xref: /netbsd-src/sys/dev/raidframe/rf_raid1.c (revision 95d875fb90b1458e4f1de6950286ddcd6644bc61)
1 /*	$NetBSD: rf_raid1.c,v 1.4 1999/08/13 03:41:57 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 "rf_raid.h"
36 #include "rf_raid1.h"
37 #include "rf_dag.h"
38 #include "rf_dagffrd.h"
39 #include "rf_dagffwr.h"
40 #include "rf_dagdegrd.h"
41 #include "rf_dagutils.h"
42 #include "rf_dagfuncs.h"
43 #include "rf_threadid.h"
44 #include "rf_diskqueue.h"
45 #include "rf_general.h"
46 #include "rf_utils.h"
47 #include "rf_parityscan.h"
48 #include "rf_mcpair.h"
49 #include "rf_layout.h"
50 #include "rf_map.h"
51 #include "rf_engine.h"
52 #include "rf_reconbuffer.h"
53 
54 typedef struct RF_Raid1ConfigInfo_s {
55 	RF_RowCol_t **stripeIdentifier;
56 }       RF_Raid1ConfigInfo_t;
57 /* start of day code specific to RAID level 1 */
58 int
59 rf_ConfigureRAID1(
60     RF_ShutdownList_t ** listp,
61     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 	RF_ASSERT(raidPtr->numRow == 1);
84 
85 	/* this implementation of RAID level 1 uses one row of numCol disks
86 	 * and allows multiple (numCol / 2) stripes per row.  A stripe
87 	 * consists of a single data unit and a single parity (mirror) unit.
88 	 * stripe id = raidAddr / stripeUnitSize */
89 	raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
90 	layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
91 	layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
92 	layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit << raidPtr->logBytesPerSector;
93 	layoutPtr->numDataCol = 1;
94 	layoutPtr->numParityCol = 1;
95 	return (0);
96 }
97 
98 
99 /* returns the physical disk location of the primary copy in the mirror pair */
100 void
101 rf_MapSectorRAID1(
102     RF_Raid_t * raidPtr,
103     RF_RaidAddr_t raidSector,
104     RF_RowCol_t * row,
105     RF_RowCol_t * col,
106     RF_SectorNum_t * diskSector,
107     int remap)
108 {
109 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
110 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
111 
112 	*row = 0;
113 	*col = 2 * mirrorPair;
114 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
115 }
116 
117 
118 /* Map Parity
119  *
120  * returns the physical disk location of the secondary copy in the mirror
121  * pair
122  */
123 void
124 rf_MapParityRAID1(
125     RF_Raid_t * raidPtr,
126     RF_RaidAddr_t raidSector,
127     RF_RowCol_t * row,
128     RF_RowCol_t * col,
129     RF_SectorNum_t * diskSector,
130     int remap)
131 {
132 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
133 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
134 
135 	*row = 0;
136 	*col = (2 * mirrorPair) + 1;
137 
138 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
139 }
140 
141 
142 /* IdentifyStripeRAID1
143  *
144  * returns a list of disks for a given redundancy group
145  */
146 void
147 rf_IdentifyStripeRAID1(
148     RF_Raid_t * raidPtr,
149     RF_RaidAddr_t addr,
150     RF_RowCol_t ** diskids,
151     RF_RowCol_t * outRow)
152 {
153 	RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
154 	RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
155 	RF_ASSERT(stripeID >= 0);
156 	RF_ASSERT(addr >= 0);
157 	*outRow = 0;
158 	*diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
159 	RF_ASSERT(*diskids);
160 }
161 
162 
163 /* MapSIDToPSIDRAID1
164  *
165  * maps a logical stripe to a stripe in the redundant array
166  */
167 void
168 rf_MapSIDToPSIDRAID1(
169     RF_RaidLayout_t * layoutPtr,
170     RF_StripeNum_t stripeID,
171     RF_StripeNum_t * psID,
172     RF_ReconUnitNum_t * which_ru)
173 {
174 	*which_ru = 0;
175 	*psID = stripeID;
176 }
177 
178 
179 
180 /******************************************************************************
181  * select a graph to perform a single-stripe access
182  *
183  * Parameters:  raidPtr    - description of the physical array
184  *              type       - type of operation (read or write) requested
185  *              asmap      - logical & physical addresses for this access
186  *              createFunc - name of function to use to create the graph
187  *****************************************************************************/
188 
189 void
190 rf_RAID1DagSelect(
191     RF_Raid_t * raidPtr,
192     RF_IoType_t type,
193     RF_AccessStripeMap_t * asmap,
194     RF_VoidFuncPtr * createFunc)
195 {
196 	RF_RowCol_t frow, fcol, or, oc;
197 	RF_PhysDiskAddr_t *failedPDA;
198 	int     prior_recon, tid;
199 	RF_RowStatus_t rstat;
200 	RF_SectorNum_t oo;
201 
202 
203 	RF_ASSERT(RF_IO_IS_R_OR_W(type));
204 
205 	if (asmap->numDataFailed + asmap->numParityFailed > 1) {
206 		RF_ERRORMSG("Multiple disks failed in a single group!  Aborting I/O operation.\n");
207 		*createFunc = NULL;
208 		return;
209 	}
210 	if (asmap->numDataFailed + asmap->numParityFailed) {
211 		/*
212 	         * We've got a fault. Re-map to spare space, iff applicable.
