed %s with %d\n", RF_STRING(f), rc); \
305 		DO_RAID_FAIL(); \
306 		return(rc); \
307 	} \
308 }
309 
310 #define DO_RAID_MUTEX(_m_) { \
311 	rc = rf_create_managed_mutex(&raidPtr->shutdownList, (_m_)); \
312 	if (rc) { \
313 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", \
314 			__FILE__, __LINE__, rc); \
315 		DO_RAID_FAIL(); \
316 		return(rc); \
317 	} \
318 }
319 
320 #define DO_RAID_COND(_c_) { \
321 	rc = rf_create_managed_cond(&raidPtr->shutdownList, (_c_)); \
322 	if (rc) { \
323 		RF_ERRORMSG3("Unable to init cond file %s line %d rc=%d\n", \
324 			__FILE__, __LINE__, rc); \
325 		DO_RAID_FAIL(); \
326 		return(rc); \
327 	} \
328 }
329 
330 int
331 rf_Configure(raidPtr, cfgPtr, ac)
332 	RF_Raid_t *raidPtr;
333 	RF_Config_t *cfgPtr;
334 	RF_AutoConfig_t *ac;
335 {
336 	RF_RowCol_t row, col;
337 	int     i, rc;
338 
339 	/* XXX This check can probably be removed now, since
340 	   RAIDFRAME_CONFIGURRE now checks to make sure that the
341 	   RAID set is not already valid
342 	*/
343 	if (raidPtr->valid) {
344 		RF_ERRORMSG("RAIDframe configuration not shut down.  Aborting configure.\n");
345 		return (EINVAL);
346 	}
347 	RF_LOCK_MUTEX(configureMutex);
348 	configureCount++;
349 	if (isconfigged == 0) {
350 		rc = rf_create_managed_mutex(&globalShutdown, &rf_printf_mutex);
351 		if (rc) {
352 			RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
353 			    __LINE__, rc);
354 			rf_ShutdownList(&globalShutdown);
355 			return (rc);
356 		}
357 		/* initialize globals */
358 		printf("RAIDFRAME: protectedSectors is %ld\n",
359 		       rf_protectedSectors);
360 
361 		rf_clear_debug_print_buffer();
362 
363 		DO_INIT_CONFIGURE(rf_ConfigureAllocList);
364 
365 		/*
366 	         * Yes, this does make debugging general to the whole
367 	         * system instead of being array specific. Bummer, drag.
368 		 */
369 		rf_ConfigureDebug(cfgPtr);
370 		DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
371 		DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
372 		DO_INIT_CONFIGURE(rf_ConfigureMapModule);
373 		DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
374 		DO_INIT_CONFIGURE(rf_ConfigureCallback);
375 		DO_INIT_CONFIGURE(rf_ConfigureMemChunk);
376 		DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
377 		DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
378 		DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
379 		DO_INIT_CONFIGURE(rf_ConfigureMCPair);
380 		DO_INIT_CONFIGURE(rf_ConfigureDAGs);
381 		DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
382 		DO_INIT_CONFIGURE(rf_ConfigureDebugPrint);
383 		DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
384 		DO_INIT_CONFIGURE(rf_ConfigureCopyback);
385 		DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
386 		isconfigged = 1;
387 	}
388 	RF_UNLOCK_MUTEX(configureMutex);
389 
390 	DO_RAID_MUTEX(&raidPtr->mutex);
391 	/* set up the cleanup list.  Do this after ConfigureDebug so that
392 	 * value of memDebug will be set */
393 
394 	rf_MakeAllocList(raidPtr->cleanupList);
395 	if (raidPtr->cleanupList == NULL) {
396 		DO_RAID_FAIL();
397 		return (ENOMEM);
398 	}
399 	rc = rf_ShutdownCreate(&raidPtr->shutdownList,
400 	    (void (*) (void *)) rf_FreeAllocList,
401 	    raidPtr->cleanupList);
402 	if (rc) {
403 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n",
