rf_raid1.c revision 1.6 1 /* $NetBSD: rf_raid1.c,v 1.6 2001/07/18 06:45:34 thorpej 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 (at) 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_diskqueue.h"
44 #include "rf_general.h"
45 #include "rf_utils.h"
46 #include "rf_parityscan.h"
47 #include "rf_mcpair.h"
48 #include "rf_layout.h"
49 #include "rf_map.h"
50 #include "rf_engine.h"
51 #include "rf_reconbuffer.h"
52
53 typedef struct RF_Raid1ConfigInfo_s {
54 RF_RowCol_t **stripeIdentifier;
55 } RF_Raid1ConfigInfo_t;
56 /* start of day code specific to RAID level 1 */
57 int
58 rf_ConfigureRAID1(
59 RF_ShutdownList_t ** listp,
60 RF_Raid_t * raidPtr,
61 RF_Config_t * cfgPtr)
62 {
63 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
64 RF_Raid1ConfigInfo_t *info;
65 RF_RowCol_t i;
66
67 /* create a RAID level 1 configuration structure */
68 RF_MallocAndAdd(info, sizeof(RF_Raid1ConfigInfo_t), (RF_Raid1ConfigInfo_t *), raidPtr->cleanupList);
69 if (info == NULL)
70 return (ENOMEM);
71 layoutPtr->layoutSpecificInfo = (void *) info;
72
73 /* ... and fill it in. */
74 info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
75 if (info->stripeIdentifier == NULL)
76 return (ENOMEM);
77 for (i = 0; i < (raidPtr->numCol / 2); i++) {
78 info->stripeIdentifier[i][0] = (2 * i);
79 info->stripeIdentifier[i][1] = (2 * i) + 1;
80 }
81
82 RF_ASSERT(raidPtr->numRow == 1);
83
84 /* this implementation of RAID level 1 uses one row of numCol disks
85 * and allows multiple (numCol / 2) stripes per row. A stripe
86 * consists of a single data unit and a single parity (mirror) unit.
87 * stripe id = raidAddr / stripeUnitSize */
88 raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
89 layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
90 layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
91 layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit << raidPtr->logBytesPerSector;
92 layoutPtr->numDataCol = 1;
93 layoutPtr->numParityCol = 1;
94 return (0);
95 }
96
97
98 /* returns the physical disk location of the primary copy in the mirror pair */
99 void
100 rf_MapSectorRAID1(
101 RF_Raid_t * raidPtr,
102 RF_RaidAddr_t raidSector,
103 RF_RowCol_t * row,
104 RF_RowCol_t * col,
105 RF_SectorNum_t * diskSector,
106 int remap)
107 {
108 RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
109 RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
110
111 *row = 0;
112 *col = 2 * mirrorPair;
113 *diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
114 }
115
116
117 /* Map Parity
118 *
119 * returns the physical disk location of the secondary copy in the mirror
120 * pair
121 */
122 void
123 rf_MapParityRAID1(
124 RF_Raid_t * raidPtr,
125 RF_RaidAddr_t raidSector,
126 RF_RowCol_t * row,
127 RF_RowCol_t * col,
128 RF_SectorNum_t * diskSector,
129 int remap)
130 {
131 RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
132 RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
133
134 *row = 0;
135 *col = (2 * mirrorPair) + 1;
136
137 *diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
138 }
139
140
141 /* IdentifyStripeRAID1
142 *
143 * returns a list of disks for a given redundancy group
144 */
145 void
146 rf_IdentifyStripeRAID1(
147 RF_Raid_t * raidPtr,
148 RF_RaidAddr_t addr,
149 RF_RowCol_t ** diskids,
150 RF_RowCol_t * outRow)
151 {
152 RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
153 RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
154 RF_ASSERT(stripeID >= 0);
155 RF_ASSERT(addr >= 0);
156 *outRow = 0;
157 *diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
158 RF_ASSERT(*diskids);
159 }
160
161
162 /* MapSIDToPSIDRAID1
163 *
164 * maps a logical stripe to a stripe in the redundant array
165 */
166 void
167 rf_MapSIDToPSIDRAID1(
168 RF_RaidLayout_t * layoutPtr,
169 RF_StripeNum_t stripeID,
170 RF_StripeNum_t * psID,
171 RF_ReconUnitNum_t * which_ru)
172 {
173 *which_ru = 0;
174 *psID = stripeID;
175 }
176
177
178
179 /******************************************************************************
180 * select a graph to perform a single-stripe access
181 *
182 * Parameters: raidPtr - description of the physical array
183 * type - type of operation (read or write) requested
184 * asmap - logical & physical addresses for this access
185 * createFunc - name of function to use to create the graph
186 *****************************************************************************/
187
188 void
189 rf_RAID1DagSelect(
190 RF_Raid_t * raidPtr,
191 RF_IoType_t type,
192 RF_AccessStripeMap_t * asmap,
193 RF_VoidFuncPtr * createFunc)
194 {
195 RF_RowCol_t frow, fcol, or, oc;
196 RF_PhysDiskAddr_t *failedPDA;
197 int prior_recon;
198 RF_RowStatus_t rstat;
199 RF_SectorNum_t oo;
200
201
202 RF_ASSERT(RF_IO_IS_R_OR_W(type));
203
204 if (asmap->numDataFailed + asmap->numParityFailed > 1) {
205 RF_ERRORMSG("Multiple disks failed in a single group! Aborting I/O operation.\n");
206 *createFunc = NULL;
207 return;
208 }
209 if (asmap->numDataFailed + asmap->numParityFailed) {
210 /*
211 * We've got a fault. Re-map to spare space, iff applicable.