213 	         * Shouldn't the arch-independent code do this for us?
214 	         * Anyway, it turns out if we don't do this here, then when
215 	         * we're reconstructing, writes go only to the surviving
216 	         * original disk, and aren't reflected on the reconstructed
217 	         * spare. Oops. --jimz
218 	         */
219 		failedPDA = asmap->failedPDAs[0];
220 		frow = failedPDA->row;
221 		fcol = failedPDA->col;
222 		rstat = raidPtr->status[frow];
223 		prior_recon = (rstat == rf_rs_reconfigured) || (
224 		    (rstat == rf_rs_reconstructing) ?
225 		    rf_CheckRUReconstructed(raidPtr->reconControl[frow]->reconMap, failedPDA->startSector) : 0
226 		    );
227 		if (prior_recon) {
228 			or = frow;
229 			oc = fcol;
230 			oo = failedPDA->startSector;
231 			/*
232 		         * If we did distributed sparing, we'd monkey with that here.
233 		         * But we don't, so we'll
234 		         */
235 			failedPDA->row = raidPtr->Disks[frow][fcol].spareRow;
236 			failedPDA->col = raidPtr->Disks[frow][fcol].spareCol;
237 			/*
238 		         * Redirect other components, iff necessary. This looks
239 		         * pretty suspicious to me, but it's what the raid5
240 		         * DAG select does.
241 		         */
242 			if (asmap->parityInfo->next) {
243 				if (failedPDA == asmap->parityInfo) {
244 					failedPDA->next->row = failedPDA->row;
245 					failedPDA->next->col = failedPDA->col;
246 				} else {
247 					if (failedPDA == asmap->parityInfo->next) {
248 						asmap->parityInfo->row = failedPDA->row;
249 						asmap->parityInfo->col = failedPDA->col;
250 					}
251 				}
252 			}
253 			if (rf_dagDebug || rf_mapDebug) {
254 				rf_get_threadid(tid);
255 				printf("[%d] Redirected type '%c' r %d c %d o %ld -> r %d c %d o %ld\n",
256 				    tid, type, or, oc, (long) oo, failedPDA->row, failedPDA->col,
257 				    (long) failedPDA->startSector);
258 			}
259 			asmap->numDataFailed = asmap->numParityFailed = 0;
260 		}
261 	}
262 	if (type == RF_IO_TYPE_READ) {
263 		if (asmap->numDataFailed == 0)
264 			*createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
265 		else
266 			*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
267 	} else {
268 		*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
269 	}
270 }
271 
272 int
273 rf_VerifyParityRAID1(
274     RF_Raid_t * raidPtr,
275     RF_RaidAddr_t raidAddr,
276     RF_PhysDiskAddr_t * parityPDA,
277     int correct_it,
278     RF_RaidAccessFlags_t flags)
279 {
280 	int     nbytes, bcount, stripeWidth, ret, i, j, tid = 0, nbad, *bbufs;
281 	RF_DagNode_t *blockNode, *unblockNode, *wrBlock;
282 	RF_DagHeader_t *rd_dag_h, *wr_dag_h;
283 	RF_AccessStripeMapHeader_t *asm_h;
284 	RF_AllocListElem_t *allocList;
285 	RF_AccTraceEntry_t tracerec;
286 	RF_ReconUnitNum_t which_ru;
287 	RF_RaidLayout_t *layoutPtr;
288 	RF_AccessStripeMap_t *aasm;
289 	RF_SectorCount_t nsector;
290 	RF_RaidAddr_t startAddr;
291 	char   *buf, *buf1, *buf2;
292 	RF_PhysDiskAddr_t *pda;
293 	RF_StripeNum_t psID;
294 	RF_MCPair_t *mcpair;
295 
296 	if (rf_verifyParityDebug) {
297 		rf_get_threadid(tid);
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_verifyParityDebug) {
324 		printf("[%d] RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
325 		    tid, (long) buf, bcount, (long) buf, (long) buf + bcount);
326 	}
327 	/*
328          * Generate a DAG which will read the entire stripe- then we can
329          * just compare data chunks versus "parity" chunks.
