1 /* $NetBSD: rf_raid1.c,v 1.8 2001/11/13 07:11:16 lukem 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 1.8 2001/11/13 07:11:16 lukem 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_verifyParityDebug) { 324 printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n", 325 raidPtr->raidid, (long) buf, bcount, (long) buf, 326 (long) buf + bcount); 327 } 328 /* 329 * Generate a DAG which will read the entire stripe- then we can 330 * just compare data chunks versus "parity" chunks. 331 */ 332 333 rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf, 334 rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags, 335 RF_IO_NORMAL_PRIORITY); 336 if (rd_dag_h == NULL) 337 goto done; 338 blockNode = rd_dag_h->succedents[0]; 339 unblockNode = blockNode->succedents[0]->succedents[0]; 340 341 /* 342 * Map the access to physical disk addresses (PDAs)- this will 343 * get us both a list of data addresses, and "parity" addresses 344 * (which are really mirror copies). 345 */ 346 asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe, 347 buf, RF_DONT_REMAP); 348 aasm = asm_h->stripeMap; 349 350 buf1 = buf; 351 /* 352 * Loop through the data blocks, setting up read nodes for each. 353 */ 354 for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) { 355 RF_ASSERT(pda); 356 357 rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1); 358 359 RF_ASSERT(pda->numSector != 0); 360 if (rf_TryToRedirectPDA(raidPtr, pda, 0)) { 361 /* cannot verify parity with dead disk */ 362 goto done; 363 } 364 pda->bufPtr = buf1; 365 blockNode->succedents[i]->params[0].p = pda; 366 blockNode->succedents[i]->params[1].p = buf1; 367 blockNode->succedents[i]->params[2].v = psID; 368 blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 369 buf1 += nbytes; 370 } 371 RF_ASSERT(pda == NULL); 372 /* 373 * keep i, buf1 running 374 * 375 * Loop through parity blocks, setting up read nodes for each. 376 */ 377 for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) { 378 RF_ASSERT(pda); 379 rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1); 380 RF_ASSERT(pda->numSector != 0); 381 if (rf_TryToRedirectPDA(raidPtr, pda, 0)) { 382 /* cannot verify parity with dead disk */ 383 goto done; 384 } 385 pda->bufPtr = buf1; 386 blockNode->succedents[i]->params[0].p = pda; 387 blockNode->succedents[i]->params[1].p = buf1; 388 blockNode->succedents[i]->params[2].v = psID; 389 blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 390 buf1 += nbytes; 391 } 392 RF_ASSERT(pda == NULL); 393 394 memset((char *) &tracerec, 0, sizeof(tracerec)); 395 rd_dag_h->tracerec = &tracerec; 396 397 if (rf_verifyParityDebug > 1) { 398 printf("raid%d: RAID1 parity verify read dag:\n", 399 raidPtr->raidid); 400 rf_PrintDAGList(rd_dag_h); 401 } 402 RF_LOCK_MUTEX(mcpair->mutex); 403 mcpair->flag = 0; 404 rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 405 (void *) mcpair); 406 while (mcpair->flag == 0) { 407 RF_WAIT_MCPAIR(mcpair); 408 } 409 RF_UNLOCK_MUTEX(mcpair->mutex); 410 411 if (rd_dag_h->status != rf_enable) { 412 RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n"); 413 ret = RF_PARITY_COULD_NOT_VERIFY; 414 goto done; 415 } 416 /* 417 * buf1 is the beginning of the data blocks chunk 418 * buf2 is the beginning of the parity blocks chunk 419 */ 420 buf1 = buf; 421 buf2 = buf + (nbytes * layoutPtr->numDataCol); 422 ret = RF_PARITY_OKAY; 423 /* 424 * bbufs is "bad bufs"- an array whose entries are the data 425 * column numbers where we had miscompares. (That is, column 0 426 * and column 1 of the array are mirror copies, and are considered 427 * "data column 0" for this purpose). 428 */ 429 RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *), 430 allocList); 431 nbad = 0; 432 /* 433 * Check data vs "parity" (mirror copy). 