xref: /netbsd-src/sys/dev/raidframe/rf_raid1.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: rf_raid1.c,v 1.17 2004/01/10 00:56:28 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 1.17 2004/01/10 00:56:28 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(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
62 		  RF_Config_t *cfgPtr)
63 {
64 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
65 	RF_Raid1ConfigInfo_t *info;
66 	RF_RowCol_t i;
67 
68 	/* create a RAID level 1 configuration structure */
69 	RF_MallocAndAdd(info, sizeof(RF_Raid1ConfigInfo_t), (RF_Raid1ConfigInfo_t *), raidPtr->cleanupList);
70 	if (info == NULL)
71 		return (ENOMEM);
72 	layoutPtr->layoutSpecificInfo = (void *) info;
73 
74 	/* ... and fill it in. */
75 	info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
76 	if (info->stripeIdentifier == NULL)
77 		return (ENOMEM);
78 	for (i = 0; i < (raidPtr->numCol / 2); i++) {
79 		info->stripeIdentifier[i][0] = (2 * i);
80 		info->stripeIdentifier[i][1] = (2 * i) + 1;
81 	}
82 
83 	/* this implementation of RAID level 1 uses one row of numCol disks
84 	 * and allows multiple (numCol / 2) stripes per row.  A stripe
85 	 * consists of a single data unit and a single parity (mirror) unit.
86 	 * stripe id = raidAddr / stripeUnitSize */
87 	raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
88 	layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
89 	layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
90 	layoutPtr->numDataCol = 1;
91 	layoutPtr->numParityCol = 1;
92 	return (0);
93 }
94 
95 
96 /* returns the physical disk location of the primary copy in the mirror pair */
97 void
98 rf_MapSectorRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
99 		  RF_RowCol_t *col, RF_SectorNum_t *diskSector, int remap)
100 {
101 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
102 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
103 
104 	*col = 2 * mirrorPair;
105 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
106 }
107 
108 
109 /* Map Parity
110  *
111  * returns the physical disk location of the secondary copy in the mirror
112  * pair
113  */
114 void
115 rf_MapParityRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
116 		  RF_RowCol_t *col, RF_SectorNum_t *diskSector, int remap)
117 {
118 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
119 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
120 
121 	*col = (2 * mirrorPair) + 1;
122 
123 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
124 }
125 
126 
127 /* IdentifyStripeRAID1
128  *
129  * returns a list of disks for a given redundancy group
130  */
131 void
132 rf_IdentifyStripeRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
133 		       RF_RowCol_t **diskids)
134 {
135 	RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
136 	RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
137 	RF_ASSERT(stripeID >= 0);
138 	RF_ASSERT(addr >= 0);
139 	*diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
140 	RF_ASSERT(*diskids);
141 }
142 
143 
144 /* MapSIDToPSIDRAID1
145  *
146  * maps a logical stripe to a stripe in the redundant array
147  */
148 void
149 rf_MapSIDToPSIDRAID1(RF_RaidLayout_t *layoutPtr, RF_StripeNum_t stripeID,
150 		     RF_StripeNum_t *psID, RF_ReconUnitNum_t *which_ru)
151 {
152 	*which_ru = 0;
153 	*psID = stripeID;
154 }
155 
156 
157 
158 /******************************************************************************
159  * select a graph to perform a single-stripe access
160  *
161  * Parameters:  raidPtr    - description of the physical array
162  *              type       - type of operation (read or write) requested
163  *              asmap      - logical & physical addresses for this access
164  *              createFunc - name of function to use to create the graph
165  *****************************************************************************/
166 
167 void
168 rf_RAID1DagSelect(RF_Raid_t *raidPtr, RF_IoType_t type,
169 		  RF_AccessStripeMap_t *asmap, RF_VoidFuncPtr *createFunc)
170 {
171 	RF_RowCol_t fcol, oc;
172 	RF_PhysDiskAddr_t *failedPDA;
173 	int     prior_recon;
174 	RF_RowStatus_t rstat;
175 	RF_SectorNum_t oo;
176 
177 
178 	RF_ASSERT(RF_IO_IS_R_OR_W(type));
179 
180 	if (asmap->numDataFailed + asmap->numParityFailed > 1) {
181 		if (rf_dagDebug)
182 			RF_ERRORMSG("Multiple disks failed in a single group!  Aborting I/O operation.\n");
183 		*createFunc = NULL;
184 		return;
185 	}
186 	if (asmap->numDataFailed + asmap->numParityFailed) {
187 		/*
188 	         * We've got a fault. Re-map to spare space, iff applicable.
