1 /* $NetBSD: rf_paritylogDiskMgr.c,v 1.16 2004/01/10 00:56:28 oster Exp $ */
2 /*
3 * Copyright (c) 1995 Carnegie-Mellon University.
4 * All rights reserved.
5 *
6 * Author: William V. Courtright II
7 *
8 * Permission to use, copy, modify and distribute this software and
9 * its documentation is hereby granted, provided that both the copyright
10 * notice and this permission notice appear in all copies of the
11 * software, derivative works or modified versions, and any portions
12 * thereof, and that both notices appear in supporting documentation.
13 *
14 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17 *
18 * Carnegie Mellon requests users of this software to return to
19 *
20 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
21 * School of Computer Science
22 * Carnegie Mellon University
23 * Pittsburgh PA 15213-3890
24 *
25 * any improvements or extensions that they make and grant Carnegie the
26 * rights to redistribute these changes.
27 */
28 /* 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 1.16 2004/01/10 00:56:28 oster 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,
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, 0);
155
156 /* launch region log read dag */
157 rf_DispatchDAG(*rrd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
158 (void *) rrd_mcpair);
159 }
160
161
162
163 static void
164 WriteCoreLog(
165 RF_ParityLog_t * log,
166 RF_MCPair_t * fwr_mcpair,
167 RF_Raid_t * raidPtr,
168 RF_DagHeader_t ** fwr_dag_h,
169 RF_AllocListElem_t ** fwr_alloclist,
170 RF_PhysDiskAddr_t ** fwr_pda)
171 {
172 RF_RegionId_t regionID = log->regionID;
173 RF_AccTraceEntry_t *tracerec;
174 RF_SectorNum_t regionOffset;
175 RF_DagNode_t *fwr_wrNode;
176
177 /* Initiate the write of a core log to a region log disk. Once
178 * initiated, return to the calling routine.
179 *
180 * NON-BLOCKING */
181
182 /* create DAG to write a core log to a region log disk */
183 rf_MakeAllocList(*fwr_alloclist);
184 *fwr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, log->bufPtr,
185 rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
186 "Wcl", *fwr_alloclist, RF_DAG_FLAGS_NONE, RF_IO_NORMAL_PRIORITY);
187
188 /* create and initialize PDA for the region log */
189 /* RF_Malloc(*fwr_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t
190 * *)); */
191 *fwr_pda = rf_AllocPDAList(1);
192 regionOffset = log->diskOffset;
193 rf_MapLogParityLogging(raidPtr, regionID, regionOffset,
194 &((*fwr_pda)->col),
195 &((*fwr_pda)->startSector));
196 (*fwr_pda)->numSector = raidPtr->numSectorsPerLog;
197
198 /* initialize DAG parameters */
199 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *));
200 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t));
201 (*fwr_dag_h)->tracerec = tracerec;
202 fwr_wrNode = (*fwr_dag_h)->succedents[0]->succedents[0];
203 fwr_wrNode->params[0].p = *fwr_pda;
204 /* fwr_wrNode->params[1] = log->bufPtr; */
205 fwr_wrNode->params[2].v = 0;
206 fwr_wrNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
207
208 /* launch the dag to write the core log to disk */
209 rf_DispatchDAG(*fwr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
210 (void *) fwr_mcpair);
211 }
212
213
214 static void
215 ReadRegionParity(
216 RF_RegionId_t regionID,
217 RF_MCPair_t * prd_mcpair,
218 caddr_t parityBuffer,
219 RF_Raid_t * raidPtr,
220 RF_DagHeader_t ** prd_dag_h,
221 RF_AllocListElem_t ** prd_alloclist,
222 RF_PhysDiskAddr_t ** prd_pda)
223 {
224 /* Initiate the read region parity from disk. Once initiated, return
225 * to the calling routine.
