xref: /netbsd-src/sys/dev/raidframe/rf_raid1.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: rf_raid1.c,v 1.11 2002/09/17 03:21:40 oster Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Author: William V. Courtright II
7  *
8  * Permission to use, copy, modify and distribute this software and
9  * its documentation is hereby granted, provided that both the copyright
10  * notice and this permission notice appear in all copies of the
11  * software, derivative works or modified versions, and any portions
12  * thereof, and that both notices appear in supporting documentation.
13  *
14  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17  *
18  * Carnegie Mellon requests users of this software to return to
19  *
20  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
21  *  School of Computer Science
22  *  Carnegie Mellon University
23  *  Pittsburgh PA 15213-3890
24  *
25  * any improvements or extensions that they make and grant Carnegie the
26  * rights to redistribute these changes.
27  */
28 
29 /*****************************************************************************
30  *
31  * rf_raid1.c -- implements RAID Level 1
32  *
33  *****************************************************************************/
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: rf_raid1.c,v 1.11 2002/09/17 03:21:40 oster Exp $");
37 
38 #include "rf_raid.h"
39 #include "rf_raid1.h"
40 #include "rf_dag.h"
41 #include "rf_dagffrd.h"
42 #include "rf_dagffwr.h"
43 #include "rf_dagdegrd.h"
44 #include "rf_dagutils.h"
45 #include "rf_dagfuncs.h"
46 #include "rf_diskqueue.h"
47 #include "rf_general.h"
48 #include "rf_utils.h"
49 #include "rf_parityscan.h"
50 #include "rf_mcpair.h"
51 #include "rf_layout.h"
52 #include "rf_map.h"
53 #include "rf_engine.h"
54 #include "rf_reconbuffer.h"
55 
56 typedef struct RF_Raid1ConfigInfo_s {
57 	RF_RowCol_t **stripeIdentifier;
58 }       RF_Raid1ConfigInfo_t;
59 /* start of day code specific to RAID level 1 */
60 int
61 rf_ConfigureRAID1(
62     RF_ShutdownList_t ** listp,
63     RF_Raid_t * raidPtr,
64     RF_Config_t * cfgPtr)
65 {
66 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
67 	RF_Raid1ConfigInfo_t *info;
68 	RF_RowCol_t i;
69 
70 	/* create a RAID level 1 configuration structure */
71 	RF_MallocAndAdd(info, sizeof(RF_Raid1ConfigInfo_t), (RF_Raid1ConfigInfo_t *), raidPtr->cleanupList);
72 	if (info == NULL)
73 		return (ENOMEM);
74 	layoutPtr->layoutSpecificInfo = (void *) info;
75 
76 	/* ... and fill it in. */
77 	info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
78 	if (info->stripeIdentifier == NULL)
79 		return (ENOMEM);
80 	for (i = 0; i < (raidPtr->numCol / 2); i++) {
81 		info->stripeIdentifier[i][0] = (2 * i);
82 		info->stripeIdentifier[i][1] = (2 * i) + 1;
83 	}
84 
85 	RF_ASSERT(raidPtr->numRow == 1);
86 
87 	/* this implementation of RAID level 1 uses one row of numCol disks
88 	 * and allows multiple (numCol / 2) stripes per row.  A stripe
89 	 * consists of a single data unit and a single parity (mirror) unit.