330          */
331 
332 	rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf,
333 	    rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
334 	    RF_IO_NORMAL_PRIORITY);
335 	if (rd_dag_h == NULL)
336 		goto done;
337 	blockNode = rd_dag_h->succedents[0];
338 	unblockNode = blockNode->succedents[0]->succedents[0];
339 
340 	/*
341          * Map the access to physical disk addresses (PDAs)- this will
342          * get us both a list of data addresses, and "parity" addresses
343          * (which are really mirror copies).
344          */
345 	asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
346 	    buf, RF_DONT_REMAP);
347 	aasm = asm_h->stripeMap;
348 
349 	buf1 = buf;
350 	/*
351          * Loop through the data blocks, setting up read nodes for each.
352          */
353 	for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
354 		RF_ASSERT(pda);
355 
356 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
357 
358 		RF_ASSERT(pda->numSector != 0);
359 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
360 			/* cannot verify parity with dead disk */
361 			goto done;
362 		}
363 		pda->bufPtr = buf1;
364 		blockNode->succedents[i]->params[0].p = pda;
365 		blockNode->succedents[i]->params[1].p = buf1;
366 		blockNode->succedents[i]->params[2].v = psID;
367 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
368 		buf1 += nbytes;
369 	}
370 	RF_ASSERT(pda == NULL);
371 	/*
372          * keep i, buf1 running
373          *
374          * Loop through parity blocks, setting up read nodes for each.
375          */
376 	for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
377 		RF_ASSERT(pda);
378 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
379 		RF_ASSERT(pda->numSector != 0);
380 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
381 			/* cannot verify parity with dead disk */
382 			goto done;
383 		}
384 		pda->bufPtr = buf1;
385 		blockNode->succedents[i]->params[0].p = pda;
386 		blockNode->succedents[i]->params[1].p = buf1;
387 		blockNode->succedents[i]->params[2].v = psID;
388 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
389 		buf1 += nbytes;
390 	}
391 	RF_ASSERT(pda == NULL);
392 
393 	bzero((char *) &tracerec, sizeof(tracerec));
394 	rd_dag_h->tracerec = &tracerec;
395 
396 	if (rf_verifyParityDebug > 1) {
397 		printf("[%d] RAID1 parity verify read dag:\n", tid);
398 		rf_PrintDAGList(rd_dag_h);
399 	}
400 	RF_LOCK_MUTEX(mcpair->mutex);
401 	mcpair->flag = 0;
402 	rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
403 	    (void *) mcpair);
404 	while (mcpair->flag == 0) {
405 		RF_WAIT_MCPAIR(mcpair);
406 	}
407 	RF_UNLOCK_MUTEX(mcpair->mutex);
408 
409 	if (rd_dag_h->status != rf_enable) {
410 		RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
411 		ret = RF_PARITY_COULD_NOT_VERIFY;
412 		goto done;
413 	}
414 	/*
415          * buf1 is the beginning of the data blocks chunk
416          * buf2 is the beginning of the parity blocks chunk
417          */
418 	buf1 = buf;
419 	buf2 = buf + (nbytes * layoutPtr->numDataCol);
420 	ret = RF_PARITY_OKAY;
421 	/*
422          * bbufs is "bad bufs"- an array whose entries are the data
423          * column numbers where we had miscompares. (That is, column 0
424          * and column 1 of the array are mirror copies, and are considered
425          * "data column 0" for this purpose).
426          */
427 	RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
428 	    allocList);
429 	nbad = 0;
430 	/*
431          * Check data vs "parity" (mirror copy).
432          */
433 	for (i = 0; i < layoutPtr->numDataCol; i++) {
434 		if (rf_verifyParityDebug) {
435 			printf("[%d] RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
436 			    tid, nbytes, i, (long) buf1, (long) buf2, (long) buf);
437 		}
438 		ret = bcmp(buf1, buf2, nbytes);
439 		if (ret) {
440 			if (rf_verifyParityDebug > 1) {
441 				for (j = 0; j < nbytes; j++) {
442 					if (buf1[j] != buf2[j])
443 						break;
444 				}
445 				printf("psid=%ld j=%d\n", (long) psID, j);
446 				printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
447 				    buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
448 				printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
449 				    buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
450 			}
451 			if (rf_verifyParityDebug) {
452 				printf("[%d] RAID1: found bad parity, i=%d\n", tid, i);
453 			}
454 			/*
455 		         * Parity is bad. Keep track of which columns were bad.