226 *
227 * NON-BLOCKING */
228
229 RF_AccTraceEntry_t *tracerec;
230 RF_DagNode_t *prd_rdNode;
231
232 /* create DAG to read region parity from disk */
233 rf_MakeAllocList(*prd_alloclist);
234 *prd_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, NULL, rf_DiskReadFunc,
235 rf_DiskReadUndoFunc, "Rrp",
236 *prd_alloclist, RF_DAG_FLAGS_NONE,
237 RF_IO_NORMAL_PRIORITY);
238
239 /* create and initialize PDA for region parity */
240 /* RF_Malloc(*prd_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t
241 * *)); */
242 *prd_pda = rf_AllocPDAList(1);
243 rf_MapRegionParity(raidPtr, regionID,
244 &((*prd_pda)->col), &((*prd_pda)->startSector),
245 &((*prd_pda)->numSector));
246 if (rf_parityLogDebug)
247 printf("[reading %d sectors of parity from region %d]\n",
248 (int) (*prd_pda)->numSector, regionID);
249 if ((*prd_pda)->next) {
250 (*prd_pda)->next = NULL;
251 printf("set prd_pda->next to NULL\n");
252 }
253 /* initialize DAG parameters */
254 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *));
255 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t));
256 (*prd_dag_h)->tracerec = tracerec;
257 prd_rdNode = (*prd_dag_h)->succedents[0]->succedents[0];
258 prd_rdNode->params[0].p = *prd_pda;
259 prd_rdNode->params[1].p = parityBuffer;
260 prd_rdNode->params[2].v = 0;
261 prd_rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
262 #if RF_DEBUG_VALIDATE_DAG
263 if (rf_validateDAGDebug)
264 rf_ValidateDAG(*prd_dag_h);
265 #endif
266 /* launch region parity read dag */
267 rf_DispatchDAG(*prd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
268 (void *) prd_mcpair);
269 }
270
271 static void
272 WriteRegionParity(
273 RF_RegionId_t regionID,
274 RF_MCPair_t * pwr_mcpair,
275 caddr_t parityBuffer,
276 RF_Raid_t * raidPtr,
277 RF_DagHeader_t ** pwr_dag_h,
278 RF_AllocListElem_t ** pwr_alloclist,
279 RF_PhysDiskAddr_t ** pwr_pda)
280 {
281 /* Initiate the write of region parity to disk. Once initiated, return
282 * to the calling routine.
283 *
284 * NON-BLOCKING */
285
286 RF_AccTraceEntry_t *tracerec;
287 RF_DagNode_t *pwr_wrNode;
288
289 /* create DAG to write region log from disk */
290 rf_MakeAllocList(*pwr_alloclist);
291 *pwr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, parityBuffer,
292 rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
293 "Wrp", *pwr_alloclist,
294 RF_DAG_FLAGS_NONE,
295 RF_IO_NORMAL_PRIORITY);
296
297 /* create and initialize PDA for region parity */
298 /* RF_Malloc(*pwr_pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t
299 * *)); */
300 *pwr_pda = rf_AllocPDAList(1);
301 rf_MapRegionParity(raidPtr, regionID,
302 &((*pwr_pda)->col), &((*pwr_pda)->startSector),
303 &((*pwr_pda)->numSector));
304
305 /* initialize DAG parameters */
306 RF_Malloc(tracerec,sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *));
307 memset((char *) tracerec, 0, sizeof(RF_AccTraceEntry_t));
308 (*pwr_dag_h)->tracerec = tracerec;
309 pwr_wrNode = (*pwr_dag_h)->succedents[0]->succedents[0];
310 pwr_wrNode->params[0].p = *pwr_pda;
311 /* pwr_wrNode->params[1] = parityBuffer; */
312 pwr_wrNode->params[2].v = 0;
313 pwr_wrNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
314
315 /* launch the dag to write region parity to disk */
316 rf_DispatchDAG(*pwr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
317 (void *) pwr_mcpair);
318 }
319
320 static void
321 FlushLogsToDisk(
322 RF_Raid_t * raidPtr,
323 RF_ParityLog_t * logList)
324 {
325 /* Flush a linked list of core logs to the log disk. Logs contain the
326 * disk location where they should be written. Logs were written in
327 * FIFO order and that order must be preserved.