189 	         * Shouldn't the arch-independent code do this for us?
190 	         * Anyway, it turns out if we don't do this here, then when
191 	         * we're reconstructing, writes go only to the surviving
192 	         * original disk, and aren't reflected on the reconstructed
193 	         * spare. Oops. --jimz
194 	         */
195 		failedPDA = asmap->failedPDAs[0];
196 		fcol = failedPDA->col;
197 		rstat = raidPtr->status;
198 		prior_recon = (rstat == rf_rs_reconfigured) || (
199 		    (rstat == rf_rs_reconstructing) ?
200 		    rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, failedPDA->startSector) : 0
201 		    );
202 		if (prior_recon) {
203 			oc = fcol;
204 			oo = failedPDA->startSector;
205 			/*
206 		         * If we did distributed sparing, we'd monkey with that here.
207 		         * But we don't, so we'll
208 		         */
209 			failedPDA->col = raidPtr->Disks[fcol].spareCol;
210 			/*
211 		         * Redirect other components, iff necessary. This looks
212 		         * pretty suspicious to me, but it's what the raid5
213 		         * DAG select does.
214 		         */
215 			if (asmap->parityInfo->next) {
216 				if (failedPDA == asmap->parityInfo) {
217 					failedPDA->next->col = failedPDA->col;
218 				} else {
219 					if (failedPDA == asmap->parityInfo->next) {
220 						asmap->parityInfo->col = failedPDA->col;
221 					}
222 				}
223 			}
224 			if (rf_dagDebug || rf_mapDebug) {
225 				printf("raid%d: Redirected type '%c' c %d o %ld -> c %d o %ld\n",
226 				       raidPtr->raidid, type, oc,
227 				       (long) oo,
228 				       failedPDA->col,
229 				       (long) failedPDA->startSector);
230 			}
231 			asmap->numDataFailed = asmap->numParityFailed = 0;
232 		}
233 	}
234 	if (type == RF_IO_TYPE_READ) {
235 		if (asmap->numDataFailed == 0)
236 			*createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
237 		else
238 			*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
239 	} else {
240 		*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
241 	}
242 }
243 
244 int
245 rf_VerifyParityRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
246 		     RF_PhysDiskAddr_t *parityPDA, int correct_it,
247 		     RF_RaidAccessFlags_t flags)
248 {
249 	int     nbytes, bcount, stripeWidth, ret, i, j, nbad, *bbufs;
250 	RF_DagNode_t *blockNode, *wrBlock;
251 	RF_DagHeader_t *rd_dag_h, *wr_dag_h;
252 	RF_AccessStripeMapHeader_t *asm_h;
253 	RF_AllocListElem_t *allocList;
254 	RF_AccTraceEntry_t tracerec;
255 	RF_ReconUnitNum_t which_ru;
256 	RF_RaidLayout_t *layoutPtr;
257 	RF_AccessStripeMap_t *aasm;
258 	RF_SectorCount_t nsector;
259 	RF_RaidAddr_t startAddr;
260 	char   *buf, *buf1, *buf2;
261 	RF_PhysDiskAddr_t *pda;
262 	RF_StripeNum_t psID;
263 	RF_MCPair_t *mcpair;
264 
265 	layoutPtr = &raidPtr->Layout;
266 	startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
267 	nsector = parityPDA->numSector;
268 	nbytes = rf_RaidAddressToByte(raidPtr, nsector);
269 	psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
270 
271 	asm_h = NULL;
272 	rd_dag_h = wr_dag_h = NULL;
273 	mcpair = NULL;
274 
275 	ret = RF_PARITY_COULD_NOT_VERIFY;
276 
277 	rf_MakeAllocList(allocList);
278 	if (allocList == NULL)
279 		return (RF_PARITY_COULD_NOT_VERIFY);
280 	mcpair = rf_AllocMCPair();
281 	if (mcpair == NULL)
282 		goto done;
283 	RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
284 	stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
285 	bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
286 	RF_MallocAndAdd(buf, bcount, (char *), allocList);
287 	if (buf == NULL)
288 		goto done;
289 #if RF_DEBUG_VERIFYPARITY
290 	if (rf_verifyParityDebug) {
291 		printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
292 		       raidPtr->raidid, (long) buf, bcount, (long) buf,
293 		       (long) buf + bcount);
294 	}
295 #endif
296 	/*
297          * Generate a DAG which will read the entire stripe- then we can
298          * just compare data chunks versus "parity" chunks.
