xref: /netbsd-src/sys/dev/raidframe/rf_paritylogDiskMgr.c (revision 2cf3739a9f3a821fe3c5b66ce46809392a3766a9)
1 /*	$NetBSD: rf_paritylogDiskMgr.c,v 1.31 2021/07/23 00:54:45 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 1.31 2021/07/23 00:54:45 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 void *AcquireReintBuffer(RF_RegionBufferQueue_t *);
61 
62 static void *
AcquireReintBuffer(RF_RegionBufferQueue_t * pool)63 AcquireReintBuffer(RF_RegionBufferQueue_t *pool)
64 {
65 	void *bufPtr = NULL;
66 
67 	/* Return a region buffer from the free list (pool). If the free list
68 	 * is empty, WAIT. BLOCKING */
69 
70 	rf_lock_mutex2(pool->mutex);
71 	if (pool->availableBuffers > 0) {
72 		bufPtr = pool->buffers[pool->availBuffersIndex];
73 		pool->availableBuffers--;
74 		pool->availBuffersIndex++;
75 		if (pool->availBuffersIndex == pool->totalBuffers)
76 			pool->availBuffersIndex = 0;
77 		rf_unlock_mutex2(pool->mutex);
78 	} else {
79 		RF_PANIC();	/* should never happen in correct config,
80 				 * single reint */
81 		rf_wait_cond2(pool->cond, pool->mutex);
82 	}
83 	return (bufPtr);
84 }
85 
86 static void
ReleaseReintBuffer(RF_RegionBufferQueue_t * pool,void * bufPtr)87 ReleaseReintBuffer(
88     RF_RegionBufferQueue_t * pool,
89     void *bufPtr)
90 {
91 	/* Insert a region buffer (bufPtr) into the free list (pool).
92 	 * NON-BLOCKING */
93 
94 	rf_lock_mutex2(pool->mutex);
95 	pool->availableBuffers++;
96 	pool->buffers[pool->emptyBuffersIndex] = bufPtr;
97 	pool->emptyBuffersIndex++;
98 	if (pool->emptyBuffersIndex == pool->totalBuffers)
99 		pool->emptyBuffersIndex = 0;
100 	RF_ASSERT(pool->availableBuffers <= pool->totalBuffers);
101 	/*
102 	 * XXXmrg this signal goes with the above "shouldn't happen" wait?
103 	 */
104 	rf_signal_cond2(pool->cond);
105 	rf_unlock_mutex2(pool->mutex);
106 }
107 
108 
109 
110 static void
ReadRegionLog(RF_RegionId_t regionID,RF_MCPair_t * rrd_mcpair,void * regionBuffer,RF_Raid_t * raidPtr,RF_DagHeader_t ** rrd_dag_h,RF_AllocListElem_t ** rrd_alloclist,RF_PhysDiskAddr_t ** rrd_pda)111 ReadRegionLog(
112     RF_RegionId_t regionID,
113     RF_MCPair_t * rrd_mcpair,
114     void *regionBuffer,
115     RF_Raid_t * raidPtr,
116     RF_DagHeader_t ** rrd_dag_h,
117     RF_AllocListElem_t ** rrd_alloclist,
118     RF_PhysDiskAddr_t ** rrd_pda)
119 {
120 	/* Initiate the read a region log from disk.  Once initiated, return
121 	 * to the calling routine.