434 */ 435 for (i = 0; i < layoutPtr->numDataCol; i++) { 436 if (rf_verifyParityDebug) { 437 printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n", 438 raidPtr->raidid, nbytes, i, (long) buf1, 439 (long) buf2, (long) buf); 440 } 441 ret = memcmp(buf1, buf2, nbytes); 442 if (ret) { 443 if (rf_verifyParityDebug > 1) { 444 for (j = 0; j < nbytes; j++) { 445 if (buf1[j] != buf2[j]) 446 break; 447 } 448 printf("psid=%ld j=%d\n", (long) psID, j); 449 printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff, 450 buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff); 451 printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff, 452 buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff); 453 } 454 if (rf_verifyParityDebug) { 455 printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i); 456 } 457 /* 458 * Parity is bad. Keep track of which columns were bad. 459 */ 460 if (bbufs) 461 bbufs[nbad] = i; 462 nbad++; 463 ret = RF_PARITY_BAD; 464 } 465 buf1 += nbytes; 466 buf2 += nbytes; 467 } 468 469 if ((ret != RF_PARITY_OKAY) && correct_it) { 470 ret = RF_PARITY_COULD_NOT_CORRECT; 471 if (rf_verifyParityDebug) { 472 printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid); 473 } 474 if (bbufs == NULL) 475 goto done; 476 /* 477 * Make a DAG with one write node for each bad unit. We'll simply 478 * write the contents of the data unit onto the parity unit for 479 * correction. (It's possible that the mirror copy was the correct 480 * copy, and that we're spooging good data by writing bad over it, 481 * but there's no way we can know that. 482 */ 483 wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, buf, 484 rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags, 485 RF_IO_NORMAL_PRIORITY); 486 if (wr_dag_h == NULL) 487 goto done; 488 wrBlock = wr_dag_h->succedents[0]; 489 /* 490 * Fill in a write node for each bad compare. 491 */ 492 for (i = 0; i < nbad; i++) { 493 j = i + layoutPtr->numDataCol; 494 pda = blockNode->succedents[j]->params[0].p; 495 pda->bufPtr = blockNode->succedents[i]->params[1].p; 496 wrBlock->succedents[i]->params[0].p = pda; 497 wrBlock->succedents[i]->params[1].p = pda->bufPtr; 498 wrBlock->succedents[i]->params[2].v = psID; 499 wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 500 } 501 memset((char *) &tracerec, 0, sizeof(tracerec)); 502 wr_dag_h->tracerec = &tracerec; 503 if (rf_verifyParityDebug > 1) { 504 printf("Parity verify write dag:\n"); 505 rf_PrintDAGList(wr_dag_h); 506 } 507 RF_LOCK_MUTEX(mcpair->mutex); 508 mcpair->flag = 0; 509 /* fire off the write DAG */ 510 rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 511 (void *) mcpair); 512 while (!mcpair->flag) { 513 RF_WAIT_COND(mcpair->cond, mcpair->mutex); 514 } 515 RF_UNLOCK_MUTEX(mcpair->mutex); 516 if (wr_dag_h->status != rf_enable) { 517 RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n"); 518 goto done; 519 } 520 ret = RF_PARITY_CORRECTED; 521 } 522 done: 523 /* 524 * All done. We might've gotten here without doing part of the function, 525 * so cleanup what we have to and return our running status. 526 */ 527 if (asm_h) 528 rf_FreeAccessStripeMap(asm_h); 529 if (rd_dag_h) 530 rf_FreeDAG(rd_dag_h); 531 if (wr_dag_h) 532 rf_FreeDAG(wr_dag_h); 533 if (mcpair) 534 rf_FreeMCPair(mcpair); 535 rf_FreeAllocList(allocList); 536 if (rf_verifyParityDebug) { 537 printf("raid%d: RAID1 parity verify, returning %d\n", 538 raidPtr->raidid, ret); 539 } 540 return (ret); 541 } 542 543 int 544 rf_SubmitReconBufferRAID1(rbuf, keep_it, use_committed) 545 RF_ReconBuffer_t *rbuf; /* the recon buffer to submit */ 546 int keep_it; /* whether we can keep this buffer or we have 547 * to return it */ 548 int use_committed; /* whether to use a committed or an available 549 * recon buffer */ 550 { 551 RF_ReconParityStripeStatus_t *pssPtr; 552 RF_ReconCtrl_t *reconCtrlPtr; 553 RF_RaidLayout_t *layoutPtr; 554 int retcode, created; 555 RF_CallbackDesc_t *cb, *p; 556 RF_ReconBuffer_t *t; 557 RF_Raid_t *raidPtr; 558 caddr_t ta; 559 560 retcode = 0; 561 created = 0; 562 563 raidPtr = rbuf->raidPtr; 564 layoutPtr = &raidPtr->Layout; 565 