299          */
300 
301 	rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf,
302 	    rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
303 	    RF_IO_NORMAL_PRIORITY);
304 	if (rd_dag_h == NULL)
305 		goto done;
306 	blockNode = rd_dag_h->succedents[0];
307 
308 	/*
309          * Map the access to physical disk addresses (PDAs)- this will
310          * get us both a list of data addresses, and "parity" addresses
311          * (which are really mirror copies).
312          */
313 	asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
314 	    buf, RF_DONT_REMAP);
315 	aasm = asm_h->stripeMap;
316 
317 	buf1 = buf;
318 	/*
319          * Loop through the data blocks, setting up read nodes for each.
320          */
321 	for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
322 		RF_ASSERT(pda);
323 
324 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
325 
326 		RF_ASSERT(pda->numSector != 0);
327 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
328 			/* cannot verify parity with dead disk */
329 			goto done;
330 		}
331 		pda->bufPtr = buf1;
332 		blockNode->succedents[i]->params[0].p = pda;
333 		blockNode->succedents[i]->params[1].p = buf1;
334 		blockNode->succedents[i]->params[2].v = psID;
335 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
336 		buf1 += nbytes;
337 	}
338 	RF_ASSERT(pda == NULL);
339 	/*
340          * keep i, buf1 running
341          *
342          * Loop through parity blocks, setting up read nodes for each.
343          */
344 	for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
345 		RF_ASSERT(pda);
346 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
347 		RF_ASSERT(pda->numSector != 0);
348 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
349 			/* cannot verify parity with dead disk */
350 			goto done;
351 		}
352 		pda->bufPtr = buf1;
353 		blockNode->succedents[i]->params[0].p = pda;
354 		blockNode->succedents[i]->params[1].p = buf1;
355 		blockNode->succedents[i]->params[2].v = psID;
356 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
357 		buf1 += nbytes;
358 	}
359 	RF_ASSERT(pda == NULL);
360 
361 	memset((char *) &tracerec, 0, sizeof(tracerec));
362 	rd_dag_h->tracerec = &tracerec;
363 
364 #if 0
365 	if (rf_verifyParityDebug > 1) {
366 		printf("raid%d: RAID1 parity verify read dag:\n",
367 		       raidPtr->raidid);
368 		rf_PrintDAGList(rd_dag_h);
369 	}
370 #endif
371 	RF_LOCK_MUTEX(mcpair->mutex);
372 	mcpair->flag = 0;
373 	rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
374 	    (void *) mcpair);
375 	while (mcpair->flag == 0) {
376 		RF_WAIT_MCPAIR(mcpair);
377 	}
378 	RF_UNLOCK_MUTEX(mcpair->mutex);
379 
380 	if (rd_dag_h->status != rf_enable) {
381 		RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
382 		ret = RF_PARITY_COULD_NOT_VERIFY;
383 		goto done;
384 	}
385 	/*
386          * buf1 is the beginning of the data blocks chunk
387          * buf2 is the beginning of the parity blocks chunk
388          */
389 	buf1 = buf;
390 	buf2 = buf + (nbytes * layoutPtr->numDataCol);
391 	ret = RF_PARITY_OKAY;
392 	/*
393          * bbufs is "bad bufs"- an array whose entries are the data
394          * column numbers where we had miscompares. (That is, column 0
395          * and column 1 of the array are mirror copies, and are considered
396          * "data column 0" for this purpose).
397          */
398 	RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
399 	    allocList);
400 	nbad = 0;
401 	/*
402          * Check data vs "parity" (mirror copy).
403          */
404 	for (i = 0; i < layoutPtr->numDataCol; i++) {
405 #if RF_DEBUG_VERIFYPARITY
406 		if (rf_verifyParityDebug) {
407 			printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
408 			       raidPtr->raidid, nbytes, i, (long) buf1,
409 			       (long) buf2, (long) buf);
410 		}
411 #endif
412 		ret = memcmp(buf1, buf2, nbytes);
413 		if (ret) {
414 #if RF_DEBUG_VERIFYPARITY
415 			if (rf_verifyParityDebug > 1) {
416 				for (j = 0; j < nbytes; j++) {
417 					if (buf1[j] != buf2[j])
418 						break;
419 				}
420 				printf("psid=%ld j=%d\n", (long) psID, j);
421 				printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
422 				    buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
423 				printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
424 				    buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
425 			}
426 			if (rf_verifyParityDebug) {
427 				printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i);
428 			}
429 #endif
430 			/*
431 		         * Parity is bad. Keep track of which columns were bad.