122 	 *
123 	 * NON-BLOCKING */
124 
125 	RF_AccTraceEntry_t *tracerec;
126 	RF_DagNode_t *rrd_rdNode;
127 
128 	/* create DAG to read region log from disk */
129 	rf_MakeAllocList(*rrd_alloclist);
130 	*rrd_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, regionBuffer,
131 				      rf_DiskReadFunc, rf_DiskReadUndoFunc,
132 				      "Rrl", *rrd_alloclist,
133 				      RF_DAG_FLAGS_NONE,
134 				      RF_IO_NORMAL_PRIORITY);
135 
136 	/* create and initialize PDA for the core log */
137 	*rrd_pda = rf_AllocPDAList(raidPtr, 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 	tracerec = RF_Malloc(sizeof(*tracerec));
148 	(*rrd_dag_h)->tracerec = tracerec;
149 	rrd_rdNode = (*rrd_dag_h)->succedents[0]->succedents[0];
150 	rrd_rdNode->params[0].p = *rrd_pda;
151 /*  rrd_rdNode->params[1] = regionBuffer; */
152 	rrd_rdNode->params[2].v = 0;
153 	rrd_rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
154 
155 	/* launch region log read dag */
156 	rf_DispatchDAG(*rrd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
157 	    (void *) rrd_mcpair);
158 }
159 
160 
161 
162 static void
WriteCoreLog(RF_ParityLog_t * log,RF_MCPair_t * fwr_mcpair,RF_Raid_t * raidPtr,RF_DagHeader_t ** fwr_dag_h,RF_AllocListElem_t ** fwr_alloclist,RF_PhysDiskAddr_t ** fwr_pda)163 WriteCoreLog(
164     RF_ParityLog_t * log,
165     RF_MCPair_t * fwr_mcpair,
166     RF_Raid_t * raidPtr,
167     RF_DagHeader_t ** fwr_dag_h,
168     RF_AllocListElem_t ** fwr_alloclist,
169     RF_PhysDiskAddr_t ** fwr_pda)
170 {
171 	RF_RegionId_t regionID = log->regionID;
172 	RF_AccTraceEntry_t *tracerec;
173 	RF_SectorNum_t regionOffset;
174 	RF_DagNode_t *fwr_wrNode;
175 
176 	/* Initiate the write of a core log to a region log disk. Once
177 	 * initiated, return to the calling routine.
178 	 *
179 	 * NON-BLOCKING */
180 
181 	/* create DAG to write a core log to a region log disk */
182 	rf_MakeAllocList(*fwr_alloclist);
183 	*fwr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, log->bufPtr,
184 				      rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
185 	    "Wcl", *fwr_alloclist, RF_DAG_FLAGS_NONE, RF_IO_NORMAL_PRIORITY);
186 
187 	*fwr_pda = rf_AllocPDAList(raidPtr, 1);
188 	regionOffset = log->diskOffset;
189 	rf_MapLogParityLogging(raidPtr, regionID, regionOffset,
190 			       &((*fwr_pda)->col),
191 			       &((*fwr_pda)->startSector));
192 	(*fwr_pda)->numSector = raidPtr->numSectorsPerLog;
193 
194 	/* initialize DAG parameters */
195 	tracerec = RF_Malloc(sizeof(*tracerec));
196 	(*fwr_dag_h)->tracerec = tracerec;
197 	fwr_wrNode = (*fwr_dag_h)->succedents[0]->succedents[0];
198 	fwr_wrNode->params[0].p = *fwr_pda;
199 /*  fwr_wrNode->params[1] = log->bufPtr; */
200 	fwr_wrNode->params[2].v = 0;
201 	fwr_wrNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
202 
203 	/* launch the dag to write the core log to disk */
204 	rf_DispatchDAG(*fwr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
205 	    (void *) fwr_mcpair);
206 }
207 
208 
209 static void
ReadRegionParity(RF_RegionId_t regionID,RF_MCPair_t * prd_mcpair,void * parityBuffer,RF_Raid_t * raidPtr,RF_DagHeader_t ** prd_dag_h,RF_AllocListElem_t ** prd_alloclist,RF_PhysDiskAddr_t ** prd_pda)210 ReadRegionParity(
211     RF_RegionId_t regionID,
212     RF_MCPair_t * prd_mcpair,
213     void *parityBuffer,
214     RF_Raid_t * raidPtr,
215     RF_DagHeader_t ** prd_dag_h,
216     RF_AllocListElem_t ** prd_alloclist,
217     RF_PhysDiskAddr_t ** prd_pda)
218 {
219 	/* Initiate the read region parity from disk. Once initiated, return
220 	 * to the calling routine.