90 	 * stripe id = raidAddr / stripeUnitSize */
91 	raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
92 	layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
93 	layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
94 	layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit << raidPtr->logBytesPerSector;
95 	layoutPtr->numDataCol = 1;
96 	layoutPtr->numParityCol = 1;
97 	return (0);
98 }
99 
100 
101 /* returns the physical disk location of the primary copy in the mirror pair */
102 void
103 rf_MapSectorRAID1(
104     RF_Raid_t * raidPtr,
105     RF_RaidAddr_t raidSector,
106     RF_RowCol_t * row,
107     RF_RowCol_t * col,
108     RF_SectorNum_t * diskSector,
109     int remap)
110 {
111 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
112 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
113 
114 	*row = 0;
115 	*col = 2 * mirrorPair;
116 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
117 }
118 
119 
120 /* Map Parity
121  *
122  * returns the physical disk location of the secondary copy in the mirror
123  * pair
124  */
125 void
126 rf_MapParityRAID1(
127     RF_Raid_t * raidPtr,
128     RF_RaidAddr_t raidSector,
129     RF_RowCol_t * row,
130     RF_RowCol_t * col,
131     RF_SectorNum_t * diskSector,
132     int remap)
133 {
134 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
135 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
136 
137 	*row = 0;
138 	*col = (2 * mirrorPair) + 1;
139 
140 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
141 }
142 
143 
144 /* IdentifyStripeRAID1
145  *
146  * returns a list of disks for a given redundancy group
147  */
148 void
149 rf_IdentifyStripeRAID1(
150     RF_Raid_t * raidPtr,
151     RF_RaidAddr_t addr,
152     RF_RowCol_t ** diskids,
153     RF_RowCol_t * outRow)
154 {
155 	RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
156 	RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
157 	RF_ASSERT(stripeID >= 0);
158 	RF_ASSERT(addr >= 0);
159 	*outRow = 0;
160 	*diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
161 	RF_ASSERT(*diskids);
162 }
163 
164 
165 /* MapSIDToPSIDRAID1
166  *
167  * maps a logical stripe to a stripe in the redundant array
168  */
169 void
170 rf_MapSIDToPSIDRAID1(
171     RF_RaidLayout_t * layoutPtr,
172     RF_StripeNum_t stripeID,
173     RF_StripeNum_t * psID,
174     RF_ReconUnitNum_t * which_ru)
175 {
176 	*which_ru = 0;
177 	*psID = stripeID;
178 }
179 
180 
181 
182 /******************************************************************************
183  * select a graph to perform a single-stripe access
184  *
185  * Parameters:  raidPtr    - description of the physical array
186  *              type       - type of operation (read or write) requested
187  *              asmap      - logical & physical addresses for this access
188  *              createFunc - name of function to use to create the graph
189  *****************************************************************************/
190 
191 void
192 rf_RAID1DagSelect(
193     RF_Raid_t * raidPtr,
194     RF_IoType_t type,
195     RF_AccessStripeMap_t * asmap,
196     RF_VoidFuncPtr * createFunc)
197 {
198 	RF_RowCol_t frow, fcol, or, oc;
199 	RF_PhysDiskAddr_t *failedPDA;
200 	int     prior_recon;
201 	RF_RowStatus_t rstat;
202 	RF_SectorNum_t oo;
203 
204 
205 	RF_ASSERT(RF_IO_IS_R_OR_W(type));
206 
207 	if (asmap->numDataFailed + asmap->numParityFailed > 1) {
208 		RF_ERRORMSG("Multiple disks failed in a single group!  Aborting I/O operation.\n");
209 		*createFunc = NULL;
210 		return;
211 	}
212 	if (asmap->numDataFailed + asmap->numParityFailed) {
213 		/*
214 	         * We've got a fault. Re-map to spare space, iff applicable.
215 	         * Shouldn't the arch-independent code do this for us?
216 	         * Anyway, it turns out if we don't do this here, then when
217 	         * we're reconstructing, writes go only to the surviving
218 	         * original disk, and aren't reflected on the reconstructed
219 	         * spare. Oops. --jimz
220 	         */
221 		failedPDA = asmap->failedPDAs[0];
222 		frow = failedPDA->row;
223 		fcol = failedPDA->col;
224 		rstat = raidPtr->status[frow];
225 		prior_recon = (rstat == rf_rs_reconfigured) || (
226 		    (rstat == rf_rs_reconstructing) ?
227 		    rf_CheckRUReconstructed(raidPtr->reconControl[frow]->reconMap, failedPDA->startSector) : 0
228 		    );
229 		if (prior_recon) {
230 			or = frow;
231 			oc = fcol;
232 			oo = failedPDA->startSector;
233 			/*
234 		         * If we did distributed sparing, we'd monkey with that here.
235 		         * But we don't, so we'll
236 		         */
237 			failedPDA->row = raidPtr->Disks[frow][fcol].spareRow;
238 			failedPDA->col = raidPtr->Disks[frow][fcol].spareCol;
239 			/*
240 		         * Redirect other components, iff necessary. This looks
241 		         * pretty suspicious to me, but it's what the raid5
242 		         * DAG select does.