212 * Shouldn't the arch-independent code do this for us?
213 * Anyway, it turns out if we don't do this here, then when
214 * we're reconstructing, writes go only to the surviving
215 * original disk, and aren't reflected on the reconstructed
216 * spare. Oops. --jimz
217 */
218 failedPDA = asmap->failedPDAs[0];
219 frow = failedPDA->row;
220 fcol = failedPDA->col;
221 rstat = raidPtr->status[frow];
222 prior_recon = (rstat == rf_rs_reconfigured) || (
223 (rstat == rf_rs_reconstructing) ?
224 rf_CheckRUReconstructed(raidPtr->reconControl[frow]->reconMap, failedPDA->startSector) : 0
225 );
226 if (prior_recon) {
227 or = frow;
228 oc = fcol;
229 oo = failedPDA->startSector;
230 /*
231 * If we did distributed sparing, we'd monkey with that here.
232 * But we don't, so we'll
233 */
234 failedPDA->row = raidPtr->Disks[frow][fcol].spareRow;
235 failedPDA->col = raidPtr->Disks[frow][fcol].spareCol;
236 /*
237 * Redirect other components, iff necessary. This looks
238 * pretty suspicious to me, but it's what the raid5
239 * DAG select does.
240 */
241 if (asmap->parityInfo->next) {
242 if (failedPDA == asmap->parityInfo) {
243 failedPDA->next->row = failedPDA->row;
244 failedPDA->next->col = failedPDA->col;
245 } else {
246 if (failedPDA == asmap->parityInfo->next) {
247 asmap->parityInfo->row = failedPDA->row;
248 asmap->parityInfo->col = failedPDA->col;
249 }
250 }
251 }
252 if (rf_dagDebug || rf_mapDebug) {
253 printf("raid%d: Redirected type '%c' r %d c %d o %ld -> r %d c %d o %ld\n",
254 raidPtr->raidid, type, or, oc,
255 (long) oo, failedPDA->row,
256 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, 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 layoutPtr = &raidPtr->Layout;
297 startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
298 nsector = parityPDA->numSector;
299 nbytes = rf_RaidAddressToByte(raidPtr, nsector);
300 psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
301
302 asm_h = NULL;
303 rd_dag_h = wr_dag_h = NULL;
304 mcpair = NULL;
305
306 ret = RF_PARITY_COULD_NOT_VERIFY;
307
308 rf_MakeAllocList(allocList);
309 if (allocList == NULL)
310 return (RF_PARITY_COULD_NOT_VERIFY);
311 mcpair = rf_AllocMCPair();
312 if (mcpair == NULL)
313 goto done;
314 RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
315 stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
316 bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
317 RF_MallocAndAdd(buf, bcount, (char *), allocList);
318 if (buf == NULL)
319 goto done;
320 if (rf_verifyParityDebug) {
321 printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
322 raidPtr->raidid, (long) buf, bcount, (long) buf,
323 (long) buf + bcount);
324 }
325 /*
326 * Generate a DAG which will read the entire stripe- then we can
327 * just compare data chunks versus "parity" chunks.