221 	 *
222 	 * NON-BLOCKING */
223 
224 	RF_AccTraceEntry_t *tracerec;
225 	RF_DagNode_t *prd_rdNode;
226 
227 	/* create DAG to read region parity from disk */
228 	rf_MakeAllocList(*prd_alloclist);
229 	*prd_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, NULL, rf_DiskReadFunc,
230 				      rf_DiskReadUndoFunc, "Rrp",
231 				      *prd_alloclist, RF_DAG_FLAGS_NONE,
232 				      RF_IO_NORMAL_PRIORITY);
233 
234 	/* create and initialize PDA for region parity */
235 	*prd_pda = rf_AllocPDAList(raidPtr, 1);
236 	rf_MapRegionParity(raidPtr, regionID,
237 			   &((*prd_pda)->col), &((*prd_pda)->startSector),
238 			   &((*prd_pda)->numSector));
239 	if (rf_parityLogDebug)
240 		printf("[reading %d sectors of parity from region %d]\n",
241 		    (int) (*prd_pda)->numSector, regionID);
242 	if ((*prd_pda)->next) {
243 		(*prd_pda)->next = NULL;
244 		printf("set prd_pda->next to NULL\n");
245 	}
246 	/* initialize DAG parameters */
247 	tracerec = RF_Malloc(sizeof(*tracerec));
248 	(*prd_dag_h)->tracerec = tracerec;
249 	prd_rdNode = (*prd_dag_h)->succedents[0]->succedents[0];
250 	prd_rdNode->params[0].p = *prd_pda;
251 	prd_rdNode->params[1].p = parityBuffer;
252 	prd_rdNode->params[2].v = 0;
253 	prd_rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
254 #if RF_DEBUG_VALIDATE_DAG
255 	if (rf_validateDAGDebug)
256 		rf_ValidateDAG(*prd_dag_h);
257 #endif
258 	/* launch region parity read dag */
259 	rf_DispatchDAG(*prd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
260 	    (void *) prd_mcpair);
261 }
262 
263 static void
WriteRegionParity(RF_RegionId_t regionID,RF_MCPair_t * pwr_mcpair,void * parityBuffer,RF_Raid_t * raidPtr,RF_DagHeader_t ** pwr_dag_h,RF_AllocListElem_t ** pwr_alloclist,RF_PhysDiskAddr_t ** pwr_pda)264 WriteRegionParity(
265     RF_RegionId_t regionID,
266     RF_MCPair_t * pwr_mcpair,
267     void *parityBuffer,
268     RF_Raid_t * raidPtr,
269     RF_DagHeader_t ** pwr_dag_h,
270     RF_AllocListElem_t ** pwr_alloclist,
271     RF_PhysDiskAddr_t ** pwr_pda)
272 {
273 	/* Initiate the write of region parity to disk. Once initiated, return
274 	 * to the calling routine.
275 	 *
276 	 * NON-BLOCKING */
277 
278 	RF_AccTraceEntry_t *tracerec;
279 	RF_DagNode_t *pwr_wrNode;
280 
281 	/* create DAG to write region log from disk */
282 	rf_MakeAllocList(*pwr_alloclist);
283 	*pwr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, 0, parityBuffer,
284 				      rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
285 				      "Wrp", *pwr_alloclist,
286 				      RF_DAG_FLAGS_NONE,
287 				      RF_IO_NORMAL_PRIORITY);
288 
289 	/* create and initialize PDA for region parity */
290 	*pwr_pda = rf_AllocPDAList(raidPtr, 1);
291 	rf_MapRegionParity(raidPtr, regionID,
292 			   &((*pwr_pda)->col), &((*pwr_pda)->startSector),
293 			   &((*pwr_pda)->numSector));
294 
295 	/* initialize DAG parameters */
296 	tracerec = RF_Malloc(sizeof(*tracerec));
297 	(*pwr_dag_h)->tracerec = tracerec;
298 	pwr_wrNode = (*pwr_dag_h)->succedents[0]->succedents[0];
299 	pwr_wrNode->params[0].p = *pwr_pda;
300 /*  pwr_wrNode->params[1] = parityBuffer; */
301 	pwr_wrNode->params[2].v = 0;
302 	pwr_wrNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0);
303 
304 	/* launch the dag to write region parity to disk */
305 	rf_DispatchDAG(*pwr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
306 	    (void *) pwr_mcpair);
307 }
308 
309 static void
FlushLogsToDisk(RF_Raid_t * raidPtr,RF_ParityLog_t * logList)310 FlushLogsToDisk(
311     RF_Raid_t * raidPtr,
312     RF_ParityLog_t * logList)
313 {
314 	/* Flush a linked list of core logs to the log disk. Logs contain the
315 	 * disk location where they should be written.  Logs were written in
316 	 * FIFO order and that order must be preserved.