404 		    __FILE__, __LINE__, rc);
405 		DO_RAID_FAIL();
406 		return (rc);
407 	}
408 	raidPtr->numRow = cfgPtr->numRow;
409 	raidPtr->numCol = cfgPtr->numCol;
410 	raidPtr->numSpare = cfgPtr->numSpare;
411 
412 	/* XXX we don't even pretend to support more than one row in the
413 	 * kernel... */
414 	if (raidPtr->numRow != 1) {
415 		RF_ERRORMSG("Only one row supported in kernel.\n");
416 		DO_RAID_FAIL();
417 		return (EINVAL);
418 	}
419 	RF_CallocAndAdd(raidPtr->status, raidPtr->numRow, sizeof(RF_RowStatus_t),
420 	    (RF_RowStatus_t *), raidPtr->cleanupList);
421 	if (raidPtr->status == NULL) {
422 		DO_RAID_FAIL();
423 		return (ENOMEM);
424 	}
425 	RF_CallocAndAdd(raidPtr->reconControl, raidPtr->numRow,
426 	    sizeof(RF_ReconCtrl_t *), (RF_ReconCtrl_t **), raidPtr->cleanupList);
427 	if (raidPtr->reconControl == NULL) {
428 		DO_RAID_FAIL();
429 		return (ENOMEM);
430 	}
431 	for (i = 0; i < raidPtr->numRow; i++) {
432 		raidPtr->status[i] = rf_rs_optimal;
433 		raidPtr->reconControl[i] = NULL;
434 	}
435 
436 	DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
437 	DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
438 
439 	DO_RAID_COND(&raidPtr->outstandingCond);
440 
441 	raidPtr->nAccOutstanding = 0;
442 	raidPtr->waitShutdown = 0;
443 
444 	DO_RAID_MUTEX(&raidPtr->access_suspend_mutex);
445 	DO_RAID_COND(&raidPtr->quiescent_cond);
446 
447 	DO_RAID_COND(&raidPtr->waitForReconCond);
448 
449 	DO_RAID_MUTEX(&raidPtr->recon_done_proc_mutex);
450 
451 	if (ac!=NULL) {
452 		/* We have an AutoConfig structure..  Don't do the
453 		   normal disk configuration... call the auto config
454 		   stuff */
455 		rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
456 	} else {
457 		DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
458 		DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
459 	}
460 	/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
461 	 * no. is set */
462 	DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
463 
464 	DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
465 
466 	DO_RAID_INIT_CONFIGURE(rf_ConfigurePSStatus);
467 
468 	for (row = 0; row < raidPtr->numRow; row++) {
469 		for (col = 0; col < raidPtr->numCol; col++) {
470 			/*
471 		         * XXX better distribution
472 		         */
473 			raidPtr->hist_diskreq[row][col] = 0;
474 		}
475 	}
476 
477 	raidPtr->numNewFailures = 0;
478 	raidPtr->copyback_in_progress = 0;
479 	raidPtr->parity_rewrite_in_progress = 0;
480 	raidPtr->recon_in_progress = 0;
481 	raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
482 
483 	/* autoconfigure and root_partition will actually get filled in
484 	   after the config is done */
485 	raidPtr->autoconfigure = 0;
486 	raidPtr->root_partition = 0;
487 	raidPtr->last_unit = raidPtr->raidid;
488 	raidPtr->config_order = 0;
489 
490 	if (rf_keepAccTotals) {
491 		raidPtr->keep_acc_totals = 1;
492 	}
493 	rf_StartUserStats(raidPtr);
494 
495 	raidPtr->valid = 1;
496 	return (0);
497 }
498 
499 static int
500 init_rad(desc)
501 	RF_RaidAccessDesc_t *desc;
502 {
503 	int     rc;
504 
505 	rc = rf_mutex_init(&desc->mutex);
506 	if (rc) {
507 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
508 		    __LINE__, rc);
509 		return (rc);
510 	}
511 	rc = rf_cond_init(&desc->cond);
512 	if (rc) {
513 		RF_ERRORMSG3("Unable to init cond file %s line %d rc=%d\n", __FILE__,