328 */
329
330 rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf,
331 rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
332 RF_IO_NORMAL_PRIORITY);
333 if (rd_dag_h == NULL)
334 goto done;
335 blockNode = rd_dag_h->succedents[0];
336 unblockNode = blockNode->succedents[0]->succedents[0];
337
338 /*
339 * Map the access to physical disk addresses (PDAs)- this will
340 * get us both a list of data addresses, and "parity" addresses
341 * (which are really mirror copies).
342 */
343 asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
344 buf, RF_DONT_REMAP);
345 aasm = asm_h->stripeMap;
346
347 buf1 = buf;
348 /*
349 * Loop through the data blocks, setting up read nodes for each.
350 */
351 for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
352 RF_ASSERT(pda);
353
354 rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
355
356 RF_ASSERT(pda->numSector != 0);
357 if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
358 /* cannot verify parity with dead disk */
359 goto done;
360 }
361 pda->bufPtr = buf1;
362 blockNode->succedents[i]->params[0].p = pda;
363 blockNode->succedents[i]->params[1].p = buf1;
364 blockNode->succedents[i]->params[2].v = psID;
365 blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
366 buf1 += nbytes;
367 }
368 RF_ASSERT(pda == NULL);
369 /*
370 * keep i, buf1 running
371 *
372 * Loop through parity blocks, setting up read nodes for each.
373 */
374 for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
375 RF_ASSERT(pda);
376 rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
377 RF_ASSERT(pda->numSector != 0);
378 if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
379 /* cannot verify parity with dead disk */
380 goto done;
381 }
382 pda->bufPtr = buf1;
383 blockNode->succedents[i]->params[0].p = pda;
384 blockNode->succedents[i]->params[1].p = buf1;
385 blockNode->succedents[i]->params[2].v = psID;
386 blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
387 buf1 += nbytes;
388 }
389 RF_ASSERT(pda == NULL);
390
391 memset((char *) &tracerec, 0, sizeof(tracerec));
392 rd_dag_h->tracerec = &tracerec;
393
394 if (rf_verifyParityDebug > 1) {
395 printf("raid%d: RAID1 parity verify read dag:\n",
396 raidPtr->raidid);
397 rf_PrintDAGList(rd_dag_h);
398 }
399 RF_LOCK_MUTEX(mcpair->mutex);
400 mcpair->flag = 0;
401 rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
402 (void *) mcpair);
403 while (mcpair->flag == 0) {
404 RF_WAIT_MCPAIR(mcpair);
405 }
406 RF_UNLOCK_MUTEX(mcpair->mutex);
407
408 if (rd_dag_h->status != rf_enable) {
409 RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
410 ret = RF_PARITY_COULD_NOT_VERIFY;
411 goto done;
412 }
413 /*
414 * buf1 is the beginning of the data blocks chunk
415 * buf2 is the beginning of the parity blocks chunk
416 */
417 buf1 = buf;
418 buf2 = buf + (nbytes * layoutPtr->numDataCol);
419 ret = RF_PARITY_OKAY;
420 /*
421 * bbufs is "bad bufs"- an array whose entries are the data
422 * column numbers where we had miscompares. (That is, column 0
423 * and column 1 of the array are mirror copies, and are considered
424 * "data column 0" for this purpose).
425 */
426 RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
427 allocList);
428 nbad = 0;
429 /*
430 * Check data vs "parity" (mirror copy).