317 	 *
318 	 * Recommended optimizations: 1) allow multiple flushes to occur
319 	 * simultaneously 2) coalesce contiguous flush operations
320 	 *
321 	 * BLOCKING */
322 
323 	RF_ParityLog_t *log;
324 	RF_RegionId_t regionID;
325 	RF_MCPair_t *fwr_mcpair;
326 	RF_DagHeader_t *fwr_dag_h;
327 	RF_AllocListElem_t *fwr_alloclist;
328 	RF_PhysDiskAddr_t *fwr_pda;
329 
330 	fwr_mcpair = rf_AllocMCPair(raidPtr);
331 	RF_LOCK_MCPAIR(fwr_mcpair);
332 
333 	RF_ASSERT(logList);
334 	log = logList;
335 	while (log) {
336 		regionID = log->regionID;
337 
338 		/* create and launch a DAG to write the core log */
339 		if (rf_parityLogDebug)
340 			printf("[initiating write of core log for region %d]\n", regionID);
341 		fwr_mcpair->flag = RF_FALSE;
342 		WriteCoreLog(log, fwr_mcpair, raidPtr, &fwr_dag_h,
343 			     &fwr_alloclist, &fwr_pda);
344 
345 		/* wait for the DAG to complete */
346 		while (!fwr_mcpair->flag)
347 			RF_WAIT_MCPAIR(fwr_mcpair);
348 		if (fwr_dag_h->status != rf_enable) {
349 			RF_ERRORMSG1("Unable to write core log to disk (region %d)\n", regionID);
350 			RF_ASSERT(0);
351 		}
352 		/* RF_Free(fwr_pda, sizeof(RF_PhysDiskAddr_t)); */
353 		rf_FreePhysDiskAddr(raidPtr, fwr_pda);
354 		rf_FreeDAG(fwr_dag_h);
355 		rf_FreeAllocList(fwr_alloclist);
356 
357 		log = log->next;
358 	}
359 	RF_UNLOCK_MCPAIR(fwr_mcpair);
360 	rf_FreeMCPair(raidPtr, fwr_mcpair);
361 	rf_ReleaseParityLogs(raidPtr, logList);
362 }
363 
364 static void
ReintegrateRegion(RF_Raid_t * raidPtr,RF_RegionId_t regionID,RF_ParityLog_t * coreLog)365 ReintegrateRegion(
366     RF_Raid_t * raidPtr,
367     RF_RegionId_t regionID,
368     RF_ParityLog_t * coreLog)
369 {
370 	RF_MCPair_t *rrd_mcpair = NULL, *prd_mcpair, *pwr_mcpair;
371 	RF_DagHeader_t *rrd_dag_h = NULL, *prd_dag_h, *pwr_dag_h;
372 	RF_AllocListElem_t *rrd_alloclist = NULL, *prd_alloclist, *pwr_alloclist;
373 	RF_PhysDiskAddr_t *rrd_pda = NULL, *prd_pda, *pwr_pda;
374 	void *parityBuffer, *regionBuffer = NULL;
375 
376 	/* Reintegrate a region (regionID).
