1 /* $NetBSD: rf_paritylogDiskMgr.c,v 1.13 2001/11/13 07:11:15 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 /* Code for flushing and reintegration operations related to parity logging. 29 * 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: rf_paritylogDiskMgr.c,v 1.13 2001/11/13 07:11:15 lukem Exp $"); 34 35 #include "rf_archs.h" 36 37 #if RF_INCLUDE_PARITYLOGGING > 0 38 39 #include <dev/raidframe/raidframevar.h> 40 41 #include "rf_threadstuff.h" 42 #include "rf_mcpair.h" 43 #include "rf_raid.h" 44 #include "rf_dag.h" 45 #include "rf_dagfuncs.h" 46 #include "rf_desc.h" 47 #include "rf_layout.h" 48 #include "rf_diskqueue.h" 49 #include "rf_paritylog.h" 50 #include "rf_general.h" 51 #include "rf_etimer.h" 52 #include "rf_paritylogging.h" 53 #include "rf_engine.h" 54 #include "rf_dagutils.h" 55 #include "rf_map.h" 56 #include "rf_parityscan.h" 57 58 #include "rf_paritylogDiskMgr.h" 59 60 static caddr_t AcquireReintBuffer(RF_RegionBufferQueue_t *); 61 62 static caddr_t 63 AcquireReintBuffer(pool) 64 RF_RegionBufferQueue_t *pool; 65 { 66 caddr_t bufPtr = NULL; 67 68 /* Return a region buffer from the free list (pool). If the free list 69 * is empty, WAIT. BLOCKING */ 70 71 RF_LOCK_MUTEX(pool->mutex); 72 if (pool->availableBuffers > 0) { 73 bufPtr = pool->buffers[pool->availBuffersIndex]; 74 pool->availableBuffers--; 75 pool->availBuffersIndex++; 76 if (pool->availBuffersIndex == pool->totalBuffers) 77 pool->availBuffersIndex = 0; 78 RF_UNLOCK_MUTEX(pool->mutex); 79 } else { 80 RF_PANIC(); /* should never happen in correct config, 81 * single reint */ 82 RF_WAIT_COND(pool->cond, pool->mutex); 83 } 84 return (bufPtr); 85 } 86 87 static void 88 ReleaseReintBuffer( 89 RF_RegionBufferQueue_t * pool, 90 caddr_t bufPtr) 91 { 92 /* Insert a region buffer (bufPtr) into the free list (pool). 93 * NON-BLOCKING */ 94 95 RF_LOCK_MUTEX(pool->mutex); 96 pool->availableBuffers++; 97 pool->buffers[pool->emptyBuffersIndex] = bufPtr; 98 pool->emptyBuffersIndex++; 99 if (pool->emptyBuffersIndex == pool->totalBuffers) 100 pool->emptyBuffersIndex = 0; 101 RF_ASSERT(pool->availableBuffers <= pool->totalBuffers); 102 RF_UNLOCK_MUTEX(pool->mutex); 103 RF_SIGNAL_COND(pool->cond); 104 } 105 106 107 108 static void 109 ReadRegionLog( 110 RF_RegionId_t regionID, 111 RF_MCPair_t * rrd_mcpair, 112 caddr_t regionBuffer, 113 RF_Raid_t * raidPtr, 114 RF_DagHeader_t ** rrd_dag_h, 115 RF_AllocListElem_t ** rrd_alloclist, 116 RF_PhysDiskAddr_t ** rrd_pda) 117 { 118 /* Initiate the read a region log from disk. Once initiated, return 119 * to the calling routine. 120 * 121 * NON-BLOCKING */ 122 123 RF_AccTraceEntry_t *tracerec; 124 RF_DagNode_t *rrd_rdNode; 125 126 /* create DAG to read region log from disk */ 127 rf_MakeAllocList(*rrd_alloclist); 128 *rrd_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, regionBuffer, 129 rf_DiskReadFunc, rf_DiskReadUndoFunc, 130 "Rrl", *rrd_alloclist, 131 RF_DAG_FLAGS_NONE, 132 RF_IO_NORMAL_PRIORITY); 133 134 /* create and initialize PDA for the core log */ 135 /* RF_Malloc(*rrd_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t 136 * *)); */ 137 *rrd_pda = rf_AllocPDAList(1); 138 rf_MapLogParityLogging(raidPtr, regionID, 0, &((*rrd_pda)->row), 139 &((*rrd_pda)->col), &((*rrd_pda)->startSector)); 140 (*rrd_pda)->numSector = raidPtr->regionInfo[regionID].capacity; 141 142 if ((*rrd_pda)->next) { 143 (*rrd_pda)->next = NULL; 144 printf("set rrd_pda->next to NULL\n"); 145 } 146 /* initialize DAG parameters */ 147 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *)); 148 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t)); 149 (*rrd_dag_h)->tracerec = tracerec; 150 rrd_rdNode = (*rrd_dag_h)->succedents[0]->succedents[0]; 151 rrd_rdNode->params[0].p = *rrd_pda; 152 /* rrd_rdNode->params[1] = regionBuffer; */ 153 rrd_rdNode->params[2].v = 0; 154 rrd_rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 155 0, 0, 0); 156 157 /* launch region log read dag */ 158 rf_DispatchDAG(*rrd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 159 (void *) rrd_mcpair); 160 } 161 162 163 164 static void 165 WriteCoreLog( 166 RF_ParityLog_t * log, 167 RF_MCPair_t * fwr_mcpair, 168 RF_Raid_t * raidPtr, 169 RF_DagHeader_t ** fwr_dag_h, 170 RF_AllocListElem_t ** fwr_alloclist, 171 RF_PhysDiskAddr_t ** fwr_pda) 172 { 173 RF_RegionId_t regionID = log->regionID; 174 RF_AccTraceEntry_t *tracerec; 175 RF_SectorNum_t regionOffset; 176 RF_DagNode_t *fwr_wrNode; 177 178 /* Initiate the write of a core log to a region log disk. Once 179 * initiated, return to the calling routine. 180 * 181 * NON-BLOCKING */ 182 183 /* create DAG to write a core log to a region log disk */ 184 rf_MakeAllocList(*fwr_alloclist); 185 *fwr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, log->bufPtr, 186 rf_DiskWriteFunc, rf_DiskWriteUndoFunc, 187 "Wcl", *fwr_alloclist, RF_DAG_FLAGS_NONE, RF_IO_NORMAL_PRIORITY); 188 189 /* create and initialize PDA for the region log */ 190 /* RF_Malloc(*fwr_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t 191 * *)); */ 192 *fwr_pda = rf_AllocPDAList(1); 193 regionOffset = log->diskOffset; 194 rf_MapLogParityLogging(raidPtr, regionID, regionOffset, 195 &((*fwr_pda)->row), &((*fwr_pda)->col), 196 &((*fwr_pda)->startSector)); 197 (*fwr_pda)->numSector = raidPtr->numSectorsPerLog; 198 199 /* initialize DAG parameters */ 200 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *)); 201 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t)); 202 (*fwr_dag_h)->tracerec = tracerec; 203 fwr_wrNode = (*fwr_dag_h)->succedents[0]->succedents[0]; 204 fwr_wrNode->params[0].p = *fwr_pda; 205 /* fwr_wrNode->params[1] = log->bufPtr; */ 206 fwr_wrNode->params[2].v = 0; 207 fwr_wrNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 208 0, 0, 0); 209 210 /* launch the dag to write the core log to disk */ 211 rf_DispatchDAG(*fwr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 212 (void *) fwr_mcpair); 213 } 214 215 216 static void 217 ReadRegionParity( 218 RF_RegionId_t regionID, 219 RF_MCPair_t * prd_mcpair, 220 caddr_t parityBuffer, 221 RF_Raid_t * raidPtr, 222 RF_DagHeader_t ** prd_dag_h, 223 RF_AllocListElem_t ** prd_alloclist, 224 RF_PhysDiskAddr_t ** prd_pda) 225 { 226 /* Initiate the read region parity from disk. Once initiated, return 227 * to the calling routine. 