243 		         */
244 			if (asmap->parityInfo->next) {
245 				if (failedPDA == asmap->parityInfo) {
246 					failedPDA->next->row = failedPDA->row;
247 					failedPDA->next->col = failedPDA->col;
248 				} else {
249 					if (failedPDA == asmap->parityInfo->next) {
250 						asmap->parityInfo->row = failedPDA->row;
251 						asmap->parityInfo->col = failedPDA->col;
252 					}
253 				}
254 			}
255 			if (rf_dagDebug || rf_mapDebug) {
256 				printf("raid%d: Redirected type '%c' r %d c %d o %ld -> r %d c %d o %ld\n",
257 				       raidPtr->raidid, type, or, oc,
258 				       (long) oo, failedPDA->row,
259 				       failedPDA->col,
260 				       (long) failedPDA->startSector);
261 			}
262 			asmap->numDataFailed = asmap->numParityFailed = 0;
263 		}
264 	}
265 	if (type == RF_IO_TYPE_READ) {
266 		if (asmap->numDataFailed == 0)
267 			*createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
268 		else
269 			*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
270 	} else {
271 		*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
272 	}
273 }
274 
275 int
276 rf_VerifyParityRAID1(
277     RF_Raid_t * raidPtr,
278     RF_RaidAddr_t raidAddr,
279     RF_PhysDiskAddr_t * parityPDA,
280     int correct_it,
281     RF_RaidAccessFlags_t flags)
282 {
283 	int     nbytes, bcount, stripeWidth, ret, i, j, nbad, *bbufs;
284 	RF_DagNode_t *blockNode, *unblockNode, *wrBlock;
285 	RF_DagHeader_t *rd_dag_h, *wr_dag_h;
286 	RF_AccessStripeMapHeader_t *asm_h;
287 	RF_AllocListElem_t *allocList;
288 	RF_AccTraceEntry_t tracerec;
289 	RF_ReconUnitNum_t which_ru;
290 	RF_RaidLayout_t *layoutPtr;
291 	RF_AccessStripeMap_t *aasm;
292 	RF_SectorCount_t nsector;
293 	RF_RaidAddr_t startAddr;
294 	char   *buf, *buf1, *buf2;
295 	RF_PhysDiskAddr_t *pda;
296 	RF_StripeNum_t psID;
297 	RF_MCPair_t *mcpair;
298 
299 	layoutPtr = &raidPtr->Layout;
300 	startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
301 	nsector = parityPDA->numSector;
302 	nbytes = rf_RaidAddressToByte(raidPtr, nsector);
303 	psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
304 
305 	asm_h = NULL;
306 	rd_dag_h = wr_dag_h = NULL;
307 	mcpair = NULL;
308 
309 	ret = RF_PARITY_COULD_NOT_VERIFY;
310 
311 	rf_MakeAllocList(allocList);
312 	if (allocList == NULL)
313 		return (RF_PARITY_COULD_NOT_VERIFY);
314 	mcpair = rf_AllocMCPair();
315 	if (mcpair == NULL)
316 		goto done;
317 	RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
318 	stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
319 	bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
320 	RF_MallocAndAdd(buf, bcount, (char *), allocList);
321 	if (buf == NULL)
322 		goto done;
323 #if RF_DEBUG_VERIFYPARITY
324 	if (rf_verifyParityDebug) {
325 		printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
326 		       raidPtr->raidid, (long) buf, bcount, (long) buf,
327 		       (long) buf + bcount);
328 	}
329 #endif
330 	/*
331          * Generate a DAG which will read the entire stripe- then we can
332          * just compare data chunks versus "parity" chunks.
333          */
334 
335 	rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf,
336 	    rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
337 	    RF_IO_NORMAL_PRIORITY);
338 	if (rd_dag_h == NULL)
339 		goto done;
340 	blockNode = rd_dag_h->succedents[0];
341 	unblockNode = blockNode->succedents[0]->succedents[0];
342 
343 	/*
344          * Map the access to physical disk addresses (PDAs)- this will
345          * get us both a list of data addresses, and "parity" addresses
346          * (which are really mirror copies).
347          */
348 	asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
349 	    buf, RF_DONT_REMAP);
350 	aasm = asm_h->stripeMap;
351 
352 	buf1 = buf;
353 	/*
354          * Loop through the data blocks, setting up read nodes for each.