377 	 *
378 	 * 1. acquire region and parity buffers
379 	 * 2. read log from disk
380 	 * 3. read parity from disk
381 	 * 4. apply log to parity
382 	 * 5. apply core log to parity
383 	 * 6. write new parity to disk
384 	 *
385 	 * BLOCKING */
386 
387 	if (rf_parityLogDebug)
388 		printf("[reintegrating region %d]\n", regionID);
389 
390 	/* initiate read of region parity */
391 	if (rf_parityLogDebug)
392 		printf("[initiating read of parity for region %d]\n",regionID);
393 	parityBuffer = AcquireReintBuffer(&raidPtr->parityBufferPool);
394 	prd_mcpair = rf_AllocMCPair(raidPtr);
395 	RF_LOCK_MCPAIR(prd_mcpair);
396 	prd_mcpair->flag = RF_FALSE;
397 	ReadRegionParity(regionID, prd_mcpair, parityBuffer, raidPtr,
398 			 &prd_dag_h, &prd_alloclist, &prd_pda);
399 
400 	/* if region log nonempty, initiate read */
401 	if (raidPtr->regionInfo[regionID].diskCount > 0) {
402 		if (rf_parityLogDebug)
403 			printf("[initiating read of disk log for region %d]\n",
404 			       regionID);
405 		regionBuffer = AcquireReintBuffer(&raidPtr->regionBufferPool);
406 		rrd_mcpair = rf_AllocMCPair(raidPtr);
407 		RF_LOCK_MCPAIR(rrd_mcpair);
408 		rrd_mcpair->flag = RF_FALSE;
409 		ReadRegionLog(regionID, rrd_mcpair, regionBuffer, raidPtr,
410 			      &rrd_dag_h, &rrd_alloclist, &rrd_pda);
411 	}
412 	/* wait on read of region parity to complete */
413 	while (!prd_mcpair->flag) {
414 		RF_WAIT_MCPAIR(prd_mcpair);
415 	}
416 	RF_UNLOCK_MCPAIR(prd_mcpair);
417 	if (prd_dag_h->status != rf_enable) {
418 		RF_ERRORMSG("Unable to read parity from disk\n");
419 		/* add code to fail the parity disk */
420 		RF_ASSERT(0);
421 	}
422 	/* apply core log to parity */
423 	/* if (coreLog) ApplyLogsToParity(coreLog, parityBuffer); */
424 
425 	if (raidPtr->regionInfo[regionID].diskCount > 0) {
426 		/* wait on read of region log to complete */
427 		while (!rrd_mcpair->flag)
428 			RF_WAIT_MCPAIR(rrd_mcpair);
429 		RF_UNLOCK_MCPAIR(rrd_mcpair);
430 		if (rrd_dag_h->status != rf_enable) {
431 			RF_ERRORMSG("Unable to read region log from disk\n");
432 			/* add code to fail the log disk */
433 			RF_ASSERT(0);
434 		}
435 		/* apply region log to parity */
436 		/* ApplyRegionToParity(regionID, regionBuffer, parityBuffer); */
437 		/* release resources associated with region log */
438 		/* RF_Free(rrd_pda, sizeof(RF_PhysDiskAddr_t)); */
439 		rf_FreePhysDiskAddr(raidPtr, rrd_pda);
440 		rf_FreeDAG(rrd_dag_h);
441 		rf_FreeAllocList(rrd_alloclist);
442 		rf_FreeMCPair(raidPtr, rrd_mcpair);
443 		ReleaseReintBuffer(&raidPtr->regionBufferPool, regionBuffer);
444 	}
445 	/* write reintegrated parity to disk */
446 	if (rf_parityLogDebug)
447 		printf("[initiating write of parity for region %d]\n",
448 		       regionID);
449 	pwr_mcpair = rf_AllocMCPair(raidPtr);
450 	RF_LOCK_MCPAIR(pwr_mcpair);
451 	pwr_mcpair->flag = RF_FALSE;