228 * 229 * NON-BLOCKING */ 230 231 RF_AccTraceEntry_t *tracerec; 232 RF_DagNode_t *prd_rdNode; 233 234 /* create DAG to read region parity from disk */ 235 rf_MakeAllocList(*prd_alloclist); 236 *prd_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, NULL, rf_DiskReadFunc, 237 rf_DiskReadUndoFunc, "Rrp", 238 *prd_alloclist, RF_DAG_FLAGS_NONE, 239 RF_IO_NORMAL_PRIORITY); 240 241 /* create and initialize PDA for region parity */ 242 /* RF_Malloc(*prd_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t 243 * *)); */ 244 *prd_pda = rf_AllocPDAList(1); 245 rf_MapRegionParity(raidPtr, regionID, &((*prd_pda)->row), 246 &((*prd_pda)->col), &((*prd_pda)->startSector), 247 &((*prd_pda)->numSector)); 248 if (rf_parityLogDebug) 249 printf("[reading %d sectors of parity from region %d]\n", 250 (int) (*prd_pda)->numSector, regionID); 251 if ((*prd_pda)->next) { 252 (*prd_pda)->next = NULL; 253 printf("set prd_pda->next to NULL\n"); 254 } 255 /* initialize DAG parameters */ 256 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *)); 257 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t)); 258 (*prd_dag_h)->tracerec = tracerec; 259 prd_rdNode = (*prd_dag_h)->succedents[0]->succedents[0]; 260 prd_rdNode->params[0].p = *prd_pda; 261 prd_rdNode->params[1].p = parityBuffer; 262 prd_rdNode->params[2].v = 0; 263 prd_rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 264 0, 0, 0); 265 if (rf_validateDAGDebug) 266 rf_ValidateDAG(*prd_dag_h); 267 /* launch region parity read dag */ 268 rf_DispatchDAG(*prd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 269 (void *) prd_mcpair); 270 } 271 272 static void 273 WriteRegionParity( 274 RF_RegionId_t regionID, 275 RF_MCPair_t * pwr_mcpair, 276 caddr_t parityBuffer, 277 RF_Raid_t * raidPtr, 278 RF_DagHeader_t ** pwr_dag_h, 279 RF_AllocListElem_t ** pwr_alloclist, 280 RF_PhysDiskAddr_t ** pwr_pda) 281 { 282 /* Initiate the write of region parity to disk. Once initiated, return 283 * to the calling routine. 284 * 285 * NON-BLOCKING */ 286 287 RF_AccTraceEntry_t *tracerec; 288 RF_DagNode_t *pwr_wrNode; 289 290 /* create DAG to write region log from disk */ 291 rf_MakeAllocList(*pwr_alloclist); 292 *pwr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, parityBuffer, 293 rf_DiskWriteFunc, rf_DiskWriteUndoFunc, 294 "Wrp", *pwr_alloclist, 295 RF_DAG_FLAGS_NONE, 296 RF_IO_NORMAL_PRIORITY); 297 298 /* create and initialize PDA for region parity */ 299 /* RF_Malloc(*pwr_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t 300 * *)); */ 301 *pwr_pda = rf_AllocPDAList(1); 302 rf_MapRegionParity(raidPtr, regionID, &((*pwr_pda)->row), 303 &((*pwr_pda)->col), &((*pwr_pda)->startSector), 304 &((*pwr_pda)->numSector)); 305 306 /* initialize DAG parameters */ 307 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *)); 308 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t)); 309 (*pwr_dag_h)->tracerec = tracerec; 310 pwr_wrNode = (*pwr_dag_h)->succedents[0]->succedents[0]; 311 pwr_wrNode->params[0].p = *pwr_pda; 312 /* pwr_wrNode->params[1] = parityBuffer; */ 313 pwr_wrNode->params[2].v = 0; 314 pwr_wrNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 315 0, 0, 0); 316 317 /* launch the dag to write region parity to disk */ 318 rf_DispatchDAG(*pwr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 319 (void *) pwr_mcpair); 320 } 321 322 static void 323 FlushLogsToDisk( 324 RF_Raid_t * raidPtr, 325 RF_ParityLog_t * logList) 326 { 327 /* Flush a linked list of core logs to the log disk. Logs contain the 328 * disk location where they should be written. Logs were written in 329 * FIFO order and that order must be preserved. 