355          */
356 	for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
357 		RF_ASSERT(pda);
358 
359 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
360 
361 		RF_ASSERT(pda->numSector != 0);
362 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
363 			/* cannot verify parity with dead disk */
364 			goto done;
365 		}
366 		pda->bufPtr = buf1;
367 		blockNode->succedents[i]->params[0].p = pda;
368 		blockNode->succedents[i]->params[1].p = buf1;
369 		blockNode->succedents[i]->params[2].v = psID;
370 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
371 		buf1 += nbytes;
372 	}
373 	RF_ASSERT(pda == NULL);
374 	/*
375          * keep i, buf1 running
376          *
377          * Loop through parity blocks, setting up read nodes for each.
378          */
379 	for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
380 		RF_ASSERT(pda);
381 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
382 		RF_ASSERT(pda->numSector != 0);
383 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
384 			/* cannot verify parity with dead disk */
385 			goto done;
386 		}
387 		pda->bufPtr = buf1;
388 		blockNode->succedents[i]->params[0].p = pda;
389 		blockNode->succedents[i]->params[1].p = buf1;
390 		blockNode->succedents[i]->params[2].v = psID;
391 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
392 		buf1 += nbytes;
393 	}
394 	RF_ASSERT(pda == NULL);
395 
396 	memset((char *) &tracerec, 0, sizeof(tracerec));
397 	rd_dag_h->tracerec = &tracerec;
398 
399 #if 0
400 	if (rf_verifyParityDebug > 1) {
401 		printf("raid%d: RAID1 parity verify read dag:\n",
402 		       raidPtr->raidid);
403 		rf_PrintDAGList(rd_dag_h);
404 	}
405 #endif
406 	RF_LOCK_MUTEX(mcpair->mutex);
407 	mcpair->flag = 0;
408 	rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
409 	    (void *) mcpair);
410 	while (mcpair->flag == 0) {
411 		RF_WAIT_MCPAIR(mcpair);
412 	}
413 	RF_UNLOCK_MUTEX(mcpair->mutex);
414 
415 	if (rd_dag_h->status != rf_enable) {
416 		RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
417 		ret = RF_PARITY_COULD_NOT_VERIFY;
418 		goto done;
419 	}
420 	/*
421          * buf1 is the beginning of the data blocks chunk
422          * buf2 is the beginning of the parity blocks chunk
423          */
424 	buf1 = buf;
425 	buf2 = buf + (nbytes * layoutPtr->numDataCol);
426 	ret = RF_PARITY_OKAY;
427 	/*
428          * bbufs is "bad bufs"- an array whose entries are the data
429          * column numbers where we had miscompares. (That is, column 0
430          * and column 1 of the array are mirror copies, and are considered
431          * "data column 0" for this purpose).
432          */
433 	RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
434 	    allocList);
435 	nbad = 0;
436 	/*
437          * Check data vs "parity" (mirror copy).
438          */
439 	for (i = 0; i < layoutPtr->numDataCol; i++) {
440 #if RF_DEBUG_VERIFYPARITY
441 		if (rf_verifyParityDebug) {
442 			printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
443 			       raidPtr->raidid, nbytes, i, (long) buf1,
444 			       (long) buf2, (long) buf);
445 		}
446 #endif
447 		ret = memcmp(buf1, buf2, nbytes);
448 		if (ret) {
449 #if RF_DEBUG_VERIFYPARITY
450 			if (rf_verifyParityDebug > 1) {
451 				for (j = 0; j < nbytes; j++) {
452 					if (buf1[j] != buf2[j])
453 						break;
454 				}
455 				printf("psid=%ld j=%d\n", (long) psID, j);
456 				printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
457 				    buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
458 				printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
459 				    buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
460 			}
461 			if (rf_verifyParityDebug) {
462 				printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i);
463 			}
464 #endif
465 			/*
466 		         * Parity is bad. Keep track of which columns were bad.
467 		         */
468 			if (bbufs)
469 				bbufs[nbad] = i;
470 			nbad++;
471 			ret = RF_PARITY_BAD;
472 		}
473 		buf1 += nbytes;
474 		buf2 += nbytes;
475 	}
476 
477 	if ((ret != RF_PARITY_OKAY) && correct_it) {
478 		ret = RF_PARITY_COULD_NOT_CORRECT;
479 #if RF_DEBUG_VERIFYPARITY
480 		if (rf_verifyParityDebug) {
481 			printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
482 		}
483 #endif
484 		if (bbufs == NULL)
485 			goto done;
486 		/*
487 	         * Make a DAG with one write node for each bad unit. We'll simply
488 	         * write the contents of the data unit onto the parity unit for
489 	         * correction. (It's possible that the mirror copy was the correct
490 	         * copy, and that we're spooging good data by writing bad over it,
491 	         * but there's no way we can know that.