514 		    __LINE__, rc);
515 		rf_mutex_destroy(&desc->mutex);
516 		return (rc);
517 	}
518 	return (0);
519 }
520 
521 static void
522 clean_rad(desc)
523 	RF_RaidAccessDesc_t *desc;
524 {
525 	rf_mutex_destroy(&desc->mutex);
526 	rf_cond_destroy(&desc->cond);
527 }
528 
529 static void
530 rf_ShutdownRDFreeList(ignored)
531 	void   *ignored;
532 {
533 	RF_FREELIST_DESTROY_CLEAN(rf_rad_freelist, next, (RF_RaidAccessDesc_t *), clean_rad);
534 }
535 
536 static int
537 rf_ConfigureRDFreeList(listp)
538 	RF_ShutdownList_t **listp;
539 {
540 	int     rc;
541 
542 	RF_FREELIST_CREATE(rf_rad_freelist, RF_MAX_FREE_RAD,
543 	    RF_RAD_INC, sizeof(RF_RaidAccessDesc_t));
544 	if (rf_rad_freelist == NULL) {
545 		return (ENOMEM);
546 	}
547 	rc = rf_ShutdownCreate(listp, rf_ShutdownRDFreeList, NULL);
548 	if (rc) {
549 		RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
550 		    __LINE__, rc);
551 		rf_ShutdownRDFreeList(NULL);
552 		return (rc);
553 	}
554 	RF_FREELIST_PRIME_INIT(rf_rad_freelist, RF_RAD_INITIAL, next,
555 	    (RF_RaidAccessDesc_t *), init_rad);
556 	return (0);
557 }
558 
559 RF_RaidAccessDesc_t *
560 rf_AllocRaidAccDesc(
561     RF_Raid_t * raidPtr,
562     RF_IoType_t type,
563     RF_RaidAddr_t raidAddress,
564     RF_SectorCount_t numBlocks,
565     caddr_t bufPtr,
566     void *bp,
567     RF_DagHeader_t ** paramDAG,
568     RF_AccessStripeMapHeader_t ** paramASM,
569     RF_RaidAccessFlags_t flags,
570     void (*cbF) (struct buf *),
571     void *cbA,
572     RF_AccessState_t * states)
573 {
574 	RF_RaidAccessDesc_t *desc;
575 
576 	RF_FREELIST_GET_INIT_NOUNLOCK(rf_rad_freelist, desc, next, (RF_RaidAccessDesc_t *), init_rad);
577 	if (raidPtr->waitShutdown) {
578 		/*
579 	         * Actually, we're shutting the array down. Free the desc
580 	         * and return NULL.
581 	         */
582 		RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
583 		RF_FREELIST_FREE_CLEAN(rf_rad_freelist, desc, next, clean_rad);
584 		return (NULL);
585 	}
586 	raidPtr->nAccOutstanding++;
587 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
588 
589 	desc->raidPtr = (void *) raidPtr;
590 	desc->type = type;
591 	desc->raidAddress = raidAddress;
592 	desc->numBlocks = numBlocks;
593 	desc->bufPtr = bufPtr;
594 	desc->bp = bp;
595 	desc->paramDAG = paramDAG;
596 	desc->paramASM = paramASM;
597 	desc->flags = flags;
598 	desc->states = states;
599 	desc->state = 0;
600 
601 	desc->status = 0;
602 	bzero((char *) &desc->tracerec, sizeof(RF_AccTraceEntry_t));
603 	desc->callbackFunc = (void (*) (RF_CBParam_t)) cbF;	/* XXX */
604 	desc->callbackArg = cbA;
605 	desc->next = NULL;
606 	desc->head = desc;
607 	desc->numPending = 0;
608 	desc->cleanupList = NULL;
609 	rf_MakeAllocList(desc->cleanupList);
610 	return (desc);
611 }
612 
613 void
614 rf_FreeRaidAccDesc(RF_RaidAccessDesc_t * desc)
615 {
616 	RF_Raid_t *raidPtr = desc->raidPtr;
617 
618 	RF_ASSERT(desc);
619 
620 	rf_FreeAllocList(desc->cleanupList);
621 	RF_FREELIST_FREE_CLEAN_NOUNLOCK(rf_rad_freelist, desc, next, clean_rad);
622 	raidPtr->nAccOutstanding--;
623 	if (raidPtr->waitShutdown) {
624 		RF_SIGNAL_COND(raidPtr->outstandingCond);
625 	}
626 	RF_FREELIST_DO_UNLOCK(rf_rad_freelist);
627 }
628 /*********************************************************************
629  * Main routine for performing an access.