330 * 331 * Recommended optimizations: 1) allow multiple flushes to occur 332 * simultaneously 2) coalesce contiguous flush operations 333 * 334 * BLOCKING */ 335 336 RF_ParityLog_t *log; 337 RF_RegionId_t regionID; 338 RF_MCPair_t *fwr_mcpair; 339 RF_DagHeader_t *fwr_dag_h; 340 RF_AllocListElem_t *fwr_alloclist; 341 RF_PhysDiskAddr_t *fwr_pda; 342 343 fwr_mcpair = rf_AllocMCPair(); 344 RF_LOCK_MUTEX(fwr_mcpair->mutex); 345 346 RF_ASSERT(logList); 347 log = logList; 348 while (log) { 349 regionID = log->regionID; 350 351 /* create and launch a DAG to write the core log */ 352 if (rf_parityLogDebug) 353 printf("[initiating write of core log for region %d]\n", regionID); 354 fwr_mcpair->flag = RF_FALSE; 355 WriteCoreLog(log, fwr_mcpair, raidPtr, &fwr_dag_h, 356 &fwr_alloclist, &fwr_pda); 357 358 /* wait for the DAG to complete */ 359 while (!fwr_mcpair->flag) 360 RF_WAIT_COND(fwr_mcpair->cond, fwr_mcpair->mutex); 361 if (fwr_dag_h->status != rf_enable) { 362 RF_ERRORMSG1("Unable to write core log to disk (region %d)\n", regionID); 363 RF_ASSERT(0); 364 } 365 /* RF_Free(fwr_pda, sizeof(RF_PhysDiskAddr_t)); */ 366 rf_FreePhysDiskAddr(fwr_pda); 367 rf_FreeDAG(fwr_dag_h); 368 rf_FreeAllocList(fwr_alloclist); 369 370 log = log->next; 371 } 372 RF_UNLOCK_MUTEX(fwr_mcpair->mutex); 373 rf_FreeMCPair(fwr_mcpair); 374 rf_ReleaseParityLogs(raidPtr, logList); 375 } 376 377 static void 378 ReintegrateRegion( 379 RF_Raid_t * raidPtr, 380 RF_RegionId_t regionID, 381 RF_ParityLog_t * coreLog) 382 { 383 RF_MCPair_t *rrd_mcpair = NULL, *prd_mcpair, *pwr_mcpair; 384 RF_DagHeader_t *rrd_dag_h, *prd_dag_h, *pwr_dag_h; 385 RF_AllocListElem_t *rrd_alloclist, *prd_alloclist, *pwr_alloclist; 386 RF_PhysDiskAddr_t *rrd_pda, *prd_pda, *pwr_pda; 387 caddr_t parityBuffer, regionBuffer = NULL; 388 389 /* Reintegrate a region (regionID). 390 * 391 * 1. acquire region and parity buffers 392 * 2. read log from disk 393 * 3. read parity from disk 394 * 4. apply log to parity 395 * 5. apply core log to parity 396 * 6. write new parity to disk 397 * 398 * BLOCKING */ 399 400 if (rf_parityLogDebug) 401 printf("[reintegrating region %d]\n", regionID); 402 403 /* initiate read of region parity */ 404 if (rf_parityLogDebug) 405 printf("[initiating read of parity for region %d]\n",regionID); 406 parityBuffer = AcquireReintBuffer(&raidPtr->parityBufferPool); 407 prd_mcpair = rf_AllocMCPair(); 408 RF_LOCK_MUTEX(prd_mcpair->mutex); 409 prd_mcpair->flag = RF_FALSE; 410 ReadRegionParity(regionID, prd_mcpair, parityBuffer, raidPtr, 411 &prd_dag_h, &prd_alloclist, &prd_pda); 412 413 /* if region log nonempty, initiate read */ 414 if (raidPtr->regionInfo[regionID].diskCount > 0) { 415 if (rf_parityLogDebug) 416 printf("[initiating read of disk log for region %d]\n", 417 regionID); 418 regionBuffer = AcquireReintBuffer(&raidPtr->regionBufferPool); 419 rrd_mcpair = rf_AllocMCPair(); 420 RF_LOCK_MUTEX(rrd_mcpair->mutex); 421 rrd_mcpair->flag = RF_FALSE; 422 ReadRegionLog(regionID, rrd_mcpair, regionBuffer, raidPtr, 423 &rrd_dag_h, &rrd_alloclist, &rrd_pda); 424 } 425 /* wait on read of region parity to complete */ 426 while (!prd_mcpair->flag) { 427 RF_WAIT_COND(prd_mcpair->cond, prd_mcpair->mutex); 