452 	WriteRegionParity(regionID, pwr_mcpair, parityBuffer, raidPtr,
453 			  &pwr_dag_h, &pwr_alloclist, &pwr_pda);
454 	while (!pwr_mcpair->flag)
455 		RF_WAIT_MCPAIR(pwr_mcpair);
456 	RF_UNLOCK_MCPAIR(pwr_mcpair);
457 	if (pwr_dag_h->status != rf_enable) {
458 		RF_ERRORMSG("Unable to write parity to disk\n");
459 		/* add code to fail the parity disk */
460 		RF_ASSERT(0);
461 	}
462 	/* release resources associated with read of old parity */
463 	/* RF_Free(prd_pda, sizeof(RF_PhysDiskAddr_t)); */
464 	rf_FreePhysDiskAddr(raidPtr, prd_pda);
465 	rf_FreeDAG(prd_dag_h);
466 	rf_FreeAllocList(prd_alloclist);
467 	rf_FreeMCPair(raidPtr, prd_mcpair);
468 
469 	/* release resources associated with write of new parity */
470 	ReleaseReintBuffer(&raidPtr->parityBufferPool, parityBuffer);
471 	/* RF_Free(pwr_pda, sizeof(RF_PhysDiskAddr_t)); */
472 	rf_FreePhysDiskAddr(raidPtr, pwr_pda);
473 	rf_FreeDAG(pwr_dag_h);
474 	rf_FreeAllocList(pwr_alloclist);
475 	rf_FreeMCPair(raidPtr, pwr_mcpair);
476 
477 	if (rf_parityLogDebug)
478 		printf("[finished reintegrating region %d]\n", regionID);
479 }
480 
481 
482 
483 static void
ReintegrateLogs(RF_Raid_t * raidPtr,RF_ParityLog_t * logList)484 ReintegrateLogs(
485     RF_Raid_t * raidPtr,
486     RF_ParityLog_t * logList)
487 {
488 	RF_ParityLog_t *log, *freeLogList = NULL;
489 	RF_ParityLogData_t *logData, *logDataList;
490 	RF_RegionId_t regionID;
491 
492 	RF_ASSERT(logList);
493 	while (logList) {
494 		log = logList;
495 		logList = logList->next;
496 		log->next = NULL;
497 		regionID = log->regionID;
498 		ReintegrateRegion(raidPtr, regionID, log);
499 		log->numRecords = 0;
500 
501 		/* remove all items which are blocked on reintegration of this
502 		 * region */
503 		rf_lock_mutex2(raidPtr->parityLogDiskQueue.mutex);
504 		logData = rf_SearchAndDequeueParityLogData(raidPtr, regionID,
505 			   &raidPtr->parityLogDiskQueue.reintBlockHead,
506 			   &raidPtr->parityLogDiskQueue.reintBlockTail,
507 							   RF_TRUE);
508 		logDataList = logData;
509 		while (logData) {
510 			logData->next = rf_SearchAndDequeueParityLogData(
511 					 raidPtr, regionID,
512 					 &raidPtr->parityLogDiskQueue.reintBlockHead,
513 					 &raidPtr->parityLogDiskQueue.reintBlockTail,
514 					 RF_TRUE);
515 			logData = logData->next;
516 		}
517 		rf_unlock_mutex2(raidPtr->parityLogDiskQueue.mutex);
518 
519 		/* process blocked log data and clear reintInProgress flag for
520 		 * this region */
521 		if (logDataList)
522 			rf_ParityLogAppend(logDataList, RF_TRUE, &log, RF_TRUE);
523 		else {
524 			/* Enable flushing for this region.  Holding both
525 			 * locks provides a synchronization barrier with
526 			 * DumpParityLogToDisk */
527 			rf_lock_mutex2(raidPtr->regionInfo[regionID].mutex);
528 			rf_lock_mutex2(raidPtr->regionInfo[regionID].reintMutex);