492 	         */
493 		wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, buf,
494 		    rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
495 		    RF_IO_NORMAL_PRIORITY);
496 		if (wr_dag_h == NULL)
497 			goto done;
498 		wrBlock = wr_dag_h->succedents[0];
499 		/*
500 	         * Fill in a write node for each bad compare.
501 	         */
502 		for (i = 0; i < nbad; i++) {
503 			j = i + layoutPtr->numDataCol;
504 			pda = blockNode->succedents[j]->params[0].p;
505 			pda->bufPtr = blockNode->succedents[i]->params[1].p;
506 			wrBlock->succedents[i]->params[0].p = pda;
507 			wrBlock->succedents[i]->params[1].p = pda->bufPtr;
508 			wrBlock->succedents[i]->params[2].v = psID;
509 			wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
510 		}
511 		memset((char *) &tracerec, 0, sizeof(tracerec));
512 		wr_dag_h->tracerec = &tracerec;
513 #if 0
514 		if (rf_verifyParityDebug > 1) {
515 			printf("Parity verify write dag:\n");
516 			rf_PrintDAGList(wr_dag_h);
517 		}
518 #endif
519 		RF_LOCK_MUTEX(mcpair->mutex);
520 		mcpair->flag = 0;
521 		/* fire off the write DAG */
522 		rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
523 		    (void *) mcpair);
524 		while (!mcpair->flag) {
525 			RF_WAIT_COND(mcpair->cond, mcpair->mutex);
526 		}
527 		RF_UNLOCK_MUTEX(mcpair->mutex);
528 		if (wr_dag_h->status != rf_enable) {
529 			RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
530 			goto done;
531 		}
532 		ret = RF_PARITY_CORRECTED;
533 	}
534 done:
535 	/*
536          * All done. We might've gotten here without doing part of the function,
537          * so cleanup what we have to and return our running status.
538          */
539 	if (asm_h)
540 		rf_FreeAccessStripeMap(asm_h);
541 	if (rd_dag_h)
542 		rf_FreeDAG(rd_dag_h);
543 	if (wr_dag_h)
544 		rf_FreeDAG(wr_dag_h);
545 	if (mcpair)
546 		rf_FreeMCPair(mcpair);
547 	rf_FreeAllocList(allocList);
548 #if RF_DEBUG_VERIFYPARITY
549 	if (rf_verifyParityDebug) {
550 		printf("raid%d: RAID1 parity verify, returning %d\n",
551 		       raidPtr->raidid, ret);
552 	}
553 #endif
554 	return (ret);
555 }
556 
557 int
558 rf_SubmitReconBufferRAID1(rbuf, keep_it, use_committed)
559 	RF_ReconBuffer_t *rbuf;	/* the recon buffer to submit */
560 	int     keep_it;	/* whether we can keep this buffer or we have
561 				 * to return it */
562 	int     use_committed;	/* whether to use a committed or an available
563 				 * recon buffer */
564 {
565 	RF_ReconParityStripeStatus_t *pssPtr;
566 	RF_ReconCtrl_t *reconCtrlPtr;
567 	RF_RaidLayout_t *layoutPtr;
568 	int     retcode, created;
569 	RF_CallbackDesc_t *cb, *p;
570 	RF_ReconBuffer_t *t;
571 	RF_Raid_t *raidPtr;
572 	caddr_t ta;
573 
574 	retcode = 0;
575 	created = 0;
576 
577 	raidPtr = rbuf->raidPtr;
578 	layoutPtr = &raidPtr->Layout;
579 	reconCtrlPtr = raidPtr->reconControl[rbuf->row];
580 
581 	RF_ASSERT(rbuf);
582 	RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
583 
584 #if RF_DEBUG_RECON
585 	if (rf_reconbufferDebug) {
586 		printf("raid%d: RAID1 reconbuffer submission r%d c%d psid %ld ru%d (failed offset %ld)\n",
587 		       raidPtr->raidid, rbuf->row, rbuf->col,
588 		       (long) rbuf->parityStripeID, rbuf->which_ru,
589 		       (long) rbuf->failedDiskSectorOffset);
590 	}
591 #endif
592 	if (rf_reconDebug) {
593 		printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
594 		    (long) rbuf->parityStripeID, (long) rbuf->buffer);
595 		printf("RAID1 psid %ld   %02x %02x %02x %02x %02x\n",
596 		    (long) rbuf->parityStripeID,
597 		    rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3],
598 		    rbuf->buffer[4]);
599 	}
600 	RF_LOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
601 
602 	RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
603 
604 	pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
605 	    rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created);
606 	RF_ASSERT(pssPtr);	/* if it didn't exist, we wouldn't have gotten
607 				 * an rbuf for it */
608 
609 	/*
610          * Since this is simple mirroring, the first submission for a stripe is also
611          * treated as the last.