428 } 429 RF_UNLOCK_MUTEX(prd_mcpair->mutex); 430 if (prd_dag_h->status != rf_enable) { 431 RF_ERRORMSG("Unable to read parity from disk\n"); 432 /* add code to fail the parity disk */ 433 RF_ASSERT(0); 434 } 435 /* apply core log to parity */ 436 /* if (coreLog) ApplyLogsToParity(coreLog, parityBuffer); */ 437 438 if (raidPtr->regionInfo[regionID].diskCount > 0) { 439 /* wait on read of region log to complete */ 440 while (!rrd_mcpair->flag) 441 RF_WAIT_COND(rrd_mcpair->cond, rrd_mcpair->mutex); 442 RF_UNLOCK_MUTEX(rrd_mcpair->mutex); 443 if (rrd_dag_h->status != rf_enable) { 444 RF_ERRORMSG("Unable to read region log from disk\n"); 445 /* add code to fail the log disk */ 446 RF_ASSERT(0); 447 } 448 /* apply region log to parity */ 449 /* ApplyRegionToParity(regionID, regionBuffer, parityBuffer); */ 450 /* release resources associated with region log */ 451 /* RF_Free(rrd_pda, sizeof(RF_PhysDiskAddr_t)); */ 452 rf_FreePhysDiskAddr(rrd_pda); 453 rf_FreeDAG(rrd_dag_h); 454 rf_FreeAllocList(rrd_alloclist); 455 rf_FreeMCPair(rrd_mcpair); 456 ReleaseReintBuffer(&raidPtr->regionBufferPool, regionBuffer); 457 } 458 /* write reintegrated parity to disk */ 459 if (rf_parityLogDebug) 460 printf("[initiating write of parity for region %d]\n", 461 regionID); 462 pwr_mcpair = rf_AllocMCPair(); 463 RF_LOCK_MUTEX(pwr_mcpair->mutex); 464 pwr_mcpair->flag = RF_FALSE; 465 WriteRegionParity(regionID, pwr_mcpair, parityBuffer, raidPtr, 466 &pwr_dag_h, &pwr_alloclist, &pwr_pda); 467 while (!pwr_mcpair->flag) 468 RF_WAIT_COND(pwr_mcpair->cond, pwr_mcpair->mutex); 469 RF_UNLOCK_MUTEX(pwr_mcpair->mutex); 470 if (pwr_dag_h->status != rf_enable) { 471 RF_ERRORMSG("Unable to write parity to disk\n"); 472 /* add code to fail the parity disk */ 473 RF_ASSERT(0); 474 } 475 /* release resources associated with read of old parity */ 476 /* RF_Free(prd_pda, sizeof(RF_PhysDiskAddr_t)); */ 477 rf_FreePhysDiskAddr(prd_pda); 478 rf_FreeDAG(prd_dag_h); 479 rf_FreeAllocList(prd_alloclist); 480 rf_FreeMCPair(prd_mcpair); 481 482 /* release resources associated with write of new parity */ 483 ReleaseReintBuffer(&raidPtr->parityBufferPool, parityBuffer); 484 /* RF_Free(pwr_pda, sizeof(RF_PhysDiskAddr_t)); */ 485 rf_FreePhysDiskAddr(pwr_pda); 486 rf_FreeDAG(pwr_dag_h); 487 rf_FreeAllocList(pwr_alloclist); 488 rf_FreeMCPair(pwr_mcpair); 489 490 if (rf_parityLogDebug) 491 printf("[finished reintegrating region %d]\n", regionID); 492 } 493 494 495 496 static void 497 ReintegrateLogs( 498 RF_Raid_t * raidPtr, 499 RF_ParityLog_t * logList) 500 { 501 RF_ParityLog_t *log, *freeLogList = NULL; 502 RF_ParityLogData_t *logData, *logDataList; 503 RF_RegionId_t regionID; 504 505 RF_ASSERT(logList); 506 while (logList) { 507 log = logList; 508 logList = logList->next; 509 log->next = NULL; 510 regionID = log->regionID; 511 ReintegrateRegion(raidPtr, regionID, log); 512 log->numRecords = 0; 513 514 /* remove all items which are blocked on reintegration of this 515 * region */ 516 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 517 logData = rf_SearchAndDequeueParityLogData(raidPtr, regionID, 518 &raidPtr->parityLogDiskQueue.reintBlockHead, 519 &raidPtr->parityLogDiskQueue.reintBlockTail, 520 RF_TRUE); 521 logDataList = logData; 522 while (logData) { 523 logData->next = rf_SearchAndDequeueParityLogData( 524 raidPtr, regionID, 525 &raidPtr->parityLogDiskQueue.reintBlockHead, 526 &raidPtr->parityLogDiskQueue.reintBlockTail, 527 RF_TRUE); 528 logData = logData->next; 529 } 530 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 531 532 /* process blocked log data and clear reintInProgress flag for 533 * this region */ 534 if (logDataList) 535 rf_ParityLogAppend(logDataList, RF_TRUE, &log, RF_TRUE); 536 else { 537 /* Enable flushing for this region. Holding both 538 * locks provides a synchronization barrier with 539 * DumpParityLogToDisk */ 540 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].mutex); 541 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].reintMutex); 542 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 543 raidPtr->regionInfo[regionID].diskCount = 0; 544 raidPtr->regionInfo[regionID].reintInProgress = RF_FALSE; 545 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].mutex); 546 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].reintMutex); /* flushing is now 547 * enabled */ 548 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 549 } 550 /* if log wasn't used, attach it to the list of logs to be 551 * returned */ 552 if (log) { 553 log->next = freeLogList; 554 freeLogList = log; 555 } 556 } 557 if (freeLogList) 558 rf_ReleaseParityLogs(raidPtr, freeLogList); 559 } 560 561 int 562 rf_ShutdownLogging(RF_Raid_t * raidPtr) 563 { 564 /* shutdown parity logging 1) disable parity logging in all regions 2) 565 * reintegrate all regions */ 566 567 RF_SectorCount_t diskCount; 568 RF_RegionId_t regionID; 569 RF_ParityLog_t *log; 570 571 if (rf_parityLogDebug) 572 printf("[shutting down parity logging]\n"); 573 /* Since parity log maps are volatile, we must reintegrate all 574 * regions. */ 575 if (rf_forceParityLogReint) { 576 for (regionID = 0; regionID < rf_numParityRegions; regionID++) { 577 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].mutex); 578 raidPtr->regionInfo[regionID].loggingEnabled = 579 RF_FALSE; 580 log = raidPtr->regionInfo[regionID].coreLog; 581 raidPtr->regionInfo[regionID].coreLog = NULL; 582 diskCount = raidPtr->regionInfo[regionID].diskCount; 583 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].mutex); 584 if (diskCount > 0 || log != NULL) 585 ReintegrateRegion(raidPtr, regionID, log); 586 if (log != NULL) 587 rf_ReleaseParityLogs(raidPtr, log); 588 } 589 } 590 if (rf_parityLogDebug) { 591 printf("[parity logging disabled]\n"); 592 printf("[should be done!]\n"); 593 } 594 return (0); 595 } 596 597 int 598 rf_ParityLoggingDiskManager(RF_Raid_t * raidPtr) 599 { 600 RF_ParityLog_t *reintQueue, *flushQueue; 601 int workNeeded, done = RF_FALSE; 602 int s; 603 604 /* Main program for parity logging disk thread. This routine waits 605 * for work to appear in either the flush or reintegration queues and 606 * is responsible for flushing core logs to the log disk as well as 607 * reintegrating parity regions. 