630  * Accesses are retried until a DAG can not be selected.  This occurs
631  * when either the DAG library is incomplete or there are too many
632  * failures in a parity group.
633  ********************************************************************/
634 int
635 rf_DoAccess(
636     RF_Raid_t * raidPtr,
637     RF_IoType_t type,
638     int async_flag,
639     RF_RaidAddr_t raidAddress,
640     RF_SectorCount_t numBlocks,
641     caddr_t bufPtr,
642     void *bp_in,
643     RF_DagHeader_t ** paramDAG,
644     RF_AccessStripeMapHeader_t ** paramASM,
645     RF_RaidAccessFlags_t flags,
646     RF_RaidAccessDesc_t ** paramDesc,
647     void (*cbF) (struct buf *),
648     void *cbA)
649 /*
650 type should be read or write
651 async_flag should be RF_TRUE or RF_FALSE
652 bp_in is a buf pointer.  void * to facilitate ignoring it outside the kernel
653 */
654 {
655 	RF_RaidAccessDesc_t *desc;
656 	caddr_t lbufPtr = bufPtr;
657 	struct buf *bp = (struct buf *) bp_in;
658 
659 	raidAddress += rf_raidSectorOffset;
660 
661 	if (!raidPtr->valid) {
662 		RF_ERRORMSG("RAIDframe driver not successfully configured.  Rejecting access.\n");
663 		IO_BUF_ERR(bp, EINVAL);
664 		return (EINVAL);
665 	}
666 
667 	if (rf_accessDebug) {
668 
669 		printf("logBytes is: %d %d %d\n", raidPtr->raidid,
670 		    raidPtr->logBytesPerSector,
671 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks));
672 		printf("raid%d: %s raidAddr %d (stripeid %d-%d) numBlocks %d (%d bytes) buf 0x%lx\n", raidPtr->raidid,
673 		    (type == RF_IO_TYPE_READ) ? "READ" : "WRITE", (int) raidAddress,
674 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress),
675 		    (int) rf_RaidAddressToStripeID(&raidPtr->Layout, raidAddress + numBlocks - 1),
676 		    (int) numBlocks,
677 		    (int) rf_RaidAddressToByte(raidPtr, numBlocks),
678 		    (long) bufPtr);
679 	}
680 	if (raidAddress + numBlocks > raidPtr->totalSectors) {
681 
682 		printf("DoAccess: raid addr %lu too large to access %lu sectors.  Max legal addr is %lu\n",
683 		    (u_long) raidAddress, (u_long) numBlocks, (u_long) raidPtr->totalSectors);
684 
685 		IO_BUF_ERR(bp, ENOSPC);
686 		return (ENOSPC);
687 	}
688 	desc = rf_AllocRaidAccDesc(raidPtr, type, raidAddress,
689 	    numBlocks, lbufPtr, bp, paramDAG, paramASM,
690 	    flags, cbF, cbA, raidPtr->Layout.map->states);
691 
692 	if (desc == NULL) {
693 		return (ENOMEM);
694 	}
695 	RF_ETIMER_START(desc->tracerec.tot_timer);
696 
697 	desc->async_flag = async_flag;
698 
699 	rf_ContinueRaidAccess(desc);
700 
701 	return (0);
702 }
703 /* force the array into reconfigured mode without doing reconstruction */
704 int
705 rf_SetReconfiguredMode(raidPtr, row, col)
706 	RF_Raid_t *raidPtr;
707 	int     row;
708 	int     col;
709 {
710 	if (!(raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
711 		printf("Can't set reconfigured mode in dedicated-spare array\n");
712 		RF_PANIC();