328 *
329 * Recommended optimizations: 1) allow multiple flushes to occur
330 * simultaneously 2) coalesce contiguous flush operations
331 *
332 * BLOCKING */
333
334 RF_ParityLog_t *log;
335 RF_RegionId_t regionID;
336 RF_MCPair_t *fwr_mcpair;
337 RF_DagHeader_t *fwr_dag_h;
338 RF_AllocListElem_t *fwr_alloclist;
339 RF_PhysDiskAddr_t *fwr_pda;
340
341 fwr_mcpair = rf_AllocMCPair();
342 RF_LOCK_MUTEX(fwr_mcpair->mutex);
343
344 RF_ASSERT(logList);
345 log = logList;
346 while (log) {
347 regionID = log->regionID;
348
349 /* create and launch a DAG to write the core log */
350 if (rf_parityLogDebug)
351 printf("[initiating write of core log for region %d]\n", regionID);
352 fwr_mcpair->flag = RF_FALSE;
353 WriteCoreLog(log, fwr_mcpair, raidPtr, &fwr_dag_h,
354 &fwr_alloclist, &fwr_pda);
355
356 /* wait for the DAG to complete */
357 while (!fwr_mcpair->flag)
358 RF_WAIT_COND(fwr_mcpair->cond, fwr_mcpair->mutex);
359 if (fwr_dag_h->status != rf_enable) {
360 RF_ERRORMSG1("Unable to write core log to disk (region %d)\n", regionID);
361 RF_ASSERT(0);
362 }
363 /* RF_Free(fwr_pda, sizeof(RF_PhysDiskAddr_t)); */
364 rf_FreePhysDiskAddr(fwr_pda);
365 rf_FreeDAG(fwr_dag_h);
366 rf_FreeAllocList(fwr_alloclist);
367
368 log = log->next;
369 }
370 RF_UNLOCK_MUTEX(fwr_mcpair->mutex);
371 rf_FreeMCPair(fwr_mcpair);
372 rf_ReleaseParityLogs(raidPtr, logList);
373 }
374
375 static void
376 ReintegrateRegion(
377 RF_Raid_t * raidPtr,
378 RF_RegionId_t regionID,
379 RF_ParityLog_t * coreLog)
380 {
381 RF_MCPair_t *rrd_mcpair = NULL, *prd_mcpair, *pwr_mcpair;
382 RF_DagHeader_t *rrd_dag_h, *prd_dag_h, *pwr_dag_h;
383 RF_AllocListElem_t *rrd_alloclist, *prd_alloclist, *pwr_alloclist;
384 RF_PhysDiskAddr_t *rrd_pda, *prd_pda, *pwr_pda;
385 caddr_t parityBuffer, regionBuffer = NULL;
386
387 /* Reintegrate a region (regionID).
388 *
389 * 1. acquire region and parity buffers
390 * 2. read log from disk
391 * 3. read parity from disk
392 * 4. apply log to parity
393 * 5. apply core log to parity
394 * 6. write new parity to disk
395 *
396 * BLOCKING */
397
398 if (rf_parityLogDebug)
399 printf("[reintegrating region %d]\n", regionID);
400
401 /* initiate read of region parity */
402 if (rf_parityLogDebug)
403 printf("[initiating read of parity for region %d]\n",regionID);
404 parityBuffer = AcquireReintBuffer(&raidPtr->parityBufferPool);
405 prd_mcpair = rf_AllocMCPair();
406 RF_LOCK_MUTEX(prd_mcpair->mutex);
407 prd_mcpair->flag = RF_FALSE;
408 ReadRegionParity(regionID, prd_mcpair, parityBuffer, raidPtr,
409 &prd_dag_h, &prd_alloclist, &prd_pda);
410
411 /* if region log nonempty, initiate read */
412 if (raidPtr->regionInfo[regionID].diskCount > 0) {
413 if (rf_parityLogDebug)
414 printf("[initiating read of disk log for region %d]\n",
415 regionID);
416 regionBuffer = AcquireReintBuffer(&raidPtr->regionBufferPool);
417 rrd_mcpair = rf_AllocMCPair();
418 RF_LOCK_MUTEX(rrd_mcpair->mutex);
419 rrd_mcpair->flag = RF_FALSE;
420 ReadRegionLog(regionID, rrd_mcpair, regionBuffer, raidPtr,
421 &rrd_dag_h, &rrd_alloclist, &rrd_pda);
422 }
423 /* wait on read of region parity to complete */
424 while (!prd_mcpair->flag) {