reconCtrlPtr = raidPtr->reconControl[rbuf->row]; 566 567 RF_ASSERT(rbuf); 568 RF_ASSERT(rbuf->col != reconCtrlPtr->fcol); 569 570 if (rf_reconbufferDebug) { 571 printf("raid%d: RAID1 reconbuffer submission r%d c%d psid %ld ru%d (failed offset %ld)\n", 572 raidPtr->raidid, rbuf->row, rbuf->col, 573 (long) rbuf->parityStripeID, rbuf->which_ru, 574 (long) rbuf->failedDiskSectorOffset); 575 } 576 if (rf_reconDebug) { 577 printf("RAID1 reconbuffer submit psid %ld buf %lx\n", 578 (long) rbuf->parityStripeID, (long) rbuf->buffer); 579 printf("RAID1 psid %ld %02x %02x %02x %02x %02x\n", 580 (long) rbuf->parityStripeID, 581 rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3], 582 rbuf->buffer[4]); 583 } 584 RF_LOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID); 585 586 RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex); 587 588 pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable, 589 rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created); 590 RF_ASSERT(pssPtr); /* if it didn't exist, we wouldn't have gotten 591 * an rbuf for it */ 592 593 /* 594 * Since this is simple mirroring, the first submission for a stripe is also 595 * treated as the last. 596 */ 597 598 t = NULL; 599 if (keep_it) { 600 if (rf_reconbufferDebug) { 601 printf("raid%d: RAID1 rbuf submission: keeping rbuf\n", 602 raidPtr->raidid); 603 } 604 t = rbuf; 605 } else { 606 if (use_committed) { 607 if (rf_reconbufferDebug) { 608 printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid); 609 } 610 t = reconCtrlPtr->committedRbufs; 611 RF_ASSERT(t); 612 reconCtrlPtr->committedRbufs = t->next; 613 t->next = NULL; 614 } else 615 if (reconCtrlPtr->floatingRbufs) { 616 if (rf_reconbufferDebug) { 617 printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid); 618 } 619 t = reconCtrlPtr->floatingRbufs; 620 reconCtrlPtr->floatingRbufs = t->next; 621 t->next = NULL; 622 } 623 } 624 if (t == NULL) { 625 if (rf_reconbufferDebug) { 626 printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid); 627 } 628 RF_ASSERT((keep_it == 0) && (use_committed == 0)); 629 raidPtr->procsInBufWait++; 630 if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1)) 631 && (raidPtr->numFullReconBuffers == 0)) { 632 /* ruh-ro */ 633 RF_ERRORMSG("Buffer wait deadlock\n"); 634 rf_PrintPSStatusTable(raidPtr, rbuf->row); 635 RF_PANIC(); 636 } 637 pssPtr->flags |= RF_PSS_BUFFERWAIT; 638 cb = rf_AllocCallbackDesc(); 639 cb->row = rbuf->row; 640 cb->col = rbuf->col; 641 cb->callbackArg.v = rbuf->parityStripeID; 642 cb->callbackArg2.v = rbuf->which_ru; 643 cb->next = NULL; 644 if (reconCtrlPtr->bufferWaitList == NULL) { 645 /* we are the wait list- lucky us */ 646 reconCtrlPtr->bufferWaitList = cb; 647 } else { 648 /* append to wait list */ 649 for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next); 650 p->next = cb; 651 } 652 retcode = 1; 653 goto out; 654 } 655 if (t != rbuf) { 656 t->row = rbuf->row; 657 t->col = reconCtrlPtr->fcol; 658 t->parityStripeID = rbuf->parityStripeID; 659 t->which_ru = rbuf->which_ru; 660 t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset; 661 t->spRow = rbuf->spRow; 662 t->spCol = rbuf->spCol; 663 t->spOffset = rbuf->spOffset; 664 /* Swap buffers. DANCE! */ 665 ta = t->buffer; 666 t->buffer = rbuf->buffer; 667 rbuf->buffer = ta; 668 } 669 /* 670 * Use the rbuf we've been given as the target. 671 */ 672 RF_ASSERT(pssPtr->rbuf == NULL); 673 pssPtr->rbuf = t; 674 675 t->count = 1; 676 /* 677 * Below, we use 1 for numDataCol (which is equal to the count in the 678 * previous line), so we'll always be done. 679 */ 680 rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1); 681 682 out: 683 RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID); 684 RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex); 685 if (rf_reconbufferDebug) { 686 printf("raid%d: RAID1 rbuf submission: returning %d\n", 687 raidPtr->raidid, retcode); 688 } 689 return (retcode); 690 } 691