713 	}
714 	RF_LOCK_MUTEX(raidPtr->mutex);
715 	raidPtr->numFailures++;
716 	raidPtr->Disks[row][col].status = rf_ds_dist_spared;
717 	raidPtr->status[row] = rf_rs_reconfigured;
718 	rf_update_component_labels(raidPtr);
719 	/* install spare table only if declustering + distributed sparing
720 	 * architecture. */
721 	if (raidPtr->Layout.map->flags & RF_BD_DECLUSTERED)
722 		rf_InstallSpareTable(raidPtr, row, col);
723 	RF_UNLOCK_MUTEX(raidPtr->mutex);
724 	return (0);
725 }
726 
727 extern int fail_row, fail_col, fail_time;
728 extern int delayed_recon;
729 
730 int
731 rf_FailDisk(
732     RF_Raid_t * raidPtr,
733     int frow,
734     int fcol,
735     int initRecon)
736 {
737 	printf("raid%d: Failing disk r%d c%d\n", raidPtr->raidid, frow, fcol);
738 	RF_LOCK_MUTEX(raidPtr->mutex);
739 	raidPtr->numFailures++;
740 	raidPtr->Disks[frow][fcol].status = rf_ds_failed;
741 	raidPtr->status[frow] = rf_rs_degraded;
742 	rf_update_component_labels(raidPtr);
743 	RF_UNLOCK_MUTEX(raidPtr->mutex);
744 	if (initRecon)
745 		rf_ReconstructFailedDisk(raidPtr, frow, fcol);
746 	return (0);
747 }
748 /* releases a thread that is waiting for the array to become quiesced.
749  * access_suspend_mutex should be locked upon calling this
750  */
751 void
752 rf_SignalQuiescenceLock(raidPtr, reconDesc)
753 	RF_Raid_t *raidPtr;
754 	RF_RaidReconDesc_t *reconDesc;
755 {
756 	if (rf_quiesceDebug) {
757 		printf("raid%d: Signalling quiescence lock\n",
758 		       raidPtr->raidid);
759 	}
760 	raidPtr->access_suspend_release = 1;
761 
762 	if (raidPtr->waiting_for_quiescence) {
763 		SIGNAL_QUIESCENT_COND(raidPtr);
764 	}
765 }
766 /* suspends all new requests to the array.  No effect on accesses that are in flight.  */
767 int
768 rf_SuspendNewRequestsAndWait(raidPtr)
769 	RF_Raid_t *raidPtr;
770 {
771 	if (rf_quiesceDebug)
772 		printf("Suspending new reqs\n");
773 
774 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
775 	raidPtr->accesses_suspended++;
776 	raidPtr->waiting_for_quiescence = (raidPtr->accs_in_flight == 0) ? 0 : 1;
777 
778 	if (raidPtr->waiting_for_quiescence) {
779 		raidPtr->access_suspend_release = 0;
780 		while (!raidPtr->access_suspend_release) {
781 			printf("Suspending: Waiting for Quiesence\n");
782 			WAIT_FOR_QUIESCENCE(raidPtr);
783 			raidPtr->waiting_for_quiescence = 0;
784 		}
785 	}
786 	printf("Quiesence reached..\n");
787 
788 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
789 	return (raidPtr->waiting_for_quiescence);
790 }
791 /* wake up everyone waiting for quiescence to be released */
792 void
793 rf_ResumeNewRequests(raidPtr)
794 	RF_Raid_t *raidPtr;
795 {
796 	RF_CallbackDesc_t *t, *cb;
797 
798 	if (rf_quiesceDebug)
799 		printf("Resuming new reqs\n");
800 
801 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
802 	raidPtr->accesses_suspended--;
803 	if (raidPtr->accesses_suspended == 0)
804 		cb = raidPtr->quiesce_wait_list;