425 RF_WAIT_COND(prd_mcpair->cond, prd_mcpair->mutex);
426 }
427 RF_UNLOCK_MUTEX(prd_mcpair->mutex);
428 if (prd_dag_h->status != rf_enable) {
429 RF_ERRORMSG("Unable to read parity from disk\n");
430 /* add code to fail the parity disk */
431 RF_ASSERT(0);
432 }
433 /* apply core log to parity */
434 /* if (coreLog) ApplyLogsToParity(coreLog, parityBuffer); */
435
436 if (raidPtr->regionInfo[regionID].diskCount > 0) {
437 /* wait on read of region log to complete */
438 while (!rrd_mcpair->flag)
439 RF_WAIT_COND(rrd_mcpair->cond, rrd_mcpair->mutex);
440 RF_UNLOCK_MUTEX(rrd_mcpair->mutex);
441 if (rrd_dag_h->status != rf_enable) {
442 RF_ERRORMSG("Unable to read region log from disk\n");
443 /* add code to fail the log disk */
444 RF_ASSERT(0);
445 }
446 /* apply region log to parity */
447 /* ApplyRegionToParity(regionID, regionBuffer, parityBuffer); */
448 /* release resources associated with region log */
449 /* RF_Free(rrd_pda, sizeof(RF_PhysDiskAddr_t)); */
450 rf_FreePhysDiskAddr(rrd_pda);
451 rf_FreeDAG(rrd_dag_h);
452 rf_FreeAllocList(rrd_alloclist);
453 rf_FreeMCPair(rrd_mcpair);
454 ReleaseReintBuffer(&raidPtr->regionBufferPool, regionBuffer);
455 }
456 /* write reintegrated parity to disk */
457 if (rf_parityLogDebug)
458 printf("[initiating write of parity for region %d]\n",
459 regionID);
460 pwr_mcpair = rf_AllocMCPair();
461 RF_LOCK_MUTEX(pwr_mcpair->mutex);
462 pwr_mcpair->flag = RF_FALSE;
463 WriteRegionParity(regionID, pwr_mcpair, parityBuffer, raidPtr,
464 &pwr_dag_h, &pwr_alloclist, &pwr_pda);
465 while (!pwr_mcpair->flag)
466 RF_WAIT_COND(pwr_mcpair->cond, pwr_mcpair->mutex);
467 RF_UNLOCK_MUTEX(pwr_mcpair->mutex);
468 if (pwr_dag_h->status != rf_enable) {
469 RF_ERRORMSG("Unable to write parity to disk\n");
470 /* add code to fail the parity disk */
471 RF_ASSERT(0);
472 }
473 /* release resources associated with read of old parity */
474 /* RF_Free(prd_pda, sizeof(RF_PhysDiskAddr_t)); */
475 rf_FreePhysDiskAddr(prd_pda);
476 rf_FreeDAG(prd_dag_h);
477 rf_FreeAllocList(prd_alloclist);
478 rf_FreeMCPair(prd_mcpair);
479
480 /* release resources associated with write of new parity */
481 ReleaseReintBuffer(&raidPtr->parityBufferPool, parityBuffer);
482 /* RF_Free(pwr_pda, sizeof(RF_PhysDiskAddr_t)); */
483 rf_FreePhysDiskAddr(pwr_pda);
484 rf_FreeDAG(pwr_dag_h);
485 rf_FreeAllocList(pwr_alloclist);
486 rf_FreeMCPair(pwr_mcpair);
487
488 if (rf_parityLogDebug)
489 printf("[finished reintegrating region %d]\n", regionID);
490 }
491
492
493
494 static void
495 ReintegrateLogs(
496 RF_Raid_t * raidPtr,
497 RF_ParityLog_t * logList)
498 {
499 RF_ParityLog_t *log, *freeLogList = NULL;
500 RF_ParityLogData_t *logData, *logDataList;
501 RF_RegionId_t regionID;
502
503 RF_ASSERT(logList);
504 while (logList) {
505 log = logList;
506 logList = logList->next;
507 log->next = NULL;
508 regionID = log->regionID;
509 ReintegrateRegion(raidPtr, regionID, log);
510 log->numRecords = 0;
511
512 /* remove all items which are blocked on reintegration of this
513 * region */
514 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
515 logData = rf_SearchAndDequeueParityLogData(raidPtr, regionID,