608 * 609 * BLOCKING */ 610 611 s = splbio(); 612 613 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 614 615 /* 616 * Inform our creator that we're running. Don't bother doing the 617 * mutex lock/unlock dance- we locked above, and we'll unlock 618 * below with nothing to do, yet. 619 */ 620 raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_RUNNING; 621 RF_SIGNAL_COND(raidPtr->parityLogDiskQueue.cond); 622 623 /* empty the work queues */ 624 flushQueue = raidPtr->parityLogDiskQueue.flushQueue; 625 raidPtr->parityLogDiskQueue.flushQueue = NULL; 626 reintQueue = raidPtr->parityLogDiskQueue.reintQueue; 627 raidPtr->parityLogDiskQueue.reintQueue = NULL; 628 workNeeded = (flushQueue || reintQueue); 629 630 while (!done) { 631 while (workNeeded) { 632 /* First, flush all logs in the flush queue, freeing 633 * buffers Second, reintegrate all regions which are 634 * reported as full. Third, append queued log data 635 * until blocked. 636 * 637 * Note: Incoming appends (ParityLogAppend) can block on 638 * either 1. empty buffer pool 2. region under 639 * reintegration To preserve a global FIFO ordering of 640 * appends, buffers are not released to the world 641 * until those appends blocked on buffers are removed 642 * from the append queue. Similarly, regions which 643 * are reintegrated are not opened for general use 644 * until the append queue has been emptied. */ 645 646 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 647 648 /* empty flushQueue, using free'd log buffers to 649 * process bufTail */ 650 if (flushQueue) 651 FlushLogsToDisk(raidPtr, flushQueue); 652 653 /* empty reintQueue, flushing from reintTail as we go */ 654 if (reintQueue) 655 ReintegrateLogs(raidPtr, reintQueue); 656 657 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 658 flushQueue = raidPtr->parityLogDiskQueue.flushQueue; 659 raidPtr->parityLogDiskQueue.flushQueue = NULL; 660 reintQueue = raidPtr->parityLogDiskQueue.reintQueue; 661 raidPtr->parityLogDiskQueue.reintQueue = NULL; 662 workNeeded = (flushQueue || reintQueue); 663 } 664 /* no work is needed at this point */ 665 if (raidPtr->parityLogDiskQueue.threadState & RF_PLOG_TERMINATE) { 666 /* shutdown parity logging 1. disable parity logging 667 * in all regions 2. reintegrate all regions */ 668 done = RF_TRUE; /* thread disabled, no work needed */ 669 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 670 rf_ShutdownLogging(raidPtr); 671 } 672 if (!done) { 673 /* thread enabled, no work needed, so sleep */ 674 if (rf_parityLogDebug) 675 printf("[parity logging disk manager sleeping]\n"); 676 RF_WAIT_COND(raidPtr->parityLogDiskQueue.cond, 677 raidPtr->parityLogDiskQueue.mutex); 678 if (rf_parityLogDebug) 679 printf("[parity logging disk manager just woke up]\n"); 680 flushQueue = raidPtr->parityLogDiskQueue.flushQueue; 681 raidPtr->parityLogDiskQueue.flushQueue = NULL; 682 reintQueue = raidPtr->parityLogDiskQueue.reintQueue; 683 raidPtr->parityLogDiskQueue.reintQueue = NULL; 684 workNeeded = (flushQueue || reintQueue); 685 } 686 } 687 /* 688 * Announce that we're done. 689 */ 690 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 691 raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_SHUTDOWN; 692 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex); 693 RF_SIGNAL_COND(raidPtr->parityLogDiskQueue.cond); 694 695 splx(s); 696 697 /* 698 * In the NetBSD kernel, the thread must exit; returning would 699 * cause the proc trampoline to attempt to return to userspace. 700 */ 701 kthread_exit(0); /* does not return */ 702 } 703 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */ 704