805 	else
806 		cb = NULL;
807 	raidPtr->quiesce_wait_list = NULL;
808 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
809 
810 	while (cb) {
811 		t = cb;
812 		cb = cb->next;
813 		(t->callbackFunc) (t->callbackArg);
814 		rf_FreeCallbackDesc(t);
815 	}
816 }
817 /*****************************************************************************************
818  *
819  * debug routines
820  *
821  ****************************************************************************************/
822 
823 static void
824 set_debug_option(name, val)
825 	char   *name;
826 	long    val;
827 {
828 	RF_DebugName_t *p;
829 
830 	for (p = rf_debugNames; p->name; p++) {
831 		if (!strcmp(p->name, name)) {
832 			*(p->ptr) = val;
833 			printf("[Set debug variable %s to %ld]\n", name, val);
834 			return;
835 		}
836 	}
837 	RF_ERRORMSG1("Unknown debug string \"%s\"\n", name);
838 }
839 
840 
841 /* would like to use sscanf here, but apparently not available in kernel */
842 /*ARGSUSED*/
843 static void
844 rf_ConfigureDebug(cfgPtr)
845 	RF_Config_t *cfgPtr;
846 {
847 	char   *val_p, *name_p, *white_p;
848 	long    val;
849 	int     i;
850 
851 	rf_ResetDebugOptions();
852 	for (i = 0; cfgPtr->debugVars[i][0] && i < RF_MAXDBGV; i++) {
853 		name_p = rf_find_non_white(&cfgPtr->debugVars[i][0]);
854 		white_p = rf_find_white(name_p);	/* skip to start of 2nd
855 							 * word */
856 		val_p = rf_find_non_white(white_p);
857 		if (*val_p == '0' && *(val_p + 1) == 'x')
858 			val = rf_htoi(val_p + 2);
859 		else
860 			val = rf_atoi(val_p);
861 		*white_p = '\0';
862 		set_debug_option(name_p, val);
863 	}
864 }
865 /* performance monitoring stuff */
866 
867 #define TIMEVAL_TO_US(t) (((long) t.tv_sec) * 1000000L + (long) t.tv_usec)
868 
869 #if !defined(_KERNEL) && !defined(SIMULATE)
870 
871 /*
872  * Throughput stats currently only used in user-level RAIDframe
873  */
874 
875 static int
876 rf_InitThroughputStats(
877     RF_ShutdownList_t ** listp,
878     RF_Raid_t * raidPtr,
879     RF_Config_t * cfgPtr)
880 {
881 	int     rc;
882 
883 	/* these used by user-level raidframe only */
884 	rc = rf_create_managed_mutex(listp, &raidPtr->throughputstats.mutex);
885 	if (rc) {
886 		RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
887 		    __LINE__, rc);
888 		return (rc);
889 	}
890 	raidPtr->throughputstats.sum_io_us = 0;
891 	raidPtr->throughputstats.num_ios = 0;
892 	raidPtr->throughputstats.num_out_ios = 0;
893 	return (0);
894 }
895 
896 void
897 rf_StartThroughputStats(RF_Raid_t * raidPtr)
898 {
899 	RF_LOCK_MUTEX(raidPtr->throughputstats.mutex);
900 	raidPtr->throughputstats.num_ios++;
901 	raidPtr->throughputstats.num_out_ios++;
902 	if (raidPtr->throughputstats.num_out_ios == 1)
903 		RF_GETTIME(raidPtr->throughputstats.start);
904 	RF_UNLOCK_MUTEX(raidPtr->throughputstats.mutex);
905 }
906 
907 static void
908 rf_StopThroughputStats(RF_Raid_t * raidPtr)