432 		         */
433 			if (bbufs)
434 				bbufs[nbad] = i;
435 			nbad++;
436 			ret = RF_PARITY_BAD;
437 		}
438 		buf1 += nbytes;
439 		buf2 += nbytes;
440 	}
441 
442 	if ((ret != RF_PARITY_OKAY) && correct_it) {
443 		ret = RF_PARITY_COULD_NOT_CORRECT;
444 #if RF_DEBUG_VERIFYPARITY
445 		if (rf_verifyParityDebug) {
446 			printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
447 		}
448 #endif
449 		if (bbufs == NULL)
450 			goto done;
451 		/*
452 	         * Make a DAG with one write node for each bad unit. We'll simply
453 	         * write the contents of the data unit onto the parity unit for
454 	         * correction. (It's possible that the mirror copy was the correct
455 	         * copy, and that we're spooging good data by writing bad over it,
456 	         * but there's no way we can know that.
457 	         */
458 		wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, buf,
459 		    rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
460 		    RF_IO_NORMAL_PRIORITY);
461 		if (wr_dag_h == NULL)
462 			goto done;
463 		wrBlock = wr_dag_h->succedents[0];
464 		/*
465 	         * Fill in a write node for each bad compare.
466 	         */
467 		for (i = 0; i < nbad; i++) {
468 			j = i + layoutPtr->numDataCol;
469 			pda = blockNode->succedents[j]->params[0].p;
470 			pda->bufPtr = blockNode->succedents[i]->params[1].p;
471 			wrBlock->succedents[i]->params[0].p = pda;
472 			wrBlock->succedents[i]->params[1].p = pda->bufPtr;
473 			wrBlock->succedents[i]->params[2].v = psID;
474 			wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
475 		}
476 		memset((char *) &tracerec, 0, sizeof(tracerec));
477 		wr_dag_h->tracerec = &tracerec;
478 #if 0
479 		if (rf_verifyParityDebug > 1) {
480 			printf("Parity verify write dag:\n");
481 			rf_PrintDAGList(wr_dag_h);
482 		}
483 #endif
484 		RF_LOCK_MUTEX(mcpair->mutex);
485 		mcpair->flag = 0;
486 		/* fire off the write DAG */
487 		rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
488 		    (void *) mcpair);
489 		while (!mcpair->flag) {
490 			RF_WAIT_COND(mcpair->cond, mcpair->mutex);
491 		}
492 		RF_UNLOCK_MUTEX(mcpair->mutex);
493 		if (wr_dag_h->status != rf_enable) {
494 			RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
495 			goto done;
496 		}
497 		ret = RF_PARITY_CORRECTED;
498 	}
499 done:
500 	/*
501          * All done. We might've gotten here without doing part of the function,
502          * so cleanup what we have to and return our running status.
503          */
504 	if (asm_h)
505 		rf_FreeAccessStripeMap(asm_h);
506 	if (rd_dag_h)
507 		rf_FreeDAG(rd_dag_h);
508 	if (wr_dag_h)
509 		rf_FreeDAG(wr_dag_h);
510 	if (mcpair)
511 		rf_FreeMCPair(mcpair);
512 	rf_FreeAllocList(allocList);
513 #if RF_DEBUG_VERIFYPARITY
514 	if (rf_verifyParityDebug) {
515 		printf("raid%d: RAID1 parity verify, returning %d\n",
516 		       raidPtr->raidid, ret);
517 	}
518 #endif
519 	return (ret);
520 }
521 
522 /* rbuf          - the recon buffer to submit
523  * keep_it       - whether we can keep this buffer or we have to return it
524  * use_committed - whether to use a committed or an available recon buffer
525  */
526 
527 int
528 rf_SubmitReconBufferRAID1(RF_ReconBuffer_t *rbuf, int keep_it,
529 			  int use_committed)
530 {
531 	RF_ReconParityStripeStatus_t *pssPtr;
532 	RF_ReconCtrl_t *reconCtrlPtr;
533 	int     retcode, created;
534 	RF_CallbackDesc_t *cb, *p;
535 	RF_ReconBuffer_t *t;
536 	RF_Raid_t *raidPtr;
537 	caddr_t ta;
538 
539 	retcode = 0;
540 	created = 0;
541 
542 	raidPtr = rbuf->raidPtr;
543 	reconCtrlPtr = raidPtr->reconControl;
544 
545 	RF_ASSERT(rbuf);
546 	RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
547 
548 #if RF_DEBUG_RECON
549 	if (rf_reconbufferDebug) {
550 		printf("raid%d: RAID1 reconbuffer submission c%d psid %ld ru%d (failed offset %ld)\n",
551 		       raidPtr->raidid, rbuf->col,
552 		       (long) rbuf->parityStripeID, rbuf->which_ru,
553 		       (long) rbuf->failedDiskSectorOffset);
554 	}
555 #endif
556 	if (rf_reconDebug) {
557 		printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
558 		    (long) rbuf->parityStripeID, (long) rbuf->buffer);
559 		printf("RAID1 psid %ld   %02x %02x %02x %02x %02x\n",
560 		    (long) rbuf->parityStripeID,
561 		    rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3],
562 		    rbuf->buffer[4]);
563 	}
564 	RF_LOCK_PSS_MUTEX(raidPtr, rbuf->parityStripeID);
565 
566 	RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
567 
568 	pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
569 	    rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created);
570 	RF_ASSERT(pssPtr);	/* if it didn't exist, we wouldn't have gotten
571 				 * an rbuf for it */
572 
573 	/*
574          * Since this is simple mirroring, the first submission for a stripe is also
575          * treated as the last.