612          */
613 
614 	t = NULL;
615 	if (keep_it) {
616 #if RF_DEBUG_RECON
617 		if (rf_reconbufferDebug) {
618 			printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
619 			       raidPtr->raidid);
620 		}
621 #endif
622 		t = rbuf;
623 	} else {
624 		if (use_committed) {
625 #if RF_DEBUG_RECON
626 			if (rf_reconbufferDebug) {
627 				printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
628 			}
629 #endif
630 			t = reconCtrlPtr->committedRbufs;
631 			RF_ASSERT(t);
632 			reconCtrlPtr->committedRbufs = t->next;
633 			t->next = NULL;
634 		} else
635 			if (reconCtrlPtr->floatingRbufs) {
636 #if RF_DEBUG_RECON
637 				if (rf_reconbufferDebug) {
638 					printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
639 				}
640 #endif
641 				t = reconCtrlPtr->floatingRbufs;
642 				reconCtrlPtr->floatingRbufs = t->next;
643 				t->next = NULL;
644 			}
645 	}
646 	if (t == NULL) {
647 #if RF_DEBUG_RECON
648 		if (rf_reconbufferDebug) {
649 			printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
650 		}
651 #endif
652 		RF_ASSERT((keep_it == 0) && (use_committed == 0));
653 		raidPtr->procsInBufWait++;
654 		if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
655 		    && (raidPtr->numFullReconBuffers == 0)) {
656 			/* ruh-ro */
657 			RF_ERRORMSG("Buffer wait deadlock\n");
658 			rf_PrintPSStatusTable(raidPtr, rbuf->row);
659 			RF_PANIC();
660 		}
661 		pssPtr->flags |= RF_PSS_BUFFERWAIT;
662 		cb = rf_AllocCallbackDesc();
663 		cb->row = rbuf->row;
664 		cb->col = rbuf->col;
665 		cb->callbackArg.v = rbuf->parityStripeID;
666 		cb->callbackArg2.v = rbuf->which_ru;
667 		cb->next = NULL;
668 		if (reconCtrlPtr->bufferWaitList == NULL) {
669 			/* we are the wait list- lucky us */
670 			reconCtrlPtr->bufferWaitList = cb;
671 		} else {
672 			/* append to wait list */
673 			for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
674 			p->next = cb;
675 		}
676 		retcode = 1;
677 		goto out;
678 	}
679 	if (t != rbuf) {
680 		t->row = rbuf->row;
681 		t->col = reconCtrlPtr->fcol;
682 		t->parityStripeID = rbuf->parityStripeID;
683 		t->which_ru = rbuf->which_ru;
684 		t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
685 		t->spRow = rbuf->spRow;
686 		t->spCol = rbuf->spCol;
687 		t->spOffset = rbuf->spOffset;
688 		/* Swap buffers. DANCE! */
689 		ta = t->buffer;
690 		t->buffer = rbuf->buffer;
691 		rbuf->buffer = ta;
692 	}
693 	/*
694          * Use the rbuf we've been given as the target.
695          */
696 	RF_ASSERT(pssPtr->rbuf == NULL);
697 	pssPtr->rbuf = t;
698 
699 	t->count = 1;
700 	/*
701          * Below, we use 1 for numDataCol (which is equal to the count in the
702          * previous line), so we'll always be done.
703          */
704 	rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
705 
706 out:
707 	RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
708 	RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
709 #if RF_DEBUG_RECON
710 	if (rf_reconbufferDebug) {
711 		printf("raid%d: RAID1 rbuf submission: returning %d\n",
712 		       raidPtr->raidid, retcode);
713 	}
714 #endif
715 	return (retcode);
716 }
717