529 			/* XXXmrg: don't need this? */
530 			rf_lock_mutex2(raidPtr->parityLogDiskQueue.mutex);
531 			raidPtr->regionInfo[regionID].diskCount = 0;
532 			raidPtr->regionInfo[regionID].reintInProgress = RF_FALSE;
533 			rf_unlock_mutex2(raidPtr->regionInfo[regionID].mutex);
534 			rf_unlock_mutex2(raidPtr->regionInfo[regionID].reintMutex);	/* flushing is now
535 											 * enabled */
536 			/* XXXmrg: don't need this? */
537 			rf_unlock_mutex2(raidPtr->parityLogDiskQueue.mutex);
538 		}
539 		/* if log wasn't used, attach it to the list of logs to be
540 		 * returned */
541 		if (log) {
542 			log->next = freeLogList;
543 			freeLogList = log;
544 		}
545 	}
546 	if (freeLogList)
547 		rf_ReleaseParityLogs(raidPtr, freeLogList);
548 }
549 
550 int
rf_ShutdownLogging(RF_Raid_t * raidPtr)551 rf_ShutdownLogging(RF_Raid_t * raidPtr)
552 {
553 	/* shutdown parity logging 1) disable parity logging in all regions 2)
554 	 * reintegrate all regions */
555 
556 	RF_SectorCount_t diskCount;
557 	RF_RegionId_t regionID;
558 	RF_ParityLog_t *log;
559 
560 	if (rf_parityLogDebug)
561 		printf("[shutting down parity logging]\n");
562 	/* Since parity log maps are volatile, we must reintegrate all
563 	 * regions. */
564 	if (rf_forceParityLogReint) {
565 		for (regionID = 0; regionID < rf_numParityRegions; regionID++) {
566 			rf_lock_mutex2(raidPtr->regionInfo[regionID].mutex);
567 			raidPtr->regionInfo[regionID].loggingEnabled =
568 				RF_FALSE;
569 			log = raidPtr->regionInfo[regionID].coreLog;
570 			raidPtr->regionInfo[regionID].coreLog = NULL;
571 			diskCount = raidPtr->regionInfo[regionID].diskCount;
572 			rf_unlock_mutex2(raidPtr->regionInfo[regionID].mutex);
573 			if (diskCount > 0 || log != NULL)
574 				ReintegrateRegion(raidPtr, regionID, log);
575 			if (log != NULL)
576 				rf_ReleaseParityLogs(raidPtr, log);
577 		}
578 	}
579 	if (rf_parityLogDebug) {
580 		printf("[parity logging disabled]\n");
581 		printf("[should be done!]\n");
582 	}
583 	return (0);
584 }
585 
586 void
rf_ParityLoggingDiskManager(void * v)587 rf_ParityLoggingDiskManager(void *v)
588 {
589 	RF_Raid_t *raidPtr = v;
590 	RF_ParityLog_t *reintQueue, *flushQueue;
591 	int     workNeeded, done = RF_FALSE;
592 	int s;
593 
594 	/* Main program for parity logging disk thread.  This routine waits
595 	 * for work to appear in either the flush or reintegration queues and
596 	 * is responsible for flushing core logs to the log disk as well as
597 	 * reintegrating parity regions.
598 	 *
599 	 * BLOCKING */
600 
601 	s = splbio();
602 
603 	rf_lock_mutex2(raidPtr->parityLogDiskQueue.mutex);
604 
605 	/*
606          * Inform our creator that we're running. Don't bother doing the
607          * mutex lock/unlock dance- we locked above, and we'll unlock
608          * below with nothing to do, yet.