516 &raidPtr->parityLogDiskQueue.reintBlockHead,
517 &raidPtr->parityLogDiskQueue.reintBlockTail,
518 RF_TRUE);
519 logDataList = logData;
520 while (logData) {
521 logData->next = rf_SearchAndDequeueParityLogData(
522 raidPtr, regionID,
523 &raidPtr->parityLogDiskQueue.reintBlockHead,
524 &raidPtr->parityLogDiskQueue.reintBlockTail,
525 RF_TRUE);
526 logData = logData->next;
527 }
528 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
529
530 /* process blocked log data and clear reintInProgress flag for
531 * this region */
532 if (logDataList)
533 rf_ParityLogAppend(logDataList, RF_TRUE, &log, RF_TRUE);
534 else {
535 /* Enable flushing for this region. Holding both
536 * locks provides a synchronization barrier with
537 * DumpParityLogToDisk */
538 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].mutex);
539 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].reintMutex);
540 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
541 raidPtr->regionInfo[regionID].diskCount = 0;
542 raidPtr->regionInfo[regionID].reintInProgress = RF_FALSE;
543 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].mutex);
544 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].reintMutex); /* flushing is now
545 * enabled */
546 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
547 }
548 /* if log wasn't used, attach it to the list of logs to be
549 * returned */
550 if (log) {
551 log->next = freeLogList;
552 freeLogList = log;
553 }
554 }
555 if (freeLogList)
556 rf_ReleaseParityLogs(raidPtr, freeLogList);
557 }
558
559 int
560 rf_ShutdownLogging(RF_Raid_t * raidPtr)
561 {
562 /* shutdown parity logging 1) disable parity logging in all regions 2)
563 * reintegrate all regions */
564
565 RF_SectorCount_t diskCount;
566 RF_RegionId_t regionID;
567 RF_ParityLog_t *log;
568
569 if (rf_parityLogDebug)
570 printf("[shutting down parity logging]\n");
571 /* Since parity log maps are volatile, we must reintegrate all
572 * regions. */
573 if (rf_forceParityLogReint) {
574 for (regionID = 0; regionID < rf_numParityRegions; regionID++) {
575 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].mutex);
576 raidPtr->regionInfo[regionID].loggingEnabled =
577 RF_FALSE;
578 log = raidPtr->regionInfo[regionID].coreLog;
579 raidPtr->regionInfo[regionID].coreLog = NULL;
580 diskCount = raidPtr->regionInfo[regionID].diskCount;
581 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].mutex);
582 if (diskCount > 0 || log != NULL)
583 ReintegrateRegion(raidPtr, regionID, log);
584 if (log != NULL)
585 rf_ReleaseParityLogs(raidPtr, log);
586 }
587 }
588 if (rf_parityLogDebug) {
589 printf("[parity logging disabled]\n");
590 printf("[should be done!]\n");
591 }
592 return (0);
593 }
594
595 int
596 rf_ParityLoggingDiskManager(RF_Raid_t * raidPtr)
597 {
598 RF_ParityLog_t *reintQueue, *flushQueue;
599 int workNeeded, done = RF_FALSE;
600 int s;
601
602 /* Main program for parity logging disk thread. This routine waits
603 * for work to appear in either the flush or reintegration queues and
604 * is responsible for flushing core logs to the log disk as well as
605 * reintegrating parity regions.