456 		         */
457 			if (bbufs)
458 				bbufs[nbad] = i;
459 			nbad++;
460 			ret = RF_PARITY_BAD;
461 		}
462 		buf1 += nbytes;
463 		buf2 += nbytes;
464 	}
465 
466 	if ((ret != RF_PARITY_OKAY) && correct_it) {
467 		ret = RF_PARITY_COULD_NOT_CORRECT;
468 		if (rf_verifyParityDebug) {
469 			printf("[%d] RAID1 parity verify: parity not correct\n", tid);
470 		}
471 		if (bbufs == NULL)
472 			goto done;
473 		/*
474 	         * Make a DAG with one write node for each bad unit. We'll simply
475 	         * write the contents of the data unit onto the parity unit for
476 	         * correction. (It's possible that the mirror copy was the correct
477 	         * copy, and that we're spooging good data by writing bad over it,
478 	         * but there's no way we can know that.
479 	         */
480 		wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, buf,
481 		    rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
482 		    RF_IO_NORMAL_PRIORITY);
483 		if (wr_dag_h == NULL)
484 			goto done;
485 		wrBlock = wr_dag_h->succedents[0];
486 		/*
487 	         * Fill in a write node for each bad compare.
488 	         */
489 		for (i = 0; i < nbad; i++) {
490 			j = i + layoutPtr->numDataCol;
491 			pda = blockNode->succedents[j]->params[0].p;
492 			pda->bufPtr = blockNode->succedents[i]->params[1].p;
493 			wrBlock->succedents[i]->params[0].p = pda;
494 			wrBlock->succedents[i]->params[1].p = pda->bufPtr;
495 			wrBlock->succedents[i]->params[2].v = psID;
496 			wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
497 		}
498 		bzero((char *) &tracerec, sizeof(tracerec));
499 		wr_dag_h->tracerec = &tracerec;
500 		if (rf_verifyParityDebug > 1) {
501 			printf("Parity verify write dag:\n");
502 			rf_PrintDAGList(wr_dag_h);
503 		}
504 		RF_LOCK_MUTEX(mcpair->mutex);
505 		mcpair->flag = 0;
506 		/* fire off the write DAG */
507 		rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
508 		    (void *) mcpair);
509 		while (!mcpair->flag) {
510 			RF_WAIT_COND(mcpair->cond, mcpair->mutex);
511 		}
512 		RF_UNLOCK_MUTEX(mcpair->mutex);
513 		if (wr_dag_h->status != rf_enable) {
514 			RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
515 			goto done;
516 		}
517 		ret = RF_PARITY_CORRECTED;
518 	}
519 done:
520 	/*
521          * All done. We might've gotten here without doing part of the function,
522          * so cleanup what we have to and return our running status.
523          */
524 	if (asm_h)
525 		rf_FreeAccessStripeMap(asm_h);
526 	if (rd_dag_h)
527 		rf_FreeDAG(rd_dag_h);
528 	if (wr_dag_h)
529 		rf_FreeDAG(wr_dag_h);
530 	if (mcpair)
531 		rf_FreeMCPair(mcpair);
532 	rf_FreeAllocList(allocList);
533 	if (rf_verifyParityDebug) {
534 		printf("[%d] RAID1 parity verify, returning %d\n", tid, ret);
535 	}
536 	return (ret);
537 }
538 
539 int
540 rf_SubmitReconBufferRAID1(rbuf, keep_it, use_committed)
541 	RF_ReconBuffer_t *rbuf;	/* the recon buffer to submit */
542 	int     keep_it;	/* whether we can keep this buffer or we have
543 				 * to return it */
544 	int     use_committed;	/* whether to use a committed or an available
545 				 * recon buffer */
546 {
547 	RF_ReconParityStripeStatus_t *pssPtr;
548 	RF_ReconCtrl_t *reconCtrlPtr;
549 	RF_RaidLayout_t *layoutPtr;
550 	int     tid = 0, retcode, created;
551 	RF_CallbackDesc_t *cb, *p;
552 	RF_ReconBuffer_t *t;
553 	RF_Raid_t *raidPtr;
554 	caddr_t ta;
555 
556 	retcode = 0;
557 	created = 0;
558 
559 	raidPtr = rbuf->raidPtr;
560 	layoutPtr = &raidPtr->Layout;
561 	reconCtrlPtr = raidPtr->reconControl[rbuf->row];
562 
563 	RF_ASSERT(rbuf);
564 	RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
565 
566 	if (rf_reconbufferDebug) {
567 		rf_get_threadid(tid);
568 		printf("[%d] RAID1 reconbuffer submission r%d c%d psid %ld ru%d (failed offset %ld)\n",
569 		    tid, rbuf->row, rbuf->col, (long) rbuf->parityStripeID, rbuf->which_ru,
570 		    (long) rbuf->failedDiskSectorOffset);
571 	}
572 	if (rf_reconDebug) {
573 		printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
574 		    (long) rbuf->parityStripeID, (long) rbuf->buffer);
575 		printf("RAID1 psid %ld   %02x %02x %02x %02x %02x\n",
576 		    (long) rbuf->parityStripeID,
577 		    rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3],
578 		    rbuf->buffer[4]);
579 	}
580 	RF_LOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
581 
582 	RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
583 
584 	pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
585 	    rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created);
586 	RF_ASSERT(pssPtr);	/* if it didn't exist, we wouldn't have gotten
587 				 * an rbuf for it */
588 
589 	/*
590          * Since this is simple mirroring, the first submission for a stripe is also
591          * treated as the last.