909 {
910 	struct timeval diff;
911 
912 	RF_LOCK_MUTEX(raidPtr->throughputstats.mutex);
913 	raidPtr->throughputstats.num_out_ios--;
914 	if (raidPtr->throughputstats.num_out_ios == 0) {
915 		RF_GETTIME(raidPtr->throughputstats.stop);
916 		RF_TIMEVAL_DIFF(&raidPtr->throughputstats.start, &raidPtr->throughputstats.stop, &diff);
917 		raidPtr->throughputstats.sum_io_us += TIMEVAL_TO_US(diff);
918 	}
919 	RF_UNLOCK_MUTEX(raidPtr->throughputstats.mutex);
920 }
921 
922 static void
923 rf_PrintThroughputStats(RF_Raid_t * raidPtr)
924 {
925 	RF_ASSERT(raidPtr->throughputstats.num_out_ios == 0);
926 	if (raidPtr->throughputstats.sum_io_us != 0) {
927 		printf("[Througphut: %8.2f IOs/second]\n", raidPtr->throughputstats.num_ios
928 		    / (raidPtr->throughputstats.sum_io_us / 1000000.0));
929 	}
930 }
931 #endif				/* !KERNEL && !SIMULATE */
932 
933 void
934 rf_StartUserStats(RF_Raid_t * raidPtr)
935 {
936 	RF_GETTIME(raidPtr->userstats.start);
937 	raidPtr->userstats.sum_io_us = 0;
938 	raidPtr->userstats.num_ios = 0;
939 	raidPtr->userstats.num_sect_moved = 0;
940 }
941 
942 void
943 rf_StopUserStats(RF_Raid_t * raidPtr)
944 {
945 	RF_GETTIME(raidPtr->userstats.stop);
946 }
947 
948 void
949 rf_UpdateUserStats(raidPtr, rt, numsect)
950 	RF_Raid_t *raidPtr;
951 	int     rt;		/* resp time in us */
952 	int     numsect;	/* number of sectors for this access */
953 {
954 	raidPtr->userstats.sum_io_us += rt;
955 	raidPtr->userstats.num_ios++;
956 	raidPtr->userstats.num_sect_moved += numsect;
957 }
958 
959 void
960 rf_PrintUserStats(RF_Raid_t * raidPtr)
961 {
962 	long    elapsed_us, mbs, mbs_frac;
963 	struct timeval diff;
964 
965 	RF_TIMEVAL_DIFF(&raidPtr->userstats.start, &raidPtr->userstats.stop, &diff);
966 	elapsed_us = TIMEVAL_TO_US(diff);
967 
968 	/* 2000 sectors per megabyte, 10000000 microseconds per second */
969 	if (elapsed_us)
970 		mbs = (raidPtr->userstats.num_sect_moved / 2000) / (elapsed_us / 1000000);
971 	else
972 		mbs = 0;
973 
974 	/* this computes only the first digit of the fractional mb/s moved */
975 	if (elapsed_us) {
976 		mbs_frac = ((raidPtr->userstats.num_sect_moved / 200) / (elapsed_us / 1000000))
977 		    - (mbs * 10);
978 	} else {
979 		mbs_frac = 0;
980 	}
981 
982 	printf("Number of I/Os:             %ld\n", raidPtr->userstats.num_ios);
983 	printf("Elapsed time (us):          %ld\n", elapsed_us);
984 	printf("User I/Os per second:       %ld\n", RF_DB0_CHECK(raidPtr->userstats.num_ios, (elapsed_us / 1000000)));
985 	printf("Average user response time: %ld us\n", RF_DB0_CHECK(raidPtr->userstats.sum_io_us, raidPtr->userstats.num_ios));
986 	printf("Total sectors moved:        %ld\n", raidPtr->userstats.num_sect_moved);
987 	printf("Average access size (sect): %ld\n", RF_DB0_CHECK(raidPtr->userstats.num_sect_moved, raidPtr->userstats.num_ios));
988 	printf("Achieved data rate:         %ld.%ld MB/sec\n", mbs, mbs_frac);
989 }
990