606 *
607 * BLOCKING */
608
609 s = splbio();
610
611 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
612
613 /*
614 * Inform our creator that we're running. Don't bother doing the
615 * mutex lock/unlock dance- we locked above, and we'll unlock
616 * below with nothing to do, yet.
617 */
618 raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_RUNNING;
619 RF_SIGNAL_COND(raidPtr->parityLogDiskQueue.cond);
620
621 /* empty the work queues */
622 flushQueue = raidPtr->parityLogDiskQueue.flushQueue;
623 raidPtr->parityLogDiskQueue.flushQueue = NULL;
624 reintQueue = raidPtr->parityLogDiskQueue.reintQueue;
625 raidPtr->parityLogDiskQueue.reintQueue = NULL;
626 workNeeded = (flushQueue || reintQueue);
627
628 while (!done) {
629 while (workNeeded) {
630 /* First, flush all logs in the flush queue, freeing
631 * buffers Second, reintegrate all regions which are
632 * reported as full. Third, append queued log data
633 * until blocked.
634 *
635 * Note: Incoming appends (ParityLogAppend) can block on
636 * either 1. empty buffer pool 2. region under
637 * reintegration To preserve a global FIFO ordering of
638 * appends, buffers are not released to the world
639 * until those appends blocked on buffers are removed
640 * from the append queue. Similarly, regions which
641 * are reintegrated are not opened for general use
642 * until the append queue has been emptied. */
643
644 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
645
646 /* empty flushQueue, using free'd log buffers to
647 * process bufTail */
648 if (flushQueue)
649 FlushLogsToDisk(raidPtr, flushQueue);
650
651 /* empty reintQueue, flushing from reintTail as we go */
652 if (reintQueue)
653 ReintegrateLogs(raidPtr, reintQueue);
654
655 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
656 flushQueue = raidPtr->parityLogDiskQueue.flushQueue;
657 raidPtr->parityLogDiskQueue.flushQueue = NULL;
658 reintQueue = raidPtr->parityLogDiskQueue.reintQueue;
659 raidPtr->parityLogDiskQueue.reintQueue = NULL;
660 workNeeded = (flushQueue || reintQueue);
661 }
662 /* no work is needed at this point */
663 if (raidPtr->parityLogDiskQueue.threadState & RF_PLOG_TERMINATE) {
664 /* shutdown parity logging 1. disable parity logging
665 * in all regions 2. reintegrate all regions */
666 done = RF_TRUE; /* thread disabled, no work needed */
667 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
668 rf_ShutdownLogging(raidPtr);
669 }
670 if (!done) {
671 /* thread enabled, no work needed, so sleep */
672 if (rf_parityLogDebug)
673 printf("[parity logging disk manager sleeping]\n");
674 RF_WAIT_COND(raidPtr->parityLogDiskQueue.cond,
675 raidPtr->parityLogDiskQueue.mutex);
676 if (rf_parityLogDebug)
677 printf("[parity logging disk manager just woke up]\n");
678 flushQueue = raidPtr->parityLogDiskQueue.flushQueue;
679 raidPtr->parityLogDiskQueue.flushQueue = NULL;
680 reintQueue = raidPtr->parityLogDiskQueue.reintQueue;
681 raidPtr->parityLogDiskQueue.reintQueue = NULL;
682 workNeeded = (flushQueue || reintQueue);
683 }
684 }
685 /*
686 * Announce that we're done.
687 */
688 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
689 raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_SHUTDOWN;
690 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
691 RF_SIGNAL_COND(raidPtr->parityLogDiskQueue.cond);
692
693 splx(s);
694
695 /*
696 * In the NetBSD kernel, the thread must exit; returning would
697 * cause the proc trampoline to attempt to return to userspace.
698 */
699 kthread_exit(0); /* does not return */
700 }
701 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */
Cache object: 6ccce550fdf1203eb9f2ba4dbdbedd3c
|