576          */
577 
578 	t = NULL;
579 	if (keep_it) {
580 #if RF_DEBUG_RECON
581 		if (rf_reconbufferDebug) {
582 			printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
583 			       raidPtr->raidid);
584 		}
585 #endif
586 		t = rbuf;
587 	} else {
588 		if (use_committed) {
589 #if RF_DEBUG_RECON
590 			if (rf_reconbufferDebug) {
591 				printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
592 			}
593 #endif
594 			t = reconCtrlPtr->committedRbufs;
595 			RF_ASSERT(t);
596 			reconCtrlPtr->committedRbufs = t->next;
597 			t->next = NULL;
598 		} else
599 			if (reconCtrlPtr->floatingRbufs) {
600 #if RF_DEBUG_RECON
601 				if (rf_reconbufferDebug) {
602 					printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
603 				}
604 #endif
605 				t = reconCtrlPtr->floatingRbufs;
606 				reconCtrlPtr->floatingRbufs = t->next;
607 				t->next = NULL;
608 			}
609 	}
610 	if (t == NULL) {
611 #if RF_DEBUG_RECON
612 		if (rf_reconbufferDebug) {
613 			printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
614 		}
615 #endif
616 		RF_ASSERT((keep_it == 0) && (use_committed == 0));
617 		raidPtr->procsInBufWait++;
618 		if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
619 		    && (raidPtr->numFullReconBuffers == 0)) {
620 			/* ruh-ro */
621 			RF_ERRORMSG("Buffer wait deadlock\n");
622 			rf_PrintPSStatusTable(raidPtr);
623 			RF_PANIC();
624 		}
625 		pssPtr->flags |= RF_PSS_BUFFERWAIT;
626 		cb = rf_AllocCallbackDesc();
627 		cb->col = rbuf->col;
628 		cb->callbackArg.v = rbuf->parityStripeID;
629 		cb->callbackArg2.v = rbuf->which_ru;
630 		cb->next = NULL;
631 		if (reconCtrlPtr->bufferWaitList == NULL) {
632 			/* we are the wait list- lucky us */
633 			reconCtrlPtr->bufferWaitList = cb;
634 		} else {
635 			/* append to wait list */
636 			for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
637 			p->next = cb;
638 		}
639 		retcode = 1;
640 		goto out;
641 	}
642 	if (t != rbuf) {
643 		t->col = reconCtrlPtr->fcol;
644 		t->parityStripeID = rbuf->parityStripeID;
645 		t->which_ru = rbuf->which_ru;
646 		t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
647 		t->spCol = rbuf->spCol;
648 		t->spOffset = rbuf->spOffset;
649 		/* Swap buffers. DANCE! */
650 		ta = t->buffer;
651 		t->buffer = rbuf->buffer;
652 		rbuf->buffer = ta;
653 	}
654 	/*
655          * Use the rbuf we've been given as the target.
656          */
657 	RF_ASSERT(pssPtr->rbuf == NULL);
658 	pssPtr->rbuf = t;
659 
660 	t->count = 1;
661 	/*
662          * Below, we use 1 for numDataCol (which is equal to the count in the
663          * previous line), so we'll always be done.
664          */
665 	rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
666 
667 out:
668 	RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->parityStripeID);
669 	RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
670 #if RF_DEBUG_RECON
671 	if (rf_reconbufferDebug) {
672 		printf("raid%d: RAID1 rbuf submission: returning %d\n",
673 		       raidPtr->raidid, retcode);
674 	}
675 #endif
676 	return (retcode);
677 }
678