609          */
610 	raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_RUNNING;
611 	rf_signal_cond2(raidPtr->parityLogDiskQueue.cond);
612 
613 	/* empty the work queues */
614 	flushQueue = raidPtr->parityLogDiskQueue.flushQueue;
615 	raidPtr->parityLogDiskQueue.flushQueue = NULL;
616 	reintQueue = raidPtr->parityLogDiskQueue.reintQueue;
617 	raidPtr->parityLogDiskQueue.reintQueue = NULL;
618 	workNeeded = (flushQueue || reintQueue);
619 
620 	while (!done) {
621 		while (workNeeded) {
622 			/* First, flush all logs in the flush queue, freeing
623 			 * buffers Second, reintegrate all regions which are
624 			 * reported as full. Third, append queued log data
625 			 * until blocked.
626 			 *
627 			 * Note: Incoming appends (ParityLogAppend) can block on
628 			 * either 1. empty buffer pool 2. region under
629 			 * reintegration To preserve a global FIFO ordering of
630 			 * appends, buffers are not released to the world
631 			 * until those appends blocked on buffers are removed
632 			 * from the append queue.  Similarly, regions which
633 			 * are reintegrated are not opened for general use
634 			 * until the append queue has been emptied. */
635 
636 			rf_unlock_mutex2(raidPtr->parityLogDiskQueue.mutex);
637 
638 			/* empty flushQueue, using free'd log buffers to
639 			 * process bufTail */
640 			if (flushQueue)
641 			       FlushLogsToDisk(raidPtr, flushQueue);
642 
643 			/* empty reintQueue, flushing from reintTail as we go */
644 			if (reintQueue)
645 				ReintegrateLogs(raidPtr, reintQueue);
646 
647 			rf_lock_mutex2(raidPtr->parityLogDiskQueue.mutex);
648 			flushQueue = raidPtr->parityLogDiskQueue.flushQueue;
649 			raidPtr->parityLogDiskQueue.flushQueue = NULL;
650 			reintQueue = raidPtr->parityLogDiskQueue.reintQueue;
651 			raidPtr->parityLogDiskQueue.reintQueue = NULL;
652 			workNeeded = (flushQueue || reintQueue);
653 		}
654 		/* no work is needed at this point */
655 		if (raidPtr->parityLogDiskQueue.threadState & RF_PLOG_TERMINATE) {
656 			/* shutdown parity logging 1. disable parity logging
657 			 * in all regions 2. reintegrate all regions */
658 			done = RF_TRUE;	/* thread disabled, no work needed */
659 			rf_unlock_mutex2(raidPtr->parityLogDiskQueue.mutex);
660 			rf_ShutdownLogging(raidPtr);
661 		}
662 		if (!done) {
663 			/* thread enabled, no work needed, so sleep */
664 			if (rf_parityLogDebug)
665 				printf("[parity logging disk manager sleeping]\n");
666 			rf_wait_cond2(raidPtr->parityLogDiskQueue.cond,
667 				      raidPtr->parityLogDiskQueue.mutex);
668 			if (rf_parityLogDebug)
669 				printf("[parity logging disk manager just woke up]\n");
670 			flushQueue = raidPtr->parityLogDiskQueue.flushQueue;
671 			raidPtr->parityLogDiskQueue.flushQueue = NULL;
672 			reintQueue = raidPtr->parityLogDiskQueue.reintQueue;
673 			raidPtr->parityLogDiskQueue.reintQueue = NULL;
674 			workNeeded = (flushQueue || reintQueue);
675 		}
676 	}
677 	/*
678          * Announce that we're done.
679          */
680 	rf_lock_mutex2(raidPtr->parityLogDiskQueue.mutex);
681 	raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_SHUTDOWN;
682 	rf_signal_cond2(raidPtr->parityLogDiskQueue.cond);
683 	rf_unlock_mutex2(raidPtr->parityLogDiskQueue.mutex);
684 
685 	splx(s);
686 
687 	/*
688          * In the NetBSD kernel, the thread must exit; returning would
689          * cause the proc trampoline to attempt to return to userspace.
690          */
691 	kthread_exit(0);	/* does not return */
692 }
693 #endif				/* RF_INCLUDE_PARITYLOGGING > 0 */
694