431 */
432 for (i = 0; i < layoutPtr->numDataCol; i++) {
433 if (rf_verifyParityDebug) {
434 printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
435 raidPtr->raidid, nbytes, i, (long) buf1,
436 (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("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, 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("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
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 memset((char *) &tracerec, 0, 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("raid%d: RAID1 parity verify, returning %d\n",
535 raidPtr->raidid, ret);
536 }
537 return (ret);
538 }
539
540 int
541 rf_SubmitReconBufferRAID1(rbuf, keep_it, use_committed)
542 RF_ReconBuffer_t *rbuf; /* the recon buffer to submit */
543 int keep_it; /* whether we can keep this buffer or we have
544 * to return it */
545 int use_committed; /* whether to use a committed or an available
546 * recon buffer */
547 {
548 RF_ReconParityStripeStatus_t *pssPtr;
549 RF_ReconCtrl_t *reconCtrlPtr;
550 RF_RaidLayout_t *layoutPtr;
551 int retcode, created;
552 RF_CallbackDesc_t *cb, *p;
553 RF_ReconBuffer_t *t;
554 RF_Raid_t *raidPtr;
555 caddr_t ta;
556
557 retcode = 0;
558 created = 0;
559
560 raidPtr = rbuf->raidPtr;
561 layoutPtr = &raidPtr->Layout;
562 reconCtrlPtr = raidPtr->reconControl[rbuf->row];
563
564 RF_ASSERT(rbuf);
565 RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
566
567 if (rf_reconbufferDebug) {
568 printf("raid%d: RAID1 reconbuffer submission r%d c%d psid %ld ru%d (failed offset %ld)\n",
569 raidPtr->raidid, rbuf->row, rbuf->col,
570 (long) rbuf->parityStripeID, rbuf->which_ru,
571 (long) rbuf->failedDiskSectorOffset);
572 }
573 if (rf_reconDebug) {
574 printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
575 (long) rbuf->parityStripeID, (long) rbuf->buffer);
576 printf("RAID1 psid %ld %02x %02x %02x %02x %02x\n",
577 (long) rbuf->parityStripeID,
578 rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3],
579 rbuf->buffer[4]);
580 }
581 RF_LOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
582
583 RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
584
585 pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
586 rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created);
587 RF_ASSERT(pssPtr); /* if it didn't exist, we wouldn't have gotten
588 * an rbuf for it */
589
590 /*
591 * Since this is simple mirroring, the first submission for a stripe is also
592 * treated as the last.
593 */
594
595 t = NULL;
596 if (keep_it) {
597 if (rf_reconbufferDebug) {
598 printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
599 raidPtr->raidid);
600 }
601 t = rbuf;
602 } else {
603 if (use_committed) {
604 if (rf_reconbufferDebug) {
605 printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
606 }
607 t = reconCtrlPtr->committedRbufs;
608 RF_ASSERT(t);
609 reconCtrlPtr->committedRbufs = t->next;
610 t->next = NULL;
611 } else
612 if (reconCtrlPtr->floatingRbufs) {
613 if (rf_reconbufferDebug) {
614 printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
615 }
616 t = reconCtrlPtr->floatingRbufs;
617 reconCtrlPtr->floatingRbufs = t->next;
618 t->next = NULL;
619 }
620 }
621 if (t == NULL) {
622 if (rf_reconbufferDebug) {
623 printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
624 }
625 RF_ASSERT((keep_it == 0) && (use_committed == 0));
626 raidPtr->procsInBufWait++;
627 if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
628 && (raidPtr->numFullReconBuffers == 0)) {
629 /* ruh-ro */
630 RF_ERRORMSG("Buffer wait deadlock\n");
631 rf_PrintPSStatusTable(raidPtr, rbuf->row);
632 RF_PANIC();
633 }
634 pssPtr->flags |= RF_PSS_BUFFERWAIT;
635 cb = rf_AllocCallbackDesc();
636 cb->row = rbuf->row;
637 cb->col = rbuf->col;
638 cb->callbackArg.v = rbuf->parityStripeID;
639 cb->callbackArg2.v = rbuf->which_ru;
640 cb->next = NULL;
641 if (reconCtrlPtr->bufferWaitList == NULL) {
642 /* we are the wait list- lucky us */
643 reconCtrlPtr->bufferWaitList = cb;
644 } else {
645 /* append to wait list */
646 for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
647 p->next = cb;
648 }
649 retcode = 1;
650 goto out;
651 }
652 if (t != rbuf) {
653 t->row = rbuf->row;
654 t->col = reconCtrlPtr->fcol;
655 t->parityStripeID = rbuf->parityStripeID;
656 t->which_ru = rbuf->which_ru;
657 t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
658 t->spRow = rbuf->spRow;
659 t->spCol = rbuf->spCol;
660 t->spOffset = rbuf->spOffset;
661 /* Swap buffers. DANCE! */
662 ta = t->buffer;
663 t->buffer = rbuf->buffer;
664 rbuf->buffer = ta;
665 }
666 /*
667 * Use the rbuf we've been given as the target.
668 */
669 RF_ASSERT(pssPtr->rbuf == NULL);
670 pssPtr->rbuf = t;
671
672 t->count = 1;
673 /*
674 * Below, we use 1 for numDataCol (which is equal to the count in the
675 * previous line), so we'll always be done.
676 */
677 rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
678
679 out:
680 RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
681 RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
682 if (rf_reconbufferDebug) {
683 printf("raid%d: RAID1 rbuf submission: returning %d\n",
684 raidPtr->raidid, retcode);
685 }
686 return (retcode);
687 }
688