592          */
593 
594 	t = NULL;
595 	if (keep_it) {
596 		if (rf_reconbufferDebug) {
597 			printf("[%d] RAID1 rbuf submission: keeping rbuf\n", tid);
598 		}
599 		t = rbuf;
600 	} else {
601 		if (use_committed) {
602 			if (rf_reconbufferDebug) {
603 				printf("[%d] RAID1 rbuf submission: using committed rbuf\n", tid);
604 			}
605 			t = reconCtrlPtr->committedRbufs;
606 			RF_ASSERT(t);
607 			reconCtrlPtr->committedRbufs = t->next;
608 			t->next = NULL;
609 		} else
610 			if (reconCtrlPtr->floatingRbufs) {
611 				if (rf_reconbufferDebug) {
612 					printf("[%d] RAID1 rbuf submission: using floating rbuf\n", tid);
613 				}
614 				t = reconCtrlPtr->floatingRbufs;
615 				reconCtrlPtr->floatingRbufs = t->next;
616 				t->next = NULL;
617 			}
618 	}
619 	if (t == NULL) {
620 		if (rf_reconbufferDebug) {
621 			printf("[%d] RAID1 rbuf submission: waiting for rbuf\n", tid);
622 		}
623 		RF_ASSERT((keep_it == 0) && (use_committed == 0));
624 		raidPtr->procsInBufWait++;
625 		if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
626 		    && (raidPtr->numFullReconBuffers == 0)) {
627 			/* ruh-ro */
628 			RF_ERRORMSG("Buffer wait deadlock\n");
629 			rf_PrintPSStatusTable(raidPtr, rbuf->row);
630 			RF_PANIC();
631 		}
632 		pssPtr->flags |= RF_PSS_BUFFERWAIT;
633 		cb = rf_AllocCallbackDesc();
634 		cb->row = rbuf->row;
635 		cb->col = rbuf->col;
636 		cb->callbackArg.v = rbuf->parityStripeID;
637 		cb->callbackArg2.v = rbuf->which_ru;
638 		cb->next = NULL;
639 		if (reconCtrlPtr->bufferWaitList == NULL) {
640 			/* we are the wait list- lucky us */
641 			reconCtrlPtr->bufferWaitList = cb;
642 		} else {
643 			/* append to wait list */
644 			for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
645 			p->next = cb;
646 		}
647 		retcode = 1;
648 		goto out;
649 	}
650 	if (t != rbuf) {
651 		t->row = rbuf->row;
652 		t->col = reconCtrlPtr->fcol;
653 		t->parityStripeID = rbuf->parityStripeID;
654 		t->which_ru = rbuf->which_ru;
655 		t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
656 		t->spRow = rbuf->spRow;
657 		t->spCol = rbuf->spCol;
658 		t->spOffset = rbuf->spOffset;
659 		/* Swap buffers. DANCE! */
660 		ta = t->buffer;
661 		t->buffer = rbuf->buffer;
662 		rbuf->buffer = ta;
663 	}
664 	/*
665          * Use the rbuf we've been given as the target.
666          */
667 	RF_ASSERT(pssPtr->rbuf == NULL);
668 	pssPtr->rbuf = t;
669 
670 	t->count = 1;
671 	/*
672          * Below, we use 1 for numDataCol (which is equal to the count in the
673          * previous line), so we'll always be done.
674          */
675 	rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
676 
677 out:
678 	RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
679 	RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
680 	if (rf_reconbufferDebug) {
681 		printf("[%d] RAID1 rbuf submission: returning %d\n", tid, retcode);
682 	}
683 